{
 "schema": 1,
 "generatedAt": "2026-10-20T09:55:30.289Z",
 "generator": "benchmarks/evidence.ts (mode full)",
 "note": "Benchmark rows come from benchmarks/results.json (Ground's report, copied to apps/web/lib/benchmark-results.json) and benchmarks/results/*.json (independent scores, speed and cost); assessment rows from docs/assessments/*.md with their raw data in benchmarks/*.json. Each row names its source. Pending rows are slots with no value.",
 "rows": [
  {"id":"adult.ground.overall","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":94,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.ground.detection-auc","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.567,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.ground.utility-real","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.ground.utility-synthetic","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.897,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.ground.rules-consistency","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"adult.ground.membership-auc-best","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.519,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.ground.dcr-share","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.507,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.ground.rare-copies","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.ground.attribute-inference-risk","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.049,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.ground.singling-out-risk","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.008,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.ground.fit-seconds","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":100.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.ground.rows-per-second","dataset":"adult","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":44118,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"adult.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":77,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.819,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.874,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.51,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.478,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.01,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":2.3,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"adult","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":110294,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"adult.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":75,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.834,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.884,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.504,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.472,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.4,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"adult","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":41667,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"adult.mostly-ai-tabularargn-local.overall","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":95,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.mostly-ai-tabularargn-local.detection-auc","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.551,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.mostly-ai-tabularargn-local.utility-real","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.mostly-ai-tabularargn-local.utility-synthetic","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.897,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.mostly-ai-tabularargn-local.rules-consistency","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"adult.mostly-ai-tabularargn-local.membership-auc-best","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.527,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.mostly-ai-tabularargn-local.dcr-share","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.494,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.mostly-ai-tabularargn-local.rare-copies","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.mostly-ai-tabularargn-local.singling-out-risk","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.005,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.mostly-ai-tabularargn-local.fit-seconds","dataset":"adult","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":235.4,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-ctgan-50-epochs.overall","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":67,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-ctgan-50-epochs.detection-auc","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.937,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-ctgan-50-epochs.utility-real","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.sdv-ctgan-50-epochs.utility-synthetic","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.865,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.sdv-ctgan-50-epochs.rules-consistency","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"adult.sdv-ctgan-50-epochs.membership-auc-best","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.5,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.sdv-ctgan-50-epochs.dcr-share","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.502,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.sdv-ctgan-50-epochs.rare-copies","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.sdv-ctgan-50-epochs.singling-out-risk","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.026,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.sdv-ctgan-50-epochs.fit-seconds","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":168.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-ctgan-50-epochs.rows-per-second","dataset":"adult","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":53571,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"adult.sdv-gaussiancopula.overall","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":65,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-gaussiancopula.detection-auc","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.963,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-gaussiancopula.utility-real","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.904,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"adult.sdv-gaussiancopula.utility-synthetic","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.867,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"adult.sdv-gaussiancopula.rules-consistency","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"adult.sdv-gaussiancopula.membership-auc-best","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.502,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"adult.sdv-gaussiancopula.dcr-share","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.485,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"adult.sdv-gaussiancopula.rare-copies","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"adult.sdv-gaussiancopula.attribute-inference-risk","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"adult.sdv-gaussiancopula.singling-out-risk","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.003,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"adult.sdv-gaussiancopula.fit-seconds","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":6.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"adult.sdv-gaussiancopula.rows-per-second","dataset":"adult","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#adult","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":38961,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"bank.ground.overall","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":90,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.ground.detection-auc","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.609,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.ground.utility-real","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.919,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.ground.utility-synthetic","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.913,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.ground.rules-consistency","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"bank.ground.membership-auc-best","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.512,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.ground.dcr-share","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.479,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.ground.rare-copies","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.ground.attribute-inference-risk","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.018,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.ground.singling-out-risk","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.003,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.ground.fit-seconds","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":110.8,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.ground.rows-per-second","dataset":"bank","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":25424,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"bank.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":64,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.998,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.917,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.869,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.502,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.489,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.019,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.008,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":4.1,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"bank","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":111524,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"bank.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":65,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.979,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.919,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.88,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.511,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.496,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.04,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.6,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"bank","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":28011,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"bank.mostly-ai-tabularargn-local.overall","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":65,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.mostly-ai-tabularargn-local.detection-auc","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":1,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.mostly-ai-tabularargn-local.utility-real","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.919,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.mostly-ai-tabularargn-local.utility-synthetic","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.896,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.mostly-ai-tabularargn-local.rules-consistency","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":77,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"bank.mostly-ai-tabularargn-local.membership-auc-best","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.508,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.mostly-ai-tabularargn-local.dcr-share","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.481,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.mostly-ai-tabularargn-local.rare-copies","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.075,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.mostly-ai-tabularargn-local.singling-out-risk","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.013,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.mostly-ai-tabularargn-local.fit-seconds","dataset":"bank","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":231.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-ctgan-50-epochs.overall","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":60,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-ctgan-50-epochs.detection-auc","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":1,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-ctgan-50-epochs.utility-real","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.919,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.sdv-ctgan-50-epochs.utility-synthetic","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.722,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.sdv-ctgan-50-epochs.rules-consistency","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":32,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"bank.sdv-ctgan-50-epochs.membership-auc-best","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.503,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.sdv-ctgan-50-epochs.dcr-share","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.526,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.sdv-ctgan-50-epochs.rare-copies","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.024,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.sdv-ctgan-50-epochs.singling-out-risk","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.sdv-ctgan-50-epochs.fit-seconds","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":190.6,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-ctgan-50-epochs.rows-per-second","dataset":"bank","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":47619,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"bank.sdv-gaussiancopula.overall","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":59,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-gaussiancopula.detection-auc","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":1,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-gaussiancopula.utility-real","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.919,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"bank.sdv-gaussiancopula.utility-synthetic","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.882,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"bank.sdv-gaussiancopula.rules-consistency","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":23,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"bank.sdv-gaussiancopula.membership-auc-best","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.489,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"bank.sdv-gaussiancopula.dcr-share","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.483,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"bank.sdv-gaussiancopula.rare-copies","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"bank.sdv-gaussiancopula.attribute-inference-risk","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.036,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"bank.sdv-gaussiancopula.singling-out-risk","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.008,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"bank.sdv-gaussiancopula.fit-seconds","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":9.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"bank.sdv-gaussiancopula.rows-per-second","dataset":"bank","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#bank","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":32258,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"credit.ground.overall","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":91,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.ground.detection-auc","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.614,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.ground.utility-real","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.ground.utility-synthetic","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.827,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.ground.rules-consistency","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"credit.ground.membership-auc-best","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.516,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.ground.dcr-share","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.518,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.ground.rare-copies","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.ground.attribute-inference-risk","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.063,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.ground.singling-out-risk","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.ground.fit-seconds","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":32.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.ground.rows-per-second","dataset":"credit","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":50099,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"credit.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":63,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.995,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.835,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.755,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.519,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.521,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.7,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"credit","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":79044,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"credit.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":66,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.974,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.835,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.776,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.554,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.518,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.026,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.1,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"credit","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":25964,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"credit.mostly-ai-tabularargn-local.overall","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":79,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.mostly-ai-tabularargn-local.detection-auc","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.798,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.mostly-ai-tabularargn-local.utility-real","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.mostly-ai-tabularargn-local.utility-synthetic","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.823,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.mostly-ai-tabularargn-local.rules-consistency","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"credit.mostly-ai-tabularargn-local.membership-auc-best","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.539,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.mostly-ai-tabularargn-local.dcr-share","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.499,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.mostly-ai-tabularargn-local.rare-copies","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.089,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.mostly-ai-tabularargn-local.singling-out-risk","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.mostly-ai-tabularargn-local.fit-seconds","dataset":"credit","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":153.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-ctgan-50-epochs.overall","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":61,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-ctgan-50-epochs.detection-auc","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.998,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-ctgan-50-epochs.utility-real","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.sdv-ctgan-50-epochs.utility-synthetic","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.456,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.sdv-ctgan-50-epochs.rules-consistency","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"credit.sdv-ctgan-50-epochs.membership-auc-best","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.528,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.sdv-ctgan-50-epochs.dcr-share","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.529,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.sdv-ctgan-50-epochs.rare-copies","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.sdv-ctgan-50-epochs.singling-out-risk","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.sdv-ctgan-50-epochs.fit-seconds","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":20.7,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-ctgan-50-epochs.rows-per-second","dataset":"credit","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":19761,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"credit.sdv-gaussiancopula.overall","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":63,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-gaussiancopula.detection-auc","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":1,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-gaussiancopula.utility-real","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"credit.sdv-gaussiancopula.utility-synthetic","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.801,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"credit.sdv-gaussiancopula.rules-consistency","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"credit.sdv-gaussiancopula.membership-auc-best","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.518,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"credit.sdv-gaussiancopula.dcr-share","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.532,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"credit.sdv-gaussiancopula.rare-copies","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"credit.sdv-gaussiancopula.attribute-inference-risk","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"credit.sdv-gaussiancopula.singling-out-risk","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.003,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"credit.sdv-gaussiancopula.fit-seconds","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":2.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"credit.sdv-gaussiancopula.rows-per-second","dataset":"credit","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#credit","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":14228,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"telco.ground.overall","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":92,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.ground.detection-auc","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.619,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.ground.utility-real","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.ground.utility-synthetic","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.829,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.ground.rules-consistency","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"telco.ground.membership-auc-best","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.524,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.ground.dcr-share","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.52,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.ground.rare-copies","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.ground.attribute-inference-risk","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.071,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.ground.singling-out-risk","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.ground.fit-seconds","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":34.3,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.ground.rows-per-second","dataset":"telco","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":24824,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"telco.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":64,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.98,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.804,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.525,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.501,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.051,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.013,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"telco","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":54183,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"telco.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":69,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.923,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.81,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.519,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.494,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.037,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.1,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"telco","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":32762,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"telco.mostly-ai-tabularargn-local.overall","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":90,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.mostly-ai-tabularargn-local.detection-auc","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.629,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.mostly-ai-tabularargn-local.utility-real","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.mostly-ai-tabularargn-local.utility-synthetic","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.821,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.mostly-ai-tabularargn-local.rules-consistency","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":98,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"telco.mostly-ai-tabularargn-local.membership-auc-best","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.526,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.mostly-ai-tabularargn-local.dcr-share","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.475,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.mostly-ai-tabularargn-local.rare-copies","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.mostly-ai-tabularargn-local.singling-out-risk","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.008,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.mostly-ai-tabularargn-local.fit-seconds","dataset":"telco","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":139.6,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-ctgan-50-epochs.overall","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":64,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-ctgan-50-epochs.detection-auc","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.965,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-ctgan-50-epochs.utility-real","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.sdv-ctgan-50-epochs.utility-synthetic","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.61,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.sdv-ctgan-50-epochs.rules-consistency","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":49,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"telco.sdv-ctgan-50-epochs.membership-auc-best","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.521,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.sdv-ctgan-50-epochs.dcr-share","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.502,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.sdv-ctgan-50-epochs.rare-copies","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.066,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.sdv-ctgan-50-epochs.singling-out-risk","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.03,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.sdv-ctgan-50-epochs.fit-seconds","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":1350.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-ctgan-50-epochs.rows-per-second","dataset":"telco","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":470,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"telco.sdv-gaussiancopula.overall","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":65,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-gaussiancopula.detection-auc","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.964,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-gaussiancopula.utility-real","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.833,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"telco.sdv-gaussiancopula.utility-synthetic","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.81,"unit":"AUC","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"telco.sdv-gaussiancopula.rules-consistency","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"telco.sdv-gaussiancopula.membership-auc-best","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.531,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"telco.sdv-gaussiancopula.dcr-share","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.462,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"telco.sdv-gaussiancopula.rare-copies","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"telco.sdv-gaussiancopula.attribute-inference-risk","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"telco.sdv-gaussiancopula.singling-out-risk","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"telco.sdv-gaussiancopula.fit-seconds","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":2.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"telco.sdv-gaussiancopula.rows-per-second","dataset":"telco","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#telco","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":22540,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"housing.ground.overall","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":95,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.ground.detection-auc","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.566,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.ground.utility-real","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.ground.utility-synthetic","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.628,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.ground.rules-consistency","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":100,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"housing.ground.membership-auc-best","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.516,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.ground.dcr-share","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.488,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.ground.rare-copies","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.ground.attribute-inference-risk","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.08,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.ground.singling-out-risk","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.017,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.ground.fit-seconds","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":53.9,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.ground.rows-per-second","dataset":"housing","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":106529,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"housing.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":75,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.868,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.533,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.521,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.49,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.02,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.7,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"housing","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":192000,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"housing.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":76,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.802,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.397,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.519,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.519,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.08,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.1,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"housing","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":48000,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"housing.mostly-ai-tabularargn-local.overall","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":93,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.mostly-ai-tabularargn-local.detection-auc","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.588,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.mostly-ai-tabularargn-local.utility-real","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.mostly-ai-tabularargn-local.utility-synthetic","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.607,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.mostly-ai-tabularargn-local.rules-consistency","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":96,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"housing.mostly-ai-tabularargn-local.membership-auc-best","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.513,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.mostly-ai-tabularargn-local.dcr-share","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.521,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.mostly-ai-tabularargn-local.rare-copies","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.12,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.mostly-ai-tabularargn-local.singling-out-risk","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.mostly-ai-tabularargn-local.fit-seconds","dataset":"housing","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":145.7,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-ctgan-50-epochs.overall","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":53,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-ctgan-50-epochs.detection-auc","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.999,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-ctgan-50-epochs.utility-real","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.sdv-ctgan-50-epochs.utility-synthetic","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.058,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.sdv-ctgan-50-epochs.rules-consistency","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":0,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"housing.sdv-ctgan-50-epochs.membership-auc-best","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.534,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.sdv-ctgan-50-epochs.dcr-share","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.527,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.sdv-ctgan-50-epochs.rare-copies","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.sdv-ctgan-50-epochs.singling-out-risk","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.016,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.sdv-ctgan-50-epochs.fit-seconds","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":73.4,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-ctgan-50-epochs.rows-per-second","dataset":"housing","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":43453,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"housing.sdv-gaussiancopula.overall","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":60,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-gaussiancopula.detection-auc","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.999,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-gaussiancopula.utility-real","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.688,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"housing.sdv-gaussiancopula.utility-synthetic","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.436,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"housing.sdv-gaussiancopula.rules-consistency","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rules_consistency","value":97,"unit":"score","scoredBy":"ground","status":"published","note":"Share of the rules the real data always follows that synthetic rows also follow; checks the rules Ground detects and enforces, so it favours Ground by design."},
  {"id":"housing.sdv-gaussiancopula.membership-auc-best","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.497,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"housing.sdv-gaussiancopula.dcr-share","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.497,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"housing.sdv-gaussiancopula.rare-copies","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"housing.sdv-gaussiancopula.attribute-inference-risk","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"housing.sdv-gaussiancopula.singling-out-risk","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"housing.sdv-gaussiancopula.fit-seconds","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":2.4,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"housing.sdv-gaussiancopula.rows-per-second","dataset":"housing","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#housing","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":40273,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"insurance.ground.overall","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":98,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground.detection-auc","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.511,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground.utility-real","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.ground.utility-synthetic","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.813,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.ground.membership-auc-best","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.507,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.ground.dcr-share","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.519,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.ground.rare-copies","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.ground.attribute-inference-risk","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.ground.singling-out-risk","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.ground.fit-seconds","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":12.3,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground.rows-per-second","dataset":"insurance","method":"Ground","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":178500,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"insurance.ground-dp-3.overall.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":80,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3.detection-auc.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.746,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3.utility-real.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.ground-dp-3.utility-synthetic.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.643,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.ground-dp-3.membership-auc-best.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.548,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.ground-dp-3.dcr-share.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.497,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.ground-dp-3.rare-copies.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.ground-dp-3.attribute-inference-risk.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.064,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.ground-dp-3.singling-out-risk.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.028,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.ground-dp-3.fit-seconds.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3.rows-per-second.epsilon-3.engine-rank-correlations","dataset":"insurance","method":"Ground (DP ε=3)","params":{"epsilon":3,"engine":"rank correlations"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":267750,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"insurance.ground-dp-3-mst.overall.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":82,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3-mst.detection-auc.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.731,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3-mst.utility-real.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.ground-dp-3-mst.utility-synthetic.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.726,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.ground-dp-3-mst.membership-auc-best.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.498,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.ground-dp-3-mst.dcr-share.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.514,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.ground-dp-3-mst.rare-copies.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.ground-dp-3-mst.attribute-inference-risk.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.ground-dp-3-mst.singling-out-risk.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.ground-dp-3-mst.fit-seconds.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.ground-dp-3-mst.rows-per-second.epsilon-3.engine-mst","dataset":"insurance","method":"Ground (DP ε=3, MST)","params":{"epsilon":3,"engine":"MST"},"source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":97364,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"insurance.mostly-ai-tabularargn-local.overall","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":77,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.mostly-ai-tabularargn-local.detection-auc","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.822,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.mostly-ai-tabularargn-local.utility-real","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.mostly-ai-tabularargn-local.utility-synthetic","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.019,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.mostly-ai-tabularargn-local.membership-auc-best","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.513,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.mostly-ai-tabularargn-local.dcr-share","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.53,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.mostly-ai-tabularargn-local.rare-copies","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.mostly-ai-tabularargn-local.attribute-inference-risk","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.mostly-ai-tabularargn-local.singling-out-risk","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.mostly-ai-tabularargn-local.fit-seconds","dataset":"insurance","method":"MOSTLY AI (TabularARGN, local)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":47,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-ctgan-50-epochs.overall","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":68,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-ctgan-50-epochs.detection-auc","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.897,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-ctgan-50-epochs.utility-real","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.sdv-ctgan-50-epochs.utility-synthetic","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.088,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.sdv-ctgan-50-epochs.membership-auc-best","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.491,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.sdv-ctgan-50-epochs.dcr-share","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.493,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.sdv-ctgan-50-epochs.rare-copies","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.sdv-ctgan-50-epochs.attribute-inference-risk","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.sdv-ctgan-50-epochs.singling-out-risk","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0.003,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.sdv-ctgan-50-epochs.fit-seconds","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":7.4,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-ctgan-50-epochs.rows-per-second","dataset":"insurance","method":"SDV CTGAN (50 epochs)","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":26775,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"insurance.sdv-gaussiancopula.overall","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"overall","value":82,"unit":"score","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-gaussiancopula.detection-auc","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"detection_auc","value":0.747,"unit":"AUC","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-gaussiancopula.utility-real","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_real","value":0.871,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on the real 80% and tested on the real 20% holdout (random forest)."},
  {"id":"insurance.sdv-gaussiancopula.utility-synthetic","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"utility_synthetic","value":0.507,"unit":"R²","scoredBy":"ground","status":"published","note":"Model trained on synthetic rows and tested on the same real 20% holdout (random forest)."},
  {"id":"insurance.sdv-gaussiancopula.membership-auc-best","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"membership_auc_best","value":0.524,"unit":"AUC","scoredBy":"ground","status":"published","note":"Best AUC of four membership attacks (distance, density ratio, neighbourhood, classifier); 0.5 = no signal, above 0.6 flags review."},
  {"id":"insurance.sdv-gaussiancopula.dcr-share","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"dcr_share","value":0.516,"unit":"share","scoredBy":"ground","status":"published","note":"Share of synthetic rows closer to a training row than to a holdout row; about 0.5 means no memorisation, above 0.6 flags review."},
  {"id":"insurance.sdv-gaussiancopula.rare-copies","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rare_copies","value":0,"unit":"count","scoredBy":"ground","status":"published","note":"Synthetic rows identical to a real record seen fewer than 5 times."},
  {"id":"insurance.sdv-gaussiancopula.attribute-inference-risk","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"attribute_inference_risk","value":0.06,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style excess success on training rows over holdout rows; above 0.1 flags review."},
  {"id":"insurance.sdv-gaussiancopula.singling-out-risk","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"singling_out_risk","value":0,"unit":"risk","scoredBy":"ground","status":"published","note":"Anonymeter-style singling-out excess risk."},
  {"id":"insurance.sdv-gaussiancopula.fit-seconds","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"fit_seconds","value":0.5,"unit":"seconds","scoredBy":"ground","status":"published"},
  {"id":"insurance.sdv-gaussiancopula.rows-per-second","dataset":"insurance","method":"SDV GaussianCopula","source":"apps/web/lib/benchmark-results.json; docs/benchmarks.md#insurance","uncertainty":{"kind":"single-run","note":"One run with fixed seeds (80/20 split, seed 7); differences of a point or two in Overall are within noise."},"metric":"rows_per_second","value":26775,"unit":"rows/s","scoredBy":"ground","status":"published","note":"Generation throughput while sampling the benchmark's synthetic rows; machine not recorded for this run. Not a per-million-row measurement."},
  {"id":"all.ground.tpr-at-1pct-fpr","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"tpr_at_1pct_fpr","unit":"share","source":"packages/core/src/tabular/evaluate.ts computes it per attack; benchmarks/run.ts does not record it yet"},
  {"id":"all.ground.canary-attack-auc","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"canary_attack_auc","unit":"AUC","source":"canary audit (docs/methods.md §4); no public-dataset run recorded"},
  {"id":"all.ground.canary-epsilon-lower-bound","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"canary_epsilon_lower_bound","unit":"ε","source":"canary audit (docs/methods.md §4); no public-dataset run recorded"},
  {"id":"all.ground.rare-class-tstr-lift","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"rare_class_tstr_lift","unit":"AUC","source":"no published run trains a model on rebalanced rare-class rows; docs/assessments/rare-cases.md measures preview coverage and exact counts"},
  {"id":"all.ground.referential-integrity","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"referential_integrity","unit":"share","source":"related-table run (packages/core/src/tabular/relational.ts); no published measurement"},
  {"id":"all.ground.text-label-leakage-auc","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"text_label_leakage_auc","unit":"AUC","source":"export text-quality summary (docs/quality-roadmap.md, label-leakage guard); no published run"},
  {"id":"all.ground.estimate-ci-coverage","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"estimate_ci_coverage","unit":"share","source":"docs/roadmap.md priority #7, research-validity report (planned)"},
  {"id":"all.ground.effective-sample-size","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"effective_sample_size","unit":"rows","source":"docs/roadmap.md priority #7, research-validity report (planned)"},
  {"id":"all.ground.real-plus-synthetic-utility","scoredBy":"ground","status":"pending","value":null,"dataset":"all","method":"Ground","metric":"real_plus_synthetic_utility","unit":"share","source":"docs/roadmap.md priority #7, research-validity report (planned)"},
  {"id":"card-transactions.baseline.previews-with-rare-row","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"baseline","metric":"previews_with_rare_row","value":0.53,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.baseline.previews-with-3-rare-rows","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"baseline","metric":"previews_with_3_rare_rows","value":0.03,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.baseline.mean-rare-rows-in-preview","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"baseline","metric":"mean_rare_rows_in_preview","value":0.71,"unit":"rows","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.baseline.rows-per-second-100k","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"baseline","metric":"rows_per_second_100k","value":108700,"unit":"rows/s","uncertainty":{"kind":"single-run","note":"Best of 3 at 100,000 rows, Node 22 on a laptop."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json"},
  {"id":"card-transactions.rare-preview-prototype.previews-with-rare-row","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"rare preview (prototype)","metric":"previews_with_rare_row","value":1,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.rare-preview-prototype.previews-with-3-rare-rows","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"rare preview (prototype)","metric":"previews_with_3_rare_rows","value":1,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.rare-preview-prototype.mean-rare-rows-in-preview","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"rare preview (prototype)","metric":"mean_rare_rows_in_preview","value":3.01,"unit":"rows","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"is_fraud (2.9%)"},
  {"id":"card-transactions.rare-preview-prototype.rows-per-second-100k","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"rare preview (prototype)","metric":"rows_per_second_100k","value":102400,"unit":"rows/s","uncertainty":{"kind":"single-run","note":"Best of 3 at 100,000 rows, Node 22 on a laptop."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json"},
  {"id":"support-tickets.baseline.previews-with-rare-row","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"previews_with_rare_row","value":0.805,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.baseline.previews-with-3-rare-rows","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"previews_with_3_rare_rows","value":0.26,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.baseline.mean-rare-rows-in-preview","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"mean_rare_rows_in_preview","value":1.71,"unit":"rows","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.baseline.rows-per-second-100k","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"rows_per_second_100k","value":100200,"unit":"rows/s","uncertainty":{"kind":"single-run","note":"Best of 3 at 100,000 rows, Node 22 on a laptop."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json"},
  {"id":"support-tickets.rare-preview-prototype.previews-with-rare-row","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"rare preview (prototype)","metric":"previews_with_rare_row","value":1,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.rare-preview-prototype.previews-with-3-rare-rows","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"rare preview (prototype)","metric":"previews_with_3_rare_rows","value":1,"unit":"share","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.rare-preview-prototype.mean-rare-rows-in-preview","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"rare preview (prototype)","metric":"mean_rare_rows_in_preview","value":3.11,"unit":"rows","uncertainty":{"kind":"seeds","note":"Share over 200 preview seeds; baseline matches the closed form 1 − 0.971²⁵ = 0.52 for card fraud."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json","note":"Priority = urgent (6.9%)"},
  {"id":"support-tickets.rare-preview-prototype.rows-per-second-100k","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"rare preview (prototype)","metric":"rows_per_second_100k","value":88300,"unit":"rows/s","uncertainty":{"kind":"single-run","note":"Best of 3 at 100,000 rows, Node 22 on a laptop."},"source":"docs/assessments/rare-cases.md §1; benchmarks/rare-cases.json"},
  {"id":"support-tickets.exact-counts-prototype.max-count-error.cell-enterprise-urgent-120-of-10-000","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"exact counts (prototype)","metric":"max_count_error","value":0,"unit":"rows","params":{"cell":"Enterprise ∧ urgent = 120 of 10,000"},"uncertainty":{"kind":"seeds","note":"Largest error over 20 seeds."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"card-transactions.exact-counts-prototype.max-count-error.cell-online-fraud-500-of-10-000","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"exact counts (prototype)","metric":"max_count_error","value":0,"unit":"rows","params":{"cell":"online ∧ fraud = 500 of 10,000"},"uncertainty":{"kind":"seeds","note":"Largest error over 20 seeds."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"loans-upload.exact-counts-prototype.max-count-error.cell-west-defaulted-150-of-3-000","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"exact counts (prototype)","metric":"max_count_error","value":0,"unit":"rows","params":{"cell":"West ∧ defaulted = 150 of 3,000"},"uncertainty":{"kind":"seeds","note":"Largest error over 20 seeds."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"loans-upload.group-shares-prototype.max-count-error.cell-30-defaulted-within-west","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"group shares (prototype)","metric":"max_count_error","value":0,"unit":"rows","params":{"cell":"30% defaulted within West"},"uncertainty":{"kind":"seeds","note":"Largest error over 20 seeds."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"card-transactions.true-conditional-400k-rows.mean-amount-fraud-online","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"true conditional (400k rows)","metric":"mean_amount_fraud_online","value":1336,"unit":"USD","uncertainty":{"kind":"noise-floor","note":"The same-size true sample shows sampling noise."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"card-transactions.same-size-true-sample.mean-amount-fraud-online","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"same-size true sample","metric":"mean_amount_fraud_online","value":1320,"unit":"USD","uncertainty":{"kind":"noise-floor","note":"The same-size true sample shows sampling noise."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"card-transactions.attempt-search-prototype.mean-amount-fraud-online","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"attempt search (prototype)","metric":"mean_amount_fraud_online","value":1364,"unit":"USD","uncertainty":{"kind":"noise-floor","note":"The same-size true sample shows sampling noise."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"card-transactions.forced-values.mean-amount-fraud-online","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"forced values","metric":"mean_amount_fraud_online","value":873,"unit":"USD","uncertainty":{"kind":"noise-floor","note":"The same-size true sample shows sampling noise."},"source":"docs/assessments/rare-cases.md §2–3; benchmarks/rare-cases.json"},
  {"id":"adult.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"adult","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":0.002,"unit":"AUC","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":0.009,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"adult.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-5000","scoredBy":"ground","status":"published","dataset":"adult","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":-0.002,"unit":"AUC","params":{"real_rows":5000},"uncertainty":{"kind":"sd","value":0.004,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"bank.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"bank","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":-0.002,"unit":"AUC","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":0.009,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"bank.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-5000","scoredBy":"ground","status":"published","dataset":"bank","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":0.006,"unit":"AUC","params":{"real_rows":5000},"uncertainty":{"kind":"sd","value":0.008,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"credit.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"credit","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":-0.003,"unit":"AUC","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":0.013,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"credit.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-3557","scoredBy":"ground","status":"published","dataset":"credit","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":0.002,"unit":"AUC","params":{"real_rows":3557},"uncertainty":{"kind":"sd","value":0.017,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"telco.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"telco","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":0,"unit":"AUC","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":0.016,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"telco.failure-sliced-plain-synthetic.slice-auc-diff.real-rows-5000","scoredBy":"ground","status":"published","dataset":"telco","method":"failure-sliced − plain synthetic","metric":"slice_auc_diff","value":0.005,"unit":"AUC","params":{"real_rows":5000},"uncertainty":{"kind":"sd","value":0.013,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"housing.failure-sliced-plain-synthetic.slice-rmse-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"housing","method":"failure-sliced − plain synthetic","metric":"slice_rmse_diff","value":3057,"unit":"target units","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":1749,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"housing.failure-sliced-repeated-real-slice-rows.slice-rmse-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"housing","method":"failure-sliced − repeated real slice rows","metric":"slice_rmse_diff","value":4802,"unit":"target units","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":1876,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"housing.failure-sliced-plain-synthetic.slice-rmse-diff.real-rows-5000","scoredBy":"ground","status":"published","dataset":"housing","method":"failure-sliced − plain synthetic","metric":"slice_rmse_diff","value":766,"unit":"target units","params":{"real_rows":5000},"uncertainty":{"kind":"sd","value":1564,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"housing.failure-sliced-repeated-real-slice-rows.slice-rmse-diff.real-rows-5000","scoredBy":"ground","status":"published","dataset":"housing","method":"failure-sliced − repeated real slice rows","metric":"slice_rmse_diff","value":3114,"unit":"target units","params":{"real_rows":5000},"uncertainty":{"kind":"sd","value":978,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"insurance.failure-sliced-plain-synthetic.slice-rmse-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"insurance","method":"failure-sliced − plain synthetic","metric":"slice_rmse_diff","value":1066,"unit":"target units","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":737,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"insurance.failure-sliced-repeated-real-slice-rows.slice-rmse-diff.real-rows-1000","scoredBy":"ground","status":"published","dataset":"insurance","method":"failure-sliced − repeated real slice rows","metric":"slice_rmse_diff","value":1514,"unit":"target units","params":{"real_rows":1000},"uncertainty":{"kind":"sd","value":1171,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"insurance.failure-sliced-plain-synthetic.slice-rmse-diff.real-rows-1071","scoredBy":"ground","status":"published","dataset":"insurance","method":"failure-sliced − plain synthetic","metric":"slice_rmse_diff","value":297,"unit":"target units","params":{"real_rows":1071},"uncertainty":{"kind":"sd","value":431,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"insurance.failure-sliced-repeated-real-slice-rows.slice-rmse-diff.real-rows-1071","scoredBy":"ground","status":"published","dataset":"insurance","method":"failure-sliced − repeated real slice rows","metric":"slice_rmse_diff","value":850,"unit":"target units","params":{"real_rows":1071},"uncertainty":{"kind":"sd","value":442,"note":"sd of the paired difference over 5 seeds; the test set is the same in every seed."},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json"},
  {"id":"all.failure-sliced-vs-plain-synthetic.runs-won-slice-auc.of-40","scoredBy":"ground","status":"published","dataset":"all","method":"failure-sliced vs plain synthetic","metric":"runs_won_slice_auc","value":21,"unit":"runs","params":{"of":40},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json","note":"Classification runs where failure-sliced rows gave the higher slice AUC."},
  {"id":"all.failure-sliced-vs-real-only.runs-slice-worse-regression.of-20","scoredBy":"ground","status":"published","dataset":"all","method":"failure-sliced vs real only","metric":"runs_slice_worse_regression","value":20,"unit":"runs","params":{"of":20},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json","note":"Regression runs where failure-sliced rows made slice RMSE worse than training on real rows alone."},
  {"id":"all.failure-sliced-vs-repeated-real-slice-rows.runs-lost-slice-rmse-regression.of-20","scoredBy":"ground","status":"published","dataset":"all","method":"failure-sliced vs repeated real slice rows","metric":"runs_lost_slice_rmse_regression","value":20,"unit":"runs","params":{"of":20},"source":"docs/assessments/failure-driven-generation.md (Results); benchmarks/failures-results.json","note":"Regression runs where repeating the real slice rows gave the lower slice RMSE."},
  {"id":"all.defects-prototype.rate-mean-relative-deviation.requested-0-005.pairs-35","scoredBy":"ground","status":"published","dataset":"all","method":"defects (prototype)","metric":"rate_mean_relative_deviation","value":0.068,"unit":"share","params":{"requested":0.005,"pairs":35},"uncertainty":{"kind":"binomial","note":"Deviations are binomial noise: at 0.5% about 95 expected hits per 19,000 cells (σ ≈ 10%)."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Largest achieved rate furthest from the request: 0.35%."},
  {"id":"all.defects-prototype.rate-mean-relative-deviation.requested-0-02.pairs-35","scoredBy":"ground","status":"published","dataset":"all","method":"defects (prototype)","metric":"rate_mean_relative_deviation","value":0.03,"unit":"share","params":{"requested":0.02,"pairs":35},"uncertainty":{"kind":"binomial","note":"Deviations are binomial noise: at 0.5% about 95 expected hits per 19,000 cells (σ ≈ 10%)."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Largest achieved rate furthest from the request: 1.75%."},
  {"id":"all.defects-prototype.rate-mean-relative-deviation.requested-0-1.pairs-35","scoredBy":"ground","status":"published","dataset":"all","method":"defects (prototype)","metric":"rate_mean_relative_deviation","value":0.014,"unit":"share","params":{"requested":0.1,"pairs":35},"uncertainty":{"kind":"binomial","note":"Deviations are binomial noise: at 0.5% about 95 expected hits per 19,000 cells (σ ≈ 10%)."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Largest achieved rate furthest from the request: 10.58%."},
  {"id":"card-transactions.defects-prototype.unlabelled-changed-cells.changed-cells-19752","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"defects (prototype)","metric":"unlabelled_changed_cells","value":0,"unit":"cells","params":{"changed_cells":19752},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Every cell that differs between the dirty and clean files has a label, and every label is a differing cell (files written and read back)."},
  {"id":"employee-directory.defects-prototype.unlabelled-changed-cells.changed-cells-19530","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"defects (prototype)","metric":"unlabelled_changed_cells","value":0,"unit":"cells","params":{"changed_cells":19530},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Every cell that differs between the dirty and clean files has a label, and every label is a differing cell (files written and read back)."},
  {"id":"loans-upload.defects-prototype.unlabelled-changed-cells.changed-cells-36037","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"defects (prototype)","metric":"unlabelled_changed_cells","value":0,"unit":"cells","params":{"changed_cells":36037},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Every cell that differs between the dirty and clean files has a label, and every label is a differing cell (files written and read back)."},
  {"id":"card-transactions.simple-validator.validator-recall.defect-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-near-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":0.68,"unit":"share","params":{"defect":"near_duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-near-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":0.79,"unit":"share","params":{"defect":"near_duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-near-duplicate.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":0.72,"unit":"share","params":{"defect":"near_duplicate","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-label-noise.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":0,"unit":"share","params":{"defect":"label_noise","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-label-noise.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":0,"unit":"share","params":{"defect":"label_noise","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-label-noise.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":0,"unit":"share","params":{"defect":"label_noise","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-rule-violation.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"rule_violation","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-rule-violation.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"rule_violation","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-swap-fields.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"swap_fields","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-swap-fields.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":0.66,"unit":"share","params":{"defect":"swap_fields","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-swap-fields.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":0.95,"unit":"share","params":{"defect":"swap_fields","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-unit-scale.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"unit_scale","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-unit-scale.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"unit_scale","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-unit-scale.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"unit_scale","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-out-of-range.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"out_of_range","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-out-of-range.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"out_of_range","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-out-of-range.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"out_of_range","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-type-corruption.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"type_corruption","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-type-corruption.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"type_corruption","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-type-corruption.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"type_corruption","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-format-drift.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"format_drift","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-format-drift.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"format_drift","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-format-drift.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":0.95,"unit":"share","params":{"defect":"format_drift","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-referential-break.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"referential_break","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-referential-break.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"referential_break","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-referential-break.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"referential_break","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-whitespace-encoding.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"whitespace_encoding","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-whitespace-encoding.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"whitespace_encoding","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-whitespace-encoding.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"whitespace_encoding","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-missing-mcar.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"missing_mcar","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-missing-mcar.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":0.93,"unit":"share","params":{"defect":"missing_mcar","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-missing-mcar.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":1,"unit":"share","params":{"defect":"missing_mcar","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-recall.defect-all.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_recall","value":0.959,"unit":"share","params":{"defect":"all","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-recall.defect-all.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_recall","value":0.914,"unit":"share","params":{"defect":"all","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-recall.defect-all.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_recall","value":0.968,"unit":"share","params":{"defect":"all","rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.validator-precision.rate-0-02","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"validator_precision","value":1,"unit":"share","params":{"rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.simple-validator.false-positives-clean-file","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"simple validator","metric":"false_positives_clean_file","value":5,"unit":"cells","uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.validator-precision.rate-0-02","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"validator_precision","value":0.879,"unit":"share","params":{"rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.simple-validator.false-positives-clean-file","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"simple validator","metric":"false_positives_clean_file","value":1,"unit":"cells","uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.validator-precision.rate-0-02","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"validator_precision","value":0.986,"unit":"share","params":{"rate":0.02},"uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.simple-validator.false-positives-clean-file","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"simple validator","metric":"false_positives_clean_file","value":7,"unit":"cells","uncertainty":{"kind":"single-run","note":"One run at 2% per defect, 20,000 rows; the validator learned its expectations from a different clean sample of the same generator, so read recall as a floor, not a score."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.edge-cases-prototype.edge-rows","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"edge cases (prototype)","metric":"edge_rows","value":29,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.edge-cases-prototype.edge-rows-breaking-rule","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"edge cases (prototype)","metric":"edge_rows_breaking_rule","value":0,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.edge-cases-prototype.edge-rows-rejected-by-validator","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"edge cases (prototype)","metric":"edge_rows_rejected_by_validator","value":15,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Valid by construction; turned away by a validator learned from typical rows."},
  {"id":"card-transactions.edge-cases-prototype.edge-case-ms","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"edge cases (prototype)","metric":"edge_case_ms","value":3,"unit":"ms","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.edge-cases-prototype.edge-rows","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"edge cases (prototype)","metric":"edge_rows","value":49,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.edge-cases-prototype.edge-rows-breaking-rule","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"edge cases (prototype)","metric":"edge_rows_breaking_rule","value":0,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.edge-cases-prototype.edge-rows-rejected-by-validator","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"edge cases (prototype)","metric":"edge_rows_rejected_by_validator","value":30,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Valid by construction; turned away by a validator learned from typical rows."},
  {"id":"employee-directory.edge-cases-prototype.edge-case-ms","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"edge cases (prototype)","metric":"edge_case_ms","value":5,"unit":"ms","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.edge-cases-prototype.edge-rows","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"edge cases (prototype)","metric":"edge_rows","value":41,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.edge-cases-prototype.edge-rows-breaking-rule","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"edge cases (prototype)","metric":"edge_rows_breaking_rule","value":0,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.edge-cases-prototype.edge-rows-rejected-by-validator","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"edge cases (prototype)","metric":"edge_rows_rejected_by_validator","value":11,"unit":"rows","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json","note":"Valid by construction; turned away by a validator learned from typical rows."},
  {"id":"loans-upload.edge-cases-prototype.edge-case-ms","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"edge cases (prototype)","metric":"edge_case_ms","value":16,"unit":"ms","source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.generation.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"generation","metric":"ms_per_20k_rows","value":198,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"card-transactions.defect-pass-12-defects-at-2.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"card-transactions","method":"defect pass (12 defects at 2%)","metric":"ms_per_20k_rows","value":186,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.generation.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"generation","metric":"ms_per_20k_rows","value":112,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"employee-directory.defect-pass-12-defects-at-2.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"employee-directory","method":"defect pass (12 defects at 2%)","metric":"ms_per_20k_rows","value":273,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.generation.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"generation","metric":"ms_per_20k_rows","value":2297,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.defect-pass-12-defects-at-2.ms-per-20k-rows","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"defect pass (12 defects at 2%)","metric":"ms_per_20k_rows","value":386,"unit":"ms","uncertainty":{"kind":"single-run","note":"Median of 5 runs, Apple M5 Pro, Node 22."},"source":"docs/assessments/defects-and-edge-cases.md (Results); benchmarks/defects-results.json"},
  {"id":"loans-upload.source.rule-break-share.rule-applied-on-decided-on","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"source","metric":"rule_break_share","value":0.00067,"unit":"share","params":{"rule":"applied_on ≤ decided_on"},"source":"docs/assessments/defects-and-edge-cases.md (\"Keep the errors\" kept the wrong rate)","note":"2 of 3,000 source rows."},
  {"id":"loans-upload.rule-off-before-the-fix.rule-break-share.rule-applied-on-decided-on","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"rule off, before the fix","metric":"rule_break_share","value":0.292,"unit":"share","params":{"rule":"applied_on ≤ decided_on"},"source":"docs/assessments/defects-and-edge-cases.md (\"Keep the errors\" kept the wrong rate)","note":"11,673 of 40,000 generated rows."},
  {"id":"loans-upload.rule-off-after-the-fix.rule-break-share.rule-applied-on-decided-on","scoredBy":"ground","status":"published","dataset":"loans (upload)","method":"rule off, after the fix","metric":"rule_break_share","value":0.00065,"unit":"share","params":{"rule":"applied_on ≤ decided_on"},"source":"docs/assessments/defects-and-edge-cases.md (\"Keep the errors\" kept the wrong rate)","note":"26 of 40,000 generated rows (27 expected)."},
  {"id":"llm-eval-set.baseline.heldout-leaves-reached.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"baseline","metric":"heldout_leaves_reached","value":27,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"llm-eval-set.baseline.heldout-leaves-3plus.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"baseline","metric":"heldout_leaves_3plus","value":22,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.baseline.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"baseline","metric":"heldout_effective_leaves","value":20.4,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"llm-eval-set.baseline.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"baseline","metric":"near_duplicate_rate","value":0.01,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"llm-eval-set.taxonomy-no-variants.heldout-leaves-reached.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy, no variants","metric":"heldout_leaves_reached","value":27,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"llm-eval-set.taxonomy-no-variants.heldout-leaves-3plus.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy, no variants","metric":"heldout_leaves_3plus","value":22,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.taxonomy-no-variants.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy, no variants","metric":"heldout_effective_leaves","value":21.3,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"llm-eval-set.taxonomy-no-variants.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy, no variants","metric":"near_duplicate_rate","value":0.041,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"llm-eval-set.taxonomy-variants.heldout-leaves-reached.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy + variants","metric":"heldout_leaves_reached","value":29,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"llm-eval-set.taxonomy-variants.heldout-leaves-3plus.of-32","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy + variants","metric":"heldout_leaves_3plus","value":25,"unit":"leaves","params":{"of":32},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.taxonomy-variants.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy + variants","metric":"heldout_effective_leaves","value":23.5,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"llm-eval-set.taxonomy-variants.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy + variants","metric":"near_duplicate_rate","value":0,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"support-tickets.baseline.heldout-leaves-reached.of-26","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"heldout_leaves_reached","value":20,"unit":"leaves","params":{"of":26},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"support-tickets.baseline.heldout-leaves-3plus.of-26","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"heldout_leaves_3plus","value":15,"unit":"leaves","params":{"of":26},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"support-tickets.baseline.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"heldout_effective_leaves","value":12.2,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"support-tickets.baseline.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"baseline","metric":"near_duplicate_rate","value":0.025,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"support-tickets.taxonomy-variants.heldout-leaves-reached.of-26","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"taxonomy + variants","metric":"heldout_leaves_reached","value":24,"unit":"leaves","params":{"of":26},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"support-tickets.taxonomy-variants.heldout-leaves-3plus.of-26","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"taxonomy + variants","metric":"heldout_leaves_3plus","value":22,"unit":"leaves","params":{"of":26},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"support-tickets.taxonomy-variants.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"taxonomy + variants","metric":"heldout_effective_leaves","value":19.8,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"support-tickets.taxonomy-variants.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"taxonomy + variants","metric":"near_duplicate_rate","value":0,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"ecommerce-reviews.baseline.heldout-leaves-reached.of-18","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"baseline","metric":"heldout_leaves_reached","value":11,"unit":"leaves","params":{"of":18},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"ecommerce-reviews.baseline.heldout-leaves-3plus.of-18","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"baseline","metric":"heldout_leaves_3plus","value":8,"unit":"leaves","params":{"of":18},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"ecommerce-reviews.baseline.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"baseline","metric":"heldout_effective_leaves","value":4.8,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"ecommerce-reviews.baseline.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"baseline","metric":"near_duplicate_rate","value":0,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"ecommerce-reviews.taxonomy-variants.heldout-leaves-reached.of-18","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"taxonomy + variants","metric":"heldout_leaves_reached","value":17,"unit":"leaves","params":{"of":18},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"ecommerce-reviews.taxonomy-variants.heldout-leaves-3plus.of-18","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"taxonomy + variants","metric":"heldout_leaves_3plus","value":14,"unit":"leaves","params":{"of":18},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"ecommerce-reviews.taxonomy-variants.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"taxonomy + variants","metric":"heldout_effective_leaves","value":10.6,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"ecommerce-reviews.taxonomy-variants.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"taxonomy + variants","metric":"near_duplicate_rate","value":0.02,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"patient-visit-notes.baseline.heldout-leaves-reached.of-29","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"baseline","metric":"heldout_leaves_reached","value":25,"unit":"leaves","params":{"of":29},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"patient-visit-notes.baseline.heldout-leaves-3plus.of-29","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"baseline","metric":"heldout_leaves_3plus","value":16,"unit":"leaves","params":{"of":29},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"patient-visit-notes.baseline.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"baseline","metric":"heldout_effective_leaves","value":15.2,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"patient-visit-notes.baseline.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"baseline","metric":"near_duplicate_rate","value":0,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"patient-visit-notes.taxonomy-variants.heldout-leaves-reached.of-29","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"taxonomy + variants","metric":"heldout_leaves_reached","value":25,"unit":"leaves","params":{"of":29},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Judged against a held-out taxonomy never used to generate anything."},
  {"id":"patient-visit-notes.taxonomy-variants.heldout-leaves-3plus.of-29","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"taxonomy + variants","metric":"heldout_leaves_3plus","value":19,"unit":"leaves","params":{"of":29},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"patient-visit-notes.taxonomy-variants.heldout-effective-leaves","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"taxonomy + variants","metric":"heldout_effective_leaves","value":19.8,"unit":"leaves","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"exp(entropy) of the judged leaf mix."},
  {"id":"patient-visit-notes.taxonomy-variants.near-duplicate-rate","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"taxonomy + variants","metric":"near_duplicate_rate","value":0,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Share of texts with another text at word-pair Jaccard ≥ 0.4."},
  {"id":"llm-eval-set.taxonomy-variants.text-in-assigned-leaf","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"taxonomy + variants","metric":"text_in_assigned_leaf","value":0.96,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Blind judge (Haiku) put the text in the leaf it was written for."},
  {"id":"support-tickets.taxonomy-variants.text-in-assigned-leaf","scoredBy":"ground","status":"published","dataset":"support-tickets","method":"taxonomy + variants","metric":"text_in_assigned_leaf","value":0.97,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Blind judge (Haiku) put the text in the leaf it was written for."},
  {"id":"ecommerce-reviews.taxonomy-variants.text-in-assigned-leaf","scoredBy":"ground","status":"published","dataset":"ecommerce-reviews","method":"taxonomy + variants","metric":"text_in_assigned_leaf","value":0.85,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Blind judge (Haiku) put the text in the leaf it was written for."},
  {"id":"patient-visit-notes.taxonomy-variants.text-in-assigned-leaf","scoredBy":"ground","status":"published","dataset":"patient-visit-notes","method":"taxonomy + variants","metric":"text_in_assigned_leaf","value":0.95,"unit":"share","uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Blind judge (Haiku) put the text in the leaf it was written for."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judged-difficulty.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judged_difficulty","value":1.93,"unit":"1–5","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judge-category-accuracy.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judge_category_accuracy","value":0.985,"unit":"share","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.label-correct.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"label_correct","value":0.985,"unit":"share","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.answer-score.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"answer_score","value":3.43,"unit":"1–5","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judged-difficulty.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judged_difficulty","value":2.2,"unit":"1–5","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judge-category-accuracy.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judge_category_accuracy","value":0.938,"unit":"share","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.label-correct.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"label_correct","value":0.969,"unit":"share","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.answer-score.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"answer_score","value":3.38,"unit":"1–5","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judged-difficulty.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judged_difficulty","value":3.15,"unit":"1–5","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.judge-category-accuracy.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"judge_category_accuracy","value":0.941,"unit":"share","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.label-correct.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"label_correct","value":1,"unit":"share","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.naive-level-named-one-line-hint.answer-score.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"naive (level named, one-line hint)","metric":"answer_score","value":3.21,"unit":"1–5","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"llm-eval-set.difficulty-rubric.judged-difficulty.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judged_difficulty","value":1.93,"unit":"1–5","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.judge-category-accuracy.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judge_category_accuracy","value":0.97,"unit":"share","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.label-correct.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"label_correct","value":0.985,"unit":"share","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.difficulty-rubric.answer-score.level-easy.rows-67","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"answer_score","value":3.46,"unit":"1–5","params":{"level":"easy","rows":67},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"llm-eval-set.difficulty-rubric.judged-difficulty.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judged_difficulty","value":2.2,"unit":"1–5","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.judge-category-accuracy.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judge_category_accuracy","value":0.923,"unit":"share","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.label-correct.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"label_correct","value":0.985,"unit":"share","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.difficulty-rubric.answer-score.level-medium.rows-65","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"answer_score","value":3.66,"unit":"1–5","params":{"level":"medium","rows":65},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"llm-eval-set.difficulty-rubric.judged-difficulty.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judged_difficulty","value":3.38,"unit":"1–5","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.judge-category-accuracy.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"judge_category_accuracy","value":0.5,"unit":"share","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json"},
  {"id":"llm-eval-set.difficulty-rubric.label-correct.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"label_correct","value":0.765,"unit":"share","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Sonnet accepted the assigned intent as a correct reading of the text."},
  {"id":"llm-eval-set.difficulty-rubric.answer-score.level-hard.rows-68","scoredBy":"ground","status":"published","dataset":"llm-eval-set","method":"difficulty rubric","metric":"answer_score","value":3.72,"unit":"1–5","params":{"level":"hard","rows":68},"uncertainty":{"kind":"single-run","note":"One run, 200 rows per arm, no confidence intervals: treat differences of one or two leaves as noise."},"source":"docs/assessments/taxonomy-and-difficulty.md (Results); benchmarks/coverage-results.json","note":"Haiku's answer graded 1–5 against the reference answer; a weak measure (same small model grades)."},
  {"id":"all.ground.independent-score.sdmetrics","dataset":"all","method":"Ground","metric":"independent_score","value":null,"unit":"score","scoredBy":"sdmetrics","status":"pending","source":"benchmarks/rescore.py --scorer sdmetrics (not yet run on a full leaderboard run)"},
  {"id":"all.ground.independent-score.syntheval","dataset":"all","method":"Ground","metric":"independent_score","value":null,"unit":"score","scoredBy":"syntheval","status":"pending","source":"benchmarks/rescore.py --scorer syntheval (not yet run on a full leaderboard run)"},
  {"id":"all.ground.independent-score.synthcity","dataset":"all","method":"Ground","metric":"independent_score","value":null,"unit":"score","scoredBy":"synthcity","status":"pending","source":"benchmarks/rescore.py --scorer synthcity (not yet run on a full leaderboard run)"},
  {"id":"all.ground.seconds-per-million-rows","dataset":"all","method":"Ground","metric":"seconds_per_million_rows","value":null,"unit":"seconds","scoredBy":"ground","status":"pending","source":"benchmarks/collect.ts + benchmarks/evidence.ts after a full leaderboard run"},
  {"id":"all.ground.cost-per-million-rows","dataset":"all","method":"Ground","metric":"cost_per_million_rows","value":null,"unit":"USD","scoredBy":"ground","status":"pending","source":"benchmarks/collect.ts + benchmarks/evidence.ts after a full leaderboard run"}
 ]
}