modelrig 0.1.0 → 0.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (309) hide show
  1. package/CHANGELOG.md +94 -0
  2. package/README.md +41 -11
  3. package/data/PROVENANCE.md +19 -0
  4. package/data/pricing-manual.json +22 -0
  5. package/data/pricing-map.json +46362 -0
  6. package/data/pricing-overlay.json +419 -0
  7. package/dist/adapters/anthropic.d.ts +50 -6
  8. package/dist/adapters/anthropic.d.ts.map +1 -1
  9. package/dist/adapters/anthropic.js +123 -29
  10. package/dist/adapters/build-adapters.d.ts +11 -0
  11. package/dist/adapters/build-adapters.d.ts.map +1 -0
  12. package/dist/adapters/build-adapters.js +45 -0
  13. package/dist/adapters/deepseek.d.ts +6 -0
  14. package/dist/adapters/deepseek.d.ts.map +1 -1
  15. package/dist/adapters/deepseek.js +54 -1
  16. package/dist/adapters/gemini.d.ts.map +1 -1
  17. package/dist/adapters/gemini.js +46 -6
  18. package/dist/adapters/grok.d.ts +9 -0
  19. package/dist/adapters/grok.d.ts.map +1 -1
  20. package/dist/adapters/grok.js +67 -3
  21. package/dist/adapters/openai-compat.d.ts +6 -0
  22. package/dist/adapters/openai-compat.d.ts.map +1 -1
  23. package/dist/adapters/openai-compat.js +58 -1
  24. package/dist/adapters/openai-stream.d.ts +58 -0
  25. package/dist/adapters/openai-stream.d.ts.map +1 -0
  26. package/dist/adapters/openai-stream.js +120 -0
  27. package/dist/adapters/openai.d.ts +6 -0
  28. package/dist/adapters/openai.d.ts.map +1 -1
  29. package/dist/adapters/openai.js +60 -1
  30. package/dist/adapters/sampling.d.ts +28 -0
  31. package/dist/adapters/sampling.d.ts.map +1 -0
  32. package/dist/adapters/sampling.js +34 -0
  33. package/dist/adapters/tool-mapping.d.ts +92 -0
  34. package/dist/adapters/tool-mapping.d.ts.map +1 -0
  35. package/dist/adapters/tool-mapping.js +143 -0
  36. package/dist/artifacts/artifact.d.ts +137 -0
  37. package/dist/artifacts/artifact.d.ts.map +1 -0
  38. package/dist/artifacts/artifact.js +377 -0
  39. package/dist/artifacts/content-uploader.d.ts +87 -0
  40. package/dist/artifacts/content-uploader.d.ts.map +1 -0
  41. package/dist/artifacts/content-uploader.js +124 -0
  42. package/dist/artifacts/index.d.ts +63 -0
  43. package/dist/artifacts/index.d.ts.map +1 -0
  44. package/dist/artifacts/index.js +80 -0
  45. package/dist/artifacts/run-context.d.ts +129 -0
  46. package/dist/artifacts/run-context.d.ts.map +1 -0
  47. package/dist/artifacts/run-context.js +208 -0
  48. package/dist/artifacts/scrub.d.ts +60 -0
  49. package/dist/artifacts/scrub.d.ts.map +1 -0
  50. package/dist/artifacts/scrub.js +139 -0
  51. package/dist/artifacts/step.d.ts +26 -0
  52. package/dist/artifacts/step.d.ts.map +1 -0
  53. package/dist/artifacts/step.js +37 -0
  54. package/dist/artifacts/taint.d.ts +63 -0
  55. package/dist/artifacts/taint.d.ts.map +1 -0
  56. package/dist/artifacts/taint.js +76 -0
  57. package/dist/bakeoff/cache-sim.d.ts +99 -0
  58. package/dist/bakeoff/cache-sim.d.ts.map +1 -0
  59. package/dist/bakeoff/cache-sim.js +126 -0
  60. package/dist/bakeoff/eval-cases.d.ts +96 -0
  61. package/dist/bakeoff/eval-cases.d.ts.map +1 -0
  62. package/dist/bakeoff/eval-cases.js +105 -0
  63. package/dist/bakeoff/report.d.ts +41 -0
  64. package/dist/bakeoff/report.d.ts.map +1 -0
  65. package/dist/bakeoff/report.js +142 -0
  66. package/dist/bakeoff/samples.d.ts +15 -0
  67. package/dist/bakeoff/samples.d.ts.map +1 -1
  68. package/dist/bakeoff/samples.js +20 -0
  69. package/dist/cli.d.ts +1 -3
  70. package/dist/cli.d.ts.map +1 -1
  71. package/dist/cli.js +134 -116
  72. package/dist/config.d.ts +100 -0
  73. package/dist/config.d.ts.map +1 -1
  74. package/dist/config.js +104 -0
  75. package/dist/deprecations.d.ts +32 -0
  76. package/dist/deprecations.d.ts.map +1 -0
  77. package/dist/deprecations.js +59 -0
  78. package/dist/gateway/http-protocol.d.ts +175 -0
  79. package/dist/gateway/http-protocol.d.ts.map +1 -0
  80. package/dist/gateway/http-protocol.js +105 -0
  81. package/dist/index.d.ts +105 -18
  82. package/dist/index.d.ts.map +1 -1
  83. package/dist/index.js +266 -138
  84. package/dist/init/finder.d.ts +15 -0
  85. package/dist/init/finder.d.ts.map +1 -1
  86. package/dist/init/finder.js +46 -0
  87. package/dist/init/init-cli.d.ts +10 -0
  88. package/dist/init/init-cli.d.ts.map +1 -0
  89. package/dist/init/init-cli.js +86 -0
  90. package/dist/init/observe-cli.d.ts +19 -0
  91. package/dist/init/observe-cli.d.ts.map +1 -0
  92. package/dist/init/observe-cli.js +73 -0
  93. package/dist/init/observe.d.ts +81 -0
  94. package/dist/init/observe.d.ts.map +1 -0
  95. package/dist/init/observe.js +245 -0
  96. package/dist/registry/build.d.ts +94 -1
  97. package/dist/registry/build.d.ts.map +1 -1
  98. package/dist/registry/build.js +125 -4
  99. package/dist/registry/cache-policy.d.ts +94 -0
  100. package/dist/registry/cache-policy.d.ts.map +1 -0
  101. package/dist/registry/cache-policy.js +238 -0
  102. package/dist/registry/call-notes.d.ts.map +1 -1
  103. package/dist/registry/call-notes.js +40 -2
  104. package/dist/registry/effective-rate.d.ts +151 -0
  105. package/dist/registry/effective-rate.d.ts.map +1 -0
  106. package/dist/registry/effective-rate.js +156 -0
  107. package/dist/registry/facts.d.ts.map +1 -1
  108. package/dist/registry/facts.js +16 -0
  109. package/dist/registry/parity.d.ts +26 -0
  110. package/dist/registry/parity.d.ts.map +1 -1
  111. package/dist/registry/parity.js +43 -2
  112. package/dist/registry/pricing-shape.d.ts +113 -0
  113. package/dist/registry/pricing-shape.d.ts.map +1 -0
  114. package/dist/registry/pricing-shape.js +123 -0
  115. package/dist/registry/pricing.d.ts +25 -1
  116. package/dist/registry/pricing.d.ts.map +1 -1
  117. package/dist/registry/pricing.js +64 -6
  118. package/dist/rigindex/bootstrap.d.ts +36 -0
  119. package/dist/rigindex/bootstrap.d.ts.map +1 -0
  120. package/dist/rigindex/bootstrap.js +97 -0
  121. package/dist/rigindex/fit.d.ts +24 -0
  122. package/dist/rigindex/fit.d.ts.map +1 -0
  123. package/dist/rigindex/fit.js +76 -0
  124. package/dist/rigindex/gates.d.ts +52 -0
  125. package/dist/rigindex/gates.d.ts.map +1 -0
  126. package/dist/rigindex/gates.js +109 -0
  127. package/dist/rigindex/graph.d.ts +31 -0
  128. package/dist/rigindex/graph.d.ts.map +1 -0
  129. package/dist/rigindex/graph.js +120 -0
  130. package/dist/rigindex/index.d.ts +23 -0
  131. package/dist/rigindex/index.d.ts.map +1 -0
  132. package/dist/rigindex/index.js +112 -0
  133. package/dist/rigindex/types.d.ts +76 -0
  134. package/dist/rigindex/types.d.ts.map +1 -0
  135. package/dist/rigindex/types.js +8 -0
  136. package/dist/routes/cache-lint.d.ts +41 -0
  137. package/dist/routes/cache-lint.d.ts.map +1 -0
  138. package/dist/routes/cache-lint.js +81 -0
  139. package/dist/routes/loader.d.ts +37 -0
  140. package/dist/routes/loader.d.ts.map +1 -1
  141. package/dist/routes/loader.js +138 -0
  142. package/dist/routes/resolved-bundle.d.ts +79 -0
  143. package/dist/routes/resolved-bundle.d.ts.map +1 -0
  144. package/dist/routes/resolved-bundle.js +148 -0
  145. package/dist/routes/sampling-config.d.ts +32 -0
  146. package/dist/routes/sampling-config.d.ts.map +1 -0
  147. package/dist/routes/sampling-config.js +91 -0
  148. package/dist/run/ajv.d.ts +22 -0
  149. package/dist/run/ajv.d.ts.map +1 -0
  150. package/dist/run/ajv.js +32 -0
  151. package/dist/run/cache-handle.d.ts +58 -0
  152. package/dist/run/cache-handle.d.ts.map +1 -0
  153. package/dist/run/cache-handle.js +54 -0
  154. package/dist/run/cache-stamp.d.ts +65 -0
  155. package/dist/run/cache-stamp.d.ts.map +1 -0
  156. package/dist/run/cache-stamp.js +0 -0
  157. package/dist/run/capture-write.d.ts +33 -0
  158. package/dist/run/capture-write.d.ts.map +1 -0
  159. package/dist/run/capture-write.js +55 -0
  160. package/dist/run/cost.d.ts.map +1 -1
  161. package/dist/run/cost.js +1 -1
  162. package/dist/run/dispatch.d.ts +99 -0
  163. package/dist/run/dispatch.d.ts.map +1 -0
  164. package/dist/run/dispatch.js +191 -0
  165. package/dist/run/raw.d.ts +88 -0
  166. package/dist/run/raw.d.ts.map +1 -0
  167. package/dist/run/raw.js +286 -0
  168. package/dist/run/record-attempt.d.ts +40 -0
  169. package/dist/run/record-attempt.d.ts.map +1 -0
  170. package/dist/run/record-attempt.js +99 -0
  171. package/dist/run/resolve.d.ts +75 -2
  172. package/dist/run/resolve.d.ts.map +1 -1
  173. package/dist/run/resolve.js +126 -4
  174. package/dist/run/run.d.ts +14 -0
  175. package/dist/run/run.d.ts.map +1 -1
  176. package/dist/run/run.js +0 -0
  177. package/dist/run/stream.d.ts +49 -0
  178. package/dist/run/stream.d.ts.map +1 -0
  179. package/dist/run/stream.js +87 -0
  180. package/dist/run/validate.d.ts.map +1 -1
  181. package/dist/run/validate.js +6 -7
  182. package/dist/savings/cache-ledger.d.ts +108 -0
  183. package/dist/savings/cache-ledger.d.ts.map +1 -0
  184. package/dist/savings/cache-ledger.js +174 -0
  185. package/dist/savings/cache-reconcile.d.ts +64 -0
  186. package/dist/savings/cache-reconcile.d.ts.map +1 -0
  187. package/dist/savings/cache-reconcile.js +59 -0
  188. package/dist/savings/cache-stats.d.ts +63 -0
  189. package/dist/savings/cache-stats.d.ts.map +1 -0
  190. package/dist/savings/cache-stats.js +108 -0
  191. package/dist/savings/emit.d.ts +13 -0
  192. package/dist/savings/emit.d.ts.map +1 -1
  193. package/dist/savings/emit.js +25 -0
  194. package/dist/savings/swap-execute.d.ts +14 -0
  195. package/dist/savings/swap-execute.d.ts.map +1 -1
  196. package/dist/savings/swap-execute.js +44 -1
  197. package/dist/savings/swaps-client.d.ts +14 -0
  198. package/dist/savings/swaps-client.d.ts.map +1 -1
  199. package/dist/savings/swaps-client.js +16 -2
  200. package/dist/settings/api-fetcher.d.ts +39 -0
  201. package/dist/settings/api-fetcher.d.ts.map +1 -0
  202. package/dist/settings/api-fetcher.js +92 -0
  203. package/dist/settings/index.d.ts +10 -0
  204. package/dist/settings/index.d.ts.map +1 -0
  205. package/dist/settings/index.js +15 -0
  206. package/dist/settings/task-settings.d.ts +91 -0
  207. package/dist/settings/task-settings.d.ts.map +1 -0
  208. package/dist/settings/task-settings.js +116 -0
  209. package/dist/status.d.ts +77 -0
  210. package/dist/status.d.ts.map +1 -0
  211. package/dist/status.js +113 -0
  212. package/dist/telemetry/artifact-store.d.ts +48 -0
  213. package/dist/telemetry/artifact-store.d.ts.map +1 -0
  214. package/dist/telemetry/artifact-store.js +189 -0
  215. package/dist/telemetry/cache-miss-streak.d.ts +79 -0
  216. package/dist/telemetry/cache-miss-streak.d.ts.map +1 -0
  217. package/dist/telemetry/cache-miss-streak.js +99 -0
  218. package/dist/telemetry/capture.d.ts +3 -0
  219. package/dist/telemetry/capture.d.ts.map +1 -1
  220. package/dist/telemetry/capture.js +15 -8
  221. package/dist/telemetry/exporter-setup.d.ts +28 -0
  222. package/dist/telemetry/exporter-setup.d.ts.map +1 -0
  223. package/dist/telemetry/exporter-setup.js +34 -0
  224. package/dist/telemetry/exporter.d.ts +75 -0
  225. package/dist/telemetry/exporter.d.ts.map +1 -1
  226. package/dist/telemetry/exporter.js +324 -47
  227. package/dist/telemetry/health.d.ts +76 -0
  228. package/dist/telemetry/health.d.ts.map +1 -0
  229. package/dist/telemetry/health.js +92 -0
  230. package/dist/telemetry/ingest-protocol.d.ts +152 -3
  231. package/dist/telemetry/ingest-protocol.d.ts.map +1 -1
  232. package/dist/telemetry/ingest-protocol.js +204 -4
  233. package/dist/telemetry/load-error-report.d.ts +24 -0
  234. package/dist/telemetry/load-error-report.d.ts.map +1 -0
  235. package/dist/telemetry/load-error-report.js +50 -0
  236. package/dist/telemetry/otel-map.d.ts +121 -0
  237. package/dist/telemetry/otel-map.d.ts.map +1 -0
  238. package/dist/telemetry/otel-map.js +224 -0
  239. package/dist/telemetry/otel-normalize.d.ts +104 -0
  240. package/dist/telemetry/otel-normalize.d.ts.map +1 -0
  241. package/dist/telemetry/otel-normalize.js +530 -0
  242. package/dist/telemetry/otlp-emitter.d.ts +52 -0
  243. package/dist/telemetry/otlp-emitter.d.ts.map +1 -0
  244. package/dist/telemetry/otlp-emitter.js +144 -0
  245. package/dist/telemetry/otlp-proto.d.ts +29 -0
  246. package/dist/telemetry/otlp-proto.d.ts.map +1 -0
  247. package/dist/telemetry/otlp-proto.js +266 -0
  248. package/dist/telemetry/sinks.d.ts +37 -1
  249. package/dist/telemetry/sinks.d.ts.map +1 -1
  250. package/dist/telemetry/sinks.js +121 -30
  251. package/dist/telemetry/store.d.ts +28 -1
  252. package/dist/telemetry/store.d.ts.map +1 -1
  253. package/dist/telemetry/store.js +218 -3
  254. package/dist/tenancy/api-key-format.d.ts +15 -5
  255. package/dist/tenancy/api-key-format.d.ts.map +1 -1
  256. package/dist/tenancy/api-key-format.js +24 -5
  257. package/dist/types/config-teaching.d.ts +62 -0
  258. package/dist/types/config-teaching.d.ts.map +1 -0
  259. package/dist/types/config-teaching.js +183 -0
  260. package/dist/types/provider.d.ts +122 -1
  261. package/dist/types/provider.d.ts.map +1 -1
  262. package/dist/types/provider.js +21 -1
  263. package/dist/types/route.d.ts +79 -1
  264. package/dist/types/route.d.ts.map +1 -1
  265. package/dist/types/telemetry.d.ts +117 -0
  266. package/dist/types/telemetry.d.ts.map +1 -1
  267. package/dist/validate-cli.d.ts +12 -0
  268. package/dist/validate-cli.d.ts.map +1 -0
  269. package/dist/validate-cli.js +57 -0
  270. package/dist/verify/capture.d.ts +66 -0
  271. package/dist/verify/capture.d.ts.map +1 -0
  272. package/dist/verify/capture.js +180 -0
  273. package/dist/verify/cli-run.d.ts +23 -0
  274. package/dist/verify/cli-run.d.ts.map +1 -0
  275. package/dist/verify/cli-run.js +57 -0
  276. package/dist/verify/diff.d.ts +48 -0
  277. package/dist/verify/diff.d.ts.map +1 -0
  278. package/dist/verify/diff.js +157 -0
  279. package/dist/verify/index.d.ts +17 -0
  280. package/dist/verify/index.d.ts.map +1 -0
  281. package/dist/verify/index.js +24 -0
  282. package/dist/verify/report.d.ts +14 -0
  283. package/dist/verify/report.d.ts.map +1 -0
  284. package/dist/verify/report.js +88 -0
  285. package/dist/verify/types.d.ts +101 -0
  286. package/dist/verify/types.d.ts.map +1 -0
  287. package/dist/verify/types.js +17 -0
  288. package/dist/verify/verify.d.ts +29 -0
  289. package/dist/verify/verify.d.ts.map +1 -0
  290. package/dist/verify/verify.js +148 -0
  291. package/dist/watch/evaluate.d.ts +162 -3
  292. package/dist/watch/evaluate.d.ts.map +1 -1
  293. package/dist/watch/evaluate.js +61 -0
  294. package/dist/watch/monitor.d.ts +116 -0
  295. package/dist/watch/monitor.d.ts.map +1 -0
  296. package/dist/watch/monitor.js +199 -0
  297. package/dist/watch/proposals.d.ts +20 -4
  298. package/dist/watch/proposals.d.ts.map +1 -1
  299. package/dist/watch/proposals.js +31 -3
  300. package/dist/watch/run.d.ts +33 -0
  301. package/dist/watch/run.d.ts.map +1 -1
  302. package/dist/watch/run.js +39 -0
  303. package/package.json +7 -1
  304. package/registry/LICENSE +20 -0
  305. package/registry/leaderboard.json +3639 -0
  306. package/registry/registry.json +7263 -0
  307. package/dist/telemetry/supabase-exporter.d.ts +0 -64
  308. package/dist/telemetry/supabase-exporter.d.ts.map +0 -1
  309. package/dist/telemetry/supabase-exporter.js +0 -211
package/registry/leaderboard.json ADDED
@@ -0,0 +1,3639 @@
1
+ [
2
+ {
3
+ "model_key": "deepinfra/mistralai/Mistral-Small-3.2-24B-Instruct-2506",
4
+ "conform_rate": 0.9473684210526315,
5
+ "conform_ci95": [
6
+ 0.882651138829621,
7
+ 0.9773118869508504
8
+ ],
9
+ "value_accuracy_mean": 0.8947368421052632,
10
+ "native_rung_rate": 0,
11
+ "hard_conform_rate": 0.8,
12
+ "hard_value_accuracy_mean": 0.6,
13
+ "hard_samples": 25,
14
+ "hard_native_rung_rate": 0,
15
+ "effective_usd_per_1k_conformant": 0.04881361111111111,
16
+ "grounded_rate": 0,
17
+ "cache_hit_rate": 0,
18
+ "samples": 95,
19
+ "as_of": "2026-08-18T23:57:53.763Z",
20
+ "families": [
21
+ "demo-rig",
22
+ "probe-suite"
23
+ ],
24
+ "fixture_counts": {
25
+ "demo-rig": 20,
26
+ "probe-suite": 75
27
+ },
28
+ "by_family": {
29
+ "demo-rig": {
30
+ "samples": 20,
31
+ "conform_rate": 1,
32
+ "value_accuracy_mean": 1,
33
+ "native_rung_rate": 0,
34
+ "effective_usd_per_1k_conformant": 0.04018875
35
+ },
36
+ "probe-suite": {
37
+ "samples": 75,
38
+ "conform_rate": 0.9333333333333333,
39
+ "value_accuracy_mean": 0.8666666666666667,
40
+ "native_rung_rate": 0,
41
+ "effective_usd_per_1k_conformant": 0.05127785714285711
42
+ }
43
+ },
44
+ "by_subtype": {
45
+ "classification": {
46
+ "samples": 10,
47
+ "conform_rate": 1,
48
+ "value_accuracy_mean": 1,
49
+ "native_rung_rate": 0,
50
+ "effective_usd_per_1k_conformant": 0.030812499999999996
51
+ },
52
+ "extraction.tabular": {
53
+ "samples": 15,
54
+ "conform_rate": 1,
55
+ "value_accuracy_mean": 1,
56
+ "native_rung_rate": 0,
57
+ "effective_usd_per_1k_conformant": 0.05940833333333332
58
+ },
59
+ "generation.prose": {
60
+ "samples": 15,
61
+ "conform_rate": 0.6666666666666666,
62
+ "value_accuracy_mean": 1,
63
+ "native_rung_rate": 0,
64
+ "effective_usd_per_1k_conformant": 0.12657000000000002
65
+ },
66
+ "qa.docs": {
67
+ "samples": 15,
68
+ "conform_rate": 1,
69
+ "value_accuracy_mean": 1,
70
+ "native_rung_rate": 0,
71
+ "effective_usd_per_1k_conformant": 0.03405333333333333
72
+ },
73
+ "reasoning.numeric": {
74
+ "samples": 15,
75
+ "conform_rate": 1,
76
+ "value_accuracy_mean": 0.3333333333333333,
77
+ "native_rung_rate": 0,
78
+ "effective_usd_per_1k_conformant": 0.02698833333333333
79
+ },
80
+ "structure.strict": {
81
+ "samples": 10,
82
+ "conform_rate": 1,
83
+ "value_accuracy_mean": 1,
84
+ "native_rung_rate": 0,
85
+ "effective_usd_per_1k_conformant": 0.0416875
86
+ },
87
+ "summarize": {
88
+ "samples": 15,
89
+ "conform_rate": 1,
90
+ "value_accuracy_mean": 1,
91
+ "native_rung_rate": 0,
92
+ "effective_usd_per_1k_conformant": 0.039718333333333335
93
+ }
94
+ },
95
+ "discrepancies": [
96
+ {
97
+ "kind": "undeclared_schema_capable",
98
+ "message": "no declared schema support, but probed conformance is 95% — capability exists undeclared"
99
+ }
100
+ ]
101
+ },
102
+ {
103
+ "model_key": "deepinfra/meta-llama/Llama-4-Scout-17B-16E-Instruct",
104
+ "conform_rate": 0.9263157894736842,
105
+ "conform_ci95": [
106
+ 0.8556431471175041,
107
+ 0.9638510303909453
108
+ ],
109
+ "value_accuracy_mean": 1,
110
+ "native_rung_rate": 0,
111
+ "hard_conform_rate": 0.72,
112
+ "hard_value_accuracy_mean": 1,
113
+ "hard_samples": 25,
114
+ "hard_native_rung_rate": 0,
115
+ "effective_usd_per_1k_conformant": 0.0601749999999999,
116
+ "grounded_rate": 1,
117
+ "cache_hit_rate": 0,
118
+ "samples": 95,
119
+ "as_of": "2026-08-18T23:52:34.057Z",
120
+ "families": [
121
+ "demo-rig",
122
+ "probe-suite"
123
+ ],
124
+ "fixture_counts": {
125
+ "demo-rig": 20,
126
+ "probe-suite": 75
127
+ },
128
+ "by_family": {
129
+ "demo-rig": {
130
+ "samples": 20,
131
+ "conform_rate": 1,
132
+ "value_accuracy_mean": 1,
133
+ "native_rung_rate": 0,
134
+ "effective_usd_per_1k_conformant": 0.04660000000000001
135
+ },
136
+ "probe-suite": {
137
+ "samples": 75,
138
+ "conform_rate": 0.9066666666666666,
139
+ "value_accuracy_mean": 1,
140
+ "native_rung_rate": 0,
141
+ "effective_usd_per_1k_conformant": 0.06416764705882347
142
+ }
143
+ },
144
+ "by_subtype": {
145
+ "classification": {
146
+ "samples": 10,
147
+ "conform_rate": 1,
148
+ "value_accuracy_mean": 1,
149
+ "native_rung_rate": 0,
150
+ "effective_usd_per_1k_conformant": 0.03506000000000001
151
+ },
152
+ "extraction.tabular": {
153
+ "samples": 15,
154
+ "conform_rate": 1,
155
+ "value_accuracy_mean": 1,
156
+ "native_rung_rate": 0,
157
+ "effective_usd_per_1k_conformant": 0.06653333333333333
158
+ },
159
+ "generation.prose": {
160
+ "samples": 15,
161
+ "conform_rate": 0.8666666666666667,
162
+ "value_accuracy_mean": 1,
163
+ "native_rung_rate": 0,
164
+ "effective_usd_per_1k_conformant": 0.10449999999999998
165
+ },
166
+ "qa.docs": {
167
+ "samples": 15,
168
+ "conform_rate": 1,
169
+ "value_accuracy_mean": 1,
170
+ "native_rung_rate": 0,
171
+ "effective_usd_per_1k_conformant": 0.039786666666666665
172
+ },
173
+ "reasoning.numeric": {
174
+ "samples": 15,
175
+ "conform_rate": 0.6666666666666666,
176
+ "value_accuracy_mean": 1,
177
+ "native_rung_rate": 0,
178
+ "effective_usd_per_1k_conformant": 0.08385000000000001
179
+ },
180
+ "structure.strict": {
181
+ "samples": 10,
182
+ "conform_rate": 1,
183
+ "value_accuracy_mean": 1,
184
+ "native_rung_rate": 0,
185
+ "effective_usd_per_1k_conformant": 0.04498999999999999
186
+ },
187
+ "summarize": {
188
+ "samples": 15,
189
+ "conform_rate": 1,
190
+ "value_accuracy_mean": 1,
191
+ "native_rung_rate": 0,
192
+ "effective_usd_per_1k_conformant": 0.04687333333333332
193
+ }
194
+ },
195
+ "discrepancies": [
196
+ {
197
+ "kind": "undeclared_schema_capable",
198
+ "message": "no declared schema support, but probed conformance is 93% — capability exists undeclared"
199
+ },
200
+ {
201
+ "kind": "citations_without_declared_search",
202
+ "message": "no declared web-search support, but 100% of probed answers cited URLs — verify whether these are real or fabricated sources"
203
+ }
204
+ ]
205
+ },
206
+ {
207
+ "model_key": "deepinfra/google/gemma-4-31B-it",
208
+ "conform_rate": 1,
209
+ "conform_ci95": [
210
+ 0.9611351464605271,
211
+ 1
212
+ ],
213
+ "value_accuracy_mean": 0.9789473684210527,
214
+ "native_rung_rate": 0,
215
+ "hard_conform_rate": 1,
216
+ "hard_value_accuracy_mean": 1,
217
+ "hard_samples": 25,
218
+ "hard_native_rung_rate": 0,
219
+ "effective_usd_per_1k_conformant": 0.08219515789473689,
220
+ "grounded_rate": null,
221
+ "cache_hit_rate": 0,
222
+ "samples": 95,
223
+ "as_of": "2026-08-18T23:40:53.724Z",
224
+ "families": [
225
+ "demo-rig",
226
+ "probe-suite"
227
+ ],
228
+ "fixture_counts": {
229
+ "demo-rig": 20,
230
+ "probe-suite": 75
231
+ },
232
+ "by_family": {
233
+ "demo-rig": {
234
+ "samples": 20,
235
+ "conform_rate": 1,
236
+ "value_accuracy_mean": 1,
237
+ "native_rung_rate": 0,
238
+ "effective_usd_per_1k_conformant": 0.07670100000000002
239
+ },
240
+ "probe-suite": {
241
+ "samples": 75,
242
+ "conform_rate": 1,
243
+ "value_accuracy_mean": 0.9733333333333334,
244
+ "native_rung_rate": 0,
245
+ "effective_usd_per_1k_conformant": 0.08366026666666669
246
+ }
247
+ },
248
+ "by_subtype": {
249
+ "classification": {
250
+ "samples": 10,
251
+ "conform_rate": 1,
252
+ "value_accuracy_mean": 1,
253
+ "native_rung_rate": 0,
254
+ "effective_usd_per_1k_conformant": 0.05132000000000001
255
+ },
256
+ "extraction.tabular": {
257
+ "samples": 15,
258
+ "conform_rate": 1,
259
+ "value_accuracy_mean": 1,
260
+ "native_rung_rate": 0,
261
+ "effective_usd_per_1k_conformant": 0.12285733333333333
262
+ },
263
+ "generation.prose": {
264
+ "samples": 15,
265
+ "conform_rate": 1,
266
+ "value_accuracy_mean": 0.8666666666666667,
267
+ "native_rung_rate": 0,
268
+ "effective_usd_per_1k_conformant": 0.123994
269
+ },
270
+ "qa.docs": {
271
+ "samples": 15,
272
+ "conform_rate": 1,
273
+ "value_accuracy_mean": 1,
274
+ "native_rung_rate": 0,
275
+ "effective_usd_per_1k_conformant": 0.062358666666666666
276
+ },
277
+ "reasoning.numeric": {
278
+ "samples": 15,
279
+ "conform_rate": 1,
280
+ "value_accuracy_mean": 1,
281
+ "native_rung_rate": 0,
282
+ "effective_usd_per_1k_conformant": 0.04666666666666665
283
+ },
284
+ "structure.strict": {
285
+ "samples": 10,
286
+ "conform_rate": 1,
287
+ "value_accuracy_mean": 1,
288
+ "native_rung_rate": 0,
289
+ "effective_usd_per_1k_conformant": 0.08510599999999999
290
+ },
291
+ "summarize": {
292
+ "samples": 15,
293
+ "conform_rate": 1,
294
+ "value_accuracy_mean": 1,
295
+ "native_rung_rate": 0,
296
+ "effective_usd_per_1k_conformant": 0.07374199999999999
297
+ }
298
+ },
299
+ "discrepancies": [
300
+ {
301
+ "kind": "undeclared_schema_capable",
302
+ "message": "no declared schema support, but probed conformance is 100% — capability exists undeclared"
303
+ }
304
+ ]
305
+ },
306
+ {
307
+ "model_key": "deepseek/deepseek-chat",
308
+ "conform_rate": 0.9473684210526315,
309
+ "conform_ci95": [
310
+ 0.882651138829621,
311
+ 0.9773118869508504
312
+ ],
313
+ "value_accuracy_mean": 0.9473684210526315,
314
+ "native_rung_rate": 0,
315
+ "hard_conform_rate": 0.8,
316
+ "hard_value_accuracy_mean": 0.8,
317
+ "hard_samples": 25,
318
+ "hard_native_rung_rate": 0,
319
+ "effective_usd_per_1k_conformant": 0.08375764444444445,
320
+ "grounded_rate": 1,
321
+ "cache_hit_rate": 1,
322
+ "samples": 95,
323
+ "as_of": "2026-08-19T02:17:57.106Z",
324
+ "families": [
325
+ "demo-rig",
326
+ "probe-suite"
327
+ ],
328
+ "fixture_counts": {
329
+ "demo-rig": 20,
330
+ "probe-suite": 75
331
+ },
332
+ "by_family": {
333
+ "demo-rig": {
334
+ "samples": 20,
335
+ "conform_rate": 1,
336
+ "value_accuracy_mean": 1,
337
+ "native_rung_rate": 0,
338
+ "effective_usd_per_1k_conformant": 0.062279000000000015
339
+ },
340
+ "probe-suite": {
341
+ "samples": 75,
342
+ "conform_rate": 0.9333333333333333,
343
+ "value_accuracy_mean": 0.9333333333333333,
344
+ "native_rung_rate": 0,
345
+ "effective_usd_per_1k_conformant": 0.08989439999999999
346
+ }
347
+ },
348
+ "by_subtype": {
349
+ "classification": {
350
+ "samples": 10,
351
+ "conform_rate": 1,
352
+ "value_accuracy_mean": 1,
353
+ "native_rung_rate": 0,
354
+ "effective_usd_per_1k_conformant": 0.047860399999999984
355
+ },
356
+ "extraction.tabular": {
357
+ "samples": 15,
358
+ "conform_rate": 1,
359
+ "value_accuracy_mean": 1,
360
+ "native_rung_rate": 0,
361
+ "effective_usd_per_1k_conformant": 0.08931066666666669
362
+ },
363
+ "generation.prose": {
364
+ "samples": 15,
365
+ "conform_rate": 0.6666666666666666,
366
+ "value_accuracy_mean": 1,
367
+ "native_rung_rate": 0,
368
+ "effective_usd_per_1k_conformant": 0.2557856
369
+ },
370
+ "qa.docs": {
371
+ "samples": 15,
372
+ "conform_rate": 1,
373
+ "value_accuracy_mean": 1,
374
+ "native_rung_rate": 0,
375
+ "effective_usd_per_1k_conformant": 0.05801973333333334
376
+ },
377
+ "reasoning.numeric": {
378
+ "samples": 15,
379
+ "conform_rate": 1,
380
+ "value_accuracy_mean": 0.6666666666666666,
381
+ "native_rung_rate": 0,
382
+ "effective_usd_per_1k_conformant": 0.04099199999999999
383
+ },
384
+ "structure.strict": {
385
+ "samples": 10,
386
+ "conform_rate": 1,
387
+ "value_accuracy_mean": 1,
388
+ "native_rung_rate": 0,
389
+ "effective_usd_per_1k_conformant": 0.06134800000000002
390
+ },
391
+ "summarize": {
392
+ "samples": 15,
393
+ "conform_rate": 1,
394
+ "value_accuracy_mean": 1,
395
+ "native_rung_rate": 0,
396
+ "effective_usd_per_1k_conformant": 0.07089413333333333
397
+ }
398
+ },
399
+ "discrepancies": [
400
+ {
401
+ "kind": "schema_served_via_json_mode",
402
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
403
+ },
404
+ {
405
+ "kind": "citations_without_declared_search",
406
+ "message": "no declared web-search support, but 100% of probed answers cited URLs — verify whether these are real or fabricated sources"
407
+ }
408
+ ]
409
+ },
410
+ {
411
+ "model_key": "deepseek/deepseek-v4-flash",
412
+ "conform_rate": 0.9578947368421052,
413
+ "conform_ci95": [
414
+ 0.8966912227790326,
415
+ 0.9835062271374501
416
+ ],
417
+ "value_accuracy_mean": 1,
418
+ "native_rung_rate": 0,
419
+ "hard_conform_rate": 0.84,
420
+ "hard_value_accuracy_mean": 1,
421
+ "hard_samples": 25,
422
+ "hard_native_rung_rate": 0,
423
+ "effective_usd_per_1k_conformant": 0.09266338461538463,
424
+ "grounded_rate": 0.8,
425
+ "cache_hit_rate": 1,
426
+ "samples": 95,
427
+ "as_of": "2026-08-19T02:28:44.368Z",
428
+ "families": [
429
+ "demo-rig",
430
+ "probe-suite"
431
+ ],
432
+ "fixture_counts": {
433
+ "demo-rig": 20,
434
+ "probe-suite": 75
435
+ },
436
+ "by_family": {
437
+ "demo-rig": {
438
+ "samples": 20,
439
+ "conform_rate": 1,
440
+ "value_accuracy_mean": 1,
441
+ "native_rung_rate": 0,
442
+ "effective_usd_per_1k_conformant": 0.0522536
443
+ },
444
+ "probe-suite": {
445
+ "samples": 75,
446
+ "conform_rate": 0.9466666666666667,
447
+ "value_accuracy_mean": 1,
448
+ "native_rung_rate": 0,
449
+ "effective_usd_per_1k_conformant": 0.10404642253521124
450
+ }
451
+ },
452
+ "by_subtype": {
453
+ "classification": {
454
+ "samples": 10,
455
+ "conform_rate": 1,
456
+ "value_accuracy_mean": 1,
457
+ "native_rung_rate": 0,
458
+ "effective_usd_per_1k_conformant": 0.040362
459
+ },
460
+ "extraction.tabular": {
461
+ "samples": 15,
462
+ "conform_rate": 1,
463
+ "value_accuracy_mean": 1,
464
+ "native_rung_rate": 0,
465
+ "effective_usd_per_1k_conformant": 0.07868746666666668
466
+ },
467
+ "generation.prose": {
468
+ "samples": 15,
469
+ "conform_rate": 0.7333333333333333,
470
+ "value_accuracy_mean": 1,
471
+ "native_rung_rate": 0,
472
+ "effective_usd_per_1k_conformant": 0.3591407272727273
473
+ },
474
+ "qa.docs": {
475
+ "samples": 15,
476
+ "conform_rate": 1,
477
+ "value_accuracy_mean": 1,
478
+ "native_rung_rate": 0,
479
+ "effective_usd_per_1k_conformant": 0.03925973333333333
480
+ },
481
+ "reasoning.numeric": {
482
+ "samples": 15,
483
+ "conform_rate": 1,
484
+ "value_accuracy_mean": 1,
485
+ "native_rung_rate": 0,
486
+ "effective_usd_per_1k_conformant": 0.0392056
487
+ },
488
+ "structure.strict": {
489
+ "samples": 10,
490
+ "conform_rate": 1,
491
+ "value_accuracy_mean": 1,
492
+ "native_rung_rate": 0,
493
+ "effective_usd_per_1k_conformant": 0.054992000000000006
494
+ },
495
+ "summarize": {
496
+ "samples": 15,
497
+ "conform_rate": 1,
498
+ "value_accuracy_mean": 1,
499
+ "native_rung_rate": 0,
500
+ "effective_usd_per_1k_conformant": 0.07806586666666665
501
+ }
502
+ },
503
+ "discrepancies": [
504
+ {
505
+ "kind": "schema_served_via_json_mode",
506
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
507
+ },
508
+ {
509
+ "kind": "citations_without_declared_search",
510
+ "message": "no declared web-search support, but 80% of probed answers cited URLs — verify whether these are real or fabricated sources"
511
+ }
512
+ ]
513
+ },
514
+ {
515
+ "model_key": "grok/grok-4-1-fast",
516
+ "conform_rate": 1,
517
+ "conform_ci95": [
518
+ 0.8864866068260312,
519
+ 0.9999999999999999
520
+ ],
521
+ "value_accuracy_mean": 1,
522
+ "native_rung_rate": 1,
523
+ "hard_conform_rate": null,
524
+ "hard_value_accuracy_mean": null,
525
+ "hard_samples": 0,
526
+ "hard_native_rung_rate": null,
527
+ "effective_usd_per_1k_conformant": 0.10421333333333337,
528
+ "grounded_rate": 0.26666666666666666,
529
+ "cache_hit_rate": 1,
530
+ "samples": 30,
531
+ "as_of": "2026-08-02T20:29:07.180Z",
532
+ "families": [
533
+ "probe-suite"
534
+ ],
535
+ "fixture_counts": {
536
+ "probe-suite": 30
537
+ },
538
+ "by_family": {
539
+ "probe-suite": {
540
+ "samples": 30,
541
+ "conform_rate": 1,
542
+ "value_accuracy_mean": 1,
543
+ "native_rung_rate": 1,
544
+ "effective_usd_per_1k_conformant": 0.10421333333333337
545
+ }
546
+ },
547
+ "by_subtype": {
548
+ "classification": {
549
+ "samples": 5,
550
+ "conform_rate": null,
551
+ "value_accuracy_mean": null,
552
+ "native_rung_rate": 1,
553
+ "effective_usd_per_1k_conformant": null
554
+ },
555
+ "extraction.tabular": {
556
+ "samples": 10,
557
+ "conform_rate": 1,
558
+ "value_accuracy_mean": 1,
559
+ "native_rung_rate": 1,
560
+ "effective_usd_per_1k_conformant": 0.12037
561
+ },
562
+ "generation.prose": {
563
+ "samples": 5,
564
+ "conform_rate": null,
565
+ "value_accuracy_mean": null,
566
+ "native_rung_rate": 1,
567
+ "effective_usd_per_1k_conformant": null
568
+ },
569
+ "structure.strict": {
570
+ "samples": 10,
571
+ "conform_rate": 1,
572
+ "value_accuracy_mean": 1,
573
+ "native_rung_rate": 1,
574
+ "effective_usd_per_1k_conformant": 0.07872999999999998
575
+ }
576
+ },
577
+ "discrepancies": [
578
+ {
579
+ "kind": "declared_search_ungrounded",
580
+ "message": "declared supports_web_search, but only 27% of probed answers carried citations (bar: 50%)"
581
+ }
582
+ ]
583
+ },
584
+ {
585
+ "model_key": "deepinfra/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
586
+ "conform_rate": 0.8315789473684211,
587
+ "conform_ci95": [
588
+ 0.743796910314561,
589
+ 0.8935874499697887
590
+ ],
591
+ "value_accuracy_mean": 1,
592
+ "native_rung_rate": 0,
593
+ "hard_conform_rate": 0.36,
594
+ "hard_value_accuracy_mean": 1,
595
+ "hard_samples": 25,
596
+ "hard_native_rung_rate": 0,
597
+ "effective_usd_per_1k_conformant": 0.1401493670886076,
598
+ "grounded_rate": 1,
599
+ "cache_hit_rate": 0,
600
+ "samples": 95,
601
+ "as_of": "2026-08-18T23:44:50.284Z",
602
+ "families": [
603
+ "demo-rig",
604
+ "probe-suite"
605
+ ],
606
+ "fixture_counts": {
607
+ "demo-rig": 20,
608
+ "probe-suite": 75
609
+ },
610
+ "by_family": {
611
+ "demo-rig": {
612
+ "samples": 20,
613
+ "conform_rate": 1,
614
+ "value_accuracy_mean": 1,
615
+ "native_rung_rate": 0,
616
+ "effective_usd_per_1k_conformant": 0.08479500000000002
617
+ },
618
+ "probe-suite": {
619
+ "samples": 75,
620
+ "conform_rate": 0.7866666666666666,
621
+ "value_accuracy_mean": 1,
622
+ "native_rung_rate": 0,
623
+ "effective_usd_per_1k_conformant": 0.1589135593220339
624
+ }
625
+ },
626
+ "by_subtype": {
627
+ "classification": {
628
+ "samples": 10,
629
+ "conform_rate": 1,
630
+ "value_accuracy_mean": 1,
631
+ "native_rung_rate": 0,
632
+ "effective_usd_per_1k_conformant": 0.06712499999999999
633
+ },
634
+ "extraction.tabular": {
635
+ "samples": 15,
636
+ "conform_rate": 1,
637
+ "value_accuracy_mean": 1,
638
+ "native_rung_rate": 0,
639
+ "effective_usd_per_1k_conformant": 0.11855000000000002
640
+ },
641
+ "generation.prose": {
642
+ "samples": 15,
643
+ "conform_rate": 0.9333333333333333,
644
+ "value_accuracy_mean": 1,
645
+ "native_rung_rate": 0,
646
+ "effective_usd_per_1k_conformant": 0.17464285714285716
647
+ },
648
+ "qa.docs": {
649
+ "samples": 15,
650
+ "conform_rate": 1,
651
+ "value_accuracy_mean": 1,
652
+ "native_rung_rate": 0,
653
+ "effective_usd_per_1k_conformant": 0.06892
654
+ },
655
+ "reasoning.numeric": {
656
+ "samples": 15,
657
+ "conform_rate": 0,
658
+ "value_accuracy_mean": null,
659
+ "native_rung_rate": 0,
660
+ "effective_usd_per_1k_conformant": null
661
+ },
662
+ "structure.strict": {
663
+ "samples": 10,
664
+ "conform_rate": 1,
665
+ "value_accuracy_mean": 1,
666
+ "native_rung_rate": 0,
667
+ "effective_usd_per_1k_conformant": 0.0849
668
+ },
669
+ "summarize": {
670
+ "samples": 15,
671
+ "conform_rate": 1,
672
+ "value_accuracy_mean": 1,
673
+ "native_rung_rate": 0,
674
+ "effective_usd_per_1k_conformant": 0.09215000000000002
675
+ }
676
+ },
677
+ "discrepancies": [
678
+ {
679
+ "kind": "citations_without_declared_search",
680
+ "message": "no declared web-search support, but 100% of probed answers cited URLs — verify whether these are real or fabricated sources"
681
+ }
682
+ ]
683
+ },
684
+ {
685
+ "model_key": "deepseek/deepseek-reasoner",
686
+ "conform_rate": 0.9473684210526315,
687
+ "conform_ci95": [
688
+ 0.882651138829621,
689
+ 0.9773118869508504
690
+ ],
691
+ "value_accuracy_mean": 1,
692
+ "native_rung_rate": 0,
693
+ "hard_conform_rate": 0.8,
694
+ "hard_value_accuracy_mean": 1,
695
+ "hard_samples": 25,
696
+ "hard_native_rung_rate": 0,
697
+ "effective_usd_per_1k_conformant": 0.1480976,
698
+ "grounded_rate": 0.9333333333333333,
699
+ "cache_hit_rate": 1,
700
+ "samples": 95,
701
+ "as_of": "2026-08-19T02:23:10.350Z",
702
+ "families": [
703
+ "demo-rig",
704
+ "probe-suite"
705
+ ],
706
+ "fixture_counts": {
707
+ "demo-rig": 20,
708
+ "probe-suite": 75
709
+ },
710
+ "by_family": {
711
+ "demo-rig": {
712
+ "samples": 20,
713
+ "conform_rate": 1,
714
+ "value_accuracy_mean": 1,
715
+ "native_rung_rate": 0,
716
+ "effective_usd_per_1k_conformant": 0.11320680000000001
717
+ },
718
+ "probe-suite": {
719
+ "samples": 75,
720
+ "conform_rate": 0.9333333333333333,
721
+ "value_accuracy_mean": 1,
722
+ "native_rung_rate": 0,
723
+ "effective_usd_per_1k_conformant": 0.15806639999999997
724
+ }
725
+ },
726
+ "by_subtype": {
727
+ "classification": {
728
+ "samples": 10,
729
+ "conform_rate": 1,
730
+ "value_accuracy_mean": 1,
731
+ "native_rung_rate": 0,
732
+ "effective_usd_per_1k_conformant": 0.08883000000000002
733
+ },
734
+ "extraction.tabular": {
735
+ "samples": 15,
736
+ "conform_rate": 1,
737
+ "value_accuracy_mean": 1,
738
+ "native_rung_rate": 0,
739
+ "effective_usd_per_1k_conformant": 0.17154106666666666
740
+ },
741
+ "generation.prose": {
742
+ "samples": 15,
743
+ "conform_rate": 0.6666666666666666,
744
+ "value_accuracy_mean": 1,
745
+ "native_rung_rate": 0,
746
+ "effective_usd_per_1k_conformant": 0.3892784
747
+ },
748
+ "qa.docs": {
749
+ "samples": 15,
750
+ "conform_rate": 1,
751
+ "value_accuracy_mean": 1,
752
+ "native_rung_rate": 0,
753
+ "effective_usd_per_1k_conformant": 0.09396053333333333
754
+ },
755
+ "reasoning.numeric": {
756
+ "samples": 15,
757
+ "conform_rate": 1,
758
+ "value_accuracy_mean": 1,
759
+ "native_rung_rate": 0,
760
+ "effective_usd_per_1k_conformant": 0.0839888
761
+ },
762
+ "structure.strict": {
763
+ "samples": 10,
764
+ "conform_rate": 1,
765
+ "value_accuracy_mean": 1,
766
+ "native_rung_rate": 0,
767
+ "effective_usd_per_1k_conformant": 0.11736200000000001
768
+ },
769
+ "summarize": {
770
+ "samples": 15,
771
+ "conform_rate": 1,
772
+ "value_accuracy_mean": 1,
773
+ "native_rung_rate": 0,
774
+ "effective_usd_per_1k_conformant": 0.14211493333333333
775
+ }
776
+ },
777
+ "discrepancies": [
778
+ {
779
+ "kind": "schema_served_via_json_mode",
780
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
781
+ },
782
+ {
783
+ "kind": "citations_without_declared_search",
784
+ "message": "no declared web-search support, but 93% of probed answers cited URLs — verify whether these are real or fabricated sources"
785
+ }
786
+ ]
787
+ },
788
+ {
789
+ "model_key": "openai/gpt-5.4-nano",
790
+ "conform_rate": 1,
791
+ "conform_ci95": [
792
+ 0.9611351464605271,
793
+ 1
794
+ ],
795
+ "value_accuracy_mean": 0.8842105263157894,
796
+ "native_rung_rate": 1,
797
+ "hard_conform_rate": 1,
798
+ "hard_value_accuracy_mean": 0.56,
799
+ "hard_samples": 25,
800
+ "hard_native_rung_rate": 1,
801
+ "effective_usd_per_1k_conformant": 0.15002631578947373,
802
+ "grounded_rate": 0.5333333333333333,
803
+ "cache_hit_rate": 0.6,
804
+ "samples": 95,
805
+ "as_of": "2026-08-19T04:31:30.888Z",
806
+ "families": [
807
+ "demo-rig",
808
+ "probe-suite"
809
+ ],
810
+ "fixture_counts": {
811
+ "demo-rig": 20,
812
+ "probe-suite": 75
813
+ },
814
+ "by_family": {
815
+ "demo-rig": {
816
+ "samples": 20,
817
+ "conform_rate": 1,
818
+ "value_accuracy_mean": 1,
819
+ "native_rung_rate": 1,
820
+ "effective_usd_per_1k_conformant": 0.12881250000000002
821
+ },
822
+ "probe-suite": {
823
+ "samples": 75,
824
+ "conform_rate": 1,
825
+ "value_accuracy_mean": 0.8533333333333334,
826
+ "native_rung_rate": 1,
827
+ "effective_usd_per_1k_conformant": 0.15568333333333353
828
+ }
829
+ },
830
+ "by_subtype": {
831
+ "classification": {
832
+ "samples": 10,
833
+ "conform_rate": 1,
834
+ "value_accuracy_mean": 1,
835
+ "native_rung_rate": 1,
836
+ "effective_usd_per_1k_conformant": 0.08467500000000001
837
+ },
838
+ "extraction.tabular": {
839
+ "samples": 15,
840
+ "conform_rate": 1,
841
+ "value_accuracy_mean": 1,
842
+ "native_rung_rate": 1,
843
+ "effective_usd_per_1k_conformant": 0.17641666666666672
844
+ },
845
+ "generation.prose": {
846
+ "samples": 15,
847
+ "conform_rate": 1,
848
+ "value_accuracy_mean": 1,
849
+ "native_rung_rate": 1,
850
+ "effective_usd_per_1k_conformant": 0.30548333333333333
851
+ },
852
+ "qa.docs": {
853
+ "samples": 15,
854
+ "conform_rate": 1,
855
+ "value_accuracy_mean": 1,
856
+ "native_rung_rate": 1,
857
+ "effective_usd_per_1k_conformant": 0.11193333333333333
858
+ },
859
+ "reasoning.numeric": {
860
+ "samples": 15,
861
+ "conform_rate": 1,
862
+ "value_accuracy_mean": 0.26666666666666666,
863
+ "native_rung_rate": 1,
864
+ "effective_usd_per_1k_conformant": 0.05993333333333334
865
+ },
866
+ "structure.strict": {
867
+ "samples": 10,
868
+ "conform_rate": 1,
869
+ "value_accuracy_mean": 1,
870
+ "native_rung_rate": 1,
871
+ "effective_usd_per_1k_conformant": 0.15255
872
+ },
873
+ "summarize": {
874
+ "samples": 15,
875
+ "conform_rate": 1,
876
+ "value_accuracy_mean": 1,
877
+ "native_rung_rate": 1,
878
+ "effective_usd_per_1k_conformant": 0.13825
879
+ }
880
+ },
881
+ "discrepancies": []
882
+ },
883
+ {
884
+ "model_key": "deepinfra/Qwen/Qwen3-30B-A3B",
885
+ "conform_rate": 0.968421052631579,
886
+ "conform_ci95": [
887
+ 0.9112290135269181,
888
+ 0.9892028605255756
889
+ ],
890
+ "value_accuracy_mean": 0.9921052631578947,
891
+ "native_rung_rate": 0,
892
+ "hard_conform_rate": 0.88,
893
+ "hard_value_accuracy_mean": 1,
894
+ "hard_samples": 25,
895
+ "hard_native_rung_rate": 0,
896
+ "effective_usd_per_1k_conformant": 0.17057804347826086,
897
+ "grounded_rate": 0.8666666666666667,
898
+ "cache_hit_rate": 0,
899
+ "samples": 95,
900
+ "as_of": "2026-08-18T23:15:34.247Z",
901
+ "families": [
902
+ "demo-rig",
903
+ "probe-suite"
904
+ ],
905
+ "fixture_counts": {
906
+ "demo-rig": 20,
907
+ "probe-suite": 75
908
+ },
909
+ "by_family": {
910
+ "demo-rig": {
911
+ "samples": 20,
912
+ "conform_rate": 1,
913
+ "value_accuracy_mean": 0.9875,
914
+ "native_rung_rate": 0,
915
+ "effective_usd_per_1k_conformant": 0.13501700000000003
916
+ },
917
+ "probe-suite": {
918
+ "samples": 75,
919
+ "conform_rate": 0.96,
920
+ "value_accuracy_mean": 0.9933333333333333,
921
+ "native_rung_rate": 0,
922
+ "effective_usd_per_1k_conformant": 0.180456111111111
923
+ }
924
+ },
925
+ "by_subtype": {
926
+ "classification": {
927
+ "samples": 10,
928
+ "conform_rate": 1,
929
+ "value_accuracy_mean": 1,
930
+ "native_rung_rate": 0,
931
+ "effective_usd_per_1k_conformant": 0.10238299999999999
932
+ },
933
+ "extraction.tabular": {
934
+ "samples": 15,
935
+ "conform_rate": 1,
936
+ "value_accuracy_mean": 1,
937
+ "native_rung_rate": 0,
938
+ "effective_usd_per_1k_conformant": 0.17585666666666663
939
+ },
940
+ "generation.prose": {
941
+ "samples": 15,
942
+ "conform_rate": 0.8,
943
+ "value_accuracy_mean": 0.9666666666666667,
944
+ "native_rung_rate": 0,
945
+ "effective_usd_per_1k_conformant": 0.3904558333333334
946
+ },
947
+ "qa.docs": {
948
+ "samples": 15,
949
+ "conform_rate": 1,
950
+ "value_accuracy_mean": 1,
951
+ "native_rung_rate": 0,
952
+ "effective_usd_per_1k_conformant": 0.10034266666666668
953
+ },
954
+ "reasoning.numeric": {
955
+ "samples": 15,
956
+ "conform_rate": 1,
957
+ "value_accuracy_mean": 1,
958
+ "native_rung_rate": 0,
959
+ "effective_usd_per_1k_conformant": 0.13603799999999996
960
+ },
961
+ "structure.strict": {
962
+ "samples": 10,
963
+ "conform_rate": 1,
964
+ "value_accuracy_mean": 1,
965
+ "native_rung_rate": 0,
966
+ "effective_usd_per_1k_conformant": 0.14819800000000002
967
+ },
968
+ "summarize": {
969
+ "samples": 15,
970
+ "conform_rate": 1,
971
+ "value_accuracy_mean": 0.9833333333333333,
972
+ "native_rung_rate": 0,
973
+ "effective_usd_per_1k_conformant": 0.154556
974
+ }
975
+ },
976
+ "discrepancies": [
977
+ {
978
+ "kind": "undeclared_schema_capable",
979
+ "message": "no declared schema support, but probed conformance is 97% — capability exists undeclared"
980
+ },
981
+ {
982
+ "kind": "citations_without_declared_search",
983
+ "message": "no declared web-search support, but 87% of probed answers cited URLs — verify whether these are real or fabricated sources"
984
+ }
985
+ ]
986
+ },
987
+ {
988
+ "model_key": "deepinfra/openai/gpt-oss-120b",
989
+ "conform_rate": 0.9894736842105263,
990
+ "conform_ci95": [
991
+ 0.9427612985868637,
992
+ 0.9981394237376521
993
+ ],
994
+ "value_accuracy_mean": 1,
995
+ "native_rung_rate": 0,
996
+ "hard_conform_rate": 0.96,
997
+ "hard_value_accuracy_mean": 1,
998
+ "hard_samples": 25,
999
+ "hard_native_rung_rate": 0,
1000
+ "effective_usd_per_1k_conformant": 0.1706031914893617,
1001
+ "grounded_rate": 0,
1002
+ "cache_hit_rate": 0,
1003
+ "samples": 95,
1004
+ "as_of": "2026-08-19T01:41:22.518Z",
1005
+ "families": [
1006
+ "demo-rig",
1007
+ "probe-suite"
1008
+ ],
1009
+ "fixture_counts": {
1010
+ "demo-rig": 20,
1011
+ "probe-suite": 75
1012
+ },
1013
+ "by_family": {
1014
+ "demo-rig": {
1015
+ "samples": 20,
1016
+ "conform_rate": 1,
1017
+ "value_accuracy_mean": 1,
1018
+ "native_rung_rate": 0,
1019
+ "effective_usd_per_1k_conformant": 0.13035000000000005
1020
+ },
1021
+ "probe-suite": {
1022
+ "samples": 75,
1023
+ "conform_rate": 0.9866666666666667,
1024
+ "value_accuracy_mean": 1,
1025
+ "native_rung_rate": 0,
1026
+ "effective_usd_per_1k_conformant": 0.18148243243243245
1027
+ }
1028
+ },
1029
+ "by_subtype": {
1030
+ "classification": {
1031
+ "samples": 10,
1032
+ "conform_rate": 1,
1033
+ "value_accuracy_mean": 1,
1034
+ "native_rung_rate": 0,
1035
+ "effective_usd_per_1k_conformant": 0.079525
1036
+ },
1037
+ "extraction.tabular": {
1038
+ "samples": 15,
1039
+ "conform_rate": 1,
1040
+ "value_accuracy_mean": 1,
1041
+ "native_rung_rate": 0,
1042
+ "effective_usd_per_1k_conformant": 0.19753666666666664
1043
+ },
1044
+ "generation.prose": {
1045
+ "samples": 15,
1046
+ "conform_rate": 0.9333333333333333,
1047
+ "value_accuracy_mean": 1,
1048
+ "native_rung_rate": 0,
1049
+ "effective_usd_per_1k_conformant": 0.3252964285714287
1050
+ },
1051
+ "qa.docs": {
1052
+ "samples": 15,
1053
+ "conform_rate": 1,
1054
+ "value_accuracy_mean": 1,
1055
+ "native_rung_rate": 0,
1056
+ "effective_usd_per_1k_conformant": 0.11296666666666663
1057
+ },
1058
+ "reasoning.numeric": {
1059
+ "samples": 15,
1060
+ "conform_rate": 1,
1061
+ "value_accuracy_mean": 1,
1062
+ "native_rung_rate": 0,
1063
+ "effective_usd_per_1k_conformant": 0.10586666666666665
1064
+ },
1065
+ "structure.strict": {
1066
+ "samples": 10,
1067
+ "conform_rate": 1,
1068
+ "value_accuracy_mean": 1,
1069
+ "native_rung_rate": 0,
1070
+ "effective_usd_per_1k_conformant": 0.14430000000000004
1071
+ },
1072
+ "summarize": {
1073
+ "samples": 15,
1074
+ "conform_rate": 1,
1075
+ "value_accuracy_mean": 1,
1076
+ "native_rung_rate": 0,
1077
+ "effective_usd_per_1k_conformant": 0.19991666666666674
1078
+ }
1079
+ },
1080
+ "discrepancies": [
1081
+ {
1082
+ "kind": "undeclared_schema_capable",
1083
+ "message": "no declared schema support, but probed conformance is 99% — capability exists undeclared"
1084
+ }
1085
+ ]
1086
+ },
1087
+ {
1088
+ "model_key": "gemini/gemini-3.1-flash-lite",
1089
+ "conform_rate": 0.9473684210526315,
1090
+ "conform_ci95": [
1091
+ 0.882651138829621,
1092
+ 0.9773118869508504
1093
+ ],
1094
+ "value_accuracy_mean": 1,
1095
+ "native_rung_rate": 1,
1096
+ "hard_conform_rate": 0.8,
1097
+ "hard_value_accuracy_mean": 1,
1098
+ "hard_samples": 25,
1099
+ "hard_native_rung_rate": 1,
1100
+ "effective_usd_per_1k_conformant": 0.20748611111111098,
1101
+ "grounded_rate": 1,
1102
+ "cache_hit_rate": 0,
1103
+ "samples": 95,
1104
+ "as_of": "2026-08-19T03:48:20.003Z",
1105
+ "families": [
1106
+ "demo-rig",
1107
+ "probe-suite"
1108
+ ],
1109
+ "fixture_counts": {
1110
+ "demo-rig": 20,
1111
+ "probe-suite": 75
1112
+ },
1113
+ "by_family": {
1114
+ "demo-rig": {
1115
+ "samples": 20,
1116
+ "conform_rate": 1,
1117
+ "value_accuracy_mean": 1,
1118
+ "native_rung_rate": 1,
1119
+ "effective_usd_per_1k_conformant": 0.15428750000000002
1120
+ },
1121
+ "probe-suite": {
1122
+ "samples": 75,
1123
+ "conform_rate": 0.9333333333333333,
1124
+ "value_accuracy_mean": 1,
1125
+ "native_rung_rate": 1,
1126
+ "effective_usd_per_1k_conformant": 0.2226857142857143
1127
+ }
1128
+ },
1129
+ "by_subtype": {
1130
+ "classification": {
1131
+ "samples": 10,
1132
+ "conform_rate": 1,
1133
+ "value_accuracy_mean": 1,
1134
+ "native_rung_rate": 1,
1135
+ "effective_usd_per_1k_conformant": 0.107625
1136
+ },
1137
+ "extraction.tabular": {
1138
+ "samples": 15,
1139
+ "conform_rate": 1,
1140
+ "value_accuracy_mean": 1,
1141
+ "native_rung_rate": 1,
1142
+ "effective_usd_per_1k_conformant": 0.29350000000000004
1143
+ },
1144
+ "generation.prose": {
1145
+ "samples": 15,
1146
+ "conform_rate": 0.6666666666666666,
1147
+ "value_accuracy_mean": 1,
1148
+ "native_rung_rate": 1,
1149
+ "effective_usd_per_1k_conformant": 0.614075
1150
+ },
1151
+ "qa.docs": {
1152
+ "samples": 15,
1153
+ "conform_rate": 1,
1154
+ "value_accuracy_mean": 1,
1155
+ "native_rung_rate": 1,
1156
+ "effective_usd_per_1k_conformant": 0.12041666666666664
1157
+ },
1158
+ "reasoning.numeric": {
1159
+ "samples": 15,
1160
+ "conform_rate": 1,
1161
+ "value_accuracy_mean": 1,
1162
+ "native_rung_rate": 1,
1163
+ "effective_usd_per_1k_conformant": 0.07080000000000002
1164
+ },
1165
+ "structure.strict": {
1166
+ "samples": 10,
1167
+ "conform_rate": 1,
1168
+ "value_accuracy_mean": 1,
1169
+ "native_rung_rate": 1,
1170
+ "effective_usd_per_1k_conformant": 0.19932499999999997
1171
+ },
1172
+ "summarize": {
1173
+ "samples": 15,
1174
+ "conform_rate": 1,
1175
+ "value_accuracy_mean": 1,
1176
+ "native_rung_rate": 1,
1177
+ "effective_usd_per_1k_conformant": 0.14618333333333333
1178
+ }
1179
+ },
1180
+ "discrepancies": [
1181
+ {
1182
+ "kind": "declared_caching_unrealized",
1183
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
1184
+ }
1185
+ ]
1186
+ },
1187
+ {
1188
+ "model_key": "deepinfra/Qwen/Qwen3-Coder-480B-A35B-Instruct-Turbo",
1189
+ "conform_rate": 0.9473684210526315,
1190
+ "conform_ci95": [
1191
+ 0.882651138829621,
1192
+ 0.9773118869508504
1193
+ ],
1194
+ "value_accuracy_mean": 0.9631578947368421,
1195
+ "native_rung_rate": 0,
1196
+ "hard_conform_rate": 0.8,
1197
+ "hard_value_accuracy_mean": 0.88,
1198
+ "hard_samples": 25,
1199
+ "hard_native_rung_rate": 0,
1200
+ "effective_usd_per_1k_conformant": 0.22732444444444455,
1201
+ "grounded_rate": 0.7333333333333333,
1202
+ "cache_hit_rate": 0.8,
1203
+ "samples": 95,
1204
+ "as_of": "2026-08-18T23:24:40.751Z",
1205
+ "families": [
1206
+ "demo-rig",
1207
+ "probe-suite"
1208
+ ],
1209
+ "fixture_counts": {
1210
+ "demo-rig": 20,
1211
+ "probe-suite": 75
1212
+ },
1213
+ "by_family": {
1214
+ "demo-rig": {
1215
+ "samples": 20,
1216
+ "conform_rate": 1,
1217
+ "value_accuracy_mean": 1,
1218
+ "native_rung_rate": 0,
1219
+ "effective_usd_per_1k_conformant": 0.18485499999999996
1220
+ },
1221
+ "probe-suite": {
1222
+ "samples": 75,
1223
+ "conform_rate": 0.9333333333333333,
1224
+ "value_accuracy_mean": 0.9533333333333334,
1225
+ "native_rung_rate": 0,
1226
+ "effective_usd_per_1k_conformant": 0.23945857142857152
1227
+ }
1228
+ },
1229
+ "by_subtype": {
1230
+ "classification": {
1231
+ "samples": 10,
1232
+ "conform_rate": 1,
1233
+ "value_accuracy_mean": 1,
1234
+ "native_rung_rate": 0,
1235
+ "effective_usd_per_1k_conformant": 0.12136499999999999
1236
+ },
1237
+ "extraction.tabular": {
1238
+ "samples": 15,
1239
+ "conform_rate": 1,
1240
+ "value_accuracy_mean": 1,
1241
+ "native_rung_rate": 0,
1242
+ "effective_usd_per_1k_conformant": 0.30405333333333323
1243
+ },
1244
+ "generation.prose": {
1245
+ "samples": 15,
1246
+ "conform_rate": 0.6666666666666666,
1247
+ "value_accuracy_mean": 0.9666666666666667,
1248
+ "native_rung_rate": 0,
1249
+ "effective_usd_per_1k_conformant": 0.610865
1250
+ },
1251
+ "qa.docs": {
1252
+ "samples": 15,
1253
+ "conform_rate": 1,
1254
+ "value_accuracy_mean": 1,
1255
+ "native_rung_rate": 0,
1256
+ "effective_usd_per_1k_conformant": 0.15368333333333328
1257
+ },
1258
+ "reasoning.numeric": {
1259
+ "samples": 15,
1260
+ "conform_rate": 1,
1261
+ "value_accuracy_mean": 0.8,
1262
+ "native_rung_rate": 0,
1263
+ "effective_usd_per_1k_conformant": 0.10605666666666667
1264
+ },
1265
+ "structure.strict": {
1266
+ "samples": 10,
1267
+ "conform_rate": 1,
1268
+ "value_accuracy_mean": 1,
1269
+ "native_rung_rate": 0,
1270
+ "effective_usd_per_1k_conformant": 0.212185
1271
+ },
1272
+ "summarize": {
1273
+ "samples": 15,
1274
+ "conform_rate": 1,
1275
+ "value_accuracy_mean": 1,
1276
+ "native_rung_rate": 0,
1277
+ "effective_usd_per_1k_conformant": 0.1705433333333333
1278
+ }
1279
+ },
1280
+ "discrepancies": [
1281
+ {
1282
+ "kind": "undeclared_schema_capable",
1283
+ "message": "no declared schema support, but probed conformance is 95% — capability exists undeclared"
1284
+ },
1285
+ {
1286
+ "kind": "citations_without_declared_search",
1287
+ "message": "no declared web-search support, but 73% of probed answers cited URLs — verify whether these are real or fabricated sources"
1288
+ }
1289
+ ]
1290
+ },
1291
+ {
1292
+ "model_key": "fireworks/gpt-oss-120b",
1293
+ "conform_rate": 0.9894736842105263,
1294
+ "conform_ci95": [
1295
+ 0.9427612985868637,
1296
+ 0.9981394237376521
1297
+ ],
1298
+ "value_accuracy_mean": 1,
1299
+ "native_rung_rate": 0,
1300
+ "hard_conform_rate": 0.96,
1301
+ "hard_value_accuracy_mean": 1,
1302
+ "hard_samples": 25,
1303
+ "hard_native_rung_rate": 0,
1304
+ "effective_usd_per_1k_conformant": 0.2663190957446808,
1305
+ "grounded_rate": 0.2,
1306
+ "cache_hit_rate": 1,
1307
+ "samples": 95,
1308
+ "as_of": "2026-08-19T03:13:23.160Z",
1309
+ "families": [
1310
+ "demo-rig",
1311
+ "probe-suite"
1312
+ ],
1313
+ "fixture_counts": {
1314
+ "demo-rig": 20,
1315
+ "probe-suite": 75
1316
+ },
1317
+ "by_family": {
1318
+ "demo-rig": {
1319
+ "samples": 20,
1320
+ "conform_rate": 1,
1321
+ "value_accuracy_mean": 1,
1322
+ "native_rung_rate": 0,
1323
+ "effective_usd_per_1k_conformant": 0.21015525
1324
+ },
1325
+ "probe-suite": {
1326
+ "samples": 75,
1327
+ "conform_rate": 0.9866666666666667,
1328
+ "value_accuracy_mean": 1,
1329
+ "native_rung_rate": 0,
1330
+ "effective_usd_per_1k_conformant": 0.28149851351351346
1331
+ }
1332
+ },
1333
+ "by_subtype": {
1334
+ "classification": {
1335
+ "samples": 10,
1336
+ "conform_rate": 1,
1337
+ "value_accuracy_mean": 1,
1338
+ "native_rung_rate": 0,
1339
+ "effective_usd_per_1k_conformant": 0.1536015
1340
+ },
1341
+ "extraction.tabular": {
1342
+ "samples": 15,
1343
+ "conform_rate": 1,
1344
+ "value_accuracy_mean": 1,
1345
+ "native_rung_rate": 0,
1346
+ "effective_usd_per_1k_conformant": 0.247968
1347
+ },
1348
+ "generation.prose": {
1349
+ "samples": 15,
1350
+ "conform_rate": 0.9333333333333333,
1351
+ "value_accuracy_mean": 1,
1352
+ "native_rung_rate": 0,
1353
+ "effective_usd_per_1k_conformant": 0.5492014285714286
1354
+ },
1355
+ "qa.docs": {
1356
+ "samples": 15,
1357
+ "conform_rate": 1,
1358
+ "value_accuracy_mean": 1,
1359
+ "native_rung_rate": 0,
1360
+ "effective_usd_per_1k_conformant": 0.19312100000000001
1361
+ },
1362
+ "reasoning.numeric": {
1363
+ "samples": 15,
1364
+ "conform_rate": 1,
1365
+ "value_accuracy_mean": 1,
1366
+ "native_rung_rate": 0,
1367
+ "effective_usd_per_1k_conformant": 0.16063000000000005
1368
+ },
1369
+ "structure.strict": {
1370
+ "samples": 10,
1371
+ "conform_rate": 1,
1372
+ "value_accuracy_mean": 1,
1373
+ "native_rung_rate": 0,
1374
+ "effective_usd_per_1k_conformant": 0.20200349999999997
1375
+ },
1376
+ "summarize": {
1377
+ "samples": 15,
1378
+ "conform_rate": 1,
1379
+ "value_accuracy_mean": 1,
1380
+ "native_rung_rate": 0,
1381
+ "effective_usd_per_1k_conformant": 0.31755600000000006
1382
+ }
1383
+ },
1384
+ "discrepancies": [
1385
+ {
1386
+ "kind": "schema_served_via_json_mode",
1387
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
1388
+ }
1389
+ ]
1390
+ },
1391
+ {
1392
+ "model_key": "deepseek/deepseek-v4-pro",
1393
+ "conform_rate": 0.9894736842105263,
1394
+ "conform_ci95": [
1395
+ 0.9427612985868637,
1396
+ 0.9981394237376521
1397
+ ],
1398
+ "value_accuracy_mean": 0.9947368421052631,
1399
+ "native_rung_rate": 0,
1400
+ "hard_conform_rate": 0.96,
1401
+ "hard_value_accuracy_mean": 1,
1402
+ "hard_samples": 25,
1403
+ "hard_native_rung_rate": 0,
1404
+ "effective_usd_per_1k_conformant": 0.40443153191489356,
1405
+ "grounded_rate": 0.06666666666666667,
1406
+ "cache_hit_rate": 1,
1407
+ "samples": 95,
1408
+ "as_of": "2026-08-19T02:46:12.378Z",
1409
+ "families": [
1410
+ "demo-rig",
1411
+ "probe-suite"
1412
+ ],
1413
+ "fixture_counts": {
1414
+ "demo-rig": 20,
1415
+ "probe-suite": 75
1416
+ },
1417
+ "by_family": {
1418
+ "demo-rig": {
1419
+ "samples": 20,
1420
+ "conform_rate": 1,
1421
+ "value_accuracy_mean": 1,
1422
+ "native_rung_rate": 0,
1423
+ "effective_usd_per_1k_conformant": 0.24222830000000004
1424
+ },
1425
+ "probe-suite": {
1426
+ "samples": 75,
1427
+ "conform_rate": 0.9866666666666667,
1428
+ "value_accuracy_mean": 0.9933333333333333,
1429
+ "native_rung_rate": 0,
1430
+ "effective_usd_per_1k_conformant": 0.4482702432432434
1431
+ }
1432
+ },
1433
+ "by_subtype": {
1434
+ "classification": {
1435
+ "samples": 10,
1436
+ "conform_rate": 1,
1437
+ "value_accuracy_mean": 1,
1438
+ "native_rung_rate": 0,
1439
+ "effective_usd_per_1k_conformant": 0.14339049999999998
1440
+ },
1441
+ "extraction.tabular": {
1442
+ "samples": 15,
1443
+ "conform_rate": 1,
1444
+ "value_accuracy_mean": 1,
1445
+ "native_rung_rate": 0,
1446
+ "effective_usd_per_1k_conformant": 0.29252106666666666
1447
+ },
1448
+ "generation.prose": {
1449
+ "samples": 15,
1450
+ "conform_rate": 0.9333333333333333,
1451
+ "value_accuracy_mean": 0.9666666666666667,
1452
+ "native_rung_rate": 0,
1453
+ "effective_usd_per_1k_conformant": 1.3953992142857143
1454
+ },
1455
+ "qa.docs": {
1456
+ "samples": 15,
1457
+ "conform_rate": 1,
1458
+ "value_accuracy_mean": 1,
1459
+ "native_rung_rate": 0,
1460
+ "effective_usd_per_1k_conformant": 0.23345386666666665
1461
+ },
1462
+ "reasoning.numeric": {
1463
+ "samples": 15,
1464
+ "conform_rate": 1,
1465
+ "value_accuracy_mean": 1,
1466
+ "native_rung_rate": 0,
1467
+ "effective_usd_per_1k_conformant": 0.1623091333333333
1468
+ },
1469
+ "structure.strict": {
1470
+ "samples": 10,
1471
+ "conform_rate": 1,
1472
+ "value_accuracy_mean": 1,
1473
+ "native_rung_rate": 0,
1474
+ "effective_usd_per_1k_conformant": 0.22042899999999999
1475
+ },
1476
+ "summarize": {
1477
+ "samples": 15,
1478
+ "conform_rate": 1,
1479
+ "value_accuracy_mean": 1,
1480
+ "native_rung_rate": 0,
1481
+ "effective_usd_per_1k_conformant": 0.3012346000000001
1482
+ }
1483
+ },
1484
+ "discrepancies": [
1485
+ {
1486
+ "kind": "schema_served_via_json_mode",
1487
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
1488
+ }
1489
+ ]
1490
+ },
1491
+ {
1492
+ "model_key": "openai/gpt-5.4-mini",
1493
+ "conform_rate": 1,
1494
+ "conform_ci95": [
1495
+ 0.9611351464605271,
1496
+ 1
1497
+ ],
1498
+ "value_accuracy_mean": 0.9789473684210527,
1499
+ "native_rung_rate": 1,
1500
+ "hard_conform_rate": 1,
1501
+ "hard_value_accuracy_mean": 0.92,
1502
+ "hard_samples": 25,
1503
+ "hard_native_rung_rate": 1,
1504
+ "effective_usd_per_1k_conformant": 0.5249842105263154,
1505
+ "grounded_rate": 0.6,
1506
+ "cache_hit_rate": 1,
1507
+ "samples": 95,
1508
+ "as_of": "2026-08-19T04:29:27.154Z",
1509
+ "families": [
1510
+ "demo-rig",
1511
+ "probe-suite"
1512
+ ],
1513
+ "fixture_counts": {
1514
+ "demo-rig": 20,
1515
+ "probe-suite": 75
1516
+ },
1517
+ "by_family": {
1518
+ "demo-rig": {
1519
+ "samples": 20,
1520
+ "conform_rate": 1,
1521
+ "value_accuracy_mean": 1,
1522
+ "native_rung_rate": 1,
1523
+ "effective_usd_per_1k_conformant": 0.44355
1524
+ },
1525
+ "probe-suite": {
1526
+ "samples": 75,
1527
+ "conform_rate": 1,
1528
+ "value_accuracy_mean": 0.9733333333333334,
1529
+ "native_rung_rate": 1,
1530
+ "effective_usd_per_1k_conformant": 0.5466999999999997
1531
+ }
1532
+ },
1533
+ "by_subtype": {
1534
+ "classification": {
1535
+ "samples": 10,
1536
+ "conform_rate": 1,
1537
+ "value_accuracy_mean": 1,
1538
+ "native_rung_rate": 1,
1539
+ "effective_usd_per_1k_conformant": 0.3105000000000001
1540
+ },
1541
+ "extraction.tabular": {
1542
+ "samples": 15,
1543
+ "conform_rate": 1,
1544
+ "value_accuracy_mean": 1,
1545
+ "native_rung_rate": 1,
1546
+ "effective_usd_per_1k_conformant": 0.6442500000000002
1547
+ },
1548
+ "generation.prose": {
1549
+ "samples": 15,
1550
+ "conform_rate": 1,
1551
+ "value_accuracy_mean": 1,
1552
+ "native_rung_rate": 1,
1553
+ "effective_usd_per_1k_conformant": 1.0529500000000003
1554
+ },
1555
+ "qa.docs": {
1556
+ "samples": 15,
1557
+ "conform_rate": 1,
1558
+ "value_accuracy_mean": 1,
1559
+ "native_rung_rate": 1,
1560
+ "effective_usd_per_1k_conformant": 0.36200000000000004
1561
+ },
1562
+ "reasoning.numeric": {
1563
+ "samples": 15,
1564
+ "conform_rate": 1,
1565
+ "value_accuracy_mean": 0.8666666666666667,
1566
+ "native_rung_rate": 1,
1567
+ "effective_usd_per_1k_conformant": 0.23175
1568
+ },
1569
+ "structure.strict": {
1570
+ "samples": 10,
1571
+ "conform_rate": 1,
1572
+ "value_accuracy_mean": 1,
1573
+ "native_rung_rate": 1,
1574
+ "effective_usd_per_1k_conformant": 0.4653750000000001
1575
+ },
1576
+ "summarize": {
1577
+ "samples": 15,
1578
+ "conform_rate": 1,
1579
+ "value_accuracy_mean": 1,
1580
+ "native_rung_rate": 1,
1581
+ "effective_usd_per_1k_conformant": 0.5167
1582
+ }
1583
+ },
1584
+ "discrepancies": []
1585
+ },
1586
+ {
1587
+ "model_key": "grok/grok-4.3",
1588
+ "conform_rate": 1,
1589
+ "conform_ci95": [
1590
+ 0.8864866068260312,
1591
+ 0.9999999999999999
1592
+ ],
1593
+ "value_accuracy_mean": 1,
1594
+ "native_rung_rate": 1,
1595
+ "hard_conform_rate": null,
1596
+ "hard_value_accuracy_mean": null,
1597
+ "hard_samples": 0,
1598
+ "hard_native_rung_rate": null,
1599
+ "effective_usd_per_1k_conformant": 0.5436599999999999,
1600
+ "grounded_rate": 0.2,
1601
+ "cache_hit_rate": 1,
1602
+ "samples": 30,
1603
+ "as_of": "2026-08-02T20:25:45.769Z",
1604
+ "families": [
1605
+ "probe-suite"
1606
+ ],
1607
+ "fixture_counts": {
1608
+ "probe-suite": 30
1609
+ },
1610
+ "by_family": {
1611
+ "probe-suite": {
1612
+ "samples": 30,
1613
+ "conform_rate": 1,
1614
+ "value_accuracy_mean": 1,
1615
+ "native_rung_rate": 1,
1616
+ "effective_usd_per_1k_conformant": 0.5436599999999999
1617
+ }
1618
+ },
1619
+ "by_subtype": {
1620
+ "classification": {
1621
+ "samples": 5,
1622
+ "conform_rate": null,
1623
+ "value_accuracy_mean": null,
1624
+ "native_rung_rate": 1,
1625
+ "effective_usd_per_1k_conformant": null
1626
+ },
1627
+ "extraction.tabular": {
1628
+ "samples": 10,
1629
+ "conform_rate": 1,
1630
+ "value_accuracy_mean": 1,
1631
+ "native_rung_rate": 1,
1632
+ "effective_usd_per_1k_conformant": 0.6411650000000002
1633
+ },
1634
+ "generation.prose": {
1635
+ "samples": 5,
1636
+ "conform_rate": null,
1637
+ "value_accuracy_mean": null,
1638
+ "native_rung_rate": 1,
1639
+ "effective_usd_per_1k_conformant": null
1640
+ },
1641
+ "structure.strict": {
1642
+ "samples": 10,
1643
+ "conform_rate": 1,
1644
+ "value_accuracy_mean": 1,
1645
+ "native_rung_rate": 1,
1646
+ "effective_usd_per_1k_conformant": 0.4360350000000001
1647
+ }
1648
+ },
1649
+ "discrepancies": [
1650
+ {
1651
+ "kind": "declared_search_ungrounded",
1652
+ "message": "declared supports_web_search, but only 20% of probed answers carried citations (bar: 50%)"
1653
+ }
1654
+ ]
1655
+ },
1656
+ {
1657
+ "model_key": "deepinfra/MiniMaxAI/MiniMax-M3",
1658
+ "conform_rate": 0.9894736842105263,
1659
+ "conform_ci95": [
1660
+ 0.9427612985868637,
1661
+ 0.9981394237376521
1662
+ ],
1663
+ "value_accuracy_mean": 1,
1664
+ "native_rung_rate": 0,
1665
+ "hard_conform_rate": 0.96,
1666
+ "hard_value_accuracy_mean": 1,
1667
+ "hard_samples": 25,
1668
+ "hard_native_rung_rate": 0,
1669
+ "effective_usd_per_1k_conformant": 0.6967627659574466,
1670
+ "grounded_rate": 0.8666666666666667,
1671
+ "cache_hit_rate": 1,
1672
+ "samples": 95,
1673
+ "as_of": "2026-08-18T23:05:15.308Z",
1674
+ "families": [
1675
+ "demo-rig",
1676
+ "probe-suite"
1677
+ ],
1678
+ "fixture_counts": {
1679
+ "demo-rig": 20,
1680
+ "probe-suite": 75
1681
+ },
1682
+ "by_family": {
1683
+ "demo-rig": {
1684
+ "samples": 20,
1685
+ "conform_rate": 1,
1686
+ "value_accuracy_mean": 1,
1687
+ "native_rung_rate": 0,
1688
+ "effective_usd_per_1k_conformant": 0.36619000000000007
1689
+ },
1690
+ "probe-suite": {
1691
+ "samples": 75,
1692
+ "conform_rate": 0.9866666666666667,
1693
+ "value_accuracy_mean": 1,
1694
+ "native_rung_rate": 0,
1695
+ "effective_usd_per_1k_conformant": 0.7861067567567567
1696
+ }
1697
+ },
1698
+ "by_subtype": {
1699
+ "classification": {
1700
+ "samples": 10,
1701
+ "conform_rate": 1,
1702
+ "value_accuracy_mean": 1,
1703
+ "native_rung_rate": 0,
1704
+ "effective_usd_per_1k_conformant": 0.25671
1705
+ },
1706
+ "extraction.tabular": {
1707
+ "samples": 15,
1708
+ "conform_rate": 1,
1709
+ "value_accuracy_mean": 1,
1710
+ "native_rung_rate": 0,
1711
+ "effective_usd_per_1k_conformant": 0.43009333333333344
1712
+ },
1713
+ "generation.prose": {
1714
+ "samples": 15,
1715
+ "conform_rate": 0.9333333333333333,
1716
+ "value_accuracy_mean": 1,
1717
+ "native_rung_rate": 0,
1718
+ "effective_usd_per_1k_conformant": 2.6062142857142856
1719
+ },
1720
+ "qa.docs": {
1721
+ "samples": 15,
1722
+ "conform_rate": 1,
1723
+ "value_accuracy_mean": 1,
1724
+ "native_rung_rate": 0,
1725
+ "effective_usd_per_1k_conformant": 0.33156
1726
+ },
1727
+ "reasoning.numeric": {
1728
+ "samples": 15,
1729
+ "conform_rate": 1,
1730
+ "value_accuracy_mean": 1,
1731
+ "native_rung_rate": 0,
1732
+ "effective_usd_per_1k_conformant": 0.31377999999999995
1733
+ },
1734
+ "structure.strict": {
1735
+ "samples": 10,
1736
+ "conform_rate": 1,
1737
+ "value_accuracy_mean": 1,
1738
+ "native_rung_rate": 0,
1739
+ "effective_usd_per_1k_conformant": 0.4170000000000001
1740
+ },
1741
+ "summarize": {
1742
+ "samples": 15,
1743
+ "conform_rate": 1,
1744
+ "value_accuracy_mean": 1,
1745
+ "native_rung_rate": 0,
1746
+ "effective_usd_per_1k_conformant": 0.40934000000000004
1747
+ }
1748
+ },
1749
+ "discrepancies": [
1750
+ {
1751
+ "kind": "undeclared_schema_capable",
1752
+ "message": "no declared schema support, but probed conformance is 99% — capability exists undeclared"
1753
+ },
1754
+ {
1755
+ "kind": "citations_without_declared_search",
1756
+ "message": "no declared web-search support, but 87% of probed answers cited URLs — verify whether these are real or fabricated sources"
1757
+ }
1758
+ ]
1759
+ },
1760
+ {
1761
+ "model_key": "openai/gpt-5-mini",
1762
+ "conform_rate": 1,
1763
+ "conform_ci95": [
1764
+ 0.9611351464605271,
1765
+ 1
1766
+ ],
1767
+ "value_accuracy_mean": 1,
1768
+ "native_rung_rate": 1,
1769
+ "hard_conform_rate": 1,
1770
+ "hard_value_accuracy_mean": 1,
1771
+ "hard_samples": 25,
1772
+ "hard_native_rung_rate": 1,
1773
+ "effective_usd_per_1k_conformant": 1.0472315789473683,
1774
+ "grounded_rate": 0.5333333333333333,
1775
+ "cache_hit_rate": 1,
1776
+ "samples": 95,
1777
+ "as_of": "2026-08-19T04:21:20.089Z",
1778
+ "families": [
1779
+ "demo-rig",
1780
+ "probe-suite"
1781
+ ],
1782
+ "fixture_counts": {
1783
+ "demo-rig": 20,
1784
+ "probe-suite": 75
1785
+ },
1786
+ "by_family": {
1787
+ "demo-rig": {
1788
+ "samples": 20,
1789
+ "conform_rate": 1,
1790
+ "value_accuracy_mean": 1,
1791
+ "native_rung_rate": 1,
1792
+ "effective_usd_per_1k_conformant": 0.5944999999999999
1793
+ },
1794
+ "probe-suite": {
1795
+ "samples": 75,
1796
+ "conform_rate": 1,
1797
+ "value_accuracy_mean": 1,
1798
+ "native_rung_rate": 1,
1799
+ "effective_usd_per_1k_conformant": 1.16796
1800
+ }
1801
+ },
1802
+ "by_subtype": {
1803
+ "classification": {
1804
+ "samples": 10,
1805
+ "conform_rate": 1,
1806
+ "value_accuracy_mean": 1,
1807
+ "native_rung_rate": 1,
1808
+ "effective_usd_per_1k_conformant": 0.5310499999999999
1809
+ },
1810
+ "extraction.tabular": {
1811
+ "samples": 15,
1812
+ "conform_rate": 1,
1813
+ "value_accuracy_mean": 1,
1814
+ "native_rung_rate": 1,
1815
+ "effective_usd_per_1k_conformant": 0.64825
1816
+ },
1817
+ "generation.prose": {
1818
+ "samples": 15,
1819
+ "conform_rate": 1,
1820
+ "value_accuracy_mean": 1,
1821
+ "native_rung_rate": 1,
1822
+ "effective_usd_per_1k_conformant": 3.22455
1823
+ },
1824
+ "qa.docs": {
1825
+ "samples": 15,
1826
+ "conform_rate": 1,
1827
+ "value_accuracy_mean": 1,
1828
+ "native_rung_rate": 1,
1829
+ "effective_usd_per_1k_conformant": 0.8489666666666665
1830
+ },
1831
+ "reasoning.numeric": {
1832
+ "samples": 15,
1833
+ "conform_rate": 1,
1834
+ "value_accuracy_mean": 1,
1835
+ "native_rung_rate": 1,
1836
+ "effective_usd_per_1k_conformant": 0.49078333333333335
1837
+ },
1838
+ "structure.strict": {
1839
+ "samples": 10,
1840
+ "conform_rate": 1,
1841
+ "value_accuracy_mean": 1,
1842
+ "native_rung_rate": 1,
1843
+ "effective_usd_per_1k_conformant": 0.6320250000000001
1844
+ },
1845
+ "summarize": {
1846
+ "samples": 15,
1847
+ "conform_rate": 1,
1848
+ "value_accuracy_mean": 1,
1849
+ "native_rung_rate": 1,
1850
+ "effective_usd_per_1k_conformant": 0.6445333333333334
1851
+ }
1852
+ },
1853
+ "discrepancies": []
1854
+ },
1855
+ {
1856
+ "model_key": "deepinfra/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B",
1857
+ "conform_rate": 1,
1858
+ "conform_ci95": [
1859
+ 0.9611351464605271,
1860
+ 1
1861
+ ],
1862
+ "value_accuracy_mean": 1,
1863
+ "native_rung_rate": 0,
1864
+ "hard_conform_rate": 1,
1865
+ "hard_value_accuracy_mean": 1,
1866
+ "hard_samples": 25,
1867
+ "hard_native_rung_rate": 0,
1868
+ "effective_usd_per_1k_conformant": 1.1873010526315793,
1869
+ "grounded_rate": 0.3333333333333333,
1870
+ "cache_hit_rate": 0,
1871
+ "samples": 95,
1872
+ "as_of": "2026-08-19T01:25:37.101Z",
1873
+ "families": [
1874
+ "demo-rig",
1875
+ "probe-suite"
1876
+ ],
1877
+ "fixture_counts": {
1878
+ "demo-rig": 20,
1879
+ "probe-suite": 75
1880
+ },
1881
+ "by_family": {
1882
+ "demo-rig": {
1883
+ "samples": 20,
1884
+ "conform_rate": 1,
1885
+ "value_accuracy_mean": 1,
1886
+ "native_rung_rate": 0,
1887
+ "effective_usd_per_1k_conformant": 0.9907050000000002
1888
+ },
1889
+ "probe-suite": {
1890
+ "samples": 75,
1891
+ "conform_rate": 1,
1892
+ "value_accuracy_mean": 1,
1893
+ "native_rung_rate": 0,
1894
+ "effective_usd_per_1k_conformant": 1.2397266666666673
1895
+ }
1896
+ },
1897
+ "by_subtype": {
1898
+ "classification": {
1899
+ "samples": 10,
1900
+ "conform_rate": 1,
1901
+ "value_accuracy_mean": 1,
1902
+ "native_rung_rate": 0,
1903
+ "effective_usd_per_1k_conformant": 0.74061
1904
+ },
1905
+ "extraction.tabular": {
1906
+ "samples": 15,
1907
+ "conform_rate": 1,
1908
+ "value_accuracy_mean": 1,
1909
+ "native_rung_rate": 0,
1910
+ "effective_usd_per_1k_conformant": 1.4415733333333334
1911
+ },
1912
+ "generation.prose": {
1913
+ "samples": 15,
1914
+ "conform_rate": 1,
1915
+ "value_accuracy_mean": 1,
1916
+ "native_rung_rate": 0,
1917
+ "effective_usd_per_1k_conformant": 2.3358200000000005
1918
+ },
1919
+ "qa.docs": {
1920
+ "samples": 15,
1921
+ "conform_rate": 1,
1922
+ "value_accuracy_mean": 1,
1923
+ "native_rung_rate": 0,
1924
+ "effective_usd_per_1k_conformant": 0.8641800000000001
1925
+ },
1926
+ "reasoning.numeric": {
1927
+ "samples": 15,
1928
+ "conform_rate": 1,
1929
+ "value_accuracy_mean": 1,
1930
+ "native_rung_rate": 0,
1931
+ "effective_usd_per_1k_conformant": 0.6694
1932
+ },
1933
+ "structure.strict": {
1934
+ "samples": 10,
1935
+ "conform_rate": 1,
1936
+ "value_accuracy_mean": 1,
1937
+ "native_rung_rate": 0,
1938
+ "effective_usd_per_1k_conformant": 0.9782700000000001
1939
+ },
1940
+ "summarize": {
1941
+ "samples": 15,
1942
+ "conform_rate": 1,
1943
+ "value_accuracy_mean": 1,
1944
+ "native_rung_rate": 0,
1945
+ "effective_usd_per_1k_conformant": 1.06268
1946
+ }
1947
+ },
1948
+ "discrepancies": [
1949
+ {
1950
+ "kind": "undeclared_schema_capable",
1951
+ "message": "no declared schema support, but probed conformance is 100% — capability exists undeclared"
1952
+ }
1953
+ ]
1954
+ },
1955
+ {
1956
+ "model_key": "grok/grok-4.5",
1957
+ "conform_rate": 1,
1958
+ "conform_ci95": [
1959
+ 0.8864866068260312,
1960
+ 0.9999999999999999
1961
+ ],
1962
+ "value_accuracy_mean": 1,
1963
+ "native_rung_rate": 1,
1964
+ "hard_conform_rate": null,
1965
+ "hard_value_accuracy_mean": null,
1966
+ "hard_samples": 0,
1967
+ "hard_native_rung_rate": null,
1968
+ "effective_usd_per_1k_conformant": 1.2777999999999998,
1969
+ "grounded_rate": 0.26666666666666666,
1970
+ "cache_hit_rate": 1,
1971
+ "samples": 30,
1972
+ "as_of": "2026-08-02T20:22:46.268Z",
1973
+ "families": [
1974
+ "probe-suite"
1975
+ ],
1976
+ "fixture_counts": {
1977
+ "probe-suite": 30
1978
+ },
1979
+ "by_family": {
1980
+ "probe-suite": {
1981
+ "samples": 30,
1982
+ "conform_rate": 1,
1983
+ "value_accuracy_mean": 1,
1984
+ "native_rung_rate": 1,
1985
+ "effective_usd_per_1k_conformant": 1.2777999999999998
1986
+ }
1987
+ },
1988
+ "by_subtype": {
1989
+ "classification": {
1990
+ "samples": 5,
1991
+ "conform_rate": null,
1992
+ "value_accuracy_mean": null,
1993
+ "native_rung_rate": 1,
1994
+ "effective_usd_per_1k_conformant": null
1995
+ },
1996
+ "extraction.tabular": {
1997
+ "samples": 10,
1998
+ "conform_rate": 1,
1999
+ "value_accuracy_mean": 1,
2000
+ "native_rung_rate": 1,
2001
+ "effective_usd_per_1k_conformant": 1.5486000000000002
2002
+ },
2003
+ "generation.prose": {
2004
+ "samples": 5,
2005
+ "conform_rate": null,
2006
+ "value_accuracy_mean": null,
2007
+ "native_rung_rate": 1,
2008
+ "effective_usd_per_1k_conformant": null
2009
+ },
2010
+ "structure.strict": {
2011
+ "samples": 10,
2012
+ "conform_rate": 1,
2013
+ "value_accuracy_mean": 1,
2014
+ "native_rung_rate": 1,
2015
+ "effective_usd_per_1k_conformant": 0.8667999999999998
2016
+ }
2017
+ },
2018
+ "discrepancies": [
2019
+ {
2020
+ "kind": "declared_search_ungrounded",
2021
+ "message": "declared supports_web_search, but only 27% of probed answers carried citations (bar: 50%)"
2022
+ }
2023
+ ]
2024
+ },
2025
+ {
2026
+ "model_key": "gemini/gemini-2.5-flash",
2027
+ "conform_rate": 0.9473684210526315,
2028
+ "conform_ci95": [
2029
+ 0.882651138829621,
2030
+ 0.9773118869508504
2031
+ ],
2032
+ "value_accuracy_mean": 1,
2033
+ "native_rung_rate": 1,
2034
+ "hard_conform_rate": 0.8,
2035
+ "hard_value_accuracy_mean": 1,
2036
+ "hard_samples": 25,
2037
+ "hard_native_rung_rate": 1,
2038
+ "effective_usd_per_1k_conformant": 1.2811757777777781,
2039
+ "grounded_rate": 0.9333333333333333,
2040
+ "cache_hit_rate": 0.4,
2041
+ "samples": 95,
2042
+ "as_of": "2026-08-19T03:36:45.162Z",
2043
+ "families": [
2044
+ "demo-rig",
2045
+ "probe-suite"
2046
+ ],
2047
+ "fixture_counts": {
2048
+ "demo-rig": 20,
2049
+ "probe-suite": 75
2050
+ },
2051
+ "by_family": {
2052
+ "demo-rig": {
2053
+ "samples": 20,
2054
+ "conform_rate": 1,
2055
+ "value_accuracy_mean": 1,
2056
+ "native_rung_rate": 1,
2057
+ "effective_usd_per_1k_conformant": 0.7699500000000001
2058
+ },
2059
+ "probe-suite": {
2060
+ "samples": 75,
2061
+ "conform_rate": 0.9333333333333333,
2062
+ "value_accuracy_mean": 1,
2063
+ "native_rung_rate": 1,
2064
+ "effective_usd_per_1k_conformant": 1.427240285714286
2065
+ }
2066
+ },
2067
+ "by_subtype": {
2068
+ "classification": {
2069
+ "samples": 10,
2070
+ "conform_rate": 1,
2071
+ "value_accuracy_mean": 1,
2072
+ "native_rung_rate": 1,
2073
+ "effective_usd_per_1k_conformant": 0.6468000000000002
2074
+ },
2075
+ "extraction.tabular": {
2076
+ "samples": 15,
2077
+ "conform_rate": 1,
2078
+ "value_accuracy_mean": 1,
2079
+ "native_rung_rate": 1,
2080
+ "effective_usd_per_1k_conformant": 1.1068213333333332
2081
+ },
2082
+ "generation.prose": {
2083
+ "samples": 15,
2084
+ "conform_rate": 0.6666666666666666,
2085
+ "value_accuracy_mean": 1,
2086
+ "native_rung_rate": 1,
2087
+ "effective_usd_per_1k_conformant": 4.477050000000001
2088
+ },
2089
+ "qa.docs": {
2090
+ "samples": 15,
2091
+ "conform_rate": 1,
2092
+ "value_accuracy_mean": 1,
2093
+ "native_rung_rate": 1,
2094
+ "effective_usd_per_1k_conformant": 0.6075333333333335
2095
+ },
2096
+ "reasoning.numeric": {
2097
+ "samples": 15,
2098
+ "conform_rate": 1,
2099
+ "value_accuracy_mean": 1,
2100
+ "native_rung_rate": 1,
2101
+ "effective_usd_per_1k_conformant": 1.0529333333333335
2102
+ },
2103
+ "structure.strict": {
2104
+ "samples": 10,
2105
+ "conform_rate": 1,
2106
+ "value_accuracy_mean": 1,
2107
+ "native_rung_rate": 1,
2108
+ "effective_usd_per_1k_conformant": 0.84945
2109
+ },
2110
+ "summarize": {
2111
+ "samples": 15,
2112
+ "conform_rate": 1,
2113
+ "value_accuracy_mean": 1,
2114
+ "native_rung_rate": 1,
2115
+ "effective_usd_per_1k_conformant": 0.9375666666666667
2116
+ }
2117
+ },
2118
+ "discrepancies": []
2119
+ },
2120
+ {
2121
+ "model_key": "openai/gpt-5.2",
2122
+ "conform_rate": 1,
2123
+ "conform_ci95": [
2124
+ 0.9611351464605271,
2125
+ 1
2126
+ ],
2127
+ "value_accuracy_mean": 1,
2128
+ "native_rung_rate": 1,
2129
+ "hard_conform_rate": 1,
2130
+ "hard_value_accuracy_mean": 1,
2131
+ "hard_samples": 25,
2132
+ "hard_native_rung_rate": 1,
2133
+ "effective_usd_per_1k_conformant": 1.5723473684210518,
2134
+ "grounded_rate": 0.06666666666666667,
2135
+ "cache_hit_rate": 0.8,
2136
+ "samples": 95,
2137
+ "as_of": "2026-08-19T04:24:16.638Z",
2138
+ "families": [
2139
+ "demo-rig",
2140
+ "probe-suite"
2141
+ ],
2142
+ "fixture_counts": {
2143
+ "demo-rig": 20,
2144
+ "probe-suite": 75
2145
+ },
2146
+ "by_family": {
2147
+ "demo-rig": {
2148
+ "samples": 20,
2149
+ "conform_rate": 1,
2150
+ "value_accuracy_mean": 1,
2151
+ "native_rung_rate": 1,
2152
+ "effective_usd_per_1k_conformant": 1.3069000000000002
2153
+ },
2154
+ "probe-suite": {
2155
+ "samples": 75,
2156
+ "conform_rate": 1,
2157
+ "value_accuracy_mean": 1,
2158
+ "native_rung_rate": 1,
2159
+ "effective_usd_per_1k_conformant": 1.6431333333333333
2160
+ }
2161
+ },
2162
+ "by_subtype": {
2163
+ "classification": {
2164
+ "samples": 10,
2165
+ "conform_rate": 1,
2166
+ "value_accuracy_mean": 1,
2167
+ "native_rung_rate": 1,
2168
+ "effective_usd_per_1k_conformant": 0.85575
2169
+ },
2170
+ "extraction.tabular": {
2171
+ "samples": 15,
2172
+ "conform_rate": 1,
2173
+ "value_accuracy_mean": 1,
2174
+ "native_rung_rate": 1,
2175
+ "effective_usd_per_1k_conformant": 1.826416666666667
2176
+ },
2177
+ "generation.prose": {
2178
+ "samples": 15,
2179
+ "conform_rate": 1,
2180
+ "value_accuracy_mean": 1,
2181
+ "native_rung_rate": 1,
2182
+ "effective_usd_per_1k_conformant": 3.5374499999999998
2183
+ },
2184
+ "qa.docs": {
2185
+ "samples": 15,
2186
+ "conform_rate": 1,
2187
+ "value_accuracy_mean": 1,
2188
+ "native_rung_rate": 1,
2189
+ "effective_usd_per_1k_conformant": 1.1379666666666666
2190
+ },
2191
+ "reasoning.numeric": {
2192
+ "samples": 15,
2193
+ "conform_rate": 1,
2194
+ "value_accuracy_mean": 1,
2195
+ "native_rung_rate": 1,
2196
+ "effective_usd_per_1k_conformant": 0.6037500000000002
2197
+ },
2198
+ "structure.strict": {
2199
+ "samples": 10,
2200
+ "conform_rate": 1,
2201
+ "value_accuracy_mean": 1,
2202
+ "native_rung_rate": 1,
2203
+ "effective_usd_per_1k_conformant": 1.327375
2204
+ },
2205
+ "summarize": {
2206
+ "samples": 15,
2207
+ "conform_rate": 1,
2208
+ "value_accuracy_mean": 1,
2209
+ "native_rung_rate": 1,
2210
+ "effective_usd_per_1k_conformant": 1.3972000000000002
2211
+ }
2212
+ },
2213
+ "discrepancies": [
2214
+ {
2215
+ "kind": "declared_search_ungrounded",
2216
+ "message": "declared supports_web_search, but only 7% of probed answers carried citations (bar: 50%)"
2217
+ }
2218
+ ]
2219
+ },
2220
+ {
2221
+ "model_key": "deepinfra/moonshotai/Kimi-K2.7-Code",
2222
+ "conform_rate": 0.968421052631579,
2223
+ "conform_ci95": [
2224
+ 0.9112290135269181,
2225
+ 0.9892028605255756
2226
+ ],
2227
+ "value_accuracy_mean": 1,
2228
+ "native_rung_rate": 0,
2229
+ "hard_conform_rate": 0.88,
2230
+ "hard_value_accuracy_mean": 1,
2231
+ "hard_samples": 25,
2232
+ "hard_native_rung_rate": 0,
2233
+ "effective_usd_per_1k_conformant": 1.6180673913043473,
2234
+ "grounded_rate": 0.13333333333333333,
2235
+ "cache_hit_rate": 1,
2236
+ "samples": 95,
2237
+ "as_of": "2026-08-19T01:14:46.362Z",
2238
+ "families": [
2239
+ "demo-rig",
2240
+ "probe-suite"
2241
+ ],
2242
+ "fixture_counts": {
2243
+ "demo-rig": 20,
2244
+ "probe-suite": 75
2245
+ },
2246
+ "by_family": {
2247
+ "demo-rig": {
2248
+ "samples": 20,
2249
+ "conform_rate": 1,
2250
+ "value_accuracy_mean": 1,
2251
+ "native_rung_rate": 0,
2252
+ "effective_usd_per_1k_conformant": 1.1633099999999998
2253
+ },
2254
+ "probe-suite": {
2255
+ "samples": 75,
2256
+ "conform_rate": 0.96,
2257
+ "value_accuracy_mean": 1,
2258
+ "native_rung_rate": 0,
2259
+ "effective_usd_per_1k_conformant": 1.7443888888888885
2260
+ }
2261
+ },
2262
+ "by_subtype": {
2263
+ "classification": {
2264
+ "samples": 10,
2265
+ "conform_rate": 1,
2266
+ "value_accuracy_mean": 1,
2267
+ "native_rung_rate": 0,
2268
+ "effective_usd_per_1k_conformant": 0.7690799999999999
2269
+ },
2270
+ "extraction.tabular": {
2271
+ "samples": 15,
2272
+ "conform_rate": 1,
2273
+ "value_accuracy_mean": 1,
2274
+ "native_rung_rate": 0,
2275
+ "effective_usd_per_1k_conformant": 1.3425466666666666
2276
+ },
2277
+ "generation.prose": {
2278
+ "samples": 15,
2279
+ "conform_rate": 0.8,
2280
+ "value_accuracy_mean": 1,
2281
+ "native_rung_rate": 0,
2282
+ "effective_usd_per_1k_conformant": 5.729566666666668
2283
+ },
2284
+ "qa.docs": {
2285
+ "samples": 15,
2286
+ "conform_rate": 1,
2287
+ "value_accuracy_mean": 1,
2288
+ "native_rung_rate": 0,
2289
+ "effective_usd_per_1k_conformant": 0.8676800000000001
2290
+ },
2291
+ "reasoning.numeric": {
2292
+ "samples": 15,
2293
+ "conform_rate": 1,
2294
+ "value_accuracy_mean": 1,
2295
+ "native_rung_rate": 0,
2296
+ "effective_usd_per_1k_conformant": 0.6006666666666665
2297
+ },
2298
+ "structure.strict": {
2299
+ "samples": 10,
2300
+ "conform_rate": 1,
2301
+ "value_accuracy_mean": 1,
2302
+ "native_rung_rate": 0,
2303
+ "effective_usd_per_1k_conformant": 0.9989199999999999
2304
+ },
2305
+ "summarize": {
2306
+ "samples": 15,
2307
+ "conform_rate": 1,
2308
+ "value_accuracy_mean": 1,
2309
+ "native_rung_rate": 0,
2310
+ "effective_usd_per_1k_conformant": 1.350933333333333
2311
+ }
2312
+ },
2313
+ "discrepancies": [
2314
+ {
2315
+ "kind": "undeclared_schema_capable",
2316
+ "message": "no declared schema support, but probed conformance is 97% — capability exists undeclared"
2317
+ }
2318
+ ]
2319
+ },
2320
+ {
2321
+ "model_key": "anthropic/claude-haiku-4-5",
2322
+ "conform_rate": 1,
2323
+ "conform_ci95": [
2324
+ 0.9611351464605271,
2325
+ 1
2326
+ ],
2327
+ "value_accuracy_mean": 0.8631578947368421,
2328
+ "native_rung_rate": 1,
2329
+ "hard_conform_rate": 1,
2330
+ "hard_value_accuracy_mean": 0.48,
2331
+ "hard_samples": 25,
2332
+ "hard_native_rung_rate": 1,
2333
+ "effective_usd_per_1k_conformant": 1.786105263157895,
2334
+ "grounded_rate": 0.8666666666666667,
2335
+ "cache_hit_rate": 0,
2336
+ "samples": 95,
2337
+ "as_of": "2026-08-18T21:36:12.554Z",
2338
+ "families": [
2339
+ "demo-rig",
2340
+ "probe-suite"
2341
+ ],
2342
+ "fixture_counts": {
2343
+ "demo-rig": 20,
2344
+ "probe-suite": 75
2345
+ },
2346
+ "by_family": {
2347
+ "demo-rig": {
2348
+ "samples": 20,
2349
+ "conform_rate": 1,
2350
+ "value_accuracy_mean": 1,
2351
+ "native_rung_rate": 1,
2352
+ "effective_usd_per_1k_conformant": 1.7394999999999998
2353
+ },
2354
+ "probe-suite": {
2355
+ "samples": 75,
2356
+ "conform_rate": 1,
2357
+ "value_accuracy_mean": 0.8266666666666667,
2358
+ "native_rung_rate": 1,
2359
+ "effective_usd_per_1k_conformant": 1.7985333333333335
2360
+ }
2361
+ },
2362
+ "by_subtype": {
2363
+ "classification": {
2364
+ "samples": 10,
2365
+ "conform_rate": 1,
2366
+ "value_accuracy_mean": 1,
2367
+ "native_rung_rate": 1,
2368
+ "effective_usd_per_1k_conformant": 1.345
2369
+ },
2370
+ "extraction.tabular": {
2371
+ "samples": 15,
2372
+ "conform_rate": 1,
2373
+ "value_accuracy_mean": 1,
2374
+ "native_rung_rate": 1,
2375
+ "effective_usd_per_1k_conformant": 2.271666666666666
2376
+ },
2377
+ "generation.prose": {
2378
+ "samples": 15,
2379
+ "conform_rate": 1,
2380
+ "value_accuracy_mean": 1,
2381
+ "native_rung_rate": 1,
2382
+ "effective_usd_per_1k_conformant": 2.6576666666666666
2383
+ },
2384
+ "qa.docs": {
2385
+ "samples": 15,
2386
+ "conform_rate": 1,
2387
+ "value_accuracy_mean": 1,
2388
+ "native_rung_rate": 1,
2389
+ "effective_usd_per_1k_conformant": 1.502
2390
+ },
2391
+ "reasoning.numeric": {
2392
+ "samples": 15,
2393
+ "conform_rate": 1,
2394
+ "value_accuracy_mean": 0.13333333333333333,
2395
+ "native_rung_rate": 1,
2396
+ "effective_usd_per_1k_conformant": 1.1583333333333334
2397
+ },
2398
+ "structure.strict": {
2399
+ "samples": 10,
2400
+ "conform_rate": 1,
2401
+ "value_accuracy_mean": 1,
2402
+ "native_rung_rate": 1,
2403
+ "effective_usd_per_1k_conformant": 1.7515
2404
+ },
2405
+ "summarize": {
2406
+ "samples": 15,
2407
+ "conform_rate": 1,
2408
+ "value_accuracy_mean": 1,
2409
+ "native_rung_rate": 1,
2410
+ "effective_usd_per_1k_conformant": 1.6580000000000004
2411
+ }
2412
+ },
2413
+ "discrepancies": [
2414
+ {
2415
+ "kind": "citations_without_declared_search",
2416
+ "message": "no declared web-search support, but 87% of probed answers cited URLs — verify whether these are real or fabricated sources"
2417
+ },
2418
+ {
2419
+ "kind": "declared_caching_unrealized",
2420
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
2421
+ }
2422
+ ]
2423
+ },
2424
+ {
2425
+ "model_key": "openai/gpt-5.4",
2426
+ "conform_rate": 1,
2427
+ "conform_ci95": [
2428
+ 0.9611351464605271,
2429
+ 1
2430
+ ],
2431
+ "value_accuracy_mean": 1,
2432
+ "native_rung_rate": 1,
2433
+ "hard_conform_rate": 1,
2434
+ "hard_value_accuracy_mean": 1,
2435
+ "hard_samples": 25,
2436
+ "hard_native_rung_rate": 1,
2437
+ "effective_usd_per_1k_conformant": 1.8063157894736839,
2438
+ "grounded_rate": 0.4,
2439
+ "cache_hit_rate": 1,
2440
+ "samples": 95,
2441
+ "as_of": "2026-08-19T04:27:15.709Z",
2442
+ "families": [
2443
+ "demo-rig",
2444
+ "probe-suite"
2445
+ ],
2446
+ "fixture_counts": {
2447
+ "demo-rig": 20,
2448
+ "probe-suite": 75
2449
+ },
2450
+ "by_family": {
2451
+ "demo-rig": {
2452
+ "samples": 20,
2453
+ "conform_rate": 1,
2454
+ "value_accuracy_mean": 1,
2455
+ "native_rung_rate": 1,
2456
+ "effective_usd_per_1k_conformant": 1.5160000000000005
2457
+ },
2458
+ "probe-suite": {
2459
+ "samples": 75,
2460
+ "conform_rate": 1,
2461
+ "value_accuracy_mean": 1,
2462
+ "native_rung_rate": 1,
2463
+ "effective_usd_per_1k_conformant": 1.8837333333333328
2464
+ }
2465
+ },
2466
+ "by_subtype": {
2467
+ "classification": {
2468
+ "samples": 10,
2469
+ "conform_rate": 1,
2470
+ "value_accuracy_mean": 1,
2471
+ "native_rung_rate": 1,
2472
+ "effective_usd_per_1k_conformant": 1.035
2473
+ },
2474
+ "extraction.tabular": {
2475
+ "samples": 15,
2476
+ "conform_rate": 1,
2477
+ "value_accuracy_mean": 1,
2478
+ "native_rung_rate": 1,
2479
+ "effective_usd_per_1k_conformant": 2.1474999999999995
2480
+ },
2481
+ "generation.prose": {
2482
+ "samples": 15,
2483
+ "conform_rate": 1,
2484
+ "value_accuracy_mean": 1,
2485
+ "native_rung_rate": 1,
2486
+ "effective_usd_per_1k_conformant": 3.899833333333333
2487
+ },
2488
+ "qa.docs": {
2489
+ "samples": 15,
2490
+ "conform_rate": 1,
2491
+ "value_accuracy_mean": 1,
2492
+ "native_rung_rate": 1,
2493
+ "effective_usd_per_1k_conformant": 1.206666666666667
2494
+ },
2495
+ "reasoning.numeric": {
2496
+ "samples": 15,
2497
+ "conform_rate": 1,
2498
+ "value_accuracy_mean": 1,
2499
+ "native_rung_rate": 1,
2500
+ "effective_usd_per_1k_conformant": 0.7425000000000002
2501
+ },
2502
+ "structure.strict": {
2503
+ "samples": 10,
2504
+ "conform_rate": 1,
2505
+ "value_accuracy_mean": 1,
2506
+ "native_rung_rate": 1,
2507
+ "effective_usd_per_1k_conformant": 1.55125
2508
+ },
2509
+ "summarize": {
2510
+ "samples": 15,
2511
+ "conform_rate": 1,
2512
+ "value_accuracy_mean": 1,
2513
+ "native_rung_rate": 1,
2514
+ "effective_usd_per_1k_conformant": 1.7193333333333338
2515
+ }
2516
+ },
2517
+ "discrepancies": []
2518
+ },
2519
+ {
2520
+ "model_key": "deepinfra/zai-org/GLM-5.2",
2521
+ "conform_rate": 0.9789473684210527,
2522
+ "conform_ci95": [
2523
+ 0.9264588050529105,
2524
+ 0.9942074931355945
2525
+ ],
2526
+ "value_accuracy_mean": 1,
2527
+ "native_rung_rate": 0,
2528
+ "hard_conform_rate": 0.96,
2529
+ "hard_value_accuracy_mean": 1,
2530
+ "hard_samples": 25,
2531
+ "hard_native_rung_rate": 0,
2532
+ "effective_usd_per_1k_conformant": 1.8604080645161287,
2533
+ "grounded_rate": 0.6,
2534
+ "cache_hit_rate": 1,
2535
+ "samples": 95,
2536
+ "as_of": "2026-08-19T02:14:34.948Z",
2537
+ "families": [
2538
+ "demo-rig",
2539
+ "probe-suite"
2540
+ ],
2541
+ "fixture_counts": {
2542
+ "demo-rig": 20,
2543
+ "probe-suite": 75
2544
+ },
2545
+ "by_family": {
2546
+ "demo-rig": {
2547
+ "samples": 20,
2548
+ "conform_rate": 1,
2549
+ "value_accuracy_mean": 1,
2550
+ "native_rung_rate": 0,
2551
+ "effective_usd_per_1k_conformant": 1.3809224999999998
2552
+ },
2553
+ "probe-suite": {
2554
+ "samples": 75,
2555
+ "conform_rate": 0.9733333333333334,
2556
+ "value_accuracy_mean": 1,
2557
+ "native_rung_rate": 0,
2558
+ "effective_usd_per_1k_conformant": 1.9917739726027388
2559
+ }
2560
+ },
2561
+ "by_subtype": {
2562
+ "classification": {
2563
+ "samples": 10,
2564
+ "conform_rate": 1,
2565
+ "value_accuracy_mean": 1,
2566
+ "native_rung_rate": 0,
2567
+ "effective_usd_per_1k_conformant": 1.2350999999999999
2568
+ },
2569
+ "extraction.tabular": {
2570
+ "samples": 15,
2571
+ "conform_rate": 0.9333333333333333,
2572
+ "value_accuracy_mean": 1,
2573
+ "native_rung_rate": 0,
2574
+ "effective_usd_per_1k_conformant": 1.9478999999999997
2575
+ },
2576
+ "generation.prose": {
2577
+ "samples": 15,
2578
+ "conform_rate": 0.9333333333333333,
2579
+ "value_accuracy_mean": 1,
2580
+ "native_rung_rate": 0,
2581
+ "effective_usd_per_1k_conformant": 5.648367857142857
2582
+ },
2583
+ "qa.docs": {
2584
+ "samples": 15,
2585
+ "conform_rate": 1,
2586
+ "value_accuracy_mean": 1,
2587
+ "native_rung_rate": 0,
2588
+ "effective_usd_per_1k_conformant": 0.9080099999999998
2589
+ },
2590
+ "reasoning.numeric": {
2591
+ "samples": 15,
2592
+ "conform_rate": 1,
2593
+ "value_accuracy_mean": 1,
2594
+ "native_rung_rate": 0,
2595
+ "effective_usd_per_1k_conformant": 0.5982799999999999
2596
+ },
2597
+ "structure.strict": {
2598
+ "samples": 10,
2599
+ "conform_rate": 1,
2600
+ "value_accuracy_mean": 1,
2601
+ "native_rung_rate": 0,
2602
+ "effective_usd_per_1k_conformant": 1.2784049999999998
2603
+ },
2604
+ "summarize": {
2605
+ "samples": 15,
2606
+ "conform_rate": 1,
2607
+ "value_accuracy_mean": 1,
2608
+ "native_rung_rate": 0,
2609
+ "effective_usd_per_1k_conformant": 1.2627199999999998
2610
+ }
2611
+ },
2612
+ "discrepancies": [
2613
+ {
2614
+ "kind": "undeclared_schema_capable",
2615
+ "message": "no declared schema support, but probed conformance is 98% — capability exists undeclared"
2616
+ },
2617
+ {
2618
+ "kind": "citations_without_declared_search",
2619
+ "message": "no declared web-search support, but 60% of probed answers cited URLs — verify whether these are real or fabricated sources"
2620
+ }
2621
+ ]
2622
+ },
2623
+ {
2624
+ "model_key": "gemini/gemini-3-flash-preview",
2625
+ "conform_rate": 0.9789473684210527,
2626
+ "conform_ci95": [
2627
+ 0.9264588050529105,
2628
+ 0.9942074931355945
2629
+ ],
2630
+ "value_accuracy_mean": 1,
2631
+ "native_rung_rate": 1,
2632
+ "hard_conform_rate": 0.92,
2633
+ "hard_value_accuracy_mean": 1,
2634
+ "hard_samples": 25,
2635
+ "hard_native_rung_rate": 1,
2636
+ "effective_usd_per_1k_conformant": 2.636876344086021,
2637
+ "grounded_rate": 1,
2638
+ "cache_hit_rate": 0,
2639
+ "samples": 95,
2640
+ "as_of": "2026-08-19T03:46:20.459Z",
2641
+ "families": [
2642
+ "demo-rig",
2643
+ "probe-suite"
2644
+ ],
2645
+ "fixture_counts": {
2646
+ "demo-rig": 20,
2647
+ "probe-suite": 75
2648
+ },
2649
+ "by_family": {
2650
+ "demo-rig": {
2651
+ "samples": 20,
2652
+ "conform_rate": 1,
2653
+ "value_accuracy_mean": 1,
2654
+ "native_rung_rate": 1,
2655
+ "effective_usd_per_1k_conformant": 2.181625
2656
+ },
2657
+ "probe-suite": {
2658
+ "samples": 75,
2659
+ "conform_rate": 0.9733333333333334,
2660
+ "value_accuracy_mean": 1,
2661
+ "native_rung_rate": 1,
2662
+ "effective_usd_per_1k_conformant": 2.761602739726027
2663
+ }
2664
+ },
2665
+ "by_subtype": {
2666
+ "classification": {
2667
+ "samples": 10,
2668
+ "conform_rate": 1,
2669
+ "value_accuracy_mean": 1,
2670
+ "native_rung_rate": 1,
2671
+ "effective_usd_per_1k_conformant": 1.2856499999999997
2672
+ },
2673
+ "extraction.tabular": {
2674
+ "samples": 15,
2675
+ "conform_rate": 1,
2676
+ "value_accuracy_mean": 1,
2677
+ "native_rung_rate": 1,
2678
+ "effective_usd_per_1k_conformant": 2.5542000000000002
2679
+ },
2680
+ "generation.prose": {
2681
+ "samples": 15,
2682
+ "conform_rate": 0.8666666666666667,
2683
+ "value_accuracy_mean": 1,
2684
+ "native_rung_rate": 1,
2685
+ "effective_usd_per_1k_conformant": 7.10673076923077
2686
+ },
2687
+ "qa.docs": {
2688
+ "samples": 15,
2689
+ "conform_rate": 1,
2690
+ "value_accuracy_mean": 1,
2691
+ "native_rung_rate": 1,
2692
+ "effective_usd_per_1k_conformant": 1.5518333333333332
2693
+ },
2694
+ "reasoning.numeric": {
2695
+ "samples": 15,
2696
+ "conform_rate": 1,
2697
+ "value_accuracy_mean": 1,
2698
+ "native_rung_rate": 1,
2699
+ "effective_usd_per_1k_conformant": 1.5218000000000003
2700
+ },
2701
+ "structure.strict": {
2702
+ "samples": 10,
2703
+ "conform_rate": 1,
2704
+ "value_accuracy_mean": 1,
2705
+ "native_rung_rate": 1,
2706
+ "effective_usd_per_1k_conformant": 2.03695
2707
+ },
2708
+ "summarize": {
2709
+ "samples": 15,
2710
+ "conform_rate": 1,
2711
+ "value_accuracy_mean": 1,
2712
+ "native_rung_rate": 1,
2713
+ "effective_usd_per_1k_conformant": 2.346566666666667
2714
+ }
2715
+ },
2716
+ "discrepancies": [
2717
+ {
2718
+ "kind": "declared_caching_unrealized",
2719
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
2720
+ }
2721
+ ]
2722
+ },
2723
+ {
2724
+ "model_key": "fireworks/glm-5p2",
2725
+ "conform_rate": 1,
2726
+ "conform_ci95": [
2727
+ 0.9611351464605271,
2728
+ 1
2729
+ ],
2730
+ "value_accuracy_mean": 1,
2731
+ "native_rung_rate": 0,
2732
+ "hard_conform_rate": 1,
2733
+ "hard_value_accuracy_mean": 1,
2734
+ "hard_samples": 25,
2735
+ "hard_native_rung_rate": 0,
2736
+ "effective_usd_per_1k_conformant": 2.76312947368421,
2737
+ "grounded_rate": 0.4666666666666667,
2738
+ "cache_hit_rate": 1,
2739
+ "samples": 95,
2740
+ "as_of": "2026-08-19T03:09:25.522Z",
2741
+ "families": [
2742
+ "demo-rig",
2743
+ "probe-suite"
2744
+ ],
2745
+ "fixture_counts": {
2746
+ "demo-rig": 20,
2747
+ "probe-suite": 75
2748
+ },
2749
+ "by_family": {
2750
+ "demo-rig": {
2751
+ "samples": 20,
2752
+ "conform_rate": 1,
2753
+ "value_accuracy_mean": 1,
2754
+ "native_rung_rate": 0,
2755
+ "effective_usd_per_1k_conformant": 2.2178340000000003
2756
+ },
2757
+ "probe-suite": {
2758
+ "samples": 75,
2759
+ "conform_rate": 1,
2760
+ "value_accuracy_mean": 1,
2761
+ "native_rung_rate": 0,
2762
+ "effective_usd_per_1k_conformant": 2.9085416
2763
+ }
2764
+ },
2765
+ "by_subtype": {
2766
+ "classification": {
2767
+ "samples": 10,
2768
+ "conform_rate": 1,
2769
+ "value_accuracy_mean": 1,
2770
+ "native_rung_rate": 0,
2771
+ "effective_usd_per_1k_conformant": 1.6092919999999995
2772
+ },
2773
+ "extraction.tabular": {
2774
+ "samples": 15,
2775
+ "conform_rate": 1,
2776
+ "value_accuracy_mean": 1,
2777
+ "native_rung_rate": 0,
2778
+ "effective_usd_per_1k_conformant": 3.150113333333334
2779
+ },
2780
+ "generation.prose": {
2781
+ "samples": 15,
2782
+ "conform_rate": 1,
2783
+ "value_accuracy_mean": 1,
2784
+ "native_rung_rate": 0,
2785
+ "effective_usd_per_1k_conformant": 7.958582666666667
2786
+ },
2787
+ "qa.docs": {
2788
+ "samples": 15,
2789
+ "conform_rate": 1,
2790
+ "value_accuracy_mean": 1,
2791
+ "native_rung_rate": 0,
2792
+ "effective_usd_per_1k_conformant": 1.176684
2793
+ },
2794
+ "reasoning.numeric": {
2795
+ "samples": 15,
2796
+ "conform_rate": 1,
2797
+ "value_accuracy_mean": 1,
2798
+ "native_rung_rate": 0,
2799
+ "effective_usd_per_1k_conformant": 0.8494066666666666
2800
+ },
2801
+ "structure.strict": {
2802
+ "samples": 10,
2803
+ "conform_rate": 1,
2804
+ "value_accuracy_mean": 1,
2805
+ "native_rung_rate": 0,
2806
+ "effective_usd_per_1k_conformant": 1.7351599999999998
2807
+ },
2808
+ "summarize": {
2809
+ "samples": 15,
2810
+ "conform_rate": 1,
2811
+ "value_accuracy_mean": 1,
2812
+ "native_rung_rate": 0,
2813
+ "effective_usd_per_1k_conformant": 2.135398666666667
2814
+ }
2815
+ },
2816
+ "discrepancies": [
2817
+ {
2818
+ "kind": "schema_served_via_json_mode",
2819
+ "message": "declared supports_response_schema, but every probed sample was served via json_mode coaching — no native strict enforcement exists on this serving path (the json_object-vs-schema gap)"
2820
+ }
2821
+ ]
2822
+ },
2823
+ {
2824
+ "model_key": "deepinfra/moonshotai/Kimi-K2.6",
2825
+ "conform_rate": 0.9894736842105263,
2826
+ "conform_ci95": [
2827
+ 0.9427612985868637,
2828
+ 0.9981394237376521
2829
+ ],
2830
+ "value_accuracy_mean": 1,
2831
+ "native_rung_rate": 0,
2832
+ "hard_conform_rate": 0.96,
2833
+ "hard_value_accuracy_mean": 1,
2834
+ "hard_samples": 25,
2835
+ "hard_native_rung_rate": 0,
2836
+ "effective_usd_per_1k_conformant": 3.6464680851063807,
2837
+ "grounded_rate": 0,
2838
+ "cache_hit_rate": 0.8,
2839
+ "samples": 95,
2840
+ "as_of": "2026-08-19T00:51:51.203Z",
2841
+ "families": [
2842
+ "demo-rig",
2843
+ "probe-suite"
2844
+ ],
2845
+ "fixture_counts": {
2846
+ "demo-rig": 20,
2847
+ "probe-suite": 75
2848
+ },
2849
+ "by_family": {
2850
+ "demo-rig": {
2851
+ "samples": 20,
2852
+ "conform_rate": 1,
2853
+ "value_accuracy_mean": 1,
2854
+ "native_rung_rate": 0,
2855
+ "effective_usd_per_1k_conformant": 3.0366874999999993
2856
+ },
2857
+ "probe-suite": {
2858
+ "samples": 75,
2859
+ "conform_rate": 0.9866666666666667,
2860
+ "value_accuracy_mean": 1,
2861
+ "native_rung_rate": 0,
2862
+ "effective_usd_per_1k_conformant": 3.8112736486486476
2863
+ }
2864
+ },
2865
+ "by_subtype": {
2866
+ "classification": {
2867
+ "samples": 10,
2868
+ "conform_rate": 1,
2869
+ "value_accuracy_mean": 1,
2870
+ "native_rung_rate": 0,
2871
+ "effective_usd_per_1k_conformant": 1.907
2872
+ },
2873
+ "extraction.tabular": {
2874
+ "samples": 15,
2875
+ "conform_rate": 1,
2876
+ "value_accuracy_mean": 1,
2877
+ "native_rung_rate": 0,
2878
+ "effective_usd_per_1k_conformant": 3.0454
2879
+ },
2880
+ "generation.prose": {
2881
+ "samples": 15,
2882
+ "conform_rate": 0.9333333333333333,
2883
+ "value_accuracy_mean": 1,
2884
+ "native_rung_rate": 0,
2885
+ "effective_usd_per_1k_conformant": 9.72707142857143
2886
+ },
2887
+ "qa.docs": {
2888
+ "samples": 15,
2889
+ "conform_rate": 1,
2890
+ "value_accuracy_mean": 1,
2891
+ "native_rung_rate": 0,
2892
+ "effective_usd_per_1k_conformant": 2.6239166666666662
2893
+ },
2894
+ "reasoning.numeric": {
2895
+ "samples": 15,
2896
+ "conform_rate": 1,
2897
+ "value_accuracy_mean": 1,
2898
+ "native_rung_rate": 0,
2899
+ "effective_usd_per_1k_conformant": 1.6947
2900
+ },
2901
+ "structure.strict": {
2902
+ "samples": 10,
2903
+ "conform_rate": 1,
2904
+ "value_accuracy_mean": 1,
2905
+ "native_rung_rate": 0,
2906
+ "effective_usd_per_1k_conformant": 2.212675
2907
+ },
2908
+ "summarize": {
2909
+ "samples": 15,
2910
+ "conform_rate": 1,
2911
+ "value_accuracy_mean": 1,
2912
+ "native_rung_rate": 0,
2913
+ "effective_usd_per_1k_conformant": 3.662133333333333
2914
+ }
2915
+ },
2916
+ "discrepancies": [
2917
+ {
2918
+ "kind": "undeclared_schema_capable",
2919
+ "message": "no declared schema support, but probed conformance is 99% — capability exists undeclared"
2920
+ }
2921
+ ]
2922
+ },
2923
+ {
2924
+ "model_key": "anthropic/claude-sonnet-5",
2925
+ "conform_rate": 0.9157894736842105,
2926
+ "conform_ci95": [
2927
+ 0.8425478687567327,
2928
+ 0.9567118846157054
2929
+ ],
2930
+ "value_accuracy_mean": 0.9368421052631579,
2931
+ "native_rung_rate": 1,
2932
+ "hard_conform_rate": 0.68,
2933
+ "hard_value_accuracy_mean": 0.84,
2934
+ "hard_samples": 25,
2935
+ "hard_native_rung_rate": 1,
2936
+ "effective_usd_per_1k_conformant": 4.3616091954023,
2937
+ "grounded_rate": 0.9333333333333333,
2938
+ "cache_hit_rate": 0,
2939
+ "samples": 95,
2940
+ "as_of": "2026-08-18T21:53:43.179Z",
2941
+ "families": [
2942
+ "demo-rig",
2943
+ "probe-suite"
2944
+ ],
2945
+ "fixture_counts": {
2946
+ "demo-rig": 20,
2947
+ "probe-suite": 75
2948
+ },
2949
+ "by_family": {
2950
+ "demo-rig": {
2951
+ "samples": 20,
2952
+ "conform_rate": 1,
2953
+ "value_accuracy_mean": 1,
2954
+ "native_rung_rate": 1,
2955
+ "effective_usd_per_1k_conformant": 3.7029999999999994
2956
+ },
2957
+ "probe-suite": {
2958
+ "samples": 75,
2959
+ "conform_rate": 0.8933333333333333,
2960
+ "value_accuracy_mean": 0.92,
2961
+ "native_rung_rate": 1,
2962
+ "effective_usd_per_1k_conformant": 4.558208955223882
2963
+ }
2964
+ },
2965
+ "by_subtype": {
2966
+ "classification": {
2967
+ "samples": 10,
2968
+ "conform_rate": 1,
2969
+ "value_accuracy_mean": 1,
2970
+ "native_rung_rate": 1,
2971
+ "effective_usd_per_1k_conformant": 2.7610000000000006
2972
+ },
2973
+ "extraction.tabular": {
2974
+ "samples": 15,
2975
+ "conform_rate": 1,
2976
+ "value_accuracy_mean": 1,
2977
+ "native_rung_rate": 1,
2978
+ "effective_usd_per_1k_conformant": 4.957333333333332
2979
+ },
2980
+ "generation.prose": {
2981
+ "samples": 15,
2982
+ "conform_rate": 0.4666666666666667,
2983
+ "value_accuracy_mean": 0.6,
2984
+ "native_rung_rate": 1,
2985
+ "effective_usd_per_1k_conformant": 15.292857142857143
2986
+ },
2987
+ "qa.docs": {
2988
+ "samples": 15,
2989
+ "conform_rate": 1,
2990
+ "value_accuracy_mean": 1,
2991
+ "native_rung_rate": 1,
2992
+ "effective_usd_per_1k_conformant": 3.0846666666666662
2993
+ },
2994
+ "reasoning.numeric": {
2995
+ "samples": 15,
2996
+ "conform_rate": 1,
2997
+ "value_accuracy_mean": 1,
2998
+ "native_rung_rate": 1,
2999
+ "effective_usd_per_1k_conformant": 2.218
3000
+ },
3001
+ "structure.strict": {
3002
+ "samples": 10,
3003
+ "conform_rate": 1,
3004
+ "value_accuracy_mean": 1,
3005
+ "native_rung_rate": 1,
3006
+ "effective_usd_per_1k_conformant": 3.6189999999999998
3007
+ },
3008
+ "summarize": {
3009
+ "samples": 15,
3010
+ "conform_rate": 1,
3011
+ "value_accuracy_mean": 1,
3012
+ "native_rung_rate": 1,
3013
+ "effective_usd_per_1k_conformant": 3.6473333333333335
3014
+ }
3015
+ },
3016
+ "discrepancies": [
3017
+ {
3018
+ "kind": "citations_without_declared_search",
3019
+ "message": "no declared web-search support, but 93% of probed answers cited URLs — verify whether these are real or fabricated sources"
3020
+ },
3021
+ {
3022
+ "kind": "declared_caching_unrealized",
3023
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
3024
+ }
3025
+ ]
3026
+ },
3027
+ {
3028
+ "model_key": "openai/gpt-5.5",
3029
+ "conform_rate": 1,
3030
+ "conform_ci95": [
3031
+ 0.9611351464605271,
3032
+ 1
3033
+ ],
3034
+ "value_accuracy_mean": 1,
3035
+ "native_rung_rate": 1,
3036
+ "hard_conform_rate": 1,
3037
+ "hard_value_accuracy_mean": 1,
3038
+ "hard_samples": 25,
3039
+ "hard_native_rung_rate": 1,
3040
+ "effective_usd_per_1k_conformant": 4.684105263157893,
3041
+ "grounded_rate": 0,
3042
+ "cache_hit_rate": 1,
3043
+ "samples": 95,
3044
+ "as_of": "2026-08-19T04:35:36.550Z",
3045
+ "families": [
3046
+ "demo-rig",
3047
+ "probe-suite"
3048
+ ],
3049
+ "fixture_counts": {
3050
+ "demo-rig": 20,
3051
+ "probe-suite": 75
3052
+ },
3053
+ "by_family": {
3054
+ "demo-rig": {
3055
+ "samples": 20,
3056
+ "conform_rate": 1,
3057
+ "value_accuracy_mean": 1,
3058
+ "native_rung_rate": 1,
3059
+ "effective_usd_per_1k_conformant": 3.5495000000000005
3060
+ },
3061
+ "probe-suite": {
3062
+ "samples": 75,
3063
+ "conform_rate": 1,
3064
+ "value_accuracy_mean": 1,
3065
+ "native_rung_rate": 1,
3066
+ "effective_usd_per_1k_conformant": 4.986666666666666
3067
+ }
3068
+ },
3069
+ "by_subtype": {
3070
+ "classification": {
3071
+ "samples": 10,
3072
+ "conform_rate": 1,
3073
+ "value_accuracy_mean": 1,
3074
+ "native_rung_rate": 1,
3075
+ "effective_usd_per_1k_conformant": 2.442
3076
+ },
3077
+ "extraction.tabular": {
3078
+ "samples": 15,
3079
+ "conform_rate": 1,
3080
+ "value_accuracy_mean": 1,
3081
+ "native_rung_rate": 1,
3082
+ "effective_usd_per_1k_conformant": 4.735
3083
+ },
3084
+ "generation.prose": {
3085
+ "samples": 15,
3086
+ "conform_rate": 1,
3087
+ "value_accuracy_mean": 1,
3088
+ "native_rung_rate": 1,
3089
+ "effective_usd_per_1k_conformant": 10.611666666666665
3090
+ },
3091
+ "qa.docs": {
3092
+ "samples": 15,
3093
+ "conform_rate": 1,
3094
+ "value_accuracy_mean": 1,
3095
+ "native_rung_rate": 1,
3096
+ "effective_usd_per_1k_conformant": 2.8713333333333337
3097
+ },
3098
+ "reasoning.numeric": {
3099
+ "samples": 15,
3100
+ "conform_rate": 1,
3101
+ "value_accuracy_mean": 1,
3102
+ "native_rung_rate": 1,
3103
+ "effective_usd_per_1k_conformant": 3.0129999999999995
3104
+ },
3105
+ "structure.strict": {
3106
+ "samples": 10,
3107
+ "conform_rate": 1,
3108
+ "value_accuracy_mean": 1,
3109
+ "native_rung_rate": 1,
3110
+ "effective_usd_per_1k_conformant": 3.1535
3111
+ },
3112
+ "summarize": {
3113
+ "samples": 15,
3114
+ "conform_rate": 1,
3115
+ "value_accuracy_mean": 1,
3116
+ "native_rung_rate": 1,
3117
+ "effective_usd_per_1k_conformant": 4.704666666666667
3118
+ }
3119
+ },
3120
+ "discrepancies": [
3121
+ {
3122
+ "kind": "declared_search_ungrounded",
3123
+ "message": "declared supports_web_search, but only 0% of probed answers carried citations (bar: 50%)"
3124
+ }
3125
+ ]
3126
+ },
3127
+ {
3128
+ "model_key": "fireworks/kimi-k3",
3129
+ "conform_rate": 0.9789473684210527,
3130
+ "conform_ci95": [
3131
+ 0.9264588050529105,
3132
+ 0.9942074931355945
3133
+ ],
3134
+ "value_accuracy_mean": 1,
3135
+ "native_rung_rate": 0,
3136
+ "hard_conform_rate": 0.92,
3137
+ "hard_value_accuracy_mean": 1,
3138
+ "hard_samples": 25,
3139
+ "hard_native_rung_rate": 0,
3140
+ "effective_usd_per_1k_conformant": 5.09848064516129,
3141
+ "grounded_rate": 0.6,
3142
+ "cache_hit_rate": 1,
3143
+ "samples": 95,
3144
+ "as_of": "2026-08-19T03:31:32.971Z",
3145
+ "families": [
3146
+ "demo-rig",
3147
+ "probe-suite"
3148
+ ],
3149
+ "fixture_counts": {
3150
+ "demo-rig": 20,
3151
+ "probe-suite": 75
3152
+ },
3153
+ "by_family": {
3154
+ "demo-rig": {
3155
+ "samples": 20,
3156
+ "conform_rate": 1,
3157
+ "value_accuracy_mean": 1,
3158
+ "native_rung_rate": 0,
3159
+ "effective_usd_per_1k_conformant": 3.37659
3160
+ },
3161
+ "probe-suite": {
3162
+ "samples": 75,
3163
+ "conform_rate": 0.9733333333333334,
3164
+ "value_accuracy_mean": 1,
3165
+ "native_rung_rate": 0,
3166
+ "effective_usd_per_1k_conformant": 5.570231506849315
3167
+ }
3168
+ },
3169
+ "by_subtype": {
3170
+ "classification": {
3171
+ "samples": 10,
3172
+ "conform_rate": 1,
3173
+ "value_accuracy_mean": 1,
3174
+ "native_rung_rate": 0,
3175
+ "effective_usd_per_1k_conformant": 2.01006
3176
+ },
3177
+ "extraction.tabular": {
3178
+ "samples": 15,
3179
+ "conform_rate": 1,
3180
+ "value_accuracy_mean": 1,
3181
+ "native_rung_rate": 0,
3182
+ "effective_usd_per_1k_conformant": 4.185360000000001
3183
+ },
3184
+ "generation.prose": {
3185
+ "samples": 15,
3186
+ "conform_rate": 0.8666666666666667,
3187
+ "value_accuracy_mean": 1,
3188
+ "native_rung_rate": 0,
3189
+ "effective_usd_per_1k_conformant": 18.674976923076926
3190
+ },
3191
+ "qa.docs": {
3192
+ "samples": 15,
3193
+ "conform_rate": 1,
3194
+ "value_accuracy_mean": 1,
3195
+ "native_rung_rate": 0,
3196
+ "effective_usd_per_1k_conformant": 2.31304
3197
+ },
3198
+ "reasoning.numeric": {
3199
+ "samples": 15,
3200
+ "conform_rate": 1,
3201
+ "value_accuracy_mean": 1,
3202
+ "native_rung_rate": 0,
3203
+ "effective_usd_per_1k_conformant": 2.1346000000000003
3204
+ },
3205
+ "structure.strict": {
3206
+ "samples": 10,
3207
+ "conform_rate": 1,
3208
+ "value_accuracy_mean": 1,
3209
+ "native_rung_rate": 0,
3210
+ "effective_usd_per_1k_conformant": 2.9311800000000003
3211
+ },
3212
+ "summarize": {
3213
+ "samples": 15,
3214
+ "conform_rate": 1,
3215
+ "value_accuracy_mean": 1,
3216
+ "native_rung_rate": 0,
3217
+ "effective_usd_per_1k_conformant": 3.49844
3218
+ }
3219
+ },
3220
+ "discrepancies": [
3221
+ {
3222
+ "kind": "undeclared_schema_capable",
3223
+ "message": "no declared schema support, but probed conformance is 98% — capability exists undeclared"
3224
+ },
3225
+ {
3226
+ "kind": "citations_without_declared_search",
3227
+ "message": "no declared web-search support, but 60% of probed answers cited URLs — verify whether these are real or fabricated sources"
3228
+ }
3229
+ ]
3230
+ },
3231
+ {
3232
+ "model_key": "anthropic/claude-opus-5",
3233
+ "conform_rate": 0.9894736842105263,
3234
+ "conform_ci95": [
3235
+ 0.9427612985868637,
3236
+ 0.9981394237376521
3237
+ ],
3238
+ "value_accuracy_mean": 0.973404255319149,
3239
+ "native_rung_rate": 1,
3240
+ "hard_conform_rate": 1,
3241
+ "hard_value_accuracy_mean": 1,
3242
+ "hard_samples": 25,
3243
+ "hard_native_rung_rate": 1,
3244
+ "effective_usd_per_1k_conformant": 9.705106382978727,
3245
+ "grounded_rate": 0.9333333333333333,
3246
+ "cache_hit_rate": null,
3247
+ "samples": 95,
3248
+ "as_of": "2026-08-18T21:46:28.818Z",
3249
+ "families": [
3250
+ "demo-rig",
3251
+ "probe-suite"
3252
+ ],
3253
+ "fixture_counts": {
3254
+ "demo-rig": 20,
3255
+ "probe-suite": 75
3256
+ },
3257
+ "by_family": {
3258
+ "demo-rig": {
3259
+ "samples": 20,
3260
+ "conform_rate": 1,
3261
+ "value_accuracy_mean": 1,
3262
+ "native_rung_rate": 1,
3263
+ "effective_usd_per_1k_conformant": 9.17625
3264
+ },
3265
+ "probe-suite": {
3266
+ "samples": 75,
3267
+ "conform_rate": 0.9866666666666667,
3268
+ "value_accuracy_mean": 0.9662162162162162,
3269
+ "native_rung_rate": 1,
3270
+ "effective_usd_per_1k_conformant": 9.848040540540543
3271
+ }
3272
+ },
3273
+ "by_subtype": {
3274
+ "classification": {
3275
+ "samples": 10,
3276
+ "conform_rate": 1,
3277
+ "value_accuracy_mean": 1,
3278
+ "native_rung_rate": 1,
3279
+ "effective_usd_per_1k_conformant": 6.562500000000002
3280
+ },
3281
+ "extraction.tabular": {
3282
+ "samples": 15,
3283
+ "conform_rate": 1,
3284
+ "value_accuracy_mean": 1,
3285
+ "native_rung_rate": 1,
3286
+ "effective_usd_per_1k_conformant": 12.053333333333335
3287
+ },
3288
+ "generation.prose": {
3289
+ "samples": 15,
3290
+ "conform_rate": 1,
3291
+ "value_accuracy_mean": 0.8333333333333334,
3292
+ "native_rung_rate": 1,
3293
+ "effective_usd_per_1k_conformant": 17.069999999999997
3294
+ },
3295
+ "qa.docs": {
3296
+ "samples": 15,
3297
+ "conform_rate": 0.9333333333333333,
3298
+ "value_accuracy_mean": 1,
3299
+ "native_rung_rate": 1,
3300
+ "effective_usd_per_1k_conformant": 7.66107142857143
3301
+ },
3302
+ "reasoning.numeric": {
3303
+ "samples": 15,
3304
+ "conform_rate": 1,
3305
+ "value_accuracy_mean": 1,
3306
+ "native_rung_rate": 1,
3307
+ "effective_usd_per_1k_conformant": 5.205000000000002
3308
+ },
3309
+ "structure.strict": {
3310
+ "samples": 10,
3311
+ "conform_rate": 1,
3312
+ "value_accuracy_mean": 1,
3313
+ "native_rung_rate": 1,
3314
+ "effective_usd_per_1k_conformant": 8.707499999999998
3315
+ },
3316
+ "summarize": {
3317
+ "samples": 15,
3318
+ "conform_rate": 1,
3319
+ "value_accuracy_mean": 1,
3320
+ "native_rung_rate": 1,
3321
+ "effective_usd_per_1k_conformant": 9.160000000000002
3322
+ }
3323
+ },
3324
+ "discrepancies": [
3325
+ {
3326
+ "kind": "citations_without_declared_search",
3327
+ "message": "no declared web-search support, but 93% of probed answers cited URLs — verify whether these are real or fabricated sources"
3328
+ }
3329
+ ]
3330
+ },
3331
+ {
3332
+ "model_key": "gemini/gemini-3.1-pro-preview",
3333
+ "conform_rate": 1,
3334
+ "conform_ci95": [
3335
+ 0.9611351464605271,
3336
+ 1
3337
+ ],
3338
+ "value_accuracy_mean": 1,
3339
+ "native_rung_rate": 1,
3340
+ "hard_conform_rate": 1,
3341
+ "hard_value_accuracy_mean": 1,
3342
+ "hard_samples": 25,
3343
+ "hard_native_rung_rate": 1,
3344
+ "effective_usd_per_1k_conformant": 11.558294736842107,
3345
+ "grounded_rate": 1,
3346
+ "cache_hit_rate": 0,
3347
+ "samples": 95,
3348
+ "as_of": "2026-08-19T04:09:00.900Z",
3349
+ "families": [
3350
+ "demo-rig",
3351
+ "probe-suite"
3352
+ ],
3353
+ "fixture_counts": {
3354
+ "demo-rig": 20,
3355
+ "probe-suite": 75
3356
+ },
3357
+ "by_family": {
3358
+ "demo-rig": {
3359
+ "samples": 20,
3360
+ "conform_rate": 1,
3361
+ "value_accuracy_mean": 1,
3362
+ "native_rung_rate": 1,
3363
+ "effective_usd_per_1k_conformant": 10.097499999999998
3364
+ },
3365
+ "probe-suite": {
3366
+ "samples": 75,
3367
+ "conform_rate": 1,
3368
+ "value_accuracy_mean": 1,
3369
+ "native_rung_rate": 1,
3370
+ "effective_usd_per_1k_conformant": 11.947839999999996
3371
+ }
3372
+ },
3373
+ "by_subtype": {
3374
+ "classification": {
3375
+ "samples": 10,
3376
+ "conform_rate": 1,
3377
+ "value_accuracy_mean": 1,
3378
+ "native_rung_rate": 1,
3379
+ "effective_usd_per_1k_conformant": 6.4566
3380
+ },
3381
+ "extraction.tabular": {
3382
+ "samples": 15,
3383
+ "conform_rate": 1,
3384
+ "value_accuracy_mean": 1,
3385
+ "native_rung_rate": 1,
3386
+ "effective_usd_per_1k_conformant": 11.4672
3387
+ },
3388
+ "generation.prose": {
3389
+ "samples": 15,
3390
+ "conform_rate": 1,
3391
+ "value_accuracy_mean": 1,
3392
+ "native_rung_rate": 1,
3393
+ "effective_usd_per_1k_conformant": 27.511066666666665
3394
+ },
3395
+ "qa.docs": {
3396
+ "samples": 15,
3397
+ "conform_rate": 1,
3398
+ "value_accuracy_mean": 1,
3399
+ "native_rung_rate": 1,
3400
+ "effective_usd_per_1k_conformant": 7.878533333333333
3401
+ },
3402
+ "reasoning.numeric": {
3403
+ "samples": 15,
3404
+ "conform_rate": 1,
3405
+ "value_accuracy_mean": 1,
3406
+ "native_rung_rate": 1,
3407
+ "effective_usd_per_1k_conformant": 5.3528
3408
+ },
3409
+ "structure.strict": {
3410
+ "samples": 10,
3411
+ "conform_rate": 1,
3412
+ "value_accuracy_mean": 1,
3413
+ "native_rung_rate": 1,
3414
+ "effective_usd_per_1k_conformant": 9.4642
3415
+ },
3416
+ "summarize": {
3417
+ "samples": 15,
3418
+ "conform_rate": 1,
3419
+ "value_accuracy_mean": 1,
3420
+ "native_rung_rate": 1,
3421
+ "effective_usd_per_1k_conformant": 10.379066666666667
3422
+ }
3423
+ },
3424
+ "discrepancies": [
3425
+ {
3426
+ "kind": "declared_caching_unrealized",
3427
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
3428
+ }
3429
+ ]
3430
+ },
3431
+ {
3432
+ "model_key": "anthropic/claude-fable-5",
3433
+ "conform_rate": 0.9157894736842105,
3434
+ "conform_ci95": [
3435
+ 0.8425478687567327,
3436
+ 0.9567118846157054
3437
+ ],
3438
+ "value_accuracy_mean": 0.9355889724310776,
3439
+ "native_rung_rate": 1,
3440
+ "hard_conform_rate": 1,
3441
+ "hard_value_accuracy_mean": 1,
3442
+ "hard_samples": 25,
3443
+ "hard_native_rung_rate": 1,
3444
+ "effective_usd_per_1k_conformant": 19.980459770114926,
3445
+ "grounded_rate": 0.9333333333333333,
3446
+ "cache_hit_rate": null,
3447
+ "samples": 95,
3448
+ "as_of": "2026-08-18T21:29:02.239Z",
3449
+ "families": [
3450
+ "demo-rig",
3451
+ "probe-suite"
3452
+ ],
3453
+ "fixture_counts": {
3454
+ "demo-rig": 20,
3455
+ "probe-suite": 75
3456
+ },
3457
+ "by_family": {
3458
+ "demo-rig": {
3459
+ "samples": 20,
3460
+ "conform_rate": 0.85,
3461
+ "value_accuracy_mean": 0.85,
3462
+ "native_rung_rate": 1,
3463
+ "effective_usd_per_1k_conformant": 19.56470588235294
3464
+ },
3465
+ "probe-suite": {
3466
+ "samples": 75,
3467
+ "conform_rate": 0.9333333333333333,
3468
+ "value_accuracy_mean": 0.9584126984126984,
3469
+ "native_rung_rate": 1,
3470
+ "effective_usd_per_1k_conformant": 20.081428571428553
3471
+ }
3472
+ },
3473
+ "by_subtype": {
3474
+ "classification": {
3475
+ "samples": 10,
3476
+ "conform_rate": 1,
3477
+ "value_accuracy_mean": 1,
3478
+ "native_rung_rate": 1,
3479
+ "effective_usd_per_1k_conformant": 13.165000000000001
3480
+ },
3481
+ "extraction.tabular": {
3482
+ "samples": 15,
3483
+ "conform_rate": 0.7333333333333333,
3484
+ "value_accuracy_mean": 0.8476190476190477,
3485
+ "native_rung_rate": 1,
3486
+ "effective_usd_per_1k_conformant": 31.70909090909092
3487
+ },
3488
+ "generation.prose": {
3489
+ "samples": 15,
3490
+ "conform_rate": 1,
3491
+ "value_accuracy_mean": 1,
3492
+ "native_rung_rate": 1,
3493
+ "effective_usd_per_1k_conformant": 31.826666666666664
3494
+ },
3495
+ "qa.docs": {
3496
+ "samples": 15,
3497
+ "conform_rate": 1,
3498
+ "value_accuracy_mean": 1,
3499
+ "native_rung_rate": 1,
3500
+ "effective_usd_per_1k_conformant": 14.963333333333335
3501
+ },
3502
+ "reasoning.numeric": {
3503
+ "samples": 15,
3504
+ "conform_rate": 1,
3505
+ "value_accuracy_mean": 1,
3506
+ "native_rung_rate": 1,
3507
+ "effective_usd_per_1k_conformant": 10.45
3508
+ },
3509
+ "structure.strict": {
3510
+ "samples": 10,
3511
+ "conform_rate": 0.9,
3512
+ "value_accuracy_mean": 0.9166666666666667,
3513
+ "native_rung_rate": 1,
3514
+ "effective_usd_per_1k_conformant": 18.388888888888893
3515
+ },
3516
+ "summarize": {
3517
+ "samples": 15,
3518
+ "conform_rate": 0.8,
3519
+ "value_accuracy_mean": 0.8,
3520
+ "native_rung_rate": 1,
3521
+ "effective_usd_per_1k_conformant": 19.479166666666668
3522
+ }
3523
+ },
3524
+ "discrepancies": [
3525
+ {
3526
+ "kind": "citations_without_declared_search",
3527
+ "message": "no declared web-search support, but 93% of probed answers cited URLs — verify whether these are real or fabricated sources"
3528
+ }
3529
+ ]
3530
+ },
3531
+ {
3532
+ "model_key": "openai/gpt-5.5-pro",
3533
+ "conform_rate": 1,
3534
+ "conform_ci95": [
3535
+ 0.9611351464605271,
3536
+ 1
3537
+ ],
3538
+ "value_accuracy_mean": 1,
3539
+ "native_rung_rate": 1,
3540
+ "hard_conform_rate": 1,
3541
+ "hard_value_accuracy_mean": 1,
3542
+ "hard_samples": 25,
3543
+ "hard_native_rung_rate": 1,
3544
+ "effective_usd_per_1k_conformant": 52.82652631578945,
3545
+ "grounded_rate": 0,
3546
+ "cache_hit_rate": 0,
3547
+ "samples": 95,
3548
+ "as_of": "2026-08-19T05:13:04.856Z",
3549
+ "families": [
3550
+ "demo-rig",
3551
+ "probe-suite"
3552
+ ],
3553
+ "fixture_counts": {
3554
+ "demo-rig": 20,
3555
+ "probe-suite": 75
3556
+ },
3557
+ "by_family": {
3558
+ "demo-rig": {
3559
+ "samples": 20,
3560
+ "conform_rate": 1,
3561
+ "value_accuracy_mean": 1,
3562
+ "native_rung_rate": 1,
3563
+ "effective_usd_per_1k_conformant": 35.932500000000005
3564
+ },
3565
+ "probe-suite": {
3566
+ "samples": 75,
3567
+ "conform_rate": 1,
3568
+ "value_accuracy_mean": 1,
3569
+ "native_rung_rate": 1,
3570
+ "effective_usd_per_1k_conformant": 57.331599999999966
3571
+ }
3572
+ },
3573
+ "by_subtype": {
3574
+ "classification": {
3575
+ "samples": 10,
3576
+ "conform_rate": 1,
3577
+ "value_accuracy_mean": 1,
3578
+ "native_rung_rate": 1,
3579
+ "effective_usd_per_1k_conformant": 26.622
3580
+ },
3581
+ "extraction.tabular": {
3582
+ "samples": 15,
3583
+ "conform_rate": 1,
3584
+ "value_accuracy_mean": 1,
3585
+ "native_rung_rate": 1,
3586
+ "effective_usd_per_1k_conformant": 45.75800000000001
3587
+ },
3588
+ "generation.prose": {
3589
+ "samples": 15,
3590
+ "conform_rate": 1,
3591
+ "value_accuracy_mean": 1,
3592
+ "native_rung_rate": 1,
3593
+ "effective_usd_per_1k_conformant": 152.36800000000002
3594
+ },
3595
+ "qa.docs": {
3596
+ "samples": 15,
3597
+ "conform_rate": 1,
3598
+ "value_accuracy_mean": 1,
3599
+ "native_rung_rate": 1,
3600
+ "effective_usd_per_1k_conformant": 33.410000000000004
3601
+ },
3602
+ "reasoning.numeric": {
3603
+ "samples": 15,
3604
+ "conform_rate": 1,
3605
+ "value_accuracy_mean": 1,
3606
+ "native_rung_rate": 1,
3607
+ "effective_usd_per_1k_conformant": 26.520000000000007
3608
+ },
3609
+ "structure.strict": {
3610
+ "samples": 10,
3611
+ "conform_rate": 1,
3612
+ "value_accuracy_mean": 1,
3613
+ "native_rung_rate": 1,
3614
+ "effective_usd_per_1k_conformant": 29.576999999999998
3615
+ },
3616
+ "summarize": {
3617
+ "samples": 15,
3618
+ "conform_rate": 1,
3619
+ "value_accuracy_mean": 1,
3620
+ "native_rung_rate": 1,
3621
+ "effective_usd_per_1k_conformant": 39.04600000000001
3622
+ }
3623
+ },
3624
+ "discrepancies": [
3625
+ {
3626
+ "kind": "undeclared_schema_capable",
3627
+ "message": "no declared schema support, but probed conformance is 100% — capability exists undeclared"
3628
+ },
3629
+ {
3630
+ "kind": "declared_search_ungrounded",
3631
+ "message": "declared supports_web_search, but only 0% of probed answers carried citations (bar: 50%)"
3632
+ },
3633
+ {
3634
+ "kind": "declared_caching_unrealized",
3635
+ "message": "declared supports_prompt_caching, but no eligible probed repeat call on the documented caching path reported cached tokens"
3636
+ }
3637
+ ]
3638
+ }
3639
+ ]