{"_what":"Figures shown on the demo page. Generated by scripts/build_demo_facts.py.","_regenerate":"python scripts/build_demo_facts.py","labels":["Adjustments","Agreements","Amendments","Anti-Corruption Laws","Applicable Laws","Approvals","Arbitration","Assignments","Assigns","Authority","Authorizations","Base Salary","Benefits","Binding Effects","Books","Brokers","Capitalization","Change In Control","Closings","Compliance With Laws","Confidentiality","Consent To Jurisdiction","Consents","Construction","Cooperation","Costs","Counterparts","Death","Defined Terms","Definitions","Disability","Disclosures","Duties","Effective Dates","Effectiveness","Employment","Enforceability","Enforcements","Entire Agreements","Erisa","Existence","Expenses","Fees","Financial Statements","Forfeitures","Further Assurances","General","Governing Laws","Headings","Indemnifications","Indemnity","Insurances","Integration","Intellectual Property","Interests","Interpretations","Jurisdictions","Liens","Litigations","Miscellaneous","Modifications","No Conflicts","No Defaults","No Waivers","Non-Disparagement","Notices","Organizations","Participations","Payments","Positions","Powers","Publicity","Qualifications","Records","Releases","Remedies","Representations","Sales","Sanctions","Severability","Solvency","Specific Performance","Submission To Jurisdiction","Subsidiaries","Successors","Survival","Tax Withholdings","Taxes","Terminations","Terms","Titles","Transactions With Affiliates","Use Of Proceeds","Vacations","Venues","Vesting","Waiver Of Jury Trials","Waivers","Warranties","Withholdings"],"api_comparator":{"model":"claude-sonnet-5","usd_per_clause":0.00043610433333333336,"usd_per_1k":0.43610433333333337,"basis":"MEASURED mean over the 3,000-clause Stage 1 batch, Batch API with prompt caching. Not an estimate for any particular clause: a per-clause Claude figure needs that model's own count_tokens (hard rule 9), which costs an API call.","n_requests":3000,"cost_usd_total":1.308313,"source":"results/spend_ledger.jsonl via results/cost_per_1k.json"},"served_reference":{"usd_per_clause":8.788115222574201e-7,"basis":"fp32: capital 633.86 USD / 731,170,832 lifetime requests (at 30.9137 req/s) + marginal energy 0.505668 J/req at 0.0847 USD/kWh","is_estimate":true,"tariff_is_assumed":true,"source":"configs/costs.yaml via src.serve.pricing.local_usd_per_request"},"headline":{"api_macro_f1":0.6148376610577307,"tier0_macro_f1":0.8090515439998669,"n_rows":3000,"split":"test_3000","source":"results/stage1_scored_test3000.json + results/fp32_local_onnx_ce10ep_seed1.json"},"tier0":{"macro_f1":0.8090515439998669,"accuracy":0.8733333333333333,"split":"test_3000","source":"results/fp32_local_onnx_ce10ep_seed1.json"},"platforms":{"rows":[{"host":"macOS arm64 (Apple Silicon)","machine":"arm64","isa":{"avx512_vnni":null,"avx512f":null,"avx2":null},"metric":"test_3000 macro-F1","artefact":"E1b seed 1","fp32":0.8090515439998669,"int8":0.7974432989743074,"source":"results/fp32_local_onnx_ce10ep_seed1.json, results/int8_local_ce10ep_seed1.json"},{"host":"Linux x86, Kaggle Xeon","machine":"x86_64","isa":{"avx512_vnni":false,"avx512f":true,"avx2":true},"metric":"test_3000 macro-F1","artefact":"E1b seed 1","fp32":0.8114655856105638,"int8":0.000166,"note":"INT8 at chance on a 100-class problem, no error raised.","source":"tier0_ce10ep_seed1.json (Kaggle run JSON)"},{"host":"Linux aarch64, container","machine":"aarch64","isa":{"avx512_vnni":null,"avx512f":null,"avx2":null},"metric":"canary accuracy, 200 TRAIN rows","artefact":"E1 seed 1","fp32":0.915,"int8":0.64,"note":"Same physical CPU as the arm64 host. INT8 diverged; the canary refused.","source":"PREREGISTRATION §3bf/§3bg"},{"host":"Cloud Run x86, VNNI","machine":"x86_64","isa":{"avx512_vnni":true,"avx512f":true,"avx2":true},"metric":"canary accuracy, 200 TRAIN rows","artefact":"E1b seed 1","fp32":0.955,"int8":null,"note":"12 of 13 instances observed reported VNNI.","source":"PREREGISTRATION §3bn"},{"host":"Cloud Run x86, NO VNNI","machine":"x86_64","isa":{"avx512_vnni":false,"avx512f":false,"avx2":true},"metric":"canary accuracy, 200 TRAIN rows","artefact":"E1b seed 1","fp32":0.955,"int8":null,"note":"Instance 65b67a604cb6, revision -00009-99j. FP32 does not collapse on the hardware class where INT8 scored 0.000166. CANARY-LEVEL EVIDENCE, not a test_3000 score.","source":"PREREGISTRATION §3bq"}],"source":"results/isa_matrix.json"},"int8_collapse":{"caption":"Same weights, same 3,000 rows. INT8 falls to chance on a CPU without AVX-512 VNNI, and raises no error doing it.","rows":[{"isa":"macOS arm64","vnni":null,"fp32":0.8090515439998669,"int8":0.7974432989743074},{"isa":"Linux x86 (Xeon)","vnni":false,"fp32":0.8114655856105638,"int8":0.000166}],"source":"results/isa_matrix.json"},"margin_distribution":{"n":3000,"bins":50,"range":[0.0,1.0],"counts":[8,7,6,4,3,5,11,9,5,5,1,5,6,8,3,6,5,6,3,6,2,5,1,4,2,2,2,8,1,5,8,7,12,12,7,5,12,14,12,14,18,18,13,30,29,30,56,63,153,2343],"percentiles":[0.001359,0.110211,0.188404,0.316929,0.467477,0.616472,0.676088,0.749624,0.787714,0.825524,0.863596,0.88214,0.902861,0.922701,0.931813,0.942083,0.954511,0.961803,0.965592,0.97196,0.976322,0.978568,0.980402,0.982775,0.984445,0.985728,0.987127,0.988318,0.989638,0.990759,0.99153,0.992232,0.993086,0.993663,0.994172,0.994505,0.994876,0.995205,0.995546,0.995754,0.996039,0.996262,0.996564,0.996765,0.996988,0.997161,0.997376,0.997606,0.997801,0.997945,0.998113,0.998254,0.998368,0.998464,0.998545,0.998604,0.998708,0.998821,0.998925,0.998998,0.999056,0.999118,0.999188,0.999242,0.999297,0.99933,0.99937,0.999398,0.999422,0.999449,0.999476,0.999506,0.999533,0.999562,0.999579,0.999602,0.999623,0.99964,0.999663,0.99968,0.999699,0.999717,0.999735,0.999752,0.999765,0.999777,0.999796,0.999812,0.999821,0.99983,0.999838,0.999845,0.999853,0.99986,0.999867,0.999877,0.999886,0.999897,0.999917,0.999935,0.999958],"threshold":0.5288638472557068,"escalation_rate_on_test":0.043,"note":"Margins are heavily massed near 1.0: the median is 0.9981 and 4.30% fall below the threshold. Bar heights use a sqrt scale or the tail is invisible.","source":"results/test_logits_fp32_local_ce10ep_seed1.npz (served artefact) + configs/router_threshold_fp32.json"},"escalation":{"delta":0.0008219900698291926,"ci_low":-0.005993397148057919,"ci_high":0.007201040321879485,"p":0.8572,"n_rows":3000,"n_escalated":129,"tier0_accuracy_on_escalated":0.3488372093023256,"api_accuracy_on_escalated":0.40310077519379844,"in_sample":true,"source":"results/served_config_escalation.json"},"ood":{"n_probes":20,"n_unflagged":10,"worst_case":{"input":"data policy prose","label":"Records","top1":0.9987330980261955,"margin":0.9984284043312073,"flagged":false},"in_domain_reference":{"input":"governing-law clause (in-domain control)","label":"Governing Laws","top1":0.9676666175311318,"margin":0.9369723200798035,"flagged":false},"n_distinct_labels_when_unflagged":8,"caveat":"Hand-written probes, not a sample from any distribution: these rates describe these twenty inputs and nothing wider.","source":"results/ood_probe.json"}}