@tangle-network/agent-eval 0.188.0 → 0.190.0
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.
- package/CHANGELOG.md +119 -0
- package/README.md +3 -2
- package/dist/adapters/http.d.ts +11 -2
- package/dist/adapters/http.d.ts.map +1 -1
- package/dist/adapters/http.js +75 -7
- package/dist/adapters/http.js.map +1 -1
- package/dist/{agent-profile-CfbuZk8b.d.ts → agent-profile-DNV_U6QR.d.ts} +3 -3
- package/dist/{agent-profile-CfbuZk8b.d.ts.map → agent-profile-DNV_U6QR.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +75 -12
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +213 -8
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-CLz5O01x.d.ts → analyze-runs-BvK4-wSs.d.ts} +5 -5
- package/dist/{analyze-runs-CLz5O01x.d.ts.map → analyze-runs-BvK4-wSs.d.ts.map} +1 -1
- package/dist/{analyze-runs-DCY7LMt4.js → analyze-runs-SCSWYpy0.js} +17 -9
- package/dist/analyze-runs-SCSWYpy0.js.map +1 -0
- package/dist/{benchmark-command-CdUGku3W.js → benchmark-command-TZlsuAh9.js} +12 -12
- package/dist/{benchmark-command-CdUGku3W.js.map → benchmark-command-TZlsuAh9.js.map} +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +7 -7
- package/dist/campaign/index.js +8 -8
- package/dist/{campaign-DZuysBbS.js → campaign-_29Qpc9Z.js} +258 -13
- package/dist/campaign-_29Qpc9Z.js.map +1 -0
- package/dist/{campaign-evidence-IF1p0oDa.js → campaign-evidence-CWF8pEIk.js} +9 -8
- package/dist/campaign-evidence-CWF8pEIk.js.map +1 -0
- package/dist/{chat-client-C4dg7Ng_.js → chat-client-DbH06Ek6.js} +2 -2
- package/dist/{chat-client-C4dg7Ng_.js.map → chat-client-DbH06Ek6.js.map} +1 -1
- package/dist/{chat-json-call-JVGbfx95.js → chat-json-call-B0xTEY0I.js} +2 -2
- package/dist/{chat-json-call-JVGbfx95.js.map → chat-json-call-B0xTEY0I.js.map} +1 -1
- package/dist/claim-BGOUqoMq.d.ts +38 -0
- package/dist/claim-BGOUqoMq.d.ts.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{client-CFj9dfcn.d.ts → client-BhEyAdfe.d.ts} +4 -4
- package/dist/{client-CFj9dfcn.d.ts.map → client-BhEyAdfe.d.ts.map} +1 -1
- package/dist/{client-CXE-U1SA.js → client-QwJpsNYU.js} +5 -1
- package/dist/client-QwJpsNYU.js.map +1 -0
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +22 -13
- package/dist/contract/index.js.map +1 -1
- package/dist/deep-freeze-CurcOQg3.js +16 -0
- package/dist/deep-freeze-CurcOQg3.js.map +1 -0
- package/dist/{default-registry-BtZ_t-9b.js → default-registry-16I-Bxlf.js} +4 -3
- package/dist/default-registry-16I-Bxlf.js.map +1 -0
- package/dist/{default-registry-BLsGfnoO.d.ts → default-registry-B5HrwrPM.d.ts} +6 -6
- package/dist/{default-registry-BLsGfnoO.d.ts.map → default-registry-B5HrwrPM.d.ts.map} +1 -1
- package/dist/{define-agent-eval-CsicmDxt.d.ts → define-agent-eval-BWeyIpny.d.ts} +8 -7
- package/dist/define-agent-eval-BWeyIpny.d.ts.map +1 -0
- package/dist/{define-agent-eval-ChtddasD.js → define-agent-eval-DrGfrCTZ.js} +8 -8
- package/dist/{define-agent-eval-ChtddasD.js.map → define-agent-eval-DrGfrCTZ.js.map} +1 -1
- package/dist/diagnosis/index.d.ts +3 -3
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +4 -4
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{dspy-rlm-engine-sPX7B5ac.js → dspy-rlm-engine-CTIGomgr.js} +3 -3
- package/dist/{dspy-rlm-engine-sPX7B5ac.js.map → dspy-rlm-engine-CTIGomgr.js.map} +1 -1
- package/dist/{engine-Gw2n1irt.d.ts → engine-BkLCnZQX.d.ts} +4 -4
- package/dist/{engine-Gw2n1irt.d.ts.map → engine-BkLCnZQX.d.ts.map} +1 -1
- package/dist/{eval-campaign-CeotQ1ZR.js → eval-campaign-CMKRjWl0.js} +3 -3
- package/dist/{eval-campaign-CeotQ1ZR.js.map → eval-campaign-CMKRjWl0.js.map} +1 -1
- package/dist/evaluation.d.ts +2 -2
- package/dist/{exact-types-CrhPoDp8.d.ts → exact-types-CfpP2A4Z.d.ts} +2 -2
- package/dist/{exact-types-CrhPoDp8.d.ts.map → exact-types-CfpP2A4Z.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +2 -1
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +1 -1
- package/dist/exporters-XMCYJlqT.d.ts +161 -0
- package/dist/exporters-XMCYJlqT.d.ts.map +1 -0
- package/dist/{external-optimizer-process-D3BLpZyR.js → external-optimizer-process-Mc83UoWl.js} +3 -3
- package/dist/{external-optimizer-process-D3BLpZyR.js.map → external-optimizer-process-Mc83UoWl.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-BrfsgL92.js → external-optimizer-subprocess-C0dHXH8e.js} +3 -3
- package/dist/{external-optimizer-subprocess-BrfsgL92.js.map → external-optimizer-subprocess-C0dHXH8e.js.map} +1 -1
- package/dist/{feedback-trajectory-BJi4t3Vf.d.ts → feedback-trajectory-BYjh7NEM.d.ts} +4 -4
- package/dist/{feedback-trajectory-BJi4t3Vf.d.ts.map → feedback-trajectory-BYjh7NEM.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/hosted/index.d.ts.map +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-Ca0_IjFu.d.ts → index-2e_MTznj.d.ts} +7 -7
- package/dist/{index-Ca0_IjFu.d.ts.map → index-2e_MTznj.d.ts.map} +1 -1
- package/dist/{index-BD1oSbHr.d.ts → index-BR_R1eBE.d.ts} +11 -11
- package/dist/{index-BD1oSbHr.d.ts.map → index-BR_R1eBE.d.ts.map} +1 -1
- package/dist/{index-u0d1Jp4F.d.ts → index-BqDJ6fUK.d.ts} +45 -24
- package/dist/index-BqDJ6fUK.d.ts.map +1 -0
- package/dist/{index-eTTnPe3r.d.ts → index-Ci0-EYzc.d.ts} +678 -249
- package/dist/index-Ci0-EYzc.d.ts.map +1 -0
- package/dist/{index-B-s8qDRJ.d.ts → index-D4pC7Tvd.d.ts} +4 -3
- package/dist/index-D4pC7Tvd.d.ts.map +1 -0
- package/dist/index.d.ts +81 -32
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +385 -41
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-DeE1solA.d.ts → insight-report-CvPkEIcC.d.ts} +13 -6
- package/dist/insight-report-CvPkEIcC.d.ts.map +1 -0
- package/dist/jev.d.ts +2 -2
- package/dist/jev.js +2 -1
- package/dist/jev.js.map +1 -1
- package/dist/{journal-BuRLm6ho.js → journal-CDP0Vhx_.js} +262 -115
- package/dist/journal-CDP0Vhx_.js.map +1 -0
- package/dist/{kind-factory-g1ROw67A.js → kind-factory-ByZeXDqF.js} +4 -3
- package/dist/{kind-factory-g1ROw67A.js.map → kind-factory-ByZeXDqF.js.map} +1 -1
- package/dist/ledger-core/index.d.ts +2 -2
- package/dist/ledger-core/index.js +1 -1
- package/dist/{llm-client-DF9bPNYA.js → llm-client-BqmRnibA.js} +2 -2
- package/dist/{llm-client-DF9bPNYA.js.map → llm-client-BqmRnibA.js.map} +1 -1
- package/dist/{llm-judge-DF18J3CR.js → llm-judge-bwbgDMP0.js} +1895 -1379
- package/dist/llm-judge-bwbgDMP0.js.map +1 -0
- package/dist/{matrix-BvA1WUpj.d.ts → matrix-CNypzP-v.d.ts} +2 -2
- package/dist/{matrix-BvA1WUpj.d.ts.map → matrix-CNypzP-v.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/mint-C9Lojt1V.d.ts +147 -0
- package/dist/mint-C9Lojt1V.d.ts.map +1 -0
- package/dist/{mint-B4zgSAJW.js → mint-iCD-W43s.js} +30 -12
- package/dist/mint-iCD-W43s.js.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/pipelines/index.d.ts +1 -1
- package/dist/{prime-protocol-CF8Z43Fj.d.ts → prime-protocol-Fwvctmhm.d.ts} +2 -2
- package/dist/{prime-protocol-CF8Z43Fj.d.ts.map → prime-protocol-Fwvctmhm.d.ts.map} +1 -1
- package/dist/{produced-state-xsMMjYVc.js → produced-state-zSaGDC_Q.js} +4 -4
- package/dist/{produced-state-xsMMjYVc.js.map → produced-state-zSaGDC_Q.js.map} +1 -1
- package/dist/{raw-provider-sink-C96CcqeG.js → raw-provider-sink-DIRCZjP_.js} +2 -2
- package/dist/{raw-provider-sink-C96CcqeG.js.map → raw-provider-sink-DIRCZjP_.js.map} +1 -1
- package/dist/{redact-DxwlUAvh.js → redact-8fX8-uKr.js} +16 -7
- package/dist/redact-8fX8-uKr.js.map +1 -0
- package/dist/{redact-DPoH1-OW.d.ts → redact-Du3WKUOP.d.ts} +7 -5
- package/dist/redact-Du3WKUOP.d.ts.map +1 -0
- package/dist/{registry-BM2Kdk8T.d.ts → registry-mAQVtuLx.d.ts} +3 -3
- package/dist/{registry-BM2Kdk8T.d.ts.map → registry-mAQVtuLx.d.ts.map} +1 -1
- package/dist/{release-confidence-CG4Z9QLB.d.ts → release-confidence-CkoXFx9I.d.ts} +3 -3
- package/dist/{release-confidence-CG4Z9QLB.d.ts.map → release-confidence-CkoXFx9I.d.ts.map} +1 -1
- package/dist/{release-confidence-RanRUXRW.js → release-confidence-XQm_xjhR.js} +3 -3
- package/dist/release-confidence-XQm_xjhR.js.map +1 -0
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +1 -1
- package/dist/{researcher-Bjcd_8T_.d.ts → researcher-DojEMb7E.d.ts} +3 -3
- package/dist/{researcher-Bjcd_8T_.d.ts.map → researcher-DojEMb7E.d.ts.map} +1 -1
- package/dist/{reward-hacking-DrKXJtgH.d.ts → reward-hacking-Cyk-QHk9.d.ts} +2 -2
- package/dist/{reward-hacking-DrKXJtgH.d.ts.map → reward-hacking-Cyk-QHk9.d.ts.map} +1 -1
- package/dist/rl.d.ts +10 -8
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +7 -6
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +5 -4
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-C0Ifk3vK.js → rollout-B-XBReTC.js} +2 -2
- package/dist/{rollout-C0Ifk3vK.js.map → rollout-B-XBReTC.js.map} +1 -1
- package/dist/{rubric-predictive-validity-B14yBAmM.d.ts → rubric-predictive-validity-DIUwJJ9Y.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-B14yBAmM.d.ts.map → rubric-predictive-validity-DIUwJJ9Y.d.ts.map} +1 -1
- package/dist/{run-record-tg_5H2IK.js → run-record-BRlM99rX.js} +41 -3
- package/dist/run-record-BRlM99rX.js.map +1 -0
- package/dist/{run-record-xEULpgML.d.ts → run-record-DF-fuu31.d.ts} +58 -6
- package/dist/run-record-DF-fuu31.d.ts.map +1 -0
- package/dist/{run-record-CNNzWGpB.js → run-record-DnMFyg_g.js} +26 -13
- package/dist/run-record-DnMFyg_g.js.map +1 -0
- package/dist/{schema-BzWDXhOR.d.ts → schema-B5LtYvn9.d.ts} +7 -5
- package/dist/{schema-BzWDXhOR.d.ts.map → schema-B5LtYvn9.d.ts.map} +1 -1
- package/dist/schema-C1aaAxTf.js.map +1 -1
- package/dist/search-ledger-types-C9anU2DX.d.ts +489 -0
- package/dist/search-ledger-types-C9anU2DX.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-70LxrCSL.js → semantic-concept-judge-BY0me-Jq.js} +6 -6
- package/dist/{semantic-concept-judge-70LxrCSL.js.map → semantic-concept-judge-BY0me-Jq.js.map} +1 -1
- package/dist/{server-C8XvqCg8.js → server-DnJCDLf8.js} +2 -2
- package/dist/{server-C8XvqCg8.js.map → server-DnJCDLf8.js.map} +1 -1
- package/dist/{skillopt-optimization-method-C60h0Got.js → skillopt-optimization-method-CfnQEjgT.js} +364 -19
- package/dist/skillopt-optimization-method-CfnQEjgT.js.map +1 -0
- package/dist/{spans-W0o8pbEg.d.ts → spans-D1RftP_w.d.ts} +2 -2
- package/dist/{spans-W0o8pbEg.d.ts.map → spans-D1RftP_w.d.ts.map} +1 -1
- package/dist/{statistical-heldout-CPv-y6sK.d.ts → statistical-heldout-CLpjYffJ.d.ts} +4 -38
- package/dist/statistical-heldout-CLpjYffJ.d.ts.map +1 -0
- package/dist/{store-otlp-DPgzSQKS.js → store-otlp-84DH0OEH.js} +2 -2
- package/dist/{store-otlp-DPgzSQKS.js.map → store-otlp-84DH0OEH.js.map} +1 -1
- package/dist/{store-tool-spans-C2lLwDzg.d.ts → store-tool-spans-BBXs8hDP.d.ts} +6 -6
- package/dist/{store-tool-spans-C2lLwDzg.d.ts.map → store-tool-spans-BBXs8hDP.d.ts.map} +1 -1
- package/dist/{store-tool-spans-ClPxcvh2.js → store-tool-spans-DSbYpl7f.js} +5 -5
- package/dist/{store-tool-spans-ClPxcvh2.js.map → store-tool-spans-DSbYpl7f.js.map} +1 -1
- package/dist/{summary-report-SNvJoAvD.d.ts → summary-report-DvvesgR_.d.ts} +5 -4
- package/dist/summary-report-DvvesgR_.d.ts.map +1 -0
- package/dist/supervisor-run/index.d.ts +2 -2
- package/dist/{tool-groups-CCQqAa16.d.ts → tool-groups-W7MJobX_.d.ts} +3 -3
- package/dist/tool-groups-W7MJobX_.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/traces.d.ts +6 -66
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +7 -196
- package/dist/traces.js.map +1 -1
- package/dist/{types-DDp2Mb8Y.d.ts → types-Bay1dh2o.d.ts} +5 -5
- package/dist/{types-DDp2Mb8Y.d.ts.map → types-Bay1dh2o.d.ts.map} +1 -1
- package/dist/{types-B1vDZ694.d.ts → types-BvmWNjjs.d.ts} +3 -3
- package/dist/{types-B1vDZ694.d.ts.map → types-BvmWNjjs.d.ts.map} +1 -1
- package/dist/{types-C8VBbiw6.js → types-CW5b90JH.js} +2 -15
- package/dist/types-CW5b90JH.js.map +1 -0
- package/dist/{types-5ak2PqnJ.d.ts → types-Cvqy7asY.d.ts} +3 -3
- package/dist/{types-5ak2PqnJ.d.ts.map → types-Cvqy7asY.d.ts.map} +1 -1
- package/dist/{types-Tp5WhYL9.d.ts → types-DRNlLbBR.d.ts} +2 -2
- package/dist/{types-Tp5WhYL9.d.ts.map → types-DRNlLbBR.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +1 -1
- package/dist/wire/index.js +1 -1
- package/docs/adapters-observability.md +31 -27
- package/docs/charter.md +1 -1
- package/docs/code-agent-intake.md +1 -1
- package/docs/design/mlbenchmarks-book-review.md +1 -1
- package/docs/distributed-driver.md +1 -5
- package/docs/experiment.md +15 -0
- package/docs/insight-report.md +3 -2
- package/docs/public-api.md +174 -68
- package/docs/redaction.md +7 -1
- package/docs/rollout.md +5 -5
- package/docs/search-ledger.md +174 -0
- package/docs/trace-contracts.md +48 -3
- package/package.json +1 -1
- package/dist/analyze-runs-DCY7LMt4.js.map +0 -1
- package/dist/campaign-DZuysBbS.js.map +0 -1
- package/dist/campaign-evidence-IF1p0oDa.js.map +0 -1
- package/dist/client-CXE-U1SA.js.map +0 -1
- package/dist/default-registry-BtZ_t-9b.js.map +0 -1
- package/dist/define-agent-eval-CsicmDxt.d.ts.map +0 -1
- package/dist/index-B-s8qDRJ.d.ts.map +0 -1
- package/dist/index-eTTnPe3r.d.ts.map +0 -1
- package/dist/index-u0d1Jp4F.d.ts.map +0 -1
- package/dist/insight-report-DeE1solA.d.ts.map +0 -1
- package/dist/journal-BuRLm6ho.js.map +0 -1
- package/dist/llm-judge-DF18J3CR.js.map +0 -1
- package/dist/mint-B4zgSAJW.js.map +0 -1
- package/dist/mint-CJFSYBIU.d.ts +0 -285
- package/dist/mint-CJFSYBIU.d.ts.map +0 -1
- package/dist/redact-DPoH1-OW.d.ts.map +0 -1
- package/dist/redact-DxwlUAvh.js.map +0 -1
- package/dist/release-confidence-RanRUXRW.js.map +0 -1
- package/dist/run-record-CNNzWGpB.js.map +0 -1
- package/dist/run-record-tg_5H2IK.js.map +0 -1
- package/dist/run-record-xEULpgML.d.ts.map +0 -1
- package/dist/search-ledger-types-D7KDVg-o.d.ts +0 -302
- package/dist/search-ledger-types-D7KDVg-o.d.ts.map +0 -1
- package/dist/skillopt-optimization-method-C60h0Got.js.map +0 -1
- package/dist/statistical-heldout-CPv-y6sK.d.ts.map +0 -1
- package/dist/summary-report-SNvJoAvD.d.ts.map +0 -1
- package/dist/tool-groups-CCQqAa16.d.ts.map +0 -1
- package/dist/types-C8VBbiw6.js.map +0 -1
- package/docs/search-history-receipts.md +0 -232
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mint-B4zgSAJW.js","names":[],"sources":["../src/rollout/mint.ts"],"sourcesContent":["/**\n * Rollout minting — `tangle.rollout.v1` lines joined from the records the\n * substrate ALREADY keeps. There is no separate rollout store: a rollout\n * is the JOIN of a RunRecord (identity, provenance, cost, outcome) with\n * its trace (spans share `runId`), projected into the canonical line.\n *\n * Composition, not duplication:\n * - identity/provenance → `RunRecord` (candidateId, splitTag, agentProfile, hashes)\n * - step structure → `buildTrajectory` over the shared TraceStore\n * - preference-pair export → `feedbackTrajectoryToOptimizerRow` (feedback-trajectory.ts)\n * - PRM / reward-model → `reward-model-export.ts`\n *\n * Anti-Goodhart invariant: a run whose `outcome.realness.gated` is true is\n * never exported with a positive reward OR with any of the numbers that reward\n * was computed from. The gate travels into the training data (`reward` forced\n * to 0, `realness_gated: true`) and the whole outcome is transformed by\n * `gateGamedOutcome` inside `assertMinted` below, which relocates `metrics` and\n * `verdict` to `provenance.gated_evidence`. Mint returns\n * `MintedRolloutLine[]`: the brand the training exporters require, which only\n * this function, `readRolloutLedger`, and an explicit `assertMinted` can mint.\n *\n * A record carrying NEITHER split score is REJECTED (`ValidationError`), never\n * minted at 0 — \"nobody graded this\" is not the same claim as \"graded a total\n * failure\", and a trainer reading 0 learns the second. Lines that already\n * carry `reward: null` (interchange imports, existing ledgers) remain valid on\n * the wire; only the RunRecord→line door refuses.\n *\n * Records without spans or an explicit full capture become labeled GAP LINES (messages: [],\n * provenance.gap) — present in the output AND surfaced in\n * `missingTraces`; a capture gap is a finding, never a silent omission.\n */\n\nimport { ValidationError } from '../errors'\nimport { type RunRecord, runTaskScore } from '../run-record'\nimport type { LlmSpan, Span, ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory } from '../trajectory'\nimport { rolloutRewardFields, scoreOrigin } from './reward'\nimport {\n assertMinted,\n type ChatMessage,\n type MintedRolloutLine,\n ROLLOUT_SCHEMA,\n type RolloutRole,\n type RolloutSplit,\n type RolloutStep,\n} from './schema'\n\n/** Redactor for transcript/step payload text (secrets, PII), not opaque identity. */\nexport type RolloutScrubber = (text: string) => string\n\nexport interface MintRolloutOptions {\n scrub?: RolloutScrubber\n /**\n * Full canonical capture, including turns no longer in the model's context.\n * When supplied, missing/empty capture is refused, never replaced with a\n * span summary. Use the execution owner's capture or the existing harness\n * readers; reviewer traces and compacted histories are not full capture.\n */\n messagesOf?: (\n runId: string,\n ) => readonly ChatMessage[] | undefined | Promise<readonly ChatMessage[] | undefined>\n /** Cap steps per line (longest runs first drop middle steps). Default: no cap. */\n maxSteps?: number\n /** Role recorded on every minted line. Default 'agent' (a solo eval run). */\n role?: RolloutRole\n /** Task suite label. Default: the record's `experimentId`. */\n suite?: string\n /** Injected clock for deterministic output. */\n now?: () => Date\n}\n\nexport interface MintRolloutResult {\n rows: MintedRolloutLine[]\n /** runIds that had a RunRecord but no spans — emitted as gap lines AND listed here. */\n missingTraces: string[]\n}\n\nconst asText = (v: unknown, scrub: RolloutScrubber): string => {\n const s = typeof v === 'string' ? v : JSON.stringify(v)\n return scrub(s ?? '')\n}\n\nfunction projectStep(span: Span, scrub: RolloutScrubber): RolloutStep {\n const base: RolloutStep = {\n kind: span.kind,\n name: scrub(span.name),\n status: span.status,\n durationMs: span.endedAt !== undefined ? span.endedAt - span.startedAt : undefined,\n }\n if (span.kind === 'llm') {\n const llm = span as LlmSpan\n const last = llm.messages[llm.messages.length - 1]\n if (last && last.content !== null) base.input = scrub(last.content)\n if (llm.output !== undefined) base.output = scrub(llm.output)\n } else if (span.kind === 'tool') {\n const tool = span as ToolSpan\n base.input = asText(tool.args, scrub)\n if (tool.result !== undefined) base.output = asText(tool.result, scrub)\n }\n return base\n}\n\n/** Preserve captured fields; invocation ids are opaque linkage, not generated text. */\nfunction projectMessage(message: ChatMessage, scrub: RolloutScrubber): ChatMessage {\n return {\n role: message.role,\n content: message.content === null ? null : scrub(message.content),\n ...(message.reasoning_content !== undefined\n ? { reasoning_content: scrub(message.reasoning_content) }\n : {}),\n ...(message.tool_calls !== undefined\n ? {\n tool_calls: message.tool_calls.map((call) => ({\n id: call.id,\n type: call.type,\n function: {\n name: call.function.name,\n arguments: scrub(call.function.arguments),\n },\n })),\n }\n : {}),\n ...(message.tool_call_id !== undefined ? { tool_call_id: message.tool_call_id } : {}),\n ...(message.name !== undefined ? { name: message.name } : {}),\n ...(message.is_copied_context !== undefined\n ? { is_copied_context: message.is_copied_context }\n : {}),\n }\n}\n\n/** Never recover missing calls from tool names, span order or argument summaries. */\nfunction requireToolLinkage(messages: readonly ChatMessage[], runId: string): void {\n const seen = new Set<string>()\n const pending = new Map<string, string>()\n const refuse: (detail: string) => never = (detail) => {\n throw new ValidationError(`Cannot mint rollout for run ${runId}: ${detail}`)\n }\n for (const message of messages) {\n if (message.tool_calls !== undefined) {\n if (message.role !== 'assistant' || !Array.isArray(message.tool_calls)) {\n refuse('tool calls must be captured on an assistant message')\n }\n if (pending.size > 0) refuse('new tool calls precede the prior invocation results')\n for (const call of message.tool_calls) {\n if (\n !call ||\n typeof call.id !== 'string' ||\n call.id.length === 0 ||\n seen.has(call.id) ||\n call.type !== 'function' ||\n !call.function ||\n typeof call.function.name !== 'string' ||\n call.function.name.length === 0 ||\n typeof call.function.arguments !== 'string'\n ) {\n refuse('missing, duplicate or malformed tool invocation')\n }\n seen.add(call.id)\n pending.set(call.id, call.function.name)\n }\n }\n if (message.role === 'tool') {\n const id = message.tool_call_id\n if (typeof id !== 'string' || !pending.has(id)) {\n refuse('tool result has no captured invocation (or answers it more than once)')\n }\n if (message.name !== undefined && message.name !== pending.get(id)) {\n refuse('tool result name does not match its invocation')\n }\n pending.delete(id)\n } else if (message.tool_call_id !== undefined) {\n refuse('tool_call_id is only valid on a tool result')\n } else if (pending.size > 0 && message.tool_calls === undefined) {\n refuse('conversation advances before all tool results were captured')\n }\n }\n if (pending.size > 0) refuse('tool invocation has no captured result')\n}\n\n/** Legacy span history; callers with compacted contexts supply messagesOf instead. */\nfunction finalConversation(spans: Span[]): ChatMessage[] {\n const llms = spans.filter((s): s is LlmSpan => s.kind === 'llm')\n const last = llms[llms.length - 1]\n if (!last) return []\n const messages: ChatMessage[] = [...last.messages]\n if (last.output !== undefined && last.output !== '') {\n messages.push({ role: 'assistant', content: last.output })\n }\n return messages\n}\n\n// The reward derivations live in the leaf module `./reward` so gate and\n// reporting code can import them without dragging in the trace store; they are\n// re-exported here because the derivations shipped from this path.\nexport {\n isRealnessGated,\n observedScore,\n observedSplitScore,\n type ScoreOrigin,\n type ScorePreference,\n scoreOrigin,\n trainingReward,\n trainingScore,\n} from './reward'\n\nconst REWARD_SOURCE: Record<ReturnType<typeof scoreOrigin>, string> = {\n holdout: 'run-record/holdout-score',\n search: 'run-record/search-score',\n unscored: 'run-record/unscored',\n}\n\n/**\n * The mint door refuses an execution-only record: a missing training label is\n * not a zero reward, and not a mintable line either. Lines that already carry\n * `reward: null` — interchange imports, existing ledgers — stay valid on the\n * wire and keep their labeled gap; this guard is only about the\n * RunRecord→line door, where the producer can still be told to go score the\n * run instead of shipping an unlabeled row.\n */\nfunction requireTaskScore(record: RunRecord): void {\n if (runTaskScore(record) === undefined) {\n throw new ValidationError(`Cannot mint rollout for run ${record.runId}: task score is missing`)\n }\n}\n\nconst isObject = (value: unknown): value is Record<string, unknown> =>\n typeof value === 'object' && value !== null\n\ninterface MintFieldCheck {\n /** The RunRecord path, spelled the way the caller has to fix it. */\n readonly field: string\n /** True when the record carries something the line can honestly be built from. */\n readonly present: (bag: Record<string, unknown>) => boolean\n /** What the caller writes onto the record, and why that value and not another. */\n readonly remedy: string\n}\n\n/**\n * The RunRecord fields mint reads that a record can be missing even though the\n * TYPE says it cannot. There are exactly two ways that happens:\n *\n * 1. The field was OPTIONAL when the record was serialized. `costProvenance`,\n * `terminalOutcome` and `scenarioId` were optional through agent-eval\n * 0.125 and became required in 0.126, with no on-disk migration — so every\n * ledger written before 0.126 is full of records the type calls complete.\n * 2. Mint reads a level DEEPER than the record's own type is checked at:\n * `outcome.raw`, `tokenUsage.input`, `tokenUsage.output`.\n *\n * Nothing else needs a check here. Every other field mint copies is a top-level\n * scalar landing in a typed slot on the line, where an absent value arrives as\n * `undefined` and `assertMinted` refuses it by name. These are the ones where an\n * absent value instead kills the join with `TypeError: Cannot read properties of\n * undefined`, or — worse — mints a line that reads as measured.\n *\n * This is deliberately NOT `validateRunRecord`. That validator answers \"is this\n * a valid RunRecord\", which is a wider question than \"can a rollout line be\n * built from this one\": it also enforces model-snapshot discipline, the\n * `terminalFailureReason` coupling, and the `costUsd === costProvenance.usd`\n * agreement. Routing the mint door through it would refuse records mint can\n * mint honestly today (a model alias with no snapshot date, for one), which is\n * a policy change with its own blast radius and not this bug. The door asks the\n * narrower question and answers it precisely.\n */\nconst MINT_FIELD_CHECKS: readonly MintFieldCheck[] = [\n {\n field: 'costProvenance',\n present: (bag) => isObject(bag.costProvenance) && typeof bag.costProvenance.kind === 'string',\n remedy:\n \"Records written before agent-eval 0.126 predate this field and carry `costUsd: 0` as the documented uncaptured sentinel, which is NOT an observed zero. Backfill it as costProvenance: { kind: 'uncaptured', usd: null } WITH costUsd: null — an uncaptured cost whose costUsd is non-null is rejected by validateRunRecord, so provenance alone leaves the record invalid.\",\n },\n {\n field: 'tokenUsage',\n present: (bag) => isObject(bag.tokenUsage),\n remedy:\n \"The line's cost.tokens_in and cost.tokens_out are read from it. Backfill it from the provider's usage report; mint will not write 0 for tokens nobody counted.\",\n },\n {\n field: 'tokenUsage.input',\n present: (bag) => !isObject(bag.tokenUsage) || typeof bag.tokenUsage.input === 'number',\n remedy: \"The line's cost.tokens_in is read from it, and a missing count is not a zero count.\",\n },\n {\n field: 'tokenUsage.output',\n present: (bag) => !isObject(bag.tokenUsage) || typeof bag.tokenUsage.output === 'number',\n remedy: \"The line's cost.tokens_out is read from it, and a missing count is not a zero count.\",\n },\n {\n field: 'outcome',\n present: (bag) => isObject(bag.outcome),\n remedy:\n \"The line's reward, reward_source and metrics are all read from it. A record with no outcome carries no training label at all, and mint refuses an unlabeled row.\",\n },\n {\n field: 'outcome.raw',\n // Reported only when `outcome` itself is present: one absent field should\n // produce one reason per CAUSE, not one per path that dereferences it.\n present: (bag) => !isObject(bag.outcome) || isObject(bag.outcome.raw),\n remedy:\n 'It is the metric bag copied verbatim into the line\\'s outcome.metrics. `{ ...undefined }` spreads to `{}` without complaint, so an absent bag would mint as \"this run reported no metrics\" — a different claim from \"this record predates the field\". Backfill it as {} only when that is what you mean.',\n },\n {\n field: 'terminalOutcome',\n present: (bag) => typeof bag.terminalOutcome === 'string',\n remedy:\n \"It became required in agent-eval 0.126. Backfill it from root-run or process evidence, or as 'unknown' when the producer has none — mint will not decide the line's is_completed and is_truncated for you.\",\n },\n {\n field: 'scenarioId',\n present: (bag) => typeof bag.scenarioId === 'string' && bag.scenarioId.length > 0,\n remedy:\n \"It became required in agent-eval 0.126 and becomes the line's task.instance_id, which must be a non-empty string. Backfill it from the scenario the run was dealt (pre-0.126 producers often left it in outcome.raw.scenario_id).\",\n },\n]\n\n/**\n * Why a record cannot be minted, one entry per missing field, empty when it can.\n *\n * Exported so a caller can partition a whole ledger — \"which of my 2742 records\n * predate 0.126\" — without catching an exception per record, and without\n * re-deriving the field list on their side. A re-derived list is a list that\n * drifts from the door it is supposed to predict.\n *\n * Takes a `RunRecord` because that is what the caller holds and what the\n * compiler agrees they hold. The type is precisely the thing that is wrong, so\n * the checks read the record as the untyped bag it actually is on disk.\n */\nexport function unmintableReasons(record: RunRecord): string[] {\n const bag = record as unknown as Record<string, unknown>\n return MINT_FIELD_CHECKS.filter((check) => !check.present(bag)).map(\n (check) => `${check.field} is missing. ${check.remedy}`,\n )\n}\n\n/**\n * The mint door THROWS on a record it cannot build a line from. It does NOT\n * normalise an absent `costProvenance` to `{kind:'uncaptured', usd:null}`, and\n * the choice is not stylistic:\n *\n * - Normalising cannot cover the record, only part of it. `terminalOutcome`\n * feeds `is_completed` and `is_truncated`, which the rollout schema requires\n * to be BOOLEAN — there is no null to fall back to, so every possible\n * default is a claim about how the run ended. A door that quietly fixes the\n * cost and invents the ending is a door no caller can predict.\n * - Normalising the cost requires knowing what `costUsd: 0` meant, and mint\n * cannot know. A genuinely free run and an uncaptured one are the same bytes\n * in a pre-0.126 record; only the producer can tell them apart. Guessing is\n * exactly the failure this guard exists to stop — the 0.125 optional chain\n * `record.costProvenance?.kind === 'uncaptured'` already made that guess,\n * silently, and every record it touched minted `cost.usd: 0`: an unmeasured\n * cost published as a measured zero, into a training dataset.\n * - `requireTaskScore`, directly above, already refuses an unlabeled record\n * for the same reason: \"nobody graded this\" is not \"graded zero\". \"Nobody\n * billed this\" is not \"billed zero\".\n *\n * The caller who wants historical records minted backfills them at their store,\n * in one pass, where `costUsd` can be corrected alongside `costProvenance` —\n * which is the only place that decision can be made correctly. The refusal names\n * the run, names every missing field, and spells the value to write.\n */\nfunction requireMintableRecord(record: RunRecord): void {\n const reasons = unmintableReasons(record)\n if (reasons.length === 0) return\n throw new ValidationError(`Cannot mint rollout for run ${record.runId}: ${reasons.join('\\n ')}`)\n}\n\nconst SPLIT_FROM_TAG: Record<RunRecord['splitTag'], RolloutSplit> = {\n search: 'search',\n dev: 'dev',\n holdout: 'holdout',\n}\n\nfunction mintLine(\n record: RunRecord,\n steps: RolloutStep[],\n messages: ChatMessage[],\n options: MintRolloutOptions,\n capturedAt: string,\n gap?: string,\n): MintedRolloutLine {\n // Field presence first, and BEFORE `requireTaskScore`: that guard reads\n // `record.outcome.searchScore` on its way to the answer, so an absent\n // `outcome` would throw a bare TypeError from inside the guard whose whole\n // job is to produce a clean refusal.\n //\n // Both branches of `mintRolloutRows` — the traced line and the gap line —\n // land here, which is the point: `mintLine` is the only constructor of a\n // `MintedRolloutLine` from a RunRecord, so there is no path into the waist\n // that skips the check and no way to get this wrong from the outside.\n requireMintableRecord(record)\n // A missing task score is refused before anything is built: an\n // execution-only record has no training label, and a missing label is\n // neither a zero reward nor a mintable row.\n requireTaskScore(record)\n // `reward` and `realness_gated` come out of one call, so neither door into\n // the waist can write one and forget the other.\n const rewardFields = rolloutRewardFields(record)\n const uncaptured = record.costProvenance.kind === 'uncaptured'\n const terminalOutcome = record.terminalOutcome\n const isCompleted = terminalOutcome === 'succeeded' || terminalOutcome === 'failed'\n const isTruncated = terminalOutcome === 'cancelled' || terminalOutcome === 'incomplete'\n const terminalError =\n terminalOutcome === 'failed' ||\n terminalOutcome === 'cancelled' ||\n terminalOutcome === 'incomplete'\n ? (record.terminalFailureReason ?? `run ended ${terminalOutcome}`)\n : null\n // `assertMinted` rather than a cast: mint is the producer the whole gate\n // rests on, so it proves the line it just built is valid instead of asserting\n // it by fiat. The brand is unforgeable precisely because nobody casts to it.\n return assertMinted(\n {\n schema: ROLLOUT_SCHEMA,\n rollout_id: record.runId,\n parent_rollout_id: null,\n run_id: record.runId,\n experiment_id: record.experimentId,\n candidate_id: record.candidateId,\n generation: null,\n candidate_index: null,\n role: options.role ?? 'agent',\n task: {\n suite: options.suite ?? record.experimentId,\n instance_id: record.scenarioId,\n split: SPLIT_FROM_TAG[record.splitTag],\n seed: record.seed,\n rep: 0,\n },\n policy: {\n harness: null,\n harness_version: null,\n model: record.model,\n provider: null,\n profile_commit: record.commitSha,\n prompt_hash: record.promptHash,\n config_hash: record.configHash,\n agent_profile_cell_id: record.agentProfile?.cellId ?? null,\n sampling: null,\n },\n messages,\n tool_defs: [],\n ...(steps.length > 0 ? { steps } : {}),\n outcome: {\n ...rewardFields,\n reward_source: REWARD_SOURCE[scoreOrigin(record)],\n verdict: null,\n // A verbatim bulk copy, deliberately UNFILTERED here. `outcome.raw`\n // holds the per-layer verifier scores (`layer.*`) that the reward was\n // derived from, so on a gated run this dict is the reward signal in\n // component form — but filtering it at this call site is the pattern\n // that has now leaked twice, because the next producer to write a\n // reward-bearing field forgets. The gate is applied to the whole\n // outcome once, in `assertMinted` below (`gateGamedOutcome`), which\n // moves the block to `provenance.gated_evidence` when the run is gated\n // and leaves it here untouched when it is not.\n metrics: { ...record.outcome.raw },\n is_completed: isCompleted,\n is_truncated: isTruncated,\n error: terminalError,\n },\n cost: {\n usd: uncaptured ? null : record.costUsd,\n tokens_in: record.tokenUsage.input,\n tokens_out: record.tokenUsage.output,\n tokens_reasoning: record.tokenUsage.reasoning ?? null,\n cache_read: record.tokenUsage.cached ?? null,\n cache_write: record.tokenUsage.cacheWrite ?? null,\n wall_s: Math.round(record.wallMs / 1000),\n },\n artifacts: { patch_path: null, run_dir: null, transcript_ref: null },\n provenance: {\n captured_at: capturedAt,\n capture: 'mint',\n ...(gap !== undefined ? { gap } : {}),\n },\n },\n `minted rollout line for run ${record.runId}`,\n )\n}\n\n/**\n * Join RunRecords with their traces into canonical rollout lines. Records\n * without spans or an explicit capture are emitted as labeled gap lines and reported in\n * `missingTraces`. Execution-only records without a task score are rejected\n * because a missing training label is not a zero reward.\n */\nexport async function mintRolloutRows(\n records: RunRecord[],\n store: TraceStore,\n options: MintRolloutOptions = {},\n): Promise<MintRolloutResult> {\n const scrub = options.scrub ?? ((t) => t)\n const capturedAt = (options.now?.() ?? new Date()).toISOString()\n const rows: MintedRolloutLine[] = []\n const missingTraces: string[] = []\n for (const record of records) {\n const trajectory = await buildTrajectory(store, record.runId)\n const captured = options.messagesOf ? await options.messagesOf(record.runId) : undefined\n if (options.messagesOf && (!Array.isArray(captured) || captured.length === 0)) {\n throw new ValidationError(\n `Cannot mint rollout for run ${record.runId}: full capture is missing`,\n )\n }\n if (trajectory.steps.length === 0 && !options.messagesOf) {\n missingTraces.push(record.runId)\n rows.push(\n mintLine(record, [], [], options, capturedAt, 'no trace spans recorded for this runId'),\n )\n continue\n }\n let steps = trajectory.steps.map((s) => projectStep(s.span, scrub))\n if (options.maxSteps !== undefined && steps.length > options.maxSteps) {\n // Keep the head and tail — the middle of a long run is the least\n // informative for outcome attribution.\n const head = Math.ceil(options.maxSteps / 2)\n const tail = options.maxSteps - head\n steps = [...steps.slice(0, head), ...steps.slice(steps.length - tail)]\n }\n const spans = trajectory.steps.map((s) => s.span)\n const messages = captured ?? finalConversation(spans)\n requireToolLinkage(messages, record.runId)\n const conversation = messages.map((message) => projectMessage(message, scrub))\n // SFT drops copied context. That projection must not orphan a retained result.\n requireToolLinkage(\n conversation.filter((message) => message.is_copied_context !== true),\n record.runId,\n )\n const gap =\n conversation.length === 0 ? 'trace has no llm spans — no conversation to inline' : undefined\n rows.push(mintLine(record, steps, conversation, options, capturedAt, gap))\n }\n return { rows, missingTraces }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA8EA,MAAM,UAAU,GAAY,UAAmC;CAE7D,OAAO,OADG,OAAO,MAAM,WAAW,IAAI,KAAK,UAAU,CAAC,MACpC,EAAE;AACtB;AAEA,SAAS,YAAY,MAAY,OAAqC;CACpE,MAAM,OAAoB;EACxB,MAAM,KAAK;EACX,MAAM,MAAM,KAAK,IAAI;EACrB,QAAQ,KAAK;EACb,YAAY,KAAK,YAAY,KAAA,IAAY,KAAK,UAAU,KAAK,YAAY,KAAA;CAC3E;CACA,IAAI,KAAK,SAAS,OAAO;EACvB,MAAM,MAAM;EACZ,MAAM,OAAO,IAAI,SAAS,IAAI,SAAS,SAAS;EAChD,IAAI,QAAQ,KAAK,YAAY,MAAM,KAAK,QAAQ,MAAM,KAAK,OAAO;EAClE,IAAI,IAAI,WAAW,KAAA,GAAW,KAAK,SAAS,MAAM,IAAI,MAAM;CAC9D,OAAO,IAAI,KAAK,SAAS,QAAQ;EAC/B,MAAM,OAAO;EACb,KAAK,QAAQ,OAAO,KAAK,MAAM,KAAK;EACpC,IAAI,KAAK,WAAW,KAAA,GAAW,KAAK,SAAS,OAAO,KAAK,QAAQ,KAAK;CACxE;CACA,OAAO;AACT;;AAGA,SAAS,eAAe,SAAsB,OAAqC;CACjF,OAAO;EACL,MAAM,QAAQ;EACd,SAAS,QAAQ,YAAY,OAAO,OAAO,MAAM,QAAQ,OAAO;EAChE,GAAI,QAAQ,sBAAsB,KAAA,IAC9B,EAAE,mBAAmB,MAAM,QAAQ,iBAAiB,EAAE,IACtD,CAAC;EACL,GAAI,QAAQ,eAAe,KAAA,IACvB,EACE,YAAY,QAAQ,WAAW,KAAK,UAAU;GAC5C,IAAI,KAAK;GACT,MAAM,KAAK;GACX,UAAU;IACR,MAAM,KAAK,SAAS;IACpB,WAAW,MAAM,KAAK,SAAS,SAAS;GAC1C;EACF,EAAE,EACJ,IACA,CAAC;EACL,GAAI,QAAQ,iBAAiB,KAAA,IAAY,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;EACnF,GAAI,QAAQ,SAAS,KAAA,IAAY,EAAE,MAAM,QAAQ,KAAK,IAAI,CAAC;EAC3D,GAAI,QAAQ,sBAAsB,KAAA,IAC9B,EAAE,mBAAmB,QAAQ,kBAAkB,IAC/C,CAAC;CACP;AACF;;AAGA,SAAS,mBAAmB,UAAkC,OAAqB;CACjF,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,0BAAU,IAAI,IAAoB;CACxC,MAAM,UAAqC,WAAW;EACpD,MAAM,IAAI,gBAAgB,+BAA+B,MAAM,IAAI,QAAQ;CAC7E;CACA,KAAK,MAAM,WAAW,UAAU;EAC9B,IAAI,QAAQ,eAAe,KAAA,GAAW;GACpC,IAAI,QAAQ,SAAS,eAAe,CAAC,MAAM,QAAQ,QAAQ,UAAU,GACnE,OAAO,qDAAqD;GAE9D,IAAI,QAAQ,OAAO,GAAG,OAAO,qDAAqD;GAClF,KAAK,MAAM,QAAQ,QAAQ,YAAY;IACrC,IACE,CAAC,QACD,OAAO,KAAK,OAAO,YACnB,KAAK,GAAG,WAAW,KACnB,KAAK,IAAI,KAAK,EAAE,KAChB,KAAK,SAAS,cACd,CAAC,KAAK,YACN,OAAO,KAAK,SAAS,SAAS,YAC9B,KAAK,SAAS,KAAK,WAAW,KAC9B,OAAO,KAAK,SAAS,cAAc,UAEnC,OAAO,iDAAiD;IAE1D,KAAK,IAAI,KAAK,EAAE;IAChB,QAAQ,IAAI,KAAK,IAAI,KAAK,SAAS,IAAI;GACzC;EACF;EACA,IAAI,QAAQ,SAAS,QAAQ;GAC3B,MAAM,KAAK,QAAQ;GACnB,IAAI,OAAO,OAAO,YAAY,CAAC,QAAQ,IAAI,EAAE,GAC3C,OAAO,uEAAuE;GAEhF,IAAI,QAAQ,SAAS,KAAA,KAAa,QAAQ,SAAS,QAAQ,IAAI,EAAE,GAC/D,OAAO,gDAAgD;GAEzD,QAAQ,OAAO,EAAE;EACnB,OAAO,IAAI,QAAQ,iBAAiB,KAAA,GAClC,OAAO,6CAA6C;OAC/C,IAAI,QAAQ,OAAO,KAAK,QAAQ,eAAe,KAAA,GACpD,OAAO,6DAA6D;CAExE;CACA,IAAI,QAAQ,OAAO,GAAG,OAAO,wCAAwC;AACvE;;AAGA,SAAS,kBAAkB,OAA8B;CACvD,MAAM,OAAO,MAAM,QAAQ,MAAoB,EAAE,SAAS,KAAK;CAC/D,MAAM,OAAO,KAAK,KAAK,SAAS;CAChC,IAAI,CAAC,MAAM,OAAO,CAAC;CACnB,MAAM,WAA0B,CAAC,GAAG,KAAK,QAAQ;CACjD,IAAI,KAAK,WAAW,KAAA,KAAa,KAAK,WAAW,IAC/C,SAAS,KAAK;EAAE,MAAM;EAAa,SAAS,KAAK;CAAO,CAAC;CAE3D,OAAO;AACT;AAgBA,MAAM,gBAAgE;CACpE,SAAS;CACT,QAAQ;CACR,UAAU;AACZ;;;;;;;;;AAUA,SAAS,iBAAiB,QAAyB;CACjD,IAAI,aAAa,MAAM,MAAM,KAAA,GAC3B,MAAM,IAAI,gBAAgB,+BAA+B,OAAO,MAAM,wBAAwB;AAElG;AAEA,MAAM,YAAY,UAChB,OAAO,UAAU,YAAY,UAAU;;;;;;;;;;;;;;;;;;;;;;;;;;;AAqCzC,MAAM,oBAA+C;CACnD;EACE,OAAO;EACP,UAAU,QAAQ,SAAS,IAAI,cAAc,KAAK,OAAO,IAAI,eAAe,SAAS;EACrF,QACE;CACJ;CACA;EACE,OAAO;EACP,UAAU,QAAQ,SAAS,IAAI,UAAU;EACzC,QACE;CACJ;CACA;EACE,OAAO;EACP,UAAU,QAAQ,CAAC,SAAS,IAAI,UAAU,KAAK,OAAO,IAAI,WAAW,UAAU;EAC/E,QAAQ;CACV;CACA;EACE,OAAO;EACP,UAAU,QAAQ,CAAC,SAAS,IAAI,UAAU,KAAK,OAAO,IAAI,WAAW,WAAW;EAChF,QAAQ;CACV;CACA;EACE,OAAO;EACP,UAAU,QAAQ,SAAS,IAAI,OAAO;EACtC,QACE;CACJ;CACA;EACE,OAAO;EAGP,UAAU,QAAQ,CAAC,SAAS,IAAI,OAAO,KAAK,SAAS,IAAI,QAAQ,GAAG;EACpE,QACE;CACJ;CACA;EACE,OAAO;EACP,UAAU,QAAQ,OAAO,IAAI,oBAAoB;EACjD,QACE;CACJ;CACA;EACE,OAAO;EACP,UAAU,QAAQ,OAAO,IAAI,eAAe,YAAY,IAAI,WAAW,SAAS;EAChF,QACE;CACJ;AACF;;;;;;;;;;;;;AAcA,SAAgB,kBAAkB,QAA6B;CAC7D,MAAM,MAAM;CACZ,OAAO,kBAAkB,QAAQ,UAAU,CAAC,MAAM,QAAQ,GAAG,CAAC,CAAC,CAAC,KAC7D,UAAU,GAAG,MAAM,MAAM,eAAe,MAAM,QACjD;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BA,SAAS,sBAAsB,QAAyB;CACtD,MAAM,UAAU,kBAAkB,MAAM;CACxC,IAAI,QAAQ,WAAW,GAAG;CAC1B,MAAM,IAAI,gBAAgB,+BAA+B,OAAO,MAAM,IAAI,QAAQ,KAAK,MAAM,GAAG;AAClG;AAEA,MAAM,iBAA8D;CAClE,QAAQ;CACR,KAAK;CACL,SAAS;AACX;AAEA,SAAS,SACP,QACA,OACA,UACA,SACA,YACA,KACmB;CAUnB,sBAAsB,MAAM;CAI5B,iBAAiB,MAAM;CAGvB,MAAM,eAAe,oBAAoB,MAAM;CAC/C,MAAM,aAAa,OAAO,eAAe,SAAS;CAClD,MAAM,kBAAkB,OAAO;CAC/B,MAAM,cAAc,oBAAoB,eAAe,oBAAoB;CAC3E,MAAM,cAAc,oBAAoB,eAAe,oBAAoB;CAC3E,MAAM,gBACJ,oBAAoB,YACpB,oBAAoB,eACpB,oBAAoB,eACf,OAAO,yBAAyB,aAAa,oBAC9C;CAIN,OAAO,aACL;EACE,QAAQ;EACR,YAAY,OAAO;EACnB,mBAAmB;EACnB,QAAQ,OAAO;EACf,eAAe,OAAO;EACtB,cAAc,OAAO;EACrB,YAAY;EACZ,iBAAiB;EACjB,MAAM,QAAQ,QAAQ;EACtB,MAAM;GACJ,OAAO,QAAQ,SAAS,OAAO;GAC/B,aAAa,OAAO;GACpB,OAAO,eAAe,OAAO;GAC7B,MAAM,OAAO;GACb,KAAK;EACP;EACA,QAAQ;GACN,SAAS;GACT,iBAAiB;GACjB,OAAO,OAAO;GACd,UAAU;GACV,gBAAgB,OAAO;GACvB,aAAa,OAAO;GACpB,aAAa,OAAO;GACpB,uBAAuB,OAAO,cAAc,UAAU;GACtD,UAAU;EACZ;EACA;EACA,WAAW,CAAC;EACZ,GAAI,MAAM,SAAS,IAAI,EAAE,MAAM,IAAI,CAAC;EACpC,SAAS;GACP,GAAG;GACH,eAAe,cAAc,YAAY,MAAM;GAC/C,SAAS;GAUT,SAAS,EAAE,GAAG,OAAO,QAAQ,IAAI;GACjC,cAAc;GACd,cAAc;GACd,OAAO;EACT;EACA,MAAM;GACJ,KAAK,aAAa,OAAO,OAAO;GAChC,WAAW,OAAO,WAAW;GAC7B,YAAY,OAAO,WAAW;GAC9B,kBAAkB,OAAO,WAAW,aAAa;GACjD,YAAY,OAAO,WAAW,UAAU;GACxC,aAAa,OAAO,WAAW,cAAc;GAC7C,QAAQ,KAAK,MAAM,OAAO,SAAS,GAAI;EACzC;EACA,WAAW;GAAE,YAAY;GAAM,SAAS;GAAM,gBAAgB;EAAK;EACnE,YAAY;GACV,aAAa;GACb,SAAS;GACT,GAAI,QAAQ,KAAA,IAAY,EAAE,IAAI,IAAI,CAAC;EACrC;CACF,GACA,+BAA+B,OAAO,OACxC;AACF;;;;;;;AAQA,eAAsB,gBACpB,SACA,OACA,UAA8B,CAAC,GACH;CAC5B,MAAM,QAAQ,QAAQ,WAAW,MAAM;CACvC,MAAM,cAAc,QAAQ,MAAM,qBAAK,IAAI,KAAK,EAAA,CAAG,YAAY;CAC/D,MAAM,OAA4B,CAAC;CACnC,MAAM,gBAA0B,CAAC;CACjC,KAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,aAAa,MAAM,gBAAgB,OAAO,OAAO,KAAK;EAC5D,MAAM,WAAW,QAAQ,aAAa,MAAM,QAAQ,WAAW,OAAO,KAAK,IAAI,KAAA;EAC/E,IAAI,QAAQ,eAAe,CAAC,MAAM,QAAQ,QAAQ,KAAK,SAAS,WAAW,IACzE,MAAM,IAAI,gBACR,+BAA+B,OAAO,MAAM,0BAC9C;EAEF,IAAI,WAAW,MAAM,WAAW,KAAK,CAAC,QAAQ,YAAY;GACxD,cAAc,KAAK,OAAO,KAAK;GAC/B,KAAK,KACH,SAAS,QAAQ,CAAC,GAAG,CAAC,GAAG,SAAS,YAAY,wCAAwC,CACxF;GACA;EACF;EACA,IAAI,QAAQ,WAAW,MAAM,KAAK,MAAM,YAAY,EAAE,MAAM,KAAK,CAAC;EAClE,IAAI,QAAQ,aAAa,KAAA,KAAa,MAAM,SAAS,QAAQ,UAAU;GAGrE,MAAM,OAAO,KAAK,KAAK,QAAQ,WAAW,CAAC;GAC3C,MAAM,OAAO,QAAQ,WAAW;GAChC,QAAQ,CAAC,GAAG,MAAM,MAAM,GAAG,IAAI,GAAG,GAAG,MAAM,MAAM,MAAM,SAAS,IAAI,CAAC;EACvE;EACA,MAAM,QAAQ,WAAW,MAAM,KAAK,MAAM,EAAE,IAAI;EAChD,MAAM,WAAW,YAAY,kBAAkB,KAAK;EACpD,mBAAmB,UAAU,OAAO,KAAK;EACzC,MAAM,eAAe,SAAS,KAAK,YAAY,eAAe,SAAS,KAAK,CAAC;EAE7E,mBACE,aAAa,QAAQ,YAAY,QAAQ,sBAAsB,IAAI,GACnE,OAAO,KACT;EACA,MAAM,MACJ,aAAa,WAAW,IAAI,uDAAuD,KAAA;EACrF,KAAK,KAAK,SAAS,QAAQ,OAAO,cAAc,SAAS,YAAY,GAAG,CAAC;CAC3E;CACA,OAAO;EAAE;EAAM;CAAc;AAC/B"}
|
package/dist/mint-CJFSYBIU.d.ts
DELETED
|
@@ -1,285 +0,0 @@
|
|
|
1
|
-
import { a as RunRecord } from "./run-record-xEULpgML.js";
|
|
2
|
-
import { s as TraceStore } from "./store-C4avFjIi.js";
|
|
3
|
-
import { S as ToolDef, a as MintedRolloutLine, b as RolloutStep, m as RolloutLine, t as ChatMessage, v as RolloutRole, y as RolloutSplit } from "./schema-BzWDXhOR.js";
|
|
4
|
-
//#region src/rollout/exporters.d.ts
|
|
5
|
-
/**
|
|
6
|
-
* The gate's two claims, which travel TOGETHER on every emitted row.
|
|
7
|
-
*
|
|
8
|
-
* `realness_gated` alone is ambiguous, and the ambiguity is exploitable:
|
|
9
|
-
* `false` reads as "we screened it and nothing fired", so a producer that has no
|
|
10
|
-
* screen at all emitted rows indistinguishable from screened-clean ones, and
|
|
11
|
-
* every consumer of the published dataset read them as clean. The second field
|
|
12
|
-
* is what separates the two claims, and it only removes the ambiguity if it
|
|
13
|
-
* reaches the WIRE — for a round it existed on `RolloutOutcome` and on no
|
|
14
|
-
* exported row shape at all, which left the published rows exactly as ambiguous
|
|
15
|
-
* as before.
|
|
16
|
-
*
|
|
17
|
-
* So there is one helper and every row shape spreads it. A row that states one
|
|
18
|
-
* claim without the other is not constructible by copying the pattern, and
|
|
19
|
-
* `exporters.test.ts` walks every emitted shape to prove none does.
|
|
20
|
-
*/
|
|
21
|
-
interface RealnessLabels {
|
|
22
|
-
/** The screen's VERDICT: the run faked its success signal. */
|
|
23
|
-
realness_gated: boolean;
|
|
24
|
-
/**
|
|
25
|
-
* Whether a screen RAN at all. `true` = it ran, so `realness_gated` is its
|
|
26
|
-
* verdict. `false` = the producer declares it has none. `null` = not stated
|
|
27
|
-
* (pre-unification producers), which is "unknown" and never "clean".
|
|
28
|
-
*/
|
|
29
|
-
realness_screened: boolean | null;
|
|
30
|
-
}
|
|
31
|
-
interface TrainingExportOptions {
|
|
32
|
-
/** Include held-out evaluation data in training output. Default false. */
|
|
33
|
-
allowHeldOutTrainingData?: boolean;
|
|
34
|
-
/** Require reward to be strictly greater than this value. Default 0. */
|
|
35
|
-
minimumQualityExclusive?: number;
|
|
36
|
-
/**
|
|
37
|
-
* Explicit split selection, replacing the default trainable-split rule.
|
|
38
|
-
* Use this only when producing a deliberately named non-training slice
|
|
39
|
-
* (e.g. `['holdout']` for a holdout-only evaluation bundle).
|
|
40
|
-
*/
|
|
41
|
-
splitFilter?: RolloutSplit[];
|
|
42
|
-
}
|
|
43
|
-
/**
|
|
44
|
-
* What a signed-signal exporter (verifiers, RFT) does with lines that are not
|
|
45
|
-
* clean trainable successes — realness-gated lines above all.
|
|
46
|
-
*
|
|
47
|
-
* - 'exclude' — the default, the same fail-closed policy as every
|
|
48
|
-
* other training export: positive, completed,
|
|
49
|
-
* non-gated rows on a trainable split.
|
|
50
|
-
* - 'zero-and-flag' — keep them, at their non-positive (or null) reward,
|
|
51
|
-
* with `RealnessLabels` on the row. The dataset release
|
|
52
|
-
* sets this per `FORMAT_GATE_DISPOSITION`: in these
|
|
53
|
-
* formats the reward is a signed learning signal, so a
|
|
54
|
-
* gamed trajectory at reward 0 is a correct negative,
|
|
55
|
-
* and dropping it would bias the negative population
|
|
56
|
-
* toward honest failures and leave a trainer no example
|
|
57
|
-
* of gaming being penalized. The split policy is NOT
|
|
58
|
-
* relaxed: held-out lines still need the named opt-in.
|
|
59
|
-
*
|
|
60
|
-
* SFT deliberately has no such option — an SFT row is an imitation target and
|
|
61
|
-
* a gamed trajectory must never appear in one at any weight.
|
|
62
|
-
*/
|
|
63
|
-
type GatedLineDisposition = 'exclude' | 'zero-and-flag';
|
|
64
|
-
interface SignedSignalExportOptions extends TrainingExportOptions {
|
|
65
|
-
/** Disposition for non-trainable lines. Default 'exclude'. */
|
|
66
|
-
gatedLines?: GatedLineDisposition;
|
|
67
|
-
}
|
|
68
|
-
type SftExportOptions = TrainingExportOptions;
|
|
69
|
-
interface SftRow {
|
|
70
|
-
messages: ChatMessage[];
|
|
71
|
-
metadata: {
|
|
72
|
-
rollout_id: string;
|
|
73
|
-
run_id: string;
|
|
74
|
-
candidate_id: string | null;
|
|
75
|
-
instance_id: string;
|
|
76
|
-
reward: number;
|
|
77
|
-
} & RealnessLabels;
|
|
78
|
-
}
|
|
79
|
-
/**
|
|
80
|
-
* Supervised fine-tune rows: the completed conversation of each qualifying
|
|
81
|
-
* line. Fail-closed filters: trainable split only (never holdout/canary unless
|
|
82
|
-
* an explicit `splitFilter` names a non-training slice),
|
|
83
|
-
* reward strictly above `minimumQualityExclusive` (default 0), realness-gated
|
|
84
|
-
* lines never qualify, gap lines carry no trainable content, and
|
|
85
|
-
* copied-context turns are dropped from the transcript (Harbor ATIF RFC 0001
|
|
86
|
-
* rule 7 — see `ChatMessage.is_copied_context`).
|
|
87
|
-
*
|
|
88
|
-
* `realness_gated` is therefore always `false` on an emitted row. It is carried
|
|
89
|
-
* anyway: an SFT row is a pure imitation target, so the row states its realness
|
|
90
|
-
* claims instead of making the reader know the format's policy, and carrying
|
|
91
|
-
* both flags on all four shapes is what lets the release accounting measure
|
|
92
|
-
* every config with one rule rather than skipping the one whose row shape
|
|
93
|
-
* happened to omit the field.
|
|
94
|
-
*/
|
|
95
|
-
declare function toSftRows(lines: MintedRolloutLine[], options?: SftExportOptions): SftRow[];
|
|
96
|
-
interface RewardRow {
|
|
97
|
-
/** First user turn — the task prompt. */
|
|
98
|
-
prompt: string;
|
|
99
|
-
steps: RolloutStep[];
|
|
100
|
-
reward: number;
|
|
101
|
-
metadata: {
|
|
102
|
-
rollout_id: string;
|
|
103
|
-
run_id: string;
|
|
104
|
-
candidate_id: string | null;
|
|
105
|
-
instance_id: string;
|
|
106
|
-
split: RolloutSplit;
|
|
107
|
-
} & RealnessLabels;
|
|
108
|
-
}
|
|
109
|
-
/**
|
|
110
|
-
* Reward-labeled rows for completed, positive-quality training runs.
|
|
111
|
-
*/
|
|
112
|
-
declare function toRewardRows(lines: MintedRolloutLine[], options?: TrainingExportOptions): RewardRow[];
|
|
113
|
-
interface VerifiersTokenUsage {
|
|
114
|
-
input_tokens: number | null;
|
|
115
|
-
output_tokens: number | null;
|
|
116
|
-
reasoning_tokens: number | null;
|
|
117
|
-
cache_read_tokens: number | null;
|
|
118
|
-
cache_write_tokens: number | null;
|
|
119
|
-
}
|
|
120
|
-
interface VerifiersRolloutOutput {
|
|
121
|
-
/** Messages through the last turn BEFORE the first assistant turn. */
|
|
122
|
-
prompt: ChatMessage[];
|
|
123
|
-
/** The first assistant turn onward — what the policy produced. */
|
|
124
|
-
completion: ChatMessage[];
|
|
125
|
-
reward: number | null;
|
|
126
|
-
metrics: Record<string, unknown>;
|
|
127
|
-
tool_defs: ToolDef[];
|
|
128
|
-
token_usage: VerifiersTokenUsage;
|
|
129
|
-
info: {
|
|
130
|
-
task: RolloutLine['task'];
|
|
131
|
-
policy: RolloutLine['policy'];
|
|
132
|
-
rollout_id: string;
|
|
133
|
-
run_id: string;
|
|
134
|
-
experiment_id: string | null;
|
|
135
|
-
candidate_id: string | null;
|
|
136
|
-
generation: number | null;
|
|
137
|
-
candidate_index: number | null;
|
|
138
|
-
role: RolloutLine['role'];
|
|
139
|
-
} & RealnessLabels;
|
|
140
|
-
}
|
|
141
|
-
declare function toVerifiersRolloutOutput(line: MintedRolloutLine): VerifiersRolloutOutput;
|
|
142
|
-
declare function toVerifiersRolloutOutputs(lines: MintedRolloutLine[], options?: SignedSignalExportOptions): VerifiersRolloutOutput[];
|
|
143
|
-
interface RftItem {
|
|
144
|
-
/** Prompt turns only — the graded completion is re-sampled during RFT. */
|
|
145
|
-
messages: ChatMessage[];
|
|
146
|
-
/** Verdict/label fields the grader references as item.reference.* */
|
|
147
|
-
reference: {
|
|
148
|
-
reward: number | null;
|
|
149
|
-
reward_source: string | null;
|
|
150
|
-
verdict: unknown;
|
|
151
|
-
instance_id: string;
|
|
152
|
-
suite: string;
|
|
153
|
-
split: RolloutSplit;
|
|
154
|
-
rollout_id: string;
|
|
155
|
-
} & RealnessLabels;
|
|
156
|
-
}
|
|
157
|
-
declare function toRftItem(line: MintedRolloutLine): RftItem;
|
|
158
|
-
/** RFT needs a real prompt: lines whose transcript starts with prompt turns. */
|
|
159
|
-
declare function toRftItems(lines: MintedRolloutLine[], options?: SignedSignalExportOptions): RftItem[];
|
|
160
|
-
declare function toJsonl(rows: ReadonlyArray<unknown>): string;
|
|
161
|
-
//#endregion
|
|
162
|
-
//#region src/rollout/reward.d.ts
|
|
163
|
-
/**
|
|
164
|
-
* Which split's score wins when a record carries both. `'holdout'` is the
|
|
165
|
-
* canonical "real signal" default; `'search'` exists because some callers
|
|
166
|
-
* deliberately score on the search split when both are present.
|
|
167
|
-
*/
|
|
168
|
-
type ScorePreference = 'holdout' | 'search';
|
|
169
|
-
/** Only the outcome is read, so every accessor here accepts anything carrying one. */
|
|
170
|
-
type Scored = Pick<RunRecord, 'outcome'>;
|
|
171
|
-
/** True when the authenticity gate flagged the run as gamed (`realness.gated`). */
|
|
172
|
-
declare function isRealnessGated(record: Scored): boolean;
|
|
173
|
-
/**
|
|
174
|
-
* The RAW score recorded on ONE split, with no cross-split fallback and no
|
|
175
|
-
* anti-Goodhart gate.
|
|
176
|
-
*
|
|
177
|
-
* The narrowest of the three raw readers, and the one every split-scoped
|
|
178
|
-
* consumer wants: a per-split report, a promotion gate, or a paired comparison
|
|
179
|
-
* asks "what did this run score on the split I am summarising", and answering
|
|
180
|
-
* it with the other split's number silently mixes populations. `undefined` =
|
|
181
|
-
* that split was never scored.
|
|
182
|
-
*
|
|
183
|
-
* Same warning as `observedScore`: this INCLUDES runs flagged as gamed. Never
|
|
184
|
-
* feed it into training data.
|
|
185
|
-
*/
|
|
186
|
-
declare function observedSplitScore(record: Scored, split: ScorePreference): number | undefined;
|
|
187
|
-
/**
|
|
188
|
-
* The RAW split score the run carries, with NO anti-Goodhart gate applied.
|
|
189
|
-
*
|
|
190
|
-
* INCLUDES RUNS FLAGGED AS GAMED (`outcome.realness.gated === true`); NEVER
|
|
191
|
-
* feed this into training data — a fine-tune that sees it learns from gamed
|
|
192
|
-
* successes. It is exported anyway because analysis, reporting, and
|
|
193
|
-
* reward-hacking detection legitimately need the ungated number: forcing a
|
|
194
|
-
* gamed run to 0 collapses the proxy signal toward ground truth and makes a
|
|
195
|
-
* detector report "clean" on exactly the population that is being gamed.
|
|
196
|
-
*
|
|
197
|
-
* Returns `undefined` when the record carries neither score — an unscored run
|
|
198
|
-
* is a labeled gap, not a measured zero, and each caller picks its own
|
|
199
|
-
* sentinel (`?? 0`, `?? null`, skip, throw). Non-finite values are returned
|
|
200
|
-
* as-is; callers that care keep their own `Number.isFinite` guard.
|
|
201
|
-
*/
|
|
202
|
-
declare function observedScore(record: Scored, prefer?: ScorePreference): number | undefined;
|
|
203
|
-
/** Which split actually carried the score, or that none did. */
|
|
204
|
-
type ScoreOrigin = 'holdout' | 'search' | 'unscored';
|
|
205
|
-
/**
|
|
206
|
-
* Where `observedScore` / `trainingScore` read their number from — the
|
|
207
|
-
* provenance label a rollout line's `reward_source` is built from, and the
|
|
208
|
-
* only supported way to ask "was this run scored at all" without respelling
|
|
209
|
-
* the field access.
|
|
210
|
-
*/
|
|
211
|
-
declare function scoreOrigin(record: Scored, prefer?: ScorePreference): ScoreOrigin;
|
|
212
|
-
/**
|
|
213
|
-
* The GATED score — the only derivation allowed to reach training data.
|
|
214
|
-
*
|
|
215
|
-
* A realness-gated run scores 0 no matter what it claims, so a fine-tune
|
|
216
|
-
* cannot learn from a gamed success. An unscored run stays `undefined` (a
|
|
217
|
-
* labeled gap), keeping "we never measured this" distinct from "we measured
|
|
218
|
-
* zero"; callers that need a number apply their own sentinel.
|
|
219
|
-
*/
|
|
220
|
-
declare function trainingScore(record: Scored, prefer?: ScorePreference): number | undefined;
|
|
221
|
-
/**
|
|
222
|
-
* `{reward, gated}` as written onto a minted `RolloutLine` — `trainingScore`
|
|
223
|
-
* plus the flag itself, so the gate travels into the exported row and a
|
|
224
|
-
* downstream filter can drop or down-weight the line.
|
|
225
|
-
*
|
|
226
|
-
* An unscored record yields `reward: null`, matching the schema's "no verdict
|
|
227
|
-
* exists — a labeled gap, never 0" rule. It previously collapsed to 0, which
|
|
228
|
-
* made a run nobody graded indistinguishable from one graded as a total
|
|
229
|
-
* failure, and taught any trainer reading the row that the trajectory was bad.
|
|
230
|
-
* A gated run still yields 0, because that IS a verdict: the gate decided.
|
|
231
|
-
*/
|
|
232
|
-
declare function trainingReward(record: Scored): {
|
|
233
|
-
reward: number | null;
|
|
234
|
-
gated: boolean;
|
|
235
|
-
};
|
|
236
|
-
//#endregion
|
|
237
|
-
//#region src/rollout/mint.d.ts
|
|
238
|
-
/** Redactor for transcript/step payload text (secrets, PII), not opaque identity. */
|
|
239
|
-
type RolloutScrubber = (text: string) => string;
|
|
240
|
-
interface MintRolloutOptions {
|
|
241
|
-
scrub?: RolloutScrubber;
|
|
242
|
-
/**
|
|
243
|
-
* Full canonical capture, including turns no longer in the model's context.
|
|
244
|
-
* When supplied, missing/empty capture is refused, never replaced with a
|
|
245
|
-
* span summary. Use the execution owner's capture or the existing harness
|
|
246
|
-
* readers; reviewer traces and compacted histories are not full capture.
|
|
247
|
-
*/
|
|
248
|
-
messagesOf?: (runId: string) => readonly ChatMessage[] | undefined | Promise<readonly ChatMessage[] | undefined>;
|
|
249
|
-
/** Cap steps per line (longest runs first drop middle steps). Default: no cap. */
|
|
250
|
-
maxSteps?: number;
|
|
251
|
-
/** Role recorded on every minted line. Default 'agent' (a solo eval run). */
|
|
252
|
-
role?: RolloutRole;
|
|
253
|
-
/** Task suite label. Default: the record's `experimentId`. */
|
|
254
|
-
suite?: string;
|
|
255
|
-
/** Injected clock for deterministic output. */
|
|
256
|
-
now?: () => Date;
|
|
257
|
-
}
|
|
258
|
-
interface MintRolloutResult {
|
|
259
|
-
rows: MintedRolloutLine[];
|
|
260
|
-
/** runIds that had a RunRecord but no spans — emitted as gap lines AND listed here. */
|
|
261
|
-
missingTraces: string[];
|
|
262
|
-
}
|
|
263
|
-
/**
|
|
264
|
-
* Why a record cannot be minted, one entry per missing field, empty when it can.
|
|
265
|
-
*
|
|
266
|
-
* Exported so a caller can partition a whole ledger — "which of my 2742 records
|
|
267
|
-
* predate 0.126" — without catching an exception per record, and without
|
|
268
|
-
* re-deriving the field list on their side. A re-derived list is a list that
|
|
269
|
-
* drifts from the door it is supposed to predict.
|
|
270
|
-
*
|
|
271
|
-
* Takes a `RunRecord` because that is what the caller holds and what the
|
|
272
|
-
* compiler agrees they hold. The type is precisely the thing that is wrong, so
|
|
273
|
-
* the checks read the record as the untyped bag it actually is on disk.
|
|
274
|
-
*/
|
|
275
|
-
declare function unmintableReasons(record: RunRecord): string[];
|
|
276
|
-
/**
|
|
277
|
-
* Join RunRecords with their traces into canonical rollout lines. Records
|
|
278
|
-
* without spans or an explicit capture are emitted as labeled gap lines and reported in
|
|
279
|
-
* `missingTraces`. Execution-only records without a task score are rejected
|
|
280
|
-
* because a missing training label is not a zero reward.
|
|
281
|
-
*/
|
|
282
|
-
declare function mintRolloutRows(records: RunRecord[], store: TraceStore, options?: MintRolloutOptions): Promise<MintRolloutResult>;
|
|
283
|
-
//#endregion
|
|
284
|
-
export { toRewardRows as C, toVerifiersRolloutOutput as D, toSftRows as E, toVerifiersRolloutOutputs as O, toJsonl as S, toRftItems as T, SftExportOptions as _, unmintableReasons as a, VerifiersRolloutOutput as b, isRealnessGated as c, scoreOrigin as d, trainingReward as f, RftItem as g, RewardRow as h, mintRolloutRows as i, observedScore as l, RealnessLabels as m, MintRolloutResult as n, ScoreOrigin as o, trainingScore as p, RolloutScrubber as r, ScorePreference as s, MintRolloutOptions as t, observedSplitScore as u, SftRow as v, toRftItem as w, VerifiersTokenUsage as x, TrainingExportOptions as y };
|
|
285
|
-
//# sourceMappingURL=mint-CJFSYBIU.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mint-CJFSYBIU.d.ts","names":[],"sources":["../src/rollout/exporters.ts","../src/rollout/reward.ts","../src/rollout/mint.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;UAiDiB;;EAEf;;;;;;EAMA;;UAiBe;;EAEf;;EAEA;;;;;;EAMA,cAAc;;;;;;;;;;;;;;;;;;;;;;KAuBJ;UAEK,kCAAkC;;EAEjD,aAAa;;KAGH,mBAAmB;UAEd;EACf,UAAU;EACV;IACE;IACA;IACA;IACA;IACA;MACE;;;;;;;;;;;;;;;;;;iBAmBU,UAAU,OAAO,qBAAqB,UAAS,mBAAwB;UAyBtE;;EAEf;EACA,OAAO;EACP;EACA;IACE;IACA;IACA;IACA;IACA,OAAO;MAUL;;;;;iBAMU,aACd,OAAO,qBACP,UAAS,wBACR;UAgCc;EACf;EACA;EACA;EACA;EACA;;UAGe;;EAEf,QAAQ;;EAER,YAAY;EACZ;EACA,SAAS;EACT,WAAW;EACX,aAAa;EACb;IACE,MAAM;IACN,QAAQ;IACR;IACA;IACA;IACA;IACA;IACA;IACA,MAAM;MACJ;;iBASU,yBAAyB,MAAM,oBAAoB;iBA+BnD,0BACd,OAAO,qBACP,UAAS,4BACR;UAoBc;;EAEf,UAAU;;EAEV;IACE;IACA;IACA;IACA;IACA;IACA,OAAO;IACP;MACE;;iBAGU,UAAU,MAAM,oBAAoB;;iBAmBpC,WACd,OAAO,qBACP,UAAS,4BACR;iBAkBa,QAAQ,MAAM;;;;;;;;KC3VlB;;KAGP,SAAS,KAAK;;iBAGH,gBAAgB,QAAQ;;;;;;;;;;;;;;iBAiBxB,mBAAmB,QAAQ,QAAQ,OAAO;;;;;;;;;;;;;;;;iBAmB1C,cACd,QAAQ,QACR,SAAQ;;KAOE;;;;;;;iBAQI,YAAY,QAAQ,QAAQ,SAAQ,kBAA8B;;;;;;;;;iBAelE,cACd,QAAQ,QACR,SAAQ;;;;;;;;;;;;iBAiBM,eAAe,QAAQ;EAAW;EAAuB;;;;;KCvE7D,mBAAmB;UAEd;EACf,QAAQ;;;;;;;EAOR,cACE,2BACY,4BAA4B,iBAAiB;;EAE3D;;EAEA,OAAO;;EAEP;;EAEA,YAAY;;UAGG;EACf,MAAM;;EAEN;;;;;;;;;;;;;;iBA4Pc,kBAAkB,QAAQ;;;;;;;iBA+JpB,gBACpB,SAAS,aACT,OAAO,YACP,UAAS,qBACR,QAAQ"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"redact-DPoH1-OW.d.ts","names":[],"sources":["../src/trace/redact.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAsCY;cAEC,6BAA6B;;KAG9B;;;;;;cAOC;UAEI;;EAEf;EACA,UAAU;EACV;EACA;;UAGe;EACf;EACA,SAAS;;EAET;EACA,YAAY;EACZ;EACA,UAAU;;UAGK;;EAEf,UAAU;;EAEV;;;;;;;EAOA;;EAEA,SAAS;;UAGM;;EAEf,UAAU;;EAEV;;KAGU;UAEK;EACf,UAAU;EACV;EACA;EACA;;EAEA;;UAGe;EACf,QAAQ;EACR,SAAS;EACT,UAAU;;EAEV;;;;;;;iBA6Dc,aAAa;;;;;;iBAmNb,YAAY,cAAc;;iBAyJ1B,iBAAiB;;iBAQjB,mBAAmB;;;;;;iBAsQnB,OAAO,GACrB,OAAO,GACP,UAAS;EACN,OAAO;EAAG,QAAQ;;;;;;iBAUP,WAAW,cAAc,UAAS;;iBAMlC,qBAAqB,UAAS,mBAA+B;;;;;;;iBAmH7D,kBAAkB,gBAAgB,UAAS,gBAAqB;;;;;iBAoBhE,gBACd,SAAS,kBACT,mBAAmB,sBACnB,iCACC;;iBAsCa,aAAa,SAAS;;;;;;iBAStB,eAAe,GAC7B,OAAO,GACP,UAAS;EACN,OAAO;EAAG,QAAQ;EAAiB,SAAS"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"redact-DxwlUAvh.js","names":[],"sources":["../src/trace/redact.ts"],"sourcesContent":["/**\n * The redaction core: the one place that decides what must not leave a\n * process or a machine. Every trace writer, exporter, uploader and share path\n * calls it.\n *\n * Three profiles, each a superset of the one before:\n * - `default` removes credentials, found by key name and by value shape, and\n * personal data (email addresses, card numbers, SSNs, phone numbers).\n * Content is kept. Strings are capped at 1 MiB.\n * - `share` also removes phone, IP and postal-address fields, replaces\n * account, user, tenant, session and request identifiers with pseudonyms\n * that stay consistent within one call, and caps strings at 64 KiB.\n * - `strict` also removes prompt, completion, message, tool payload and other\n * raw-content fields and embedded media, and caps strings at 4 KiB.\n *\n * A string that holds a credential is replaced whole with\n * `[REDACTED:<detector>]`. Replacing only the matched span can leave part of a\n * secret behind when a pattern's boundary is wrong. Personal data inside prose\n * is replaced in place, so the rest of the text stays readable.\n *\n * `assessShareSafety` reads a value without changing it and returns SAFE,\n * SAFE_WITH_WARNINGS, UNSAFE or UNKNOWN. UNKNOWN means the scanner could not\n * read part of the value (a cycle, binary data, nesting past the depth limit);\n * it is never reported as safe. `redactForShare` redacts and then assesses the\n * redacted output, so a detector gap shows up as UNSAFE instead of a leak.\n *\n * Reports and findings carry paths, categories and detector ids, never the\n * values they matched.\n *\n * The key classifier and the credential detector list are adapted from\n * agent-inspect (MIT, commit 3c3cbeda). `scripts/redaction-corpus.json` holds\n * the must-flag and must-not-flag cases, including the copied agent-inspect\n * safety corpus with its MIT notice; `pnpm redaction:corpus` runs this module\n * over every case.\n */\n\nimport { createHmac, randomBytes } from 'node:crypto'\n\nexport type RedactionProfile = 'default' | 'share' | 'strict'\n\nexport const REDACTION_PROFILES: readonly RedactionProfile[] = ['default', 'share', 'strict']\n\n/** What a finding is about. `media` is embedded binary content (a `data:` URI) no text detector can read. */\nexport type SafetyCategory = 'credential' | 'personal-data' | 'identifier' | 'raw-content' | 'media'\n\n/**\n * Version of the detector set and profile rules. Stored work that depends on\n * redaction behavior (saved optimizer inputs, per-span redaction stamps) keys\n * on it, so bump it whenever a detector, key list, cap or marker changes.\n */\nexport const REDACTION_VERSION = '2.0.0'\n\nexport interface RedactionFinding {\n /** JSON Pointer (RFC 6901) to the value or key that was changed. */\n path: string\n category: SafetyCategory\n detector: string\n action: 'redacted' | 'pseudonymized' | 'truncated'\n}\n\nexport interface RedactionReport {\n version: string\n profile: RedactionProfile\n /** Values and keys replaced, pseudonymized or truncated. */\n redactionCount: number\n byDetector: Record<string, number>\n truncatedCount: number\n findings: RedactionFinding[]\n}\n\nexport interface RedactOptions {\n /** Default `default`. */\n profile?: RedactionProfile\n /** Lower the profile's per-string cap. A value above the profile cap is ignored. */\n maxStringBytes?: number\n /**\n * Exact values the caller knows are secret (an API key the run was given, a\n * password from the environment). A string or key that contains one, or its\n * base64, base64url or URL-encoded form, is replaced whole with\n * `[REDACTED:known-secret]`, whatever shape the value has.\n */\n knownSecrets?: readonly string[]\n /** Add this call's findings to an existing report instead of a new one. */\n report?: RedactionReport\n}\n\nexport interface AssessOptions {\n /** Default `default`. */\n profile?: RedactionProfile\n /** As in `RedactOptions`: any surviving occurrence makes the verdict UNSAFE. */\n knownSecrets?: readonly string[]\n}\n\nexport type ShareSafetyStatus = 'SAFE' | 'SAFE_WITH_WARNINGS' | 'UNSAFE' | 'UNKNOWN'\n\nexport interface ShareSafetyFinding {\n category: SafetyCategory\n detector: string\n severity: 'error' | 'warning'\n count: number\n /** Up to five JSON Pointers where the detector fired. */\n paths: string[]\n}\n\nexport interface ShareSafetyVerdict {\n status: ShareSafetyStatus\n profile: RedactionProfile\n findings: ShareSafetyFinding[]\n /** Parts the scanner could not read, as `<path>: <reason>`. Any entry makes the status UNKNOWN. */\n unreadable: string[]\n}\n\n// ── Profiles ─────────────────────────────────────────────────────────────\n\ninterface ProfileRules {\n maxStringBytes: number\n /** Key categories whose values are removed or pseudonymized. */\n redactKeys: ReadonlySet<'share-personal' | 'identifier' | 'raw-content'>\n redactMedia: boolean\n severity: Record<SafetyCategory, 'error' | 'warning' | undefined>\n}\n\nconst KIB = 1024\n\nconst PROFILE_RULES: Record<RedactionProfile, ProfileRules> = {\n default: {\n maxStringBytes: 1024 * KIB,\n redactKeys: new Set(),\n redactMedia: false,\n severity: {\n credential: 'error',\n 'personal-data': 'warning',\n identifier: undefined,\n 'raw-content': 'warning',\n media: 'warning',\n },\n },\n share: {\n maxStringBytes: 64 * KIB,\n redactKeys: new Set(['share-personal', 'identifier']),\n redactMedia: false,\n severity: {\n credential: 'error',\n 'personal-data': 'error',\n identifier: 'warning',\n 'raw-content': 'warning',\n media: 'warning',\n },\n },\n strict: {\n maxStringBytes: 4 * KIB,\n redactKeys: new Set(['share-personal', 'identifier', 'raw-content']),\n redactMedia: true,\n severity: {\n credential: 'error',\n 'personal-data': 'error',\n identifier: 'warning',\n 'raw-content': 'error',\n media: 'error',\n },\n },\n}\n\n// ── Key classifier ───────────────────────────────────────────────────────\n\n/**\n * Canonical snake_case form of a field name, so camelCase, kebab-case, dotted\n * and snake_case spellings match one list: `userPassword`, `user-password` and\n * `user.password` all become `user_password`.\n */\nexport function normalizeKey(key: string): string {\n return key\n .replace(/([a-z0-9])([A-Z])/g, '$1_$2')\n .toLowerCase()\n .replace(/[^a-z0-9]+/g, '_')\n .replace(/^_+|_+$/g, '')\n}\n\n/** Counts, limits and model vocabulary that contain the word \"token\" but are not credentials. */\nconst NON_CREDENTIAL_TOKEN_KEYS: ReadonlySet<string> = new Set([\n 'tokens',\n 'max_tokens',\n 'min_tokens',\n 'ls_max_tokens',\n 'token_count',\n 'token_limit',\n 'token_budget',\n 'input_tokens',\n 'output_tokens',\n 'total_tokens',\n 'cached_tokens',\n 'prompt_tokens',\n 'completion_tokens',\n 'reasoning_tokens',\n 'bos_token',\n 'eos_token',\n 'pad_token',\n 'unk_token',\n 'sep_token',\n 'cls_token',\n 'mask_token',\n 'stop_token',\n // Pagination cursors are opaque but not credentials.\n 'page_token',\n 'next_page_token',\n 'prev_page_token',\n 'next_token',\n 'continuation_token',\n 'cursor_token',\n 'sync_token',\n])\n\nconst CREDENTIAL_KEYS = [\n 'authorization',\n 'auth',\n 'cookie',\n 'api_key',\n 'apikey',\n 'api_token',\n 'access_key',\n 'secret_key',\n 'private_key',\n 'signing_key',\n 'encryption_key',\n 'master_key',\n 'session_key',\n 'client_secret',\n 'secret',\n 'password',\n 'passwd',\n 'passphrase',\n 'credential',\n 'credentials',\n 'bearer',\n 'dsn',\n 'database_url',\n 'connection_string',\n 'pat',\n] as const\n\n/** Email fields are personal data in every profile. */\nconst EMAIL_KEYS = ['email', 'email_address'] as const\n\n/** Personal-data fields removed by the share and strict profiles. */\nconst SHARE_PERSONAL_KEYS = [\n 'phone',\n 'phone_number',\n 'mobile',\n 'ip',\n 'ip_address',\n 'remote_addr',\n 'street_address',\n 'postal_address',\n 'home_address',\n 'billing_address',\n 'shipping_address',\n 'mailing_address',\n 'date_of_birth',\n 'dob',\n 'ssn',\n] as const\n\n/** Identifiers pseudonymized by the share and strict profiles. */\nconst IDENTIFIER_KEYS = [\n 'user_id',\n 'user_uuid',\n 'account_id',\n 'account_uuid',\n 'customer_id',\n 'tenant_id',\n 'org_id',\n 'organization_id',\n 'organisation_id',\n 'team_id',\n 'enduser_id',\n 'actor_id',\n 'principal_id',\n 'member_id',\n 'session_id',\n 'request_id',\n 'correlation_id',\n 'device_id',\n 'installation_id',\n] as const\n\n/** Fields that carry prompts, model output, tool payloads or other free text. Removed by `strict`. */\nconst RAW_CONTENT_KEYS = [\n 'prompt',\n 'prompts',\n 'system_prompt',\n 'completion',\n 'input',\n 'output',\n 'message',\n 'messages',\n 'content',\n 'text',\n 'thinking',\n 'reasoning',\n 'transcript',\n 'context',\n 'document',\n 'documents',\n 'chunk',\n 'chunks',\n 'retrieval',\n 'query',\n 'args',\n 'arguments',\n 'result',\n 'body',\n 'command',\n 'full_command',\n 'input_value',\n 'output_value',\n 'instructions',\n 'system_instructions',\n 'stdout',\n 'stderr',\n 'stacktrace',\n 'status_description',\n 'input_preview',\n 'output_preview',\n 'current_task',\n 'user_input',\n 'request_text',\n] as const\n\n/**\n * A key matches a list entry exactly or as a suffix (`openai_api_key`,\n * `db_password`, `tool_output`). Prefix compounds (`secret_name`,\n * `password_min_length`, `auth_type`, `content_type`) do not match: they name\n * or describe a value rather than hold it, and flagging them would make the\n * share verdict refuse ordinary traces. Value detectors still catch a\n * credential-shaped value under any key.\n */\nfunction matchesKeyList(normalized: string, keys: readonly string[]): boolean {\n for (const key of keys) {\n if (normalized === key || normalized.endsWith(`_${key}`)) return true\n }\n return false\n}\n\nfunction isTokenCredentialKey(normalized: string): boolean {\n if (NON_CREDENTIAL_TOKEN_KEYS.has(normalized)) return false\n if (normalized === 'token') return true\n // access_token and refreshToken are credentials; max_tokens and input_tokens are counts.\n if (normalized.endsWith('tokens')) return false\n return normalized.endsWith('token')\n}\n\ntype KeyClass = 'credential' | 'email' | 'share-personal' | 'identifier' | 'raw-content'\n\n/** Trace keys repeat on every span, so classifications are cached. */\nconst keyClassCache = new Map<string, KeyClass | null>()\n\nfunction classifyKeyName(key: string): KeyClass | undefined {\n const cached = keyClassCache.get(key)\n if (cached !== undefined) return cached ?? undefined\n const kind = classifyNormalizedKey(normalizeKey(key))\n if (keyClassCache.size >= 10_000) keyClassCache.clear()\n keyClassCache.set(key, kind ?? null)\n return kind\n}\n\nfunction classifyNormalizedKey(normalized: string): KeyClass | undefined {\n if (!normalized || NON_CREDENTIAL_TOKEN_KEYS.has(normalized)) return undefined\n if (isTokenCredentialKey(normalized)) return 'credential'\n if (matchesKeyList(normalized, CREDENTIAL_KEYS)) return 'credential'\n if (matchesKeyList(normalized, EMAIL_KEYS)) return 'email'\n if (matchesKeyList(normalized, SHARE_PERSONAL_KEYS)) return 'share-personal'\n if (matchesKeyList(normalized, IDENTIFIER_KEYS)) return 'identifier'\n if (matchesKeyList(normalized, RAW_CONTENT_KEYS)) return 'raw-content'\n return undefined\n}\n\n/**\n * The safety category a field name puts its value in, or undefined for an\n * ordinary field. Token counts and model limits (`inputTokens`, `max_tokens`,\n * `token_count`) and names such as `author` are not credentials.\n */\nexport function classifyKey(key: string): SafetyCategory | undefined {\n const kind = classifyKeyName(key)\n if (kind === undefined) return undefined\n if (kind === 'credential') return 'credential'\n if (kind === 'identifier') return 'identifier'\n if (kind === 'raw-content') return 'raw-content'\n return 'personal-data'\n}\n\n// ── Value detectors ──────────────────────────────────────────────────────\n\ninterface ValueDetector {\n id: string\n pattern: RegExp\n}\n\n/**\n * High-confidence credential shapes. A hit replaces the whole string. Each\n * pattern is written without the `g` flag so `test` stays stateless.\n */\nconst CREDENTIAL_DETECTORS: readonly ValueDetector[] = [\n { id: 'private-key', pattern: /-----BEGIN [A-Z0-9 ]*PRIVATE KEY-----/ },\n { id: 'jwt', pattern: /\\beyJ[A-Za-z0-9_-]{8,}\\.[A-Za-z0-9_-]{8,}\\.[A-Za-z0-9_-]{8,}/ },\n { id: 'bearer', pattern: /\\bbearer\\s+[A-Za-z0-9._~+/=-]{12,}/i },\n {\n id: 'authorization-header',\n pattern: /^\\s*(?:basic|digest|apikey)\\s+[A-Za-z0-9._~+/=:-]{8,}\\s*$/i,\n },\n { id: 'anthropic-key', pattern: /\\bsk-ant-[A-Za-z0-9_-]{16,}/ },\n { id: 'provider-key', pattern: /\\bsk-(?:proj-|tan-)?[A-Za-z0-9_-]{16,}/ },\n { id: 'stripe-key', pattern: /\\b(?:sk|rk)_(?:live|test)_[A-Za-z0-9_]{10,}/ },\n { id: 'google-api-key', pattern: /\\bAIza[0-9A-Za-z_-]{20,}/ },\n { id: 'github-token', pattern: /\\b(?:gh[pousr]_[A-Za-z0-9_]{20,}|github_pat_[A-Za-z0-9_]{20,})/ },\n { id: 'slack-token', pattern: /\\bxox[abeoprs]-[A-Za-z0-9-]{10,}/ },\n { id: 'aws-access-key', pattern: /\\b(?:AKIA|ASIA)[A-Z0-9]{16}\\b/ },\n { id: 'tangle-capability', pattern: /\\b(?:hubcap_|hct_)[A-Za-z0-9_=-]{8,}/ },\n { id: 'url-credentials', pattern: /\\b[a-z][a-z0-9+.-]*:\\/\\/[^\\s/:@]+:[^\\s/@]+@/i },\n {\n id: 'url-secret-param',\n pattern:\n /[?&](?:access_token|api_key|apikey|auth|token|secret|password|sig|signature|x-amz-signature|x-goog-signature)=[^&\\s\"'#)]{8,}/i,\n },\n {\n // `GH_TOKEN=…`, `--api-key=…`, `\"client_secret\": \"…\"`, `password: …` in free\n // text. The value must mix letters and digits and be at least 12\n // characters, so prose (`token: string`), variable references (`$TOKEN`)\n // and dotenvx ciphertext (`encrypted:…`) pass. The pattern starts at the\n // key word itself, with no identifier prefix, so it scans in linear time.\n id: 'secret-assignment',\n pattern:\n /(?:api[_-]?key|apikey|access[_-]?key|secret[_-]?key|private[_-]?key|secret|token|password|passwd|passphrase|credentials?)[\"']?[ \\t]*[:=][ \\t]*[\"']?(?!encrypted:)(?![$`<[{(%])(?=[^\\s\"'`,;&|]*[0-9])(?=[^\\s\"'`,;&|]*[A-Za-z])[^\\s\"'`,;&|]{12,}/i,\n },\n]\n\n/** Personal-data shapes, replaced in place so the surrounding text survives. */\nconst PERSONAL_DATA_DETECTORS: readonly (ValueDetector & {\n /** A substring the text must contain before the pattern runs. */\n requires?: string\n accept?: (match: string) => boolean\n})[] = [\n {\n id: 'email',\n requires: '@',\n pattern: /\\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Za-z]{2,}\\b/g,\n accept: isEmailAddress,\n },\n {\n // Card numbers: a card-network prefix and a valid Luhn digit, written\n // unbroken (13-19 digits) or in printed groups (4-4-4-4, 4-4-4-4-3, Amex\n // 4-6-5). A word character or hyphen on either side rejects the match, so\n // digits inside UUIDs, dated file names and hyphenated ids never match.\n id: 'card-number',\n pattern:\n /(?<![\\w-])(?:\\d{13,19}|\\d{4}([ -])\\d{4}\\1\\d{4}\\1\\d{4}(?:\\1\\d{3})?|\\d{4}([ -])\\d{6}\\2\\d{5})(?![\\w-])/g,\n accept: (match) => {\n const digits = match.replace(/[ -]/g, '')\n return CARD_PREFIX.test(digits) && luhnValid(digits)\n },\n },\n { id: 'ssn', pattern: /\\b\\d{3}-\\d{2}-\\d{4}\\b/g },\n {\n // Separators are required: a bare ten-digit run is usually an epoch second or an id.\n id: 'phone',\n pattern: /(?:\\+1[-.\\s]?)?(?:\\(\\d{3}\\)\\s?|\\b\\d{3}[-.\\s])\\d{3}[-.\\s]\\d{4}\\b/g,\n },\n]\n\n/** Visa, Mastercard, Amex, Diners, JCB, Discover and UnionPay issuer prefixes. */\nconst CARD_PREFIX = /^(?:4|5[1-5]|2[2-7]|3[04-9]|6(?:011|4[4-9]|5|2))/\n\n/**\n * Top-level domains that are file extensions: `icon@2x.png`, `worker@3.service`\n * and `notes@AGENTS.md` are not addresses.\n */\nconst NON_EMAIL_TLDS: ReadonlySet<string> = new Set([\n 'md',\n 'mdx',\n 'txt',\n 'json',\n 'jsonl',\n 'yaml',\n 'yml',\n 'toml',\n 'lock',\n 'ts',\n 'tsx',\n 'js',\n 'jsx',\n 'mjs',\n 'cjs',\n 'py',\n 'rs',\n 'go',\n 'sh',\n 'css',\n 'html',\n 'svg',\n 'png',\n 'jpg',\n 'jpeg',\n 'gif',\n 'webp',\n 'service',\n 'socket',\n 'timer',\n 'mount',\n 'target',\n])\n\nfunction isEmailAddress(match: string): boolean {\n // `git@github.com:org/repo` is an SSH remote, not a person.\n if (match.toLowerCase().startsWith('git@')) return false\n const tld = match.slice(match.lastIndexOf('.') + 1).toLowerCase()\n return !NON_EMAIL_TLDS.has(tld)\n}\n\nfunction luhnValid(digits: string): boolean {\n if (/^(\\d)\\1+$/.test(digits)) return false\n let sum = 0\n let double = false\n for (let index = digits.length - 1; index >= 0; index -= 1) {\n let digit = digits.charCodeAt(index) - 48\n if (double) {\n digit *= 2\n if (digit > 9) digit -= 9\n }\n sum += digit\n double = !double\n }\n return sum % 10 === 0\n}\n\n/** The first credential detector that matches, or undefined. */\nexport function detectCredential(text: string): string | undefined {\n for (const detector of CREDENTIAL_DETECTORS) {\n if (detector.pattern.test(text)) return detector.id\n }\n return undefined\n}\n\n/** Personal-data detector ids that match somewhere in `text`. */\nexport function detectPersonalData(text: string): string[] {\n const hits: string[] = []\n for (const detector of PERSONAL_DATA_DETECTORS) {\n if (detector.requires && !text.includes(detector.requires)) continue\n detector.pattern.lastIndex = 0\n for (const match of text.matchAll(detector.pattern)) {\n if (!detector.accept || detector.accept(match[0])) {\n hits.push(detector.id)\n break\n }\n }\n }\n return hits\n}\n\n/**\n * Every form in which a known secret can appear in a trace: as written,\n * base64, base64url and URL-encoded. Values shorter than 4 characters are\n * ignored; they would match ordinary text.\n */\nfunction knownSecretForms(values: readonly string[] | undefined): string[] {\n const forms = new Set<string>()\n for (const value of values ?? []) {\n if (value.length < 4) continue\n const bytes = Buffer.from(value, 'utf8')\n forms.add(value)\n forms.add(bytes.toString('base64'))\n forms.add(bytes.toString('base64url'))\n forms.add(encodeURIComponent(value))\n }\n // Longest first, so a secret that contains another is reported as itself.\n return [...forms].sort((left, right) => right.length - left.length)\n}\n\nconst BASE64_TEXT = /^[A-Za-z0-9+/_-]+={0,2}$/\n\nfunction containsKnownSecret(text: string, forms: readonly string[]): boolean {\n if (forms.length === 0) return false\n if (forms.some((form) => text.includes(form))) return true\n // A whole-string base64 payload can hold a secret at any byte offset.\n if (text.length < 8 || !BASE64_TEXT.test(text)) return false\n const decoded = Buffer.from(\n text,\n text.includes('-') || text.includes('_') ? 'base64url' : 'base64',\n ).toString('utf8')\n return forms.some((form) => decoded.includes(form))\n}\n\n/** The detector that marks `text` as a credential: a known secret first, then the value shapes. */\nfunction credentialIn(text: string, knownSecrets: readonly string[]): string | undefined {\n return containsKnownSecret(text, knownSecrets) ? 'known-secret' : detectCredential(text)\n}\n\nconst MARKER_PREFIX = '[REDACTED:'\n\nfunction marker(detector: string): string {\n return `${MARKER_PREFIX}${detector}]`\n}\n\n/**\n * Placeholders other tools write for a removed value. A credential field that\n * holds one of these is already redacted and is not flagged again.\n */\nconst REDACTED_PLACEHOLDER =\n /^(?:\\[(?:REDACTED|redacted)(?::[^\\]]*)?\\]|\\[ID:[0-9a-f]{12}\\]|\\*{3,}|<redacted>|«redacted»|\\[HASH:[0-9a-f]{8,}\\])$/\n\nfunction isPlaceholder(value: string): boolean {\n return value === '' || REDACTED_PLACEHOLDER.test(value)\n}\n\nconst DATA_URI = /^data:[\\w.+-]+\\/[\\w.+-]+(?:;[\\w.+-]+=[\\w.+-]+)*;base64,/i\n\n// ── Redaction ────────────────────────────────────────────────────────────\n\nconst MAX_DEPTH = 64\n\ninterface WalkState {\n rules: ProfileRules\n maxStringBytes: number\n knownSecrets: string[]\n report: RedactionReport\n pseudonymKey: Buffer\n seen: WeakSet<object>\n}\n\nfunction record(\n state: WalkState,\n path: string,\n category: SafetyCategory,\n detector: string,\n action: RedactionFinding['action'],\n): void {\n const report = state.report\n if (action === 'truncated') report.truncatedCount += 1\n report.redactionCount += 1\n report.byDetector[detector] = (report.byDetector[detector] ?? 0) + 1\n report.findings.push({ path, category, detector, action })\n}\n\nfunction pointer(parent: string, key: string | number): string {\n return `${parent}/${String(key).replace(/~/g, '~0').replace(/\\//g, '~1')}`\n}\n\nfunction truncateUtf8(text: string, maxBytes: number): string | undefined {\n if (text.length <= maxBytes / 4) return undefined\n const bytes = Buffer.from(text, 'utf8')\n if (bytes.length <= maxBytes) return undefined\n // Step back to the first byte of a UTF-8 sequence so no character is split.\n let cut = maxBytes\n while (cut > 0 && ((bytes[cut] ?? 0) & 0xc0) === 0x80) cut -= 1\n return `${bytes.subarray(0, cut).toString('utf8')}…[truncated ${bytes.length - cut} bytes]`\n}\n\nfunction redactStringAt(text: string, path: string, state: WalkState): string {\n if (isPlaceholder(text)) return text\n if (DATA_URI.test(text)) {\n if (!state.rules.redactMedia) return text\n record(state, path, 'media', 'media', 'redacted')\n return marker('media')\n }\n const credential = credentialIn(text, state.knownSecrets)\n if (credential) {\n record(state, path, 'credential', credential, 'redacted')\n return marker(credential)\n }\n let output = text\n for (const detector of PERSONAL_DATA_DETECTORS) {\n if (detector.requires && !output.includes(detector.requires)) continue\n output = output.replace(detector.pattern, (match) => {\n if (detector.accept && !detector.accept(match)) return match\n record(state, path, 'personal-data', detector.id, 'redacted')\n return marker(detector.id)\n })\n }\n const truncated = truncateUtf8(output, state.maxStringBytes)\n if (truncated === undefined) return output\n record(state, path, 'raw-content', 'size-cap', 'truncated')\n return truncated\n}\n\nfunction pseudonym(value: string | number, state: WalkState): string {\n const digest = createHmac('sha256', state.pseudonymKey).update(String(value)).digest('hex')\n return `[ID:${digest.slice(0, 12)}]`\n}\n\nfunction redactKeyed(\n key: string,\n value: unknown,\n path: string,\n depth: number,\n state: WalkState,\n): unknown {\n const kind = classifyKeyName(key)\n if (value === null || value === undefined || typeof value === 'boolean') {\n return value\n }\n if (typeof value === 'string' && isPlaceholder(value)) return value\n // Numbers under a credential name are limits and counts (`password_max_age`), not secrets.\n if (kind === 'credential' && typeof value !== 'number') {\n record(state, path, 'credential', 'credential-key', 'redacted')\n return marker('credential-key')\n }\n if (\n kind === 'email' ||\n (kind === 'share-personal' && state.rules.redactKeys.has('share-personal'))\n ) {\n if (typeof value === 'string' || typeof value === 'number') {\n record(state, path, 'personal-data', 'personal-data-key', 'redacted')\n return marker('personal-data-key')\n }\n }\n if (kind === 'identifier' && state.rules.redactKeys.has('identifier')) {\n if (typeof value === 'string' || typeof value === 'number') {\n record(state, path, 'identifier', 'identifier-key', 'pseudonymized')\n return pseudonym(value, state)\n }\n }\n if (\n kind === 'raw-content' &&\n state.rules.redactKeys.has('raw-content') &&\n typeof value !== 'number'\n ) {\n record(state, path, 'raw-content', 'raw-content-key', 'redacted')\n return marker('raw-content-key')\n }\n return walk(value, path, depth + 1, state)\n}\n\nfunction walk(value: unknown, path: string, depth: number, state: WalkState): unknown {\n if (typeof value === 'string') return redactStringAt(value, path, state)\n if (value === null || typeof value !== 'object') return value\n if (ArrayBuffer.isView(value) || value instanceof ArrayBuffer) {\n record(state, path, 'media', 'binary', 'redacted')\n return marker('binary')\n }\n if (value instanceof Date) return value\n if (depth > MAX_DEPTH) {\n record(state, path, 'raw-content', 'depth-limit', 'truncated')\n return marker('depth-limit')\n }\n if (state.seen.has(value)) {\n record(state, path, 'raw-content', 'circular', 'redacted')\n return marker('circular')\n }\n state.seen.add(value)\n try {\n if (Array.isArray(value)) {\n return value.map((item, index) => walk(item, pointer(path, index), depth + 1, state))\n }\n const out: Record<string, unknown> = {}\n let renamed = 0\n for (const [key, item] of Object.entries(value as Record<string, unknown>)) {\n let outKey = key\n const keyCredential = credentialIn(key, state.knownSecrets)\n if (keyCredential) {\n renamed += 1\n outKey = `${marker(keyCredential)}#${renamed}`\n record(state, pointer(path, outKey), 'credential', keyCredential, 'redacted')\n }\n out[outKey] = redactKeyed(key, item, pointer(path, outKey), depth, state)\n }\n return out\n } finally {\n state.seen.delete(value)\n }\n}\n\nfunction newReport(profile: RedactionProfile): RedactionReport {\n return {\n version: REDACTION_VERSION,\n profile,\n redactionCount: 0,\n byDetector: {},\n truncatedCount: 0,\n findings: [],\n }\n}\n\nfunction newState(options: RedactOptions, report?: RedactionReport): WalkState {\n const profile = options.profile ?? 'default'\n const rules = PROFILE_RULES[profile]\n if (!rules) throw new Error(`unknown redaction profile: ${String(profile)}`)\n const requested = options.maxStringBytes\n const maxStringBytes =\n typeof requested === 'number' && Number.isFinite(requested) && requested >= 0\n ? Math.min(Math.floor(requested), rules.maxStringBytes)\n : rules.maxStringBytes\n return {\n rules,\n maxStringBytes,\n knownSecrets: knownSecretForms(options.knownSecrets),\n report: report ?? newReport(profile),\n pseudonymKey: randomBytes(32),\n seen: new WeakSet(),\n }\n}\n\n/**\n * Return a redacted deep copy of `value` and a report of what changed. Pure:\n * the input is not mutated. Never throws on the value's shape; cycles, binary\n * data and nesting past 64 levels are replaced with markers.\n */\nexport function redact<T>(\n value: T,\n options: RedactOptions = {},\n): { value: T; report: RedactionReport } {\n const state = newState(options, options.report)\n const out = walk(value, '', 0, state) as T\n return { value: out, report: state.report }\n}\n\n/**\n * Redact one free-text string with the value detectors (no field name is\n * known). Pass `report` to accumulate counts across calls.\n */\nexport function redactText(text: string, options: RedactOptions = {}): string {\n const state = newState(options, options.report)\n return redactStringAt(text, '', state)\n}\n\n/** An empty report to accumulate `redactText` calls into. */\nexport function emptyRedactionReport(profile: RedactionProfile = 'default'): RedactionReport {\n return newReport(profile)\n}\n\n// ── Share-safety assessment ──────────────────────────────────────────────\n\ninterface ScanState {\n profile: RedactionProfile\n rules: ProfileRules\n knownSecrets: string[]\n findings: Map<string, ShareSafetyFinding>\n unreadable: string[]\n seen: WeakSet<object>\n}\n\nfunction flag(state: ScanState, path: string, category: SafetyCategory, detector: string): void {\n const severity = state.rules.severity[category]\n if (!severity) return\n const id = `${category}\\u0000${detector}`\n const existing = state.findings.get(id)\n if (existing) {\n existing.count += 1\n if (existing.paths.length < 5) existing.paths.push(path)\n return\n }\n state.findings.set(id, { category, detector, severity, count: 1, paths: [path] })\n}\n\nfunction scanString(text: string, path: string, state: ScanState): void {\n if (isPlaceholder(text)) return\n if (DATA_URI.test(text)) {\n flag(state, path, 'media', 'media')\n return\n }\n const credential = credentialIn(text, state.knownSecrets)\n if (credential) {\n flag(state, path, 'credential', credential)\n return\n }\n for (const detector of detectPersonalData(text)) flag(state, path, 'personal-data', detector)\n}\n\nfunction scanKeyed(\n key: string,\n value: unknown,\n path: string,\n depth: number,\n state: ScanState,\n): void {\n if (value === null || value === undefined || typeof value === 'boolean') return\n if (typeof value === 'string' && isPlaceholder(value)) return\n const category = classifyKey(key)\n if (category === 'credential' && typeof value !== 'number') {\n flag(state, path, 'credential', 'credential-key')\n return\n }\n const scalar = typeof value === 'string' || typeof value === 'number'\n if (category === 'personal-data' && scalar) {\n const kind = classifyKeyName(key)\n // Phone, IP and address fields are flagged from the share profile up; email always.\n if (kind === 'email' || state.profile !== 'default') {\n flag(state, path, 'personal-data', 'personal-data-key')\n return\n }\n }\n if (category === 'identifier' && scalar) flag(state, path, 'identifier', 'identifier-key')\n if (category === 'raw-content' && typeof value !== 'number')\n flag(state, path, 'raw-content', 'raw-content-key')\n scan(value, path, depth + 1, state)\n}\n\nfunction scan(value: unknown, path: string, depth: number, state: ScanState): void {\n if (typeof value === 'string') {\n scanString(value, path, state)\n return\n }\n if (value === null || typeof value !== 'object') return\n if (ArrayBuffer.isView(value) || value instanceof ArrayBuffer) {\n state.unreadable.push(`${path || '/'}: binary data`)\n return\n }\n if (value instanceof Date) return\n if (depth > MAX_DEPTH) {\n state.unreadable.push(`${path || '/'}: nested deeper than ${MAX_DEPTH} levels`)\n return\n }\n if (state.seen.has(value)) {\n state.unreadable.push(`${path || '/'}: circular reference`)\n return\n }\n state.seen.add(value)\n try {\n if (Array.isArray(value)) {\n for (const [index, item] of value.entries())\n scan(item, pointer(path, index), depth + 1, state)\n return\n }\n for (const [key, item] of Object.entries(value as Record<string, unknown>)) {\n // A credential used as a key must not reach the finding paths.\n const keyCredential = credentialIn(key, state.knownSecrets)\n const segment = keyCredential ? marker(keyCredential) : key\n if (keyCredential) flag(state, pointer(path, segment), 'credential', keyCredential)\n scanKeyed(key, item, pointer(path, segment), depth, state)\n }\n } finally {\n state.seen.delete(value)\n }\n}\n\n/**\n * Read `value` without changing it and say whether it is safe to share under\n * `profile`. UNKNOWN when any part could not be read; UNSAFE when any\n * error-severity finding remains; SAFE_WITH_WARNINGS when only warnings\n * remain (raw content, embedded media, identifiers under `share`).\n */\nexport function assessShareSafety(value: unknown, options: AssessOptions = {}): ShareSafetyVerdict {\n const profile = options.profile ?? 'default'\n const rules = PROFILE_RULES[profile]\n if (!rules) throw new Error(`unknown redaction profile: ${String(profile)}`)\n const state: ScanState = {\n profile,\n rules,\n knownSecrets: knownSecretForms(options.knownSecrets),\n findings: new Map(),\n unreadable: [],\n seen: new WeakSet(),\n }\n scan(value, '', 0, state)\n return verdictFrom(profile, [...state.findings.values()], state.unreadable)\n}\n\n/**\n * One verdict over several parts (files in a directory, spans in a bundle).\n * Callers prefix each part's paths with its source before combining.\n */\nexport function combineVerdicts(\n profile: RedactionProfile,\n verdicts: readonly ShareSafetyVerdict[],\n unreadable: readonly string[] = [],\n): ShareSafetyVerdict {\n const findings = new Map<string, ShareSafetyFinding>()\n for (const verdict of verdicts) {\n for (const finding of verdict.findings) {\n const id = `${finding.category}\\u0000${finding.detector}`\n const existing = findings.get(id)\n if (!existing) {\n findings.set(id, { ...finding, paths: [...finding.paths] })\n continue\n }\n existing.count += finding.count\n for (const path of finding.paths) if (existing.paths.length < 5) existing.paths.push(path)\n }\n }\n return verdictFrom(\n profile,\n [...findings.values()],\n [...verdicts.flatMap((verdict) => verdict.unreadable), ...unreadable],\n )\n}\n\nfunction verdictFrom(\n profile: RedactionProfile,\n findings: readonly ShareSafetyFinding[],\n unreadable: readonly string[],\n): ShareSafetyVerdict {\n const status: ShareSafetyStatus =\n unreadable.length > 0\n ? 'UNKNOWN'\n : findings.some((finding) => finding.severity === 'error')\n ? 'UNSAFE'\n : findings.length > 0\n ? 'SAFE_WITH_WARNINGS'\n : 'SAFE'\n return { status, profile, findings: [...findings], unreadable: [...unreadable] }\n}\n\n/** Whether a verdict allows sharing: SAFE or SAFE_WITH_WARNINGS. UNSAFE and UNKNOWN refuse. */\nexport function shareAllowed(verdict: ShareSafetyVerdict): boolean {\n return verdict.status === 'SAFE' || verdict.status === 'SAFE_WITH_WARNINGS'\n}\n\n/**\n * Redact `value` under `profile`, then assess the redacted output. The\n * verdict describes what would actually be shared: a detector that failed to\n * remove something shows up as UNSAFE here.\n */\nexport function redactForShare<T>(\n value: T,\n options: RedactOptions = {},\n): { value: T; report: RedactionReport; verdict: ShareSafetyVerdict } {\n const redacted = redact(value, options)\n const verdict = assessShareSafety(redacted.value, {\n profile: options.profile,\n knownSecrets: options.knownSecrets,\n })\n return { value: redacted.value, report: redacted.report, verdict }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwCA,MAAa,qBAAkD;CAAC;CAAW;CAAS;AAAQ;;;;;;AAU5F,MAAa,oBAAoB;AAwEjC,MAAM,MAAM;AAEZ,MAAM,gBAAwD;CAC5D,SAAS;EACP,gBAAgB,OAAO;EACvB,4BAAY,IAAI,IAAI;EACpB,aAAa;EACb,UAAU;GACR,YAAY;GACZ,iBAAiB;GACjB,YAAY,KAAA;GACZ,eAAe;GACf,OAAO;EACT;CACF;CACA,OAAO;EACL,gBAAgB,KAAK;EACrB,4BAAY,IAAI,IAAI,CAAC,kBAAkB,YAAY,CAAC;EACpD,aAAa;EACb,UAAU;GACR,YAAY;GACZ,iBAAiB;GACjB,YAAY;GACZ,eAAe;GACf,OAAO;EACT;CACF;CACA,QAAQ;EACN,gBAAgB,IAAI;EACpB,4BAAY,IAAI,IAAI;GAAC;GAAkB;GAAc;EAAa,CAAC;EACnE,aAAa;EACb,UAAU;GACR,YAAY;GACZ,iBAAiB;GACjB,YAAY;GACZ,eAAe;GACf,OAAO;EACT;CACF;AACF;;;;;;AASA,SAAgB,aAAa,KAAqB;CAChD,OAAO,IACJ,QAAQ,sBAAsB,OAAO,CAAC,CACtC,YAAY,CAAC,CACb,QAAQ,eAAe,GAAG,CAAC,CAC3B,QAAQ,YAAY,EAAE;AAC3B;;AAGA,MAAM,4CAAiD,IAAI,IAAI;CAC7D;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAM,kBAAkB;CACtB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;AAGA,MAAM,aAAa,CAAC,SAAS,eAAe;;AAG5C,MAAM,sBAAsB;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;AAGA,MAAM,kBAAkB;CACtB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;AAGA,MAAM,mBAAmB;CACvB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;AAUA,SAAS,eAAe,YAAoB,MAAkC;CAC5E,KAAK,MAAM,OAAO,MAChB,IAAI,eAAe,OAAO,WAAW,SAAS,IAAI,KAAK,GAAG,OAAO;CAEnE,OAAO;AACT;AAEA,SAAS,qBAAqB,YAA6B;CACzD,IAAI,0BAA0B,IAAI,UAAU,GAAG,OAAO;CACtD,IAAI,eAAe,SAAS,OAAO;CAEnC,IAAI,WAAW,SAAS,QAAQ,GAAG,OAAO;CAC1C,OAAO,WAAW,SAAS,OAAO;AACpC;;AAKA,MAAM,gCAAgB,IAAI,IAA6B;AAEvD,SAAS,gBAAgB,KAAmC;CAC1D,MAAM,SAAS,cAAc,IAAI,GAAG;CACpC,IAAI,WAAW,KAAA,GAAW,OAAO,UAAU,KAAA;CAC3C,MAAM,OAAO,sBAAsB,aAAa,GAAG,CAAC;CACpD,IAAI,cAAc,QAAQ,KAAQ,cAAc,MAAM;CACtD,cAAc,IAAI,KAAK,QAAQ,IAAI;CACnC,OAAO;AACT;AAEA,SAAS,sBAAsB,YAA0C;CACvE,IAAI,CAAC,cAAc,0BAA0B,IAAI,UAAU,GAAG,OAAO,KAAA;CACrE,IAAI,qBAAqB,UAAU,GAAG,OAAO;CAC7C,IAAI,eAAe,YAAY,eAAe,GAAG,OAAO;CACxD,IAAI,eAAe,YAAY,UAAU,GAAG,OAAO;CACnD,IAAI,eAAe,YAAY,mBAAmB,GAAG,OAAO;CAC5D,IAAI,eAAe,YAAY,eAAe,GAAG,OAAO;CACxD,IAAI,eAAe,YAAY,gBAAgB,GAAG,OAAO;AAE3D;;;;;;AAOA,SAAgB,YAAY,KAAyC;CACnE,MAAM,OAAO,gBAAgB,GAAG;CAChC,IAAI,SAAS,KAAA,GAAW,OAAO,KAAA;CAC/B,IAAI,SAAS,cAAc,OAAO;CAClC,IAAI,SAAS,cAAc,OAAO;CAClC,IAAI,SAAS,eAAe,OAAO;CACnC,OAAO;AACT;;;;;AAaA,MAAM,uBAAiD;CACrD;EAAE,IAAI;EAAe,SAAS;CAAwC;CACtE;EAAE,IAAI;EAAO,SAAS;CAA+D;CACrF;EAAE,IAAI;EAAU,SAAS;CAAsC;CAC/D;EACE,IAAI;EACJ,SAAS;CACX;CACA;EAAE,IAAI;EAAiB,SAAS;CAA8B;CAC9D;EAAE,IAAI;EAAgB,SAAS;CAAyC;CACxE;EAAE,IAAI;EAAc,SAAS;CAA8C;CAC3E;EAAE,IAAI;EAAkB,SAAS;CAA2B;CAC5D;EAAE,IAAI;EAAgB,SAAS;CAAiE;CAChG;EAAE,IAAI;EAAe,SAAS;CAAmC;CACjE;EAAE,IAAI;EAAkB,SAAS;CAAgC;CACjE;EAAE,IAAI;EAAqB,SAAS;CAAuC;CAC3E;EAAE,IAAI;EAAmB,SAAS;CAA+C;CACjF;EACE,IAAI;EACJ,SACE;CACJ;CACA;EAME,IAAI;EACJ,SACE;CACJ;AACF;;AAGA,MAAM,0BAIC;CACL;EACE,IAAI;EACJ,UAAU;EACV,SAAS;EACT,QAAQ;CACV;CACA;EAKE,IAAI;EACJ,SACE;EACF,SAAS,UAAU;GACjB,MAAM,SAAS,MAAM,QAAQ,SAAS,EAAE;GACxC,OAAO,YAAY,KAAK,MAAM,KAAK,UAAU,MAAM;EACrD;CACF;CACA;EAAE,IAAI;EAAO,SAAS;CAAyB;CAC/C;EAEE,IAAI;EACJ,SAAS;CACX;AACF;;AAGA,MAAM,cAAc;;;;;AAMpB,MAAM,iCAAsC,IAAI,IAAI;CAClD;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,SAAS,eAAe,OAAwB;CAE9C,IAAI,MAAM,YAAY,CAAC,CAAC,WAAW,MAAM,GAAG,OAAO;CACnD,MAAM,MAAM,MAAM,MAAM,MAAM,YAAY,GAAG,IAAI,CAAC,CAAC,CAAC,YAAY;CAChE,OAAO,CAAC,eAAe,IAAI,GAAG;AAChC;AAEA,SAAS,UAAU,QAAyB;CAC1C,IAAI,YAAY,KAAK,MAAM,GAAG,OAAO;CACrC,IAAI,MAAM;CACV,IAAI,SAAS;CACb,KAAK,IAAI,QAAQ,OAAO,SAAS,GAAG,SAAS,GAAG,SAAS,GAAG;EAC1D,IAAI,QAAQ,OAAO,WAAW,KAAK,IAAI;EACvC,IAAI,QAAQ;GACV,SAAS;GACT,IAAI,QAAQ,GAAG,SAAS;EAC1B;EACA,OAAO;EACP,SAAS,CAAC;CACZ;CACA,OAAO,MAAM,OAAO;AACtB;;AAGA,SAAgB,iBAAiB,MAAkC;CACjE,KAAK,MAAM,YAAY,sBACrB,IAAI,SAAS,QAAQ,KAAK,IAAI,GAAG,OAAO,SAAS;AAGrD;;AAGA,SAAgB,mBAAmB,MAAwB;CACzD,MAAM,OAAiB,CAAC;CACxB,KAAK,MAAM,YAAY,yBAAyB;EAC9C,IAAI,SAAS,YAAY,CAAC,KAAK,SAAS,SAAS,QAAQ,GAAG;EAC5D,SAAS,QAAQ,YAAY;EAC7B,KAAK,MAAM,SAAS,KAAK,SAAS,SAAS,OAAO,GAChD,IAAI,CAAC,SAAS,UAAU,SAAS,OAAO,MAAM,EAAE,GAAG;GACjD,KAAK,KAAK,SAAS,EAAE;GACrB;EACF;CAEJ;CACA,OAAO;AACT;;;;;;AAOA,SAAS,iBAAiB,QAAiD;CACzE,MAAM,wBAAQ,IAAI,IAAY;CAC9B,KAAK,MAAM,SAAS,UAAU,CAAC,GAAG;EAChC,IAAI,MAAM,SAAS,GAAG;EACtB,MAAM,QAAQ,OAAO,KAAK,OAAO,MAAM;EACvC,MAAM,IAAI,KAAK;EACf,MAAM,IAAI,MAAM,SAAS,QAAQ,CAAC;EAClC,MAAM,IAAI,MAAM,SAAS,WAAW,CAAC;EACrC,MAAM,IAAI,mBAAmB,KAAK,CAAC;CACrC;CAEA,OAAO,CAAC,GAAG,KAAK,CAAC,CAAC,MAAM,MAAM,UAAU,MAAM,SAAS,KAAK,MAAM;AACpE;AAEA,MAAM,cAAc;AAEpB,SAAS,oBAAoB,MAAc,OAAmC;CAC5E,IAAI,MAAM,WAAW,GAAG,OAAO;CAC/B,IAAI,MAAM,MAAM,SAAS,KAAK,SAAS,IAAI,CAAC,GAAG,OAAO;CAEtD,IAAI,KAAK,SAAS,KAAK,CAAC,YAAY,KAAK,IAAI,GAAG,OAAO;CACvD,MAAM,UAAU,OAAO,KACrB,MACA,KAAK,SAAS,GAAG,KAAK,KAAK,SAAS,GAAG,IAAI,cAAc,QAC3D,CAAC,CAAC,SAAS,MAAM;CACjB,OAAO,MAAM,MAAM,SAAS,QAAQ,SAAS,IAAI,CAAC;AACpD;;AAGA,SAAS,aAAa,MAAc,cAAqD;CACvF,OAAO,oBAAoB,MAAM,YAAY,IAAI,iBAAiB,iBAAiB,IAAI;AACzF;AAEA,MAAM,gBAAgB;AAEtB,SAAS,OAAO,UAA0B;CACxC,OAAO,GAAG,gBAAgB,SAAS;AACrC;;;;;AAMA,MAAM,uBACJ;AAEF,SAAS,cAAc,OAAwB;CAC7C,OAAO,UAAU,MAAM,qBAAqB,KAAK,KAAK;AACxD;AAEA,MAAM,WAAW;AAIjB,MAAM,YAAY;AAWlB,SAAS,OACP,OACA,MACA,UACA,UACA,QACM;CACN,MAAM,SAAS,MAAM;CACrB,IAAI,WAAW,aAAa,OAAO,kBAAkB;CACrD,OAAO,kBAAkB;CACzB,OAAO,WAAW,aAAa,OAAO,WAAW,aAAa,KAAK;CACnE,OAAO,SAAS,KAAK;EAAE;EAAM;EAAU;EAAU;CAAO,CAAC;AAC3D;AAEA,SAAS,QAAQ,QAAgB,KAA8B;CAC7D,OAAO,GAAG,OAAO,GAAG,OAAO,GAAG,CAAC,CAAC,QAAQ,MAAM,IAAI,CAAC,CAAC,QAAQ,OAAO,IAAI;AACzE;AAEA,SAAS,aAAa,MAAc,UAAsC;CACxE,IAAI,KAAK,UAAU,WAAW,GAAG,OAAO,KAAA;CACxC,MAAM,QAAQ,OAAO,KAAK,MAAM,MAAM;CACtC,IAAI,MAAM,UAAU,UAAU,OAAO,KAAA;CAErC,IAAI,MAAM;CACV,OAAO,MAAM,OAAO,MAAM,QAAQ,KAAK,SAAU,KAAM,OAAO;CAC9D,OAAO,GAAG,MAAM,SAAS,GAAG,GAAG,CAAC,CAAC,SAAS,MAAM,EAAE,cAAc,MAAM,SAAS,IAAI;AACrF;AAEA,SAAS,eAAe,MAAc,MAAc,OAA0B;CAC5E,IAAI,cAAc,IAAI,GAAG,OAAO;CAChC,IAAI,SAAS,KAAK,IAAI,GAAG;EACvB,IAAI,CAAC,MAAM,MAAM,aAAa,OAAO;EACrC,OAAO,OAAO,MAAM,SAAS,SAAS,UAAU;EAChD,OAAO,OAAO,OAAO;CACvB;CACA,MAAM,aAAa,aAAa,MAAM,MAAM,YAAY;CACxD,IAAI,YAAY;EACd,OAAO,OAAO,MAAM,cAAc,YAAY,UAAU;EACxD,OAAO,OAAO,UAAU;CAC1B;CACA,IAAI,SAAS;CACb,KAAK,MAAM,YAAY,yBAAyB;EAC9C,IAAI,SAAS,YAAY,CAAC,OAAO,SAAS,SAAS,QAAQ,GAAG;EAC9D,SAAS,OAAO,QAAQ,SAAS,UAAU,UAAU;GACnD,IAAI,SAAS,UAAU,CAAC,SAAS,OAAO,KAAK,GAAG,OAAO;GACvD,OAAO,OAAO,MAAM,iBAAiB,SAAS,IAAI,UAAU;GAC5D,OAAO,OAAO,SAAS,EAAE;EAC3B,CAAC;CACH;CACA,MAAM,YAAY,aAAa,QAAQ,MAAM,cAAc;CAC3D,IAAI,cAAc,KAAA,GAAW,OAAO;CACpC,OAAO,OAAO,MAAM,eAAe,YAAY,WAAW;CAC1D,OAAO;AACT;AAEA,SAAS,UAAU,OAAwB,OAA0B;CAEnE,OAAO,OADQ,WAAW,UAAU,MAAM,YAAY,CAAC,CAAC,OAAO,OAAO,KAAK,CAAC,CAAC,CAAC,OAAO,KAClE,CAAC,CAAC,MAAM,GAAG,EAAE,EAAE;AACpC;AAEA,SAAS,YACP,KACA,OACA,MACA,OACA,OACS;CACT,MAAM,OAAO,gBAAgB,GAAG;CAChC,IAAI,UAAU,QAAQ,UAAU,KAAA,KAAa,OAAO,UAAU,WAC5D,OAAO;CAET,IAAI,OAAO,UAAU,YAAY,cAAc,KAAK,GAAG,OAAO;CAE9D,IAAI,SAAS,gBAAgB,OAAO,UAAU,UAAU;EACtD,OAAO,OAAO,MAAM,cAAc,kBAAkB,UAAU;EAC9D,OAAO,OAAO,gBAAgB;CAChC;CACA,IACE,SAAS,WACR,SAAS,oBAAoB,MAAM,MAAM,WAAW,IAAI,gBAAgB,GAErE;MAAA,OAAO,UAAU,YAAY,OAAO,UAAU,UAAU;GAC1D,OAAO,OAAO,MAAM,iBAAiB,qBAAqB,UAAU;GACpE,OAAO,OAAO,mBAAmB;EACnC;;CAEF,IAAI,SAAS,gBAAgB,MAAM,MAAM,WAAW,IAAI,YAAY,GAC9D;MAAA,OAAO,UAAU,YAAY,OAAO,UAAU,UAAU;GAC1D,OAAO,OAAO,MAAM,cAAc,kBAAkB,eAAe;GACnE,OAAO,UAAU,OAAO,KAAK;EAC/B;;CAEF,IACE,SAAS,iBACT,MAAM,MAAM,WAAW,IAAI,aAAa,KACxC,OAAO,UAAU,UACjB;EACA,OAAO,OAAO,MAAM,eAAe,mBAAmB,UAAU;EAChE,OAAO,OAAO,iBAAiB;CACjC;CACA,OAAO,KAAK,OAAO,MAAM,QAAQ,GAAG,KAAK;AAC3C;AAEA,SAAS,KAAK,OAAgB,MAAc,OAAe,OAA2B;CACpF,IAAI,OAAO,UAAU,UAAU,OAAO,eAAe,OAAO,MAAM,KAAK;CACvE,IAAI,UAAU,QAAQ,OAAO,UAAU,UAAU,OAAO;CACxD,IAAI,YAAY,OAAO,KAAK,KAAK,iBAAiB,aAAa;EAC7D,OAAO,OAAO,MAAM,SAAS,UAAU,UAAU;EACjD,OAAO,OAAO,QAAQ;CACxB;CACA,IAAI,iBAAiB,MAAM,OAAO;CAClC,IAAI,QAAQ,WAAW;EACrB,OAAO,OAAO,MAAM,eAAe,eAAe,WAAW;EAC7D,OAAO,OAAO,aAAa;CAC7B;CACA,IAAI,MAAM,KAAK,IAAI,KAAK,GAAG;EACzB,OAAO,OAAO,MAAM,eAAe,YAAY,UAAU;EACzD,OAAO,OAAO,UAAU;CAC1B;CACA,MAAM,KAAK,IAAI,KAAK;CACpB,IAAI;EACF,IAAI,MAAM,QAAQ,KAAK,GACrB,OAAO,MAAM,KAAK,MAAM,UAAU,KAAK,MAAM,QAAQ,MAAM,KAAK,GAAG,QAAQ,GAAG,KAAK,CAAC;EAEtF,MAAM,MAA+B,CAAC;EACtC,IAAI,UAAU;EACd,KAAK,MAAM,CAAC,KAAK,SAAS,OAAO,QAAQ,KAAgC,GAAG;GAC1E,IAAI,SAAS;GACb,MAAM,gBAAgB,aAAa,KAAK,MAAM,YAAY;GAC1D,IAAI,eAAe;IACjB,WAAW;IACX,SAAS,GAAG,OAAO,aAAa,EAAE,GAAG;IACrC,OAAO,OAAO,QAAQ,MAAM,MAAM,GAAG,cAAc,eAAe,UAAU;GAC9E;GACA,IAAI,UAAU,YAAY,KAAK,MAAM,QAAQ,MAAM,MAAM,GAAG,OAAO,KAAK;EAC1E;EACA,OAAO;CACT,UAAU;EACR,MAAM,KAAK,OAAO,KAAK;CACzB;AACF;AAEA,SAAS,UAAU,SAA4C;CAC7D,OAAO;EACL,SAAS;EACT;EACA,gBAAgB;EAChB,YAAY,CAAC;EACb,gBAAgB;EAChB,UAAU,CAAC;CACb;AACF;AAEA,SAAS,SAAS,SAAwB,QAAqC;CAC7E,MAAM,UAAU,QAAQ,WAAW;CACnC,MAAM,QAAQ,cAAc;CAC5B,IAAI,CAAC,OAAO,MAAM,IAAI,MAAM,8BAA8B,OAAO,OAAO,GAAG;CAC3E,MAAM,YAAY,QAAQ;CAK1B,OAAO;EACL;EACA,gBALA,OAAO,cAAc,YAAY,OAAO,SAAS,SAAS,KAAK,aAAa,IACxE,KAAK,IAAI,KAAK,MAAM,SAAS,GAAG,MAAM,cAAc,IACpD,MAAM;EAIV,cAAc,iBAAiB,QAAQ,YAAY;EACnD,QAAQ,UAAU,UAAU,OAAO;EACnC,cAAc,YAAY,EAAE;EAC5B,sBAAM,IAAI,QAAQ;CACpB;AACF;;;;;;AAOA,SAAgB,OACd,OACA,UAAyB,CAAC,GACa;CACvC,MAAM,QAAQ,SAAS,SAAS,QAAQ,MAAM;CAE9C,OAAO;EAAE,OADG,KAAK,OAAO,IAAI,GAAG,KACb;EAAG,QAAQ,MAAM;CAAO;AAC5C;;;;;AAMA,SAAgB,WAAW,MAAc,UAAyB,CAAC,GAAW;CAE5E,OAAO,eAAe,MAAM,IADd,SAAS,SAAS,QAAQ,MACJ,CAAC;AACvC;;AAGA,SAAgB,qBAAqB,UAA4B,WAA4B;CAC3F,OAAO,UAAU,OAAO;AAC1B;AAaA,SAAS,KAAK,OAAkB,MAAc,UAA0B,UAAwB;CAC9F,MAAM,WAAW,MAAM,MAAM,SAAS;CACtC,IAAI,CAAC,UAAU;CACf,MAAM,KAAK,GAAG,SAAS,QAAQ;CAC/B,MAAM,WAAW,MAAM,SAAS,IAAI,EAAE;CACtC,IAAI,UAAU;EACZ,SAAS,SAAS;EAClB,IAAI,SAAS,MAAM,SAAS,GAAG,SAAS,MAAM,KAAK,IAAI;EACvD;CACF;CACA,MAAM,SAAS,IAAI,IAAI;EAAE;EAAU;EAAU;EAAU,OAAO;EAAG,OAAO,CAAC,IAAI;CAAE,CAAC;AAClF;AAEA,SAAS,WAAW,MAAc,MAAc,OAAwB;CACtE,IAAI,cAAc,IAAI,GAAG;CACzB,IAAI,SAAS,KAAK,IAAI,GAAG;EACvB,KAAK,OAAO,MAAM,SAAS,OAAO;EAClC;CACF;CACA,MAAM,aAAa,aAAa,MAAM,MAAM,YAAY;CACxD,IAAI,YAAY;EACd,KAAK,OAAO,MAAM,cAAc,UAAU;EAC1C;CACF;CACA,KAAK,MAAM,YAAY,mBAAmB,IAAI,GAAG,KAAK,OAAO,MAAM,iBAAiB,QAAQ;AAC9F;AAEA,SAAS,UACP,KACA,OACA,MACA,OACA,OACM;CACN,IAAI,UAAU,QAAQ,UAAU,KAAA,KAAa,OAAO,UAAU,WAAW;CACzE,IAAI,OAAO,UAAU,YAAY,cAAc,KAAK,GAAG;CACvD,MAAM,WAAW,YAAY,GAAG;CAChC,IAAI,aAAa,gBAAgB,OAAO,UAAU,UAAU;EAC1D,KAAK,OAAO,MAAM,cAAc,gBAAgB;EAChD;CACF;CACA,MAAM,SAAS,OAAO,UAAU,YAAY,OAAO,UAAU;CAC7D,IAAI,aAAa,mBAAmB,QACrB;MAAA,gBAAgB,GAEtB,MAAM,WAAW,MAAM,YAAY,WAAW;GACnD,KAAK,OAAO,MAAM,iBAAiB,mBAAmB;GACtD;EACF;;CAEF,IAAI,aAAa,gBAAgB,QAAQ,KAAK,OAAO,MAAM,cAAc,gBAAgB;CACzF,IAAI,aAAa,iBAAiB,OAAO,UAAU,UACjD,KAAK,OAAO,MAAM,eAAe,iBAAiB;CACpD,KAAK,OAAO,MAAM,QAAQ,GAAG,KAAK;AACpC;AAEA,SAAS,KAAK,OAAgB,MAAc,OAAe,OAAwB;CACjF,IAAI,OAAO,UAAU,UAAU;EAC7B,WAAW,OAAO,MAAM,KAAK;EAC7B;CACF;CACA,IAAI,UAAU,QAAQ,OAAO,UAAU,UAAU;CACjD,IAAI,YAAY,OAAO,KAAK,KAAK,iBAAiB,aAAa;EAC7D,MAAM,WAAW,KAAK,GAAG,QAAQ,IAAI,cAAc;EACnD;CACF;CACA,IAAI,iBAAiB,MAAM;CAC3B,IAAI,QAAQ,WAAW;EACrB,MAAM,WAAW,KAAK,GAAG,QAAQ,IAAI,uBAAuB,UAAU,QAAQ;EAC9E;CACF;CACA,IAAI,MAAM,KAAK,IAAI,KAAK,GAAG;EACzB,MAAM,WAAW,KAAK,GAAG,QAAQ,IAAI,qBAAqB;EAC1D;CACF;CACA,MAAM,KAAK,IAAI,KAAK;CACpB,IAAI;EACF,IAAI,MAAM,QAAQ,KAAK,GAAG;GACxB,KAAK,MAAM,CAAC,OAAO,SAAS,MAAM,QAAQ,GACxC,KAAK,MAAM,QAAQ,MAAM,KAAK,GAAG,QAAQ,GAAG,KAAK;GACnD;EACF;EACA,KAAK,MAAM,CAAC,KAAK,SAAS,OAAO,QAAQ,KAAgC,GAAG;GAE1E,MAAM,gBAAgB,aAAa,KAAK,MAAM,YAAY;GAC1D,MAAM,UAAU,gBAAgB,OAAO,aAAa,IAAI;GACxD,IAAI,eAAe,KAAK,OAAO,QAAQ,MAAM,OAAO,GAAG,cAAc,aAAa;GAClF,UAAU,KAAK,MAAM,QAAQ,MAAM,OAAO,GAAG,OAAO,KAAK;EAC3D;CACF,UAAU;EACR,MAAM,KAAK,OAAO,KAAK;CACzB;AACF;;;;;;;AAQA,SAAgB,kBAAkB,OAAgB,UAAyB,CAAC,GAAuB;CACjG,MAAM,UAAU,QAAQ,WAAW;CACnC,MAAM,QAAQ,cAAc;CAC5B,IAAI,CAAC,OAAO,MAAM,IAAI,MAAM,8BAA8B,OAAO,OAAO,GAAG;CAC3E,MAAM,QAAmB;EACvB;EACA;EACA,cAAc,iBAAiB,QAAQ,YAAY;EACnD,0BAAU,IAAI,IAAI;EAClB,YAAY,CAAC;EACb,sBAAM,IAAI,QAAQ;CACpB;CACA,KAAK,OAAO,IAAI,GAAG,KAAK;CACxB,OAAO,YAAY,SAAS,CAAC,GAAG,MAAM,SAAS,OAAO,CAAC,GAAG,MAAM,UAAU;AAC5E;;;;;AAMA,SAAgB,gBACd,SACA,UACA,aAAgC,CAAC,GACb;CACpB,MAAM,2BAAW,IAAI,IAAgC;CACrD,KAAK,MAAM,WAAW,UACpB,KAAK,MAAM,WAAW,QAAQ,UAAU;EACtC,MAAM,KAAK,GAAG,QAAQ,SAAS,QAAQ,QAAQ;EAC/C,MAAM,WAAW,SAAS,IAAI,EAAE;EAChC,IAAI,CAAC,UAAU;GACb,SAAS,IAAI,IAAI;IAAE,GAAG;IAAS,OAAO,CAAC,GAAG,QAAQ,KAAK;GAAE,CAAC;GAC1D;EACF;EACA,SAAS,SAAS,QAAQ;EAC1B,KAAK,MAAM,QAAQ,QAAQ,OAAO,IAAI,SAAS,MAAM,SAAS,GAAG,SAAS,MAAM,KAAK,IAAI;CAC3F;CAEF,OAAO,YACL,SACA,CAAC,GAAG,SAAS,OAAO,CAAC,GACrB,CAAC,GAAG,SAAS,SAAS,YAAY,QAAQ,UAAU,GAAG,GAAG,UAAU,CACtE;AACF;AAEA,SAAS,YACP,SACA,UACA,YACoB;CASpB,OAAO;EAAE,QAPP,WAAW,SAAS,IAChB,YACA,SAAS,MAAM,YAAY,QAAQ,aAAa,OAAO,IACrD,WACA,SAAS,SAAS,IAChB,uBACA;EACO;EAAS,UAAU,CAAC,GAAG,QAAQ;EAAG,YAAY,CAAC,GAAG,UAAU;CAAE;AACjF;;AAGA,SAAgB,aAAa,SAAsC;CACjE,OAAO,QAAQ,WAAW,UAAU,QAAQ,WAAW;AACzD;;;;;;AAOA,SAAgB,eACd,OACA,UAAyB,CAAC,GAC0C;CACpE,MAAM,WAAW,OAAO,OAAO,OAAO;CACtC,MAAM,UAAU,kBAAkB,SAAS,OAAO;EAChD,SAAS,QAAQ;EACjB,cAAc,QAAQ;CACxB,CAAC;CACD,OAAO;EAAE,OAAO,SAAS;EAAO,QAAQ,SAAS;EAAQ;CAAQ;AACnE"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"release-confidence-RanRUXRW.js","names":[],"sources":["../src/promotion-gate.ts","../src/release-confidence.ts"],"sourcesContent":["/**\n * Bootstrap-CI promotion gate.\n *\n * In any iterative-improvement loop (GEPA, prompt evolution, dataset\n * curation), the question is \"did this generation actually improve, or are\n * we celebrating noise?\". With small N and noisy outcomes, point-estimate\n * deltas lie. Bootstrap confidence intervals tell the operator whether the\n * delta is real before code or prompts get promoted.\n *\n * This module is pure functions — no I/O, no model calls. Easy to unit-test\n * and to compose into any verdict gate.\n *\n * Default gate:\n * - Bootstrap mean baseline vs candidate (1k resamples).\n * - Compute the delta distribution; pass if the lower CI bound > 0.\n * - Tunable confidence (default 95%) and resample count.\n *\n * Verdict semantics intentionally match the existing `experiments.jsonl`\n * vocabulary:\n * - ADVANCE: candidate's CI lower bound > baseline mean (real win)\n * - KEEP: overlap, but candidate point estimate >= baseline (neutral)\n * - REVERT: candidate's CI upper bound < baseline mean (real regression)\n * - INCONCLUSIVE: not enough samples or CI straddles zero with no signal\n */\n\nimport { mulberry32 } from './statistics'\n\nexport type Verdict = 'ADVANCE' | 'KEEP' | 'REVERT' | 'INCONCLUSIVE'\n\nexport interface BootstrapResult {\n baselineMean: number\n candidateMean: number\n /** candidateMean - baselineMean, point estimate. */\n delta: number\n /** Lower bound of the (1 - alpha) CI on the delta. */\n ciLower: number\n /** Upper bound of the (1 - alpha) CI on the delta. */\n ciUpper: number\n /** Number of bootstrap resamples used. */\n iterations: number\n alpha: number\n verdict: Verdict\n}\n\nexport interface BootstrapOptions {\n /** Confidence level alpha (default 0.05 → 95% CI). */\n alpha?: number\n /** Number of resamples (default 1000). */\n iterations?: number\n /**\n * Minimum total samples (baseline + candidate) below which we always\n * return INCONCLUSIVE — bootstrap with too few samples is meaningless.\n * Default 6 (combined).\n */\n minTotalSamples?: number\n /** RNG seed for reproducibility. Absent, the seed is derived from the\n * compared arms, so the same inputs reproduce the same verdict. */\n seed?: number\n}\n\n/**\n * Compute the bootstrap CI on (candidateMean - baselineMean) and a verdict.\n *\n * Uses simple percentile bootstrap on the difference of resampled means.\n * That's the standard non-parametric primitive — no distributional\n * assumptions, robust to skew, easy to reason about.\n */\nexport function bootstrapCi(\n baseline: number[],\n candidate: number[],\n options: BootstrapOptions = {},\n): BootstrapResult {\n const alpha = options.alpha ?? 0.05\n const iterations = options.iterations ?? 1000\n const minTotal = options.minTotalSamples ?? 6\n const rng = mulberry32(options.seed ?? hashSeed(baseline, candidate))\n\n const baselineMean = mean(baseline)\n const candidateMean = mean(candidate)\n const delta = candidateMean - baselineMean\n\n if (\n baseline.length + candidate.length < minTotal ||\n baseline.length === 0 ||\n candidate.length === 0\n ) {\n return {\n baselineMean,\n candidateMean,\n delta,\n ciLower: -Infinity,\n ciUpper: Infinity,\n iterations: 0,\n alpha,\n verdict: 'INCONCLUSIVE',\n }\n }\n\n const deltas: number[] = new Array(iterations)\n for (let i = 0; i < iterations; i++) {\n const bResample = resample(baseline, rng)\n const cResample = resample(candidate, rng)\n deltas[i] = mean(cResample) - mean(bResample)\n }\n deltas.sort((a, b) => a - b)\n const lowerIdx = Math.floor((alpha / 2) * iterations)\n const upperIdx = Math.floor((1 - alpha / 2) * iterations) - 1\n const ciLower = deltas[Math.max(0, lowerIdx)]!\n const ciUpper = deltas[Math.min(iterations - 1, upperIdx)]!\n\n let verdict: Verdict\n // A ZERO-WIDTH interval is an absence of evidence, not a certainty. Constant\n // arms make every resample identical, so pass/fail data promotes on nothing:\n // [0,0,0] vs [1,1,1] is six samples, clears the `minTotalSamples` floor, and\n // yields [1, 1] — an \"ADVANCE\" carrying no information about how far the\n // estimate could be wrong. Same rule as `pairedDeltaTest`, one estimator over.\n if (!Number.isFinite(ciLower) || !Number.isFinite(ciUpper) || ciLower === ciUpper) {\n verdict = 'INCONCLUSIVE'\n } else if (ciLower > 0) verdict = 'ADVANCE'\n else if (ciUpper < 0) verdict = 'REVERT'\n else if (delta >= 0) verdict = 'KEEP'\n else verdict = 'INCONCLUSIVE'\n\n return {\n baselineMean,\n candidateMean,\n delta,\n ciLower,\n ciUpper,\n iterations,\n alpha,\n verdict,\n }\n}\n\nfunction mean(xs: number[]): number {\n if (xs.length === 0) return 0\n let s = 0\n for (const x of xs) s += x\n return s / xs.length\n}\n\nfunction resample(xs: number[], rng: () => number): number[] {\n const out = new Array(xs.length)\n for (let i = 0; i < xs.length; i++) out[i] = xs[Math.floor(rng() * xs.length)]\n return out\n}\n\n/** Stable seed derived from the inputs — same data → same CI bounds. */\nfunction hashSeed(a: number[], b: number[]): number {\n let h = 2166136261\n for (const x of [...a, ...b]) {\n const view = new Float64Array([x])\n const bytes = new Uint8Array(view.buffer)\n for (const byte of bytes) {\n h ^= byte\n h = Math.imul(h, 16777619)\n }\n }\n return h >>> 0\n}\n\n/**\n * Judge-replay promotion gate.\n *\n * The cheap inner-loop judge that drives an evolution run is by definition\n * fast and noisy. When you're about to promote a winning variant to the\n * canonical default, you want a STRONGER judge (a more expensive model, a\n * human grader, a separately-trained reward model) to confirm the win\n * generalises beyond the inner loop.\n *\n * This helper takes raw winner + baseline outputs, scores both through the\n * stronger judge, and applies `bootstrapCi`. ADVANCE means the stronger\n * judge agrees the winner is real with the configured confidence. Doesn't\n * matter what shape your \"output\" is — pass a string, an object, anything\n * the judge can read.\n */\nexport interface JudgeReplayGateArgs<TOutput> {\n baselineOutputs: TOutput[]\n candidateOutputs: TOutput[]\n /** Stronger judge — async to allow LLM calls. Return a 0..N scalar score. */\n judge: (output: TOutput) => Promise<number> | number\n alpha?: number\n iterations?: number\n /** RNG seed for reproducibility. */\n seed?: number\n /** Maximum concurrent judge calls. Default 4. */\n judgeConcurrency?: number\n}\n\n/**\n * Confirm a candidate's win with a stronger judge: score baseline and candidate outputs independently, then bootstrap a CI to verify the lift generalises beyond the inner loop.\n */\nexport async function judgeReplayGate<TOutput>(\n args: JudgeReplayGateArgs<TOutput>,\n): Promise<BootstrapResult & { baselineSamples: number; candidateSamples: number }> {\n const concurrency = args.judgeConcurrency ?? 4\n const baselineScores = await scoreAll(args.baselineOutputs, args.judge, concurrency)\n const candidateScores = await scoreAll(args.candidateOutputs, args.judge, concurrency)\n const ci = bootstrapCi(baselineScores, candidateScores, {\n ...(args.alpha !== undefined ? { alpha: args.alpha } : {}),\n ...(args.iterations !== undefined ? { iterations: args.iterations } : {}),\n ...(args.seed !== undefined ? { seed: args.seed } : {}),\n })\n return {\n ...ci,\n baselineSamples: baselineScores.length,\n candidateSamples: candidateScores.length,\n }\n}\n\nasync function scoreAll<TOutput>(\n outputs: TOutput[],\n judge: (output: TOutput) => Promise<number> | number,\n concurrency: number,\n): Promise<number[]> {\n const results: number[] = new Array(outputs.length)\n let next = 0\n async function worker(): Promise<void> {\n while (true) {\n const i = next++\n if (i >= outputs.length) return\n const v = await judge(outputs[i]!)\n results[i] = Number.isFinite(v) ? v : 0\n }\n }\n await Promise.all(Array.from({ length: Math.max(1, concurrency) }, () => worker()))\n return results\n}\n","/**\n * Release confidence gate.\n *\n * This is the production-facing composition layer over the lower-level\n * primitives:\n * - Dataset manifests prove corpus/version coverage.\n * - RunRecord rows prove reproducible search/holdout outcomes.\n * - Multi-shot trace evidence carries turn counts and ASI diagnostics.\n * - HeldOutGate decisions remain the paired promotion authority.\n *\n * The gate is intentionally pure and conservative. Missing declared evidence\n * fails closed instead of being treated as a neutral zero.\n */\n\nimport type { DatasetManifest, DatasetScenario, DatasetSplit } from './dataset'\nimport { ValidationError, VerificationError } from './errors'\nimport type { GateDecision } from './held-out-gate'\nimport { isRealnessGated, observedSplitScore } from './rollout/reward'\nimport { type RunRecord, type RunSplitTag, validateRunRecord } from './run-record'\nimport { FAILURE_CLASSES, type FailureClass } from './trace/schema'\n\n/** Severity of an actionable finding attached to a run/trace. */\nexport type AsiSeverity = 'info' | 'warning' | 'error' | 'critical'\n\n/** Actionable side-info — a diagnosed finding the loop can act on. */\nexport interface ActionableSideInfo {\n /** Stable expectation/check id when available. */\n expectationId?: string\n /** Human-readable diagnosis of what happened. */\n message: string\n severity?: AsiSeverity\n /** Concrete trace excerpt, file path, tool call, screenshot id, etc. */\n evidence?: string\n /** Prompt/tool/context surface likely responsible. */\n responsibleSurface?: string\n /** Suggested fix in natural language. */\n suggestion?: string\n /** Whether this expectation was satisfied. Defaults to false for ASI rows. */\n matched?: boolean\n metadata?: Record<string, unknown>\n}\n\nexport type ReleaseConfidenceStatus = 'pass' | 'warn' | 'fail'\nexport type ReleaseConfidenceAxisName =\n | 'corpus'\n | 'quality'\n | 'reliability'\n | 'generalization'\n | 'diagnostics'\n | 'efficiency'\n\nexport interface ReleaseTraceEvidence {\n scenarioId: string\n candidateId?: string\n split?: RunSplitTag\n score?: number\n ok?: boolean\n turnCount?: number\n costUsd?: number\n durationMs?: number\n /** Canonical task-failure class. Free-form detail belongs in ASI. */\n failureClass?: FailureClass\n asi?: ActionableSideInfo[]\n metadata?: Record<string, unknown>\n}\n\nexport interface ReleaseConfidenceThresholds {\n /** Require a Dataset manifest or explicit scenarios. Default true. */\n requireCorpus?: boolean\n minScenarioCount?: number\n minSearchRuns?: number\n minHoldoutRuns?: number\n /** Require at least one holdout scenario/run. Default true. */\n requireHoldout?: boolean\n minPassRate?: number\n minMeanScore?: number\n /** Search mean may exceed holdout mean by at most this much. */\n maxOverfitGap?: number\n maxMeanCostUsd?: number\n maxP95WallMs?: number\n /** Low-score/failed rows must carry ASI. Default true. */\n requireAsiForFailures?: boolean\n /** Score below this is considered a failure for ASI coverage. Default 0.5. */\n failureScoreThreshold?: number\n}\n\nexport interface ReleaseConfidenceInput {\n target: string\n candidateId?: string\n baselineId?: string\n dataset?: DatasetManifest\n scenarios?: readonly DatasetScenario[]\n runs?: readonly RunRecord[]\n traces?: readonly ReleaseTraceEvidence[]\n gateDecision?: GateDecision | null\n thresholds?: ReleaseConfidenceThresholds\n}\n\nexport interface ReleaseConfidenceAxis {\n name: ReleaseConfidenceAxisName\n status: ReleaseConfidenceStatus\n score: number | null\n detail: string\n}\n\nexport interface ReleaseConfidenceIssue {\n axis: ReleaseConfidenceAxisName\n severity: 'critical' | 'warning'\n code: string\n detail: string\n}\n\nexport interface ReleaseConfidenceMetrics {\n scenarioCount: number\n /** Search rows with a finite search score. */\n searchRuns: number\n /** Holdout rows with a finite holdout score. */\n holdoutRuns: number\n /** Runs with neither a split-matched score nor an explicit task failure. */\n unscoredRuns: number\n /** Run rows, or trace rows when no runs exist, with no classified terminal result. */\n unclassifiedTerminalRuns: number\n /** Run rows, or trace rows when no runs exist, that ended unsuccessfully. */\n terminalFailureRuns: number\n /** Success fraction when every run or fallback trace row has a classified result. */\n reliabilityRate: number | null\n passRate: number | null\n meanScore: number | null\n searchMeanScore: number | null\n holdoutMeanScore: number | null\n overfitGap: number | null\n meanCostUsd: number | null\n p95WallMs: number | null\n failedRows: number\n failuresWithAsi: number\n singleShotTraces: number\n multiShotTraces: number\n splitCounts: Record<DatasetSplit, number>\n domainCounts: Record<string, number>\n failureClassCounts: Partial<Record<FailureClass, number>>\n responsibleSurfaceCounts: Record<string, number>\n /**\n * Runs excluded from `passRate` because the authenticity gate flagged them as\n * gamed. Surfaced, never silent: a release whose pass rate is computed over a\n * shrunken denominator has to say by how much, or the exclusion is just a\n * different way of hiding the same runs.\n */\n realnessGatedRuns: number\n}\n\nexport interface ReleaseConfidenceScorecard {\n target: string\n candidateId: string | null\n baselineId: string | null\n status: ReleaseConfidenceStatus\n promote: boolean\n axes: ReleaseConfidenceAxis[]\n issues: ReleaseConfidenceIssue[]\n metrics: ReleaseConfidenceMetrics\n dataset: DatasetManifest | null\n gateDecision: GateDecision | null\n summary: string\n}\n\nconst DEFAULT_THRESHOLDS: Required<ReleaseConfidenceThresholds> = {\n requireCorpus: true,\n minScenarioCount: 1,\n minSearchRuns: 1,\n minHoldoutRuns: 1,\n requireHoldout: true,\n minPassRate: 0.8,\n minMeanScore: 0.7,\n maxOverfitGap: 0.15,\n maxMeanCostUsd: Number.POSITIVE_INFINITY,\n maxP95WallMs: Number.POSITIVE_INFINITY,\n requireAsiForFailures: true,\n failureScoreThreshold: 0.5,\n}\n\nexport function evaluateReleaseConfidence(\n input: ReleaseConfidenceInput,\n): ReleaseConfidenceScorecard {\n const thresholds = { ...DEFAULT_THRESHOLDS, ...input.thresholds }\n const candidateId = input.candidateId ?? null\n const runs = filterCandidate(\n (input.runs ?? []).map(validateRunRecord),\n candidateId,\n input.baselineId,\n )\n const traces = filterTraceCandidate(\n (input.traces ?? []).map(validateReleaseTraceEvidence),\n candidateId,\n input.baselineId,\n )\n const scenarios = input.scenarios ?? []\n const scenarioCount = input.dataset?.scenarioCount ?? scenarios.length\n const splitCounts = input.dataset?.splitCounts ?? countScenarioSplits(scenarios)\n const searchScores = scoresFor(runs, 'search')\n const holdoutScores = scoresFor(runs, 'holdout')\n const runScores = runs.map(runSplitScore).filter(isFiniteNumber)\n const traceScores = traces.map((t) => t.score).filter(isFiniteNumber)\n const scoreUniverse = runs.length > 0 ? runScores : traceScores\n const qualityRuns = runs.filter((run) => runSplitScore(run) !== undefined)\n const unscoredRuns = runs.filter(\n (run) =>\n runSplitScore(run) === undefined &&\n !hasExplicitTaskFailure(run) &&\n !isFailedTerminalOutcome(run.terminalOutcome),\n ).length\n // Realness-gated runs are EXCLUDED from the pass rate — numerator AND\n // denominator — and the count of what was dropped ships beside the rate as\n // `metrics.realnessGatedRuns`. Counting a gamed run as a pass made faking a\n // success the cheapest way to improve a release scorecard; scoring it 0\n // instead would be the other error, silently deflating the rate with a run\n // the gate says carries no usable verdict at all.\n const honestRuns = runs.filter((run) => !isRealnessGated(run))\n const passOutcomes =\n runs.length > 0\n ? honestRuns.map((run) => runPassOutcome(run, thresholds.failureScoreThreshold))\n : traces.map((trace) => tracePassOutcome(trace, thresholds.failureScoreThreshold))\n const reliabilityRows =\n runs.length > 0\n ? runs.map((run) => terminalSuccess(run.terminalOutcome))\n : traces.map((trace) => trace.ok)\n const unclassifiedTerminalRuns = reliabilityRows.filter((outcome) => outcome === undefined).length\n const terminalFailureRuns = reliabilityRows.filter((outcome) => outcome === false).length\n const reliabilityRate =\n reliabilityRows.length === 0 || unclassifiedTerminalRuns > 0\n ? null\n : (reliabilityRows.length - terminalFailureRuns) / reliabilityRows.length\n const searchRuns = qualityRuns.filter((r) => r.splitTag === 'search').length\n const holdoutRuns = qualityRuns.filter((r) => r.splitTag === 'holdout').length\n const failed = failedRows(runs, traces, thresholds.failureScoreThreshold)\n const searchMeanScore = meanOrNull(searchScores)\n const holdoutMeanScore = meanOrNull(holdoutScores)\n const runCosts = runs.flatMap((run) =>\n run.costProvenance.kind === 'uncaptured' ? [] : [run.costProvenance.usd],\n )\n const traceCosts = traces.map((trace) => trace.costUsd).filter(isFiniteNumber)\n const meanCostUsd =\n runs.length > 0\n ? runCosts.length === runs.length\n ? meanOrNull(runCosts)\n : null\n : meanOrNull(traceCosts)\n const wallTimes =\n runs.length > 0\n ? runs.map((run) => run.wallMs)\n : traces.map((trace) => trace.durationMs).filter(isFiniteNumber)\n const metrics: ReleaseConfidenceMetrics = {\n scenarioCount,\n searchRuns,\n holdoutRuns,\n unscoredRuns,\n unclassifiedTerminalRuns,\n terminalFailureRuns,\n reliabilityRate,\n passRate: passOutcomeRate(passOutcomes),\n realnessGatedRuns: runs.length - honestRuns.length,\n meanScore: meanOrNull(scoreUniverse),\n searchMeanScore,\n holdoutMeanScore,\n overfitGap: diffOrNull(searchMeanScore, holdoutMeanScore),\n meanCostUsd,\n p95WallMs: percentileOrNull(wallTimes, 0.95),\n failedRows: failed.length,\n failuresWithAsi: failed.filter((row) => row.hasAsi).length,\n singleShotTraces: traces.filter((t) => t.turnCount === 1).length,\n multiShotTraces: traces.filter((t) => (t.turnCount ?? 0) > 1).length,\n splitCounts,\n domainCounts: countDomains(scenarios),\n failureClassCounts: countFailureClasses(runs, traces, thresholds.failureScoreThreshold),\n responsibleSurfaceCounts: countResponsibleSurfaces(traces),\n }\n\n const issues: ReleaseConfidenceIssue[] = []\n checkCorpus(input, thresholds, metrics, issues)\n checkQuality(thresholds, metrics, issues)\n checkReliability(metrics, issues)\n checkGeneralization(input.gateDecision ?? null, thresholds, metrics, issues)\n checkDiagnostics(thresholds, metrics, issues)\n checkEfficiency(thresholds, metrics, issues)\n\n const axes = buildAxes(metrics, thresholds, issues)\n const status = issues.some((i) => i.severity === 'critical')\n ? 'fail'\n : issues.length > 0\n ? 'warn'\n : 'pass'\n\n return {\n target: input.target,\n candidateId,\n baselineId: input.baselineId ?? null,\n status,\n promote: status === 'pass' && (input.gateDecision ? input.gateDecision.promote : true),\n axes,\n issues,\n metrics,\n dataset: input.dataset ?? null,\n gateDecision: input.gateDecision ?? null,\n summary: renderSummary(input.target, status, metrics, issues),\n }\n}\n\nexport function assertReleaseConfidence(input: ReleaseConfidenceInput): ReleaseConfidenceScorecard {\n const scorecard = evaluateReleaseConfidence(input)\n if (scorecard.status === 'fail') {\n throw new VerificationError(scorecard.summary)\n }\n return scorecard\n}\n\nfunction filterCandidate(\n runs: readonly RunRecord[],\n candidateId: string | null,\n baselineId?: string,\n): RunRecord[] {\n if (candidateId) return runs.filter((r) => r.candidateId === candidateId)\n if (baselineId) return runs.filter((r) => r.candidateId !== baselineId)\n return [...runs]\n}\n\nfunction filterTraceCandidate(\n traces: readonly ReleaseTraceEvidence[],\n candidateId: string | null,\n baselineId?: string,\n): ReleaseTraceEvidence[] {\n if (candidateId)\n return traces.filter((t) => t.candidateId === undefined || t.candidateId === candidateId)\n if (baselineId)\n return traces.filter((t) => t.candidateId === undefined || t.candidateId !== baselineId)\n return [...traces]\n}\n\nfunction validateReleaseTraceEvidence(\n trace: ReleaseTraceEvidence,\n index: number,\n): ReleaseTraceEvidence {\n const value = trace as unknown as Record<string, unknown>\n if (Object.hasOwn(value, 'failureMode')) {\n throw new ValidationError(\n `traces[${index}].failureMode is not supported; use canonical failureClass`,\n )\n }\n if (trace.failureClass !== undefined && !FAILURE_CLASSES.includes(trace.failureClass)) {\n throw new ValidationError(\n `traces[${index}].failureClass must be one of ${FAILURE_CLASSES.join(', ')}`,\n )\n }\n return trace\n}\n\nfunction checkCorpus(\n input: ReleaseConfidenceInput,\n thresholds: Required<ReleaseConfidenceThresholds>,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (thresholds.requireCorpus && !input.dataset && (input.scenarios?.length ?? 0) === 0) {\n issues.push({\n axis: 'corpus',\n severity: 'critical',\n code: 'missing_corpus',\n detail: 'No Dataset manifest or scenarios supplied.',\n })\n }\n if (metrics.scenarioCount < thresholds.minScenarioCount) {\n issues.push({\n axis: 'corpus',\n severity: 'critical',\n code: 'few_scenarios',\n detail: `${metrics.scenarioCount} scenario(s) < min ${thresholds.minScenarioCount}.`,\n })\n }\n if (thresholds.requireHoldout && metrics.splitCounts.holdout === 0) {\n issues.push({\n axis: 'corpus',\n severity: 'critical',\n code: 'missing_holdout_split',\n detail: 'Corpus has no holdout scenarios.',\n })\n }\n}\n\nfunction checkQuality(\n thresholds: Required<ReleaseConfidenceThresholds>,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (metrics.searchRuns < thresholds.minSearchRuns) {\n issues.push({\n axis: 'quality',\n severity: 'critical',\n code: 'few_search_runs',\n detail: `${metrics.searchRuns} search run(s) < min ${thresholds.minSearchRuns}.`,\n })\n }\n if (metrics.unscoredRuns > 0) {\n issues.push({\n axis: 'quality',\n severity: 'critical',\n code: 'unscored_runs',\n detail: `${metrics.unscoredRuns} supplied run(s) have no task result.`,\n })\n }\n if (metrics.passRate === null || metrics.meanScore === null) {\n issues.push({\n axis: 'quality',\n severity: 'critical',\n code: 'missing_quality_scores',\n detail: 'No task-quality scores are available for pass-rate and mean-score checks.',\n })\n }\n if (metrics.passRate !== null && metrics.passRate < thresholds.minPassRate) {\n issues.push({\n axis: 'quality',\n severity: 'critical',\n code: 'low_pass_rate',\n detail: `passRate ${fmt(metrics.passRate)} < ${fmt(thresholds.minPassRate)}.`,\n })\n }\n if (metrics.meanScore !== null && metrics.meanScore < thresholds.minMeanScore) {\n issues.push({\n axis: 'quality',\n severity: 'critical',\n code: 'low_mean_score',\n detail: `meanScore ${fmt(metrics.meanScore)} < ${fmt(thresholds.minMeanScore)}.`,\n })\n }\n}\n\nfunction checkReliability(\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (metrics.reliabilityRate === null) {\n issues.push({\n axis: 'reliability',\n severity: 'critical',\n code: 'missing_reliability_evidence',\n detail:\n metrics.unclassifiedTerminalRuns > 0\n ? `${metrics.unclassifiedTerminalRuns} supplied run(s) have no classified terminal result.`\n : 'No classified terminal results are available.',\n })\n }\n if (metrics.terminalFailureRuns > 0) {\n issues.push({\n axis: 'reliability',\n severity: 'critical',\n code: 'terminal_run_failures',\n detail: `${metrics.terminalFailureRuns} run(s) ended failed, cancelled, or incomplete.`,\n })\n }\n}\n\nfunction checkGeneralization(\n gateDecision: GateDecision | null,\n thresholds: Required<ReleaseConfidenceThresholds>,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (thresholds.requireHoldout && metrics.holdoutRuns < thresholds.minHoldoutRuns) {\n issues.push({\n axis: 'generalization',\n severity: 'critical',\n code: 'few_holdout_runs',\n detail: `${metrics.holdoutRuns} holdout run(s) < min ${thresholds.minHoldoutRuns}.`,\n })\n }\n if (metrics.overfitGap !== null && metrics.overfitGap > thresholds.maxOverfitGap) {\n issues.push({\n axis: 'generalization',\n severity: 'critical',\n code: 'overfit_gap',\n detail: `search-holdout gap ${fmt(metrics.overfitGap)} > ${fmt(thresholds.maxOverfitGap)}.`,\n })\n }\n if (gateDecision && !gateDecision.promote) {\n issues.push({\n axis: 'generalization',\n severity: 'critical',\n code: `gate_${gateDecision.rejectionCode ?? 'reject'}`,\n detail: gateDecision.reason,\n })\n }\n}\n\nfunction checkDiagnostics(\n thresholds: Required<ReleaseConfidenceThresholds>,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (!thresholds.requireAsiForFailures) return\n if (metrics.failedRows > metrics.failuresWithAsi) {\n issues.push({\n axis: 'diagnostics',\n severity: 'critical',\n code: 'missing_failure_asi',\n detail: `${metrics.failedRows - metrics.failuresWithAsi} failed row(s) have no actionable side information.`,\n })\n }\n}\n\nfunction checkEfficiency(\n thresholds: Required<ReleaseConfidenceThresholds>,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): void {\n if (Number.isFinite(thresholds.maxMeanCostUsd) && metrics.meanCostUsd === null) {\n issues.push({\n axis: 'efficiency',\n severity: 'critical',\n code: 'missing_cost',\n detail: 'A finite cost limit was configured but no cost evidence is available.',\n })\n } else if (metrics.meanCostUsd !== null && metrics.meanCostUsd > thresholds.maxMeanCostUsd) {\n issues.push({\n axis: 'efficiency',\n severity: 'critical',\n code: 'cost_budget',\n detail: `meanCostUsd ${fmt(metrics.meanCostUsd)} > ${fmt(thresholds.maxMeanCostUsd)}.`,\n })\n }\n if (Number.isFinite(thresholds.maxP95WallMs) && metrics.p95WallMs === null) {\n issues.push({\n axis: 'efficiency',\n severity: 'critical',\n code: 'missing_latency',\n detail: 'A finite latency limit was configured but no latency evidence is available.',\n })\n } else if (metrics.p95WallMs !== null && metrics.p95WallMs > thresholds.maxP95WallMs) {\n issues.push({\n axis: 'efficiency',\n severity: 'critical',\n code: 'latency_budget',\n detail: `p95WallMs ${fmt(metrics.p95WallMs)} > ${fmt(thresholds.maxP95WallMs)}.`,\n })\n }\n}\n\nfunction buildAxes(\n metrics: ReleaseConfidenceMetrics,\n thresholds: Required<ReleaseConfidenceThresholds>,\n issues: ReleaseConfidenceIssue[],\n): ReleaseConfidenceAxis[] {\n return [\n axis(\n 'corpus',\n issues,\n bounded(metrics.scenarioCount / Math.max(1, thresholds.minScenarioCount)),\n `${metrics.scenarioCount} scenarios; holdout=${metrics.splitCounts.holdout}`,\n ),\n axis(\n 'quality',\n issues,\n metrics.passRate === null || metrics.meanScore === null\n ? null\n : Math.min(metrics.passRate, metrics.meanScore),\n `passRate=${fmt(metrics.passRate)} meanScore=${fmt(metrics.meanScore)}`,\n ),\n axis(\n 'reliability',\n issues,\n metrics.reliabilityRate,\n `successRate=${fmt(metrics.reliabilityRate)} terminalFailures=${metrics.terminalFailureRuns} unclassified=${metrics.unclassifiedTerminalRuns}`,\n ),\n axis(\n 'generalization',\n issues,\n gapScore(metrics.overfitGap, thresholds.maxOverfitGap),\n `holdoutRuns=${metrics.holdoutRuns} overfitGap=${fmt(metrics.overfitGap)}`,\n ),\n axis(\n 'diagnostics',\n issues,\n metrics.failedRows === 0 ? 1 : metrics.failuresWithAsi / metrics.failedRows,\n `failuresWithAsi=${metrics.failuresWithAsi}/${metrics.failedRows}`,\n ),\n axis(\n 'efficiency',\n issues,\n efficiencyScore(metrics, thresholds),\n `meanCostUsd=${fmt(metrics.meanCostUsd)} p95WallMs=${fmt(metrics.p95WallMs)}`,\n ),\n ]\n}\n\nfunction axis(\n name: ReleaseConfidenceAxisName,\n issues: ReleaseConfidenceIssue[],\n score: number | null,\n detail: string,\n): ReleaseConfidenceAxis {\n const own = issues.filter((i) => i.axis === name)\n const status = own.some((i) => i.severity === 'critical')\n ? 'fail'\n : own.length > 0\n ? 'warn'\n : 'pass'\n return { name, status, score: score === null ? null : bounded(score), detail }\n}\n\nfunction countScenarioSplits(scenarios: readonly DatasetScenario[]): Record<DatasetSplit, number> {\n const counts: Record<DatasetSplit, number> = { train: 0, dev: 0, test: 0, holdout: 0 }\n for (const scenario of scenarios) counts[scenario.split ?? 'train']++\n return counts\n}\n\nfunction countDomains(scenarios: readonly DatasetScenario[]): Record<string, number> {\n const out: Record<string, number> = {}\n for (const scenario of scenarios) {\n const domain = scenario.tags?.domain ?? scenario.tags?.category ?? 'uncategorized'\n out[domain] = (out[domain] ?? 0) + 1\n }\n return out\n}\n\nfunction countFailureClasses(\n runs: readonly RunRecord[],\n traces: readonly ReleaseTraceEvidence[],\n threshold: number,\n): Partial<Record<FailureClass, number>> {\n const out: Partial<Record<FailureClass, number>> = {}\n for (const run of runs) {\n // Ungated: a failure-mode census counts what the runs REPORTED, which is\n // the only way a gamed run's inflated score is visible at all. It is not a\n // promotion number — `passRate` is, and that one excludes gated runs and\n // publishes the excluded count as `metrics.realnessGatedRuns`.\n if (runPassOutcome(run, threshold) === false) {\n const failureClass =\n run.failureClass !== undefined && run.failureClass !== 'success'\n ? run.failureClass\n : 'unknown'\n out[failureClass] = (out[failureClass] ?? 0) + 1\n }\n }\n for (const trace of traces) {\n if (tracePassOutcome(trace, threshold) === false) {\n const failureClass =\n trace.failureClass !== undefined && trace.failureClass !== 'success'\n ? trace.failureClass\n : 'unknown'\n out[failureClass] = (out[failureClass] ?? 0) + 1\n }\n }\n return out\n}\n\nfunction countResponsibleSurfaces(traces: readonly ReleaseTraceEvidence[]): Record<string, number> {\n const out: Record<string, number> = {}\n for (const trace of traces) {\n for (const asi of trace.asi ?? []) {\n const surface = asi.responsibleSurface ?? 'unknown'\n out[surface] = (out[surface] ?? 0) + 1\n }\n }\n return out\n}\n\nfunction failedRows(\n runs: readonly RunRecord[],\n traces: readonly ReleaseTraceEvidence[],\n threshold: number,\n): Array<{ hasAsi: boolean }> {\n const out: Array<{ hasAsi: boolean }> = []\n for (const run of runs) {\n if (runPassOutcome(run, threshold) === false) {\n const asiMetric = run.outcome.raw.asi\n out.push({ hasAsi: typeof asiMetric === 'number' && asiMetric > 0 })\n }\n }\n for (const trace of traces) {\n if (tracePassOutcome(trace, threshold) === false) {\n out.push({ hasAsi: (trace.asi?.length ?? 0) > 0 })\n }\n }\n return out\n}\n\nfunction passOutcomeRate(outcomes: readonly (boolean | null)[]): number | null {\n const classified = outcomes.filter((outcome): outcome is boolean => outcome !== null)\n if (classified.length === 0) return null\n return classified.filter(Boolean).length / classified.length\n}\n\nfunction runPassOutcome(run: RunRecord, threshold: number): boolean | null {\n if (hasExplicitTaskFailure(run)) return false\n const score = runSplitScore(run)\n return score === undefined ? null : score >= threshold\n}\n\nfunction hasExplicitTaskFailure(run: RunRecord): boolean {\n return run.failureClass !== undefined && run.failureClass !== 'success'\n}\n\nfunction tracePassOutcome(trace: ReleaseTraceEvidence, threshold: number): boolean | null {\n if (trace.failureClass !== undefined && trace.failureClass !== 'success') return false\n if (trace.ok === false) return false\n if (isFiniteNumber(trace.score)) return trace.score >= threshold\n return trace.ok === true ? true : null\n}\n\nfunction isFailedTerminalOutcome(\n outcome: RunRecord['terminalOutcome'],\n): outcome is 'failed' | 'cancelled' | 'incomplete' {\n return outcome === 'failed' || outcome === 'cancelled' || outcome === 'incomplete'\n}\n\nfunction terminalSuccess(outcome: RunRecord['terminalOutcome']): boolean | undefined {\n if (outcome === 'succeeded') return true\n if (isFailedTerminalOutcome(outcome)) return false\n return undefined\n}\n\nfunction scoresFor(runs: readonly RunRecord[], split: RunSplitTag): number[] {\n return runs\n .filter((run) => run.splitTag === split)\n .map(runSplitScore)\n .filter(isFiniteNumber)\n}\n\n/**\n * RAW and split-exact (`observedSplitScore`): this feeds the per-split means,\n * the overfit gap, and the pass threshold — descriptions of what the runs\n * reported. A gamed run inflating them is visible next to `realnessGatedRuns`,\n * and the promotion number (`passRate`) excludes gated runs entirely.\n */\nfunction runSplitScore(run: RunRecord): number | undefined {\n const score = observedSplitScore(run, run.splitTag === 'holdout' ? 'holdout' : 'search')\n return isFiniteNumber(score) ? score : undefined\n}\n\nfunction meanOrNull(xs: readonly number[]): number | null {\n if (xs.length === 0) return null\n return xs.reduce((sum, x) => sum + x, 0) / xs.length\n}\n\nfunction percentileOrNull(xs: readonly number[], p: number): number | null {\n if (xs.length === 0) return null\n const sorted = [...xs].sort((a, b) => a - b)\n return sorted[Math.min(sorted.length - 1, Math.max(0, Math.ceil(p * sorted.length) - 1))]!\n}\n\nfunction isFiniteNumber(value: unknown): value is number {\n return typeof value === 'number' && Number.isFinite(value)\n}\n\nfunction diffOrNull(a: number | null, b: number | null): number | null {\n if (a === null || b === null) return null\n return a - b\n}\n\nfunction gapScore(gap: number | null, maxGap: number): number | null {\n if (gap === null) return null\n if (maxGap <= 0) return gap <= 0 ? 1 : 0\n return bounded(1 - Math.max(0, gap) / maxGap)\n}\n\nfunction efficiencyScore(\n metrics: ReleaseConfidenceMetrics,\n thresholds: Required<ReleaseConfidenceThresholds>,\n): number | null {\n const cost = Number.isFinite(thresholds.maxMeanCostUsd)\n ? metrics.meanCostUsd === null\n ? null\n : bounded(thresholds.maxMeanCostUsd / Math.max(metrics.meanCostUsd, 1e-12))\n : 1\n const latency = Number.isFinite(thresholds.maxP95WallMs)\n ? metrics.p95WallMs === null\n ? null\n : bounded(thresholds.maxP95WallMs / Math.max(metrics.p95WallMs, 1e-12))\n : 1\n if (cost === null || latency === null) return null\n return Math.min(cost, latency)\n}\n\nfunction bounded(x: number): number {\n if (!Number.isFinite(x)) return 0\n return Math.max(0, Math.min(1, x))\n}\n\nfunction renderSummary(\n target: string,\n status: ReleaseConfidenceStatus,\n metrics: ReleaseConfidenceMetrics,\n issues: ReleaseConfidenceIssue[],\n): string {\n const prefix = `release confidence ${status}: ${target}`\n const metricText = `scenarios=${metrics.scenarioCount} searchRuns=${metrics.searchRuns} holdoutRuns=${metrics.holdoutRuns} passRate=${fmt(metrics.passRate)} meanScore=${fmt(metrics.meanScore)}`\n if (issues.length === 0) return `${prefix}; ${metricText}`\n return `${prefix}; ${metricText}; issues=${issues.map((i) => i.code).join(',')}`\n}\n\nfunction fmt(x: number | null): string {\n if (x === null) return 'n/a'\n return x.toFixed(4)\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,SAAgB,YACd,UACA,WACA,UAA4B,CAAC,GACZ;CACjB,MAAM,QAAQ,QAAQ,SAAS;CAC/B,MAAM,aAAa,QAAQ,cAAc;CACzC,MAAM,WAAW,QAAQ,mBAAmB;CAC5C,MAAM,MAAM,WAAW,QAAQ,QAAQ,SAAS,UAAU,SAAS,CAAC;CAEpE,MAAM,eAAe,KAAK,QAAQ;CAClC,MAAM,gBAAgB,KAAK,SAAS;CACpC,MAAM,QAAQ,gBAAgB;CAE9B,IACE,SAAS,SAAS,UAAU,SAAS,YACrC,SAAS,WAAW,KACpB,UAAU,WAAW,GAErB,OAAO;EACL;EACA;EACA;EACA,SAAS;EACT,SAAS;EACT,YAAY;EACZ;EACA,SAAS;CACX;CAGF,MAAM,SAAmB,IAAI,MAAM,UAAU;CAC7C,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,KAAK;EACnC,MAAM,YAAY,SAAS,UAAU,GAAG;EAExC,OAAO,KAAK,KADM,SAAS,WAAW,GACb,CAAC,IAAI,KAAK,SAAS;CAC9C;CACA,OAAO,MAAM,GAAG,MAAM,IAAI,CAAC;CAC3B,MAAM,WAAW,KAAK,MAAO,QAAQ,IAAK,UAAU;CACpD,MAAM,WAAW,KAAK,OAAO,IAAI,QAAQ,KAAK,UAAU,IAAI;CAC5D,MAAM,UAAU,OAAO,KAAK,IAAI,GAAG,QAAQ;CAC3C,MAAM,UAAU,OAAO,KAAK,IAAI,aAAa,GAAG,QAAQ;CAExD,IAAI;CAMJ,IAAI,CAAC,OAAO,SAAS,OAAO,KAAK,CAAC,OAAO,SAAS,OAAO,KAAK,YAAY,SACxE,UAAU;MACL,IAAI,UAAU,GAAG,UAAU;MAC7B,IAAI,UAAU,GAAG,UAAU;MAC3B,IAAI,SAAS,GAAG,UAAU;MAC1B,UAAU;CAEf,OAAO;EACL;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAEA,SAAS,KAAK,IAAsB;CAClC,IAAI,GAAG,WAAW,GAAG,OAAO;CAC5B,IAAI,IAAI;CACR,KAAK,MAAM,KAAK,IAAI,KAAK;CACzB,OAAO,IAAI,GAAG;AAChB;AAEA,SAAS,SAAS,IAAc,KAA6B;CAC3D,MAAM,MAAM,IAAI,MAAM,GAAG,MAAM;CAC/B,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,QAAQ,KAAK,IAAI,KAAK,GAAG,KAAK,MAAM,IAAI,IAAI,GAAG,MAAM;CAC5E,OAAO;AACT;;AAGA,SAAS,SAAS,GAAa,GAAqB;CAClD,IAAI,IAAI;CACR,KAAK,MAAM,KAAK,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG;EAC5B,MAAM,OAAO,IAAI,aAAa,CAAC,CAAC,CAAC;EACjC,MAAM,QAAQ,IAAI,WAAW,KAAK,MAAM;EACxC,KAAK,MAAM,QAAQ,OAAO;GACxB,KAAK;GACL,IAAI,KAAK,KAAK,GAAG,QAAQ;EAC3B;CACF;CACA,OAAO,MAAM;AACf;;;;AAiCA,eAAsB,gBACpB,MACkF;CAClF,MAAM,cAAc,KAAK,oBAAoB;CAC7C,MAAM,iBAAiB,MAAM,SAAS,KAAK,iBAAiB,KAAK,OAAO,WAAW;CACnF,MAAM,kBAAkB,MAAM,SAAS,KAAK,kBAAkB,KAAK,OAAO,WAAW;CAMrF,OAAO;EACL,GANS,YAAY,gBAAgB,iBAAiB;GACtD,GAAI,KAAK,UAAU,KAAA,IAAY,EAAE,OAAO,KAAK,MAAM,IAAI,CAAC;GACxD,GAAI,KAAK,eAAe,KAAA,IAAY,EAAE,YAAY,KAAK,WAAW,IAAI,CAAC;GACvE,GAAI,KAAK,SAAS,KAAA,IAAY,EAAE,MAAM,KAAK,KAAK,IAAI,CAAC;EACvD,CAEM;EACJ,iBAAiB,eAAe;EAChC,kBAAkB,gBAAgB;CACpC;AACF;AAEA,eAAe,SACb,SACA,OACA,aACmB;CACnB,MAAM,UAAoB,IAAI,MAAM,QAAQ,MAAM;CAClD,IAAI,OAAO;CACX,eAAe,SAAwB;EACrC,OAAO,MAAM;GACX,MAAM,IAAI;GACV,IAAI,KAAK,QAAQ,QAAQ;GACzB,MAAM,IAAI,MAAM,MAAM,QAAQ,EAAG;GACjC,QAAQ,KAAK,OAAO,SAAS,CAAC,IAAI,IAAI;EACxC;CACF;CACA,MAAM,QAAQ,IAAI,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,GAAG,WAAW,EAAE,SAAS,OAAO,CAAC,CAAC;CAClF,OAAO;AACT;;;AChEA,MAAM,qBAA4D;CAChE,eAAe;CACf,kBAAkB;CAClB,eAAe;CACf,gBAAgB;CAChB,gBAAgB;CAChB,aAAa;CACb,cAAc;CACd,eAAe;CACf,gBAAgB,OAAO;CACvB,cAAc,OAAO;CACrB,uBAAuB;CACvB,uBAAuB;AACzB;AAEA,SAAgB,0BACd,OAC4B;CAC5B,MAAM,aAAa;EAAE,GAAG;EAAoB,GAAG,MAAM;CAAW;CAChE,MAAM,cAAc,MAAM,eAAe;CACzC,MAAM,OAAO,iBACV,MAAM,QAAQ,CAAC,EAAA,CAAG,IAAI,iBAAiB,GACxC,aACA,MAAM,UACR;CACA,MAAM,SAAS,sBACZ,MAAM,UAAU,CAAC,EAAA,CAAG,IAAI,4BAA4B,GACrD,aACA,MAAM,UACR;CACA,MAAM,YAAY,MAAM,aAAa,CAAC;CACtC,MAAM,gBAAgB,MAAM,SAAS,iBAAiB,UAAU;CAChE,MAAM,cAAc,MAAM,SAAS,eAAe,oBAAoB,SAAS;CAC/E,MAAM,eAAe,UAAU,MAAM,QAAQ;CAC7C,MAAM,gBAAgB,UAAU,MAAM,SAAS;CAC/C,MAAM,YAAY,KAAK,IAAI,aAAa,CAAC,CAAC,OAAO,cAAc;CAC/D,MAAM,cAAc,OAAO,KAAK,MAAM,EAAE,KAAK,CAAC,CAAC,OAAO,cAAc;CACpE,MAAM,gBAAgB,KAAK,SAAS,IAAI,YAAY;CACpD,MAAM,cAAc,KAAK,QAAQ,QAAQ,cAAc,GAAG,MAAM,KAAA,CAAS;CACzE,MAAM,eAAe,KAAK,QACvB,QACC,cAAc,GAAG,MAAM,KAAA,KACvB,CAAC,uBAAuB,GAAG,KAC3B,CAAC,wBAAwB,IAAI,eAAe,CAChD,CAAC,CAAC;CAOF,MAAM,aAAa,KAAK,QAAQ,QAAQ,CAAC,gBAAgB,GAAG,CAAC;CAC7D,MAAM,eACJ,KAAK,SAAS,IACV,WAAW,KAAK,QAAQ,eAAe,KAAK,WAAW,qBAAqB,CAAC,IAC7E,OAAO,KAAK,UAAU,iBAAiB,OAAO,WAAW,qBAAqB,CAAC;CACrF,MAAM,kBACJ,KAAK,SAAS,IACV,KAAK,KAAK,QAAQ,gBAAgB,IAAI,eAAe,CAAC,IACtD,OAAO,KAAK,UAAU,MAAM,EAAE;CACpC,MAAM,2BAA2B,gBAAgB,QAAQ,YAAY,YAAY,KAAA,CAAS,CAAC,CAAC;CAC5F,MAAM,sBAAsB,gBAAgB,QAAQ,YAAY,YAAY,KAAK,CAAC,CAAC;CACnF,MAAM,kBACJ,gBAAgB,WAAW,KAAK,2BAA2B,IACvD,QACC,gBAAgB,SAAS,uBAAuB,gBAAgB;CACvE,MAAM,aAAa,YAAY,QAAQ,MAAM,EAAE,aAAa,QAAQ,CAAC,CAAC;CACtE,MAAM,cAAc,YAAY,QAAQ,MAAM,EAAE,aAAa,SAAS,CAAC,CAAC;CACxE,MAAM,SAAS,WAAW,MAAM,QAAQ,WAAW,qBAAqB;CACxE,MAAM,kBAAkB,WAAW,YAAY;CAC/C,MAAM,mBAAmB,WAAW,aAAa;CACjD,MAAM,WAAW,KAAK,SAAS,QAC7B,IAAI,eAAe,SAAS,eAAe,CAAC,IAAI,CAAC,IAAI,eAAe,GAAG,CACzE;CACA,MAAM,aAAa,OAAO,KAAK,UAAU,MAAM,OAAO,CAAC,CAAC,OAAO,cAAc;CAC7E,MAAM,cACJ,KAAK,SAAS,IACV,SAAS,WAAW,KAAK,SACvB,WAAW,QAAQ,IACnB,OACF,WAAW,UAAU;CAC3B,MAAM,YACJ,KAAK,SAAS,IACV,KAAK,KAAK,QAAQ,IAAI,MAAM,IAC5B,OAAO,KAAK,UAAU,MAAM,UAAU,CAAC,CAAC,OAAO,cAAc;CACnE,MAAM,UAAoC;EACxC;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU,gBAAgB,YAAY;EACtC,mBAAmB,KAAK,SAAS,WAAW;EAC5C,WAAW,WAAW,aAAa;EACnC;EACA;EACA,YAAY,WAAW,iBAAiB,gBAAgB;EACxD;EACA,WAAW,iBAAiB,WAAW,GAAI;EAC3C,YAAY,OAAO;EACnB,iBAAiB,OAAO,QAAQ,QAAQ,IAAI,MAAM,CAAC,CAAC;EACpD,kBAAkB,OAAO,QAAQ,MAAM,EAAE,cAAc,CAAC,CAAC,CAAC;EAC1D,iBAAiB,OAAO,QAAQ,OAAO,EAAE,aAAa,KAAK,CAAC,CAAC,CAAC;EAC9D;EACA,cAAc,aAAa,SAAS;EACpC,oBAAoB,oBAAoB,MAAM,QAAQ,WAAW,qBAAqB;EACtF,0BAA0B,yBAAyB,MAAM;CAC3D;CAEA,MAAM,SAAmC,CAAC;CAC1C,YAAY,OAAO,YAAY,SAAS,MAAM;CAC9C,aAAa,YAAY,SAAS,MAAM;CACxC,iBAAiB,SAAS,MAAM;CAChC,oBAAoB,MAAM,gBAAgB,MAAM,YAAY,SAAS,MAAM;CAC3E,iBAAiB,YAAY,SAAS,MAAM;CAC5C,gBAAgB,YAAY,SAAS,MAAM;CAE3C,MAAM,OAAO,UAAU,SAAS,YAAY,MAAM;CAClD,MAAM,SAAS,OAAO,MAAM,MAAM,EAAE,aAAa,UAAU,IACvD,SACA,OAAO,SAAS,IACd,SACA;CAEN,OAAO;EACL,QAAQ,MAAM;EACd;EACA,YAAY,MAAM,cAAc;EAChC;EACA,SAAS,WAAW,WAAW,MAAM,eAAe,MAAM,aAAa,UAAU;EACjF;EACA;EACA;EACA,SAAS,MAAM,WAAW;EAC1B,cAAc,MAAM,gBAAgB;EACpC,SAAS,cAAc,MAAM,QAAQ,QAAQ,SAAS,MAAM;CAC9D;AACF;AAEA,SAAgB,wBAAwB,OAA2D;CACjG,MAAM,YAAY,0BAA0B,KAAK;CACjD,IAAI,UAAU,WAAW,QACvB,MAAM,IAAI,kBAAkB,UAAU,OAAO;CAE/C,OAAO;AACT;AAEA,SAAS,gBACP,MACA,aACA,YACa;CACb,IAAI,aAAa,OAAO,KAAK,QAAQ,MAAM,EAAE,gBAAgB,WAAW;CACxE,IAAI,YAAY,OAAO,KAAK,QAAQ,MAAM,EAAE,gBAAgB,UAAU;CACtE,OAAO,CAAC,GAAG,IAAI;AACjB;AAEA,SAAS,qBACP,QACA,aACA,YACwB;CACxB,IAAI,aACF,OAAO,OAAO,QAAQ,MAAM,EAAE,gBAAgB,KAAA,KAAa,EAAE,gBAAgB,WAAW;CAC1F,IAAI,YACF,OAAO,OAAO,QAAQ,MAAM,EAAE,gBAAgB,KAAA,KAAa,EAAE,gBAAgB,UAAU;CACzF,OAAO,CAAC,GAAG,MAAM;AACnB;AAEA,SAAS,6BACP,OACA,OACsB;CACtB,MAAM,QAAQ;CACd,IAAI,OAAO,OAAO,OAAO,aAAa,GACpC,MAAM,IAAI,gBACR,UAAU,MAAM,2DAClB;CAEF,IAAI,MAAM,iBAAiB,KAAA,KAAa,CAAC,gBAAgB,SAAS,MAAM,YAAY,GAClF,MAAM,IAAI,gBACR,UAAU,MAAM,gCAAgC,gBAAgB,KAAK,IAAI,GAC3E;CAEF,OAAO;AACT;AAEA,SAAS,YACP,OACA,YACA,SACA,QACM;CACN,IAAI,WAAW,iBAAiB,CAAC,MAAM,YAAY,MAAM,WAAW,UAAU,OAAO,GACnF,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ;CACV,CAAC;CAEH,IAAI,QAAQ,gBAAgB,WAAW,kBACrC,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,cAAc,qBAAqB,WAAW,iBAAiB;CACpF,CAAC;CAEH,IAAI,WAAW,kBAAkB,QAAQ,YAAY,YAAY,GAC/D,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ;CACV,CAAC;AAEL;AAEA,SAAS,aACP,YACA,SACA,QACM;CACN,IAAI,QAAQ,aAAa,WAAW,eAClC,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,WAAW,uBAAuB,WAAW,cAAc;CAChF,CAAC;CAEH,IAAI,QAAQ,eAAe,GACzB,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,aAAa;CAClC,CAAC;CAEH,IAAI,QAAQ,aAAa,QAAQ,QAAQ,cAAc,MACrD,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ;CACV,CAAC;CAEH,IAAI,QAAQ,aAAa,QAAQ,QAAQ,WAAW,WAAW,aAC7D,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,YAAY,IAAI,QAAQ,QAAQ,EAAE,KAAK,IAAI,WAAW,WAAW,EAAE;CAC7E,CAAC;CAEH,IAAI,QAAQ,cAAc,QAAQ,QAAQ,YAAY,WAAW,cAC/D,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,aAAa,IAAI,QAAQ,SAAS,EAAE,KAAK,IAAI,WAAW,YAAY,EAAE;CAChF,CAAC;AAEL;AAEA,SAAS,iBACP,SACA,QACM;CACN,IAAI,QAAQ,oBAAoB,MAC9B,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QACE,QAAQ,2BAA2B,IAC/B,GAAG,QAAQ,yBAAyB,wDACpC;CACR,CAAC;CAEH,IAAI,QAAQ,sBAAsB,GAChC,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,oBAAoB;CACzC,CAAC;AAEL;AAEA,SAAS,oBACP,cACA,YACA,SACA,QACM;CACN,IAAI,WAAW,kBAAkB,QAAQ,cAAc,WAAW,gBAChE,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,YAAY,wBAAwB,WAAW,eAAe;CACnF,CAAC;CAEH,IAAI,QAAQ,eAAe,QAAQ,QAAQ,aAAa,WAAW,eACjE,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,sBAAsB,IAAI,QAAQ,UAAU,EAAE,KAAK,IAAI,WAAW,aAAa,EAAE;CAC3F,CAAC;CAEH,IAAI,gBAAgB,CAAC,aAAa,SAChC,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM,QAAQ,aAAa,iBAAiB;EAC5C,QAAQ,aAAa;CACvB,CAAC;AAEL;AAEA,SAAS,iBACP,YACA,SACA,QACM;CACN,IAAI,CAAC,WAAW,uBAAuB;CACvC,IAAI,QAAQ,aAAa,QAAQ,iBAC/B,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,GAAG,QAAQ,aAAa,QAAQ,gBAAgB;CAC1D,CAAC;AAEL;AAEA,SAAS,gBACP,YACA,SACA,QACM;CACN,IAAI,OAAO,SAAS,WAAW,cAAc,KAAK,QAAQ,gBAAgB,MACxE,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ;CACV,CAAC;MACI,IAAI,QAAQ,gBAAgB,QAAQ,QAAQ,cAAc,WAAW,gBAC1E,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,eAAe,IAAI,QAAQ,WAAW,EAAE,KAAK,IAAI,WAAW,cAAc,EAAE;CACtF,CAAC;CAEH,IAAI,OAAO,SAAS,WAAW,YAAY,KAAK,QAAQ,cAAc,MACpE,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ;CACV,CAAC;MACI,IAAI,QAAQ,cAAc,QAAQ,QAAQ,YAAY,WAAW,cACtE,OAAO,KAAK;EACV,MAAM;EACN,UAAU;EACV,MAAM;EACN,QAAQ,aAAa,IAAI,QAAQ,SAAS,EAAE,KAAK,IAAI,WAAW,YAAY,EAAE;CAChF,CAAC;AAEL;AAEA,SAAS,UACP,SACA,YACA,QACyB;CACzB,OAAO;EACL,KACE,UACA,QACA,QAAQ,QAAQ,gBAAgB,KAAK,IAAI,GAAG,WAAW,gBAAgB,CAAC,GACxE,GAAG,QAAQ,cAAc,sBAAsB,QAAQ,YAAY,SACrE;EACA,KACE,WACA,QACA,QAAQ,aAAa,QAAQ,QAAQ,cAAc,OAC/C,OACA,KAAK,IAAI,QAAQ,UAAU,QAAQ,SAAS,GAChD,YAAY,IAAI,QAAQ,QAAQ,EAAE,aAAa,IAAI,QAAQ,SAAS,GACtE;EACA,KACE,eACA,QACA,QAAQ,iBACR,eAAe,IAAI,QAAQ,eAAe,EAAE,oBAAoB,QAAQ,oBAAoB,gBAAgB,QAAQ,0BACtH;EACA,KACE,kBACA,QACA,SAAS,QAAQ,YAAY,WAAW,aAAa,GACrD,eAAe,QAAQ,YAAY,cAAc,IAAI,QAAQ,UAAU,GACzE;EACA,KACE,eACA,QACA,QAAQ,eAAe,IAAI,IAAI,QAAQ,kBAAkB,QAAQ,YACjE,mBAAmB,QAAQ,gBAAgB,GAAG,QAAQ,YACxD;EACA,KACE,cACA,QACA,gBAAgB,SAAS,UAAU,GACnC,eAAe,IAAI,QAAQ,WAAW,EAAE,aAAa,IAAI,QAAQ,SAAS,GAC5E;CACF;AACF;AAEA,SAAS,KACP,MACA,QACA,OACA,QACuB;CACvB,MAAM,MAAM,OAAO,QAAQ,MAAM,EAAE,SAAS,IAAI;CAMhD,OAAO;EAAE;EAAM,QALA,IAAI,MAAM,MAAM,EAAE,aAAa,UAAU,IACpD,SACA,IAAI,SAAS,IACX,SACA;EACiB,OAAO,UAAU,OAAO,OAAO,QAAQ,KAAK;EAAG;CAAO;AAC/E;AAEA,SAAS,oBAAoB,WAAqE;CAChG,MAAM,SAAuC;EAAE,OAAO;EAAG,KAAK;EAAG,MAAM;EAAG,SAAS;CAAE;CACrF,KAAK,MAAM,YAAY,WAAW,OAAO,SAAS,SAAS,QAAQ;CACnE,OAAO;AACT;AAEA,SAAS,aAAa,WAA+D;CACnF,MAAM,MAA8B,CAAC;CACrC,KAAK,MAAM,YAAY,WAAW;EAChC,MAAM,SAAS,SAAS,MAAM,UAAU,SAAS,MAAM,YAAY;EACnE,IAAI,WAAW,IAAI,WAAW,KAAK;CACrC;CACA,OAAO;AACT;AAEA,SAAS,oBACP,MACA,QACA,WACuC;CACvC,MAAM,MAA6C,CAAC;CACpD,KAAK,MAAM,OAAO,MAKhB,IAAI,eAAe,KAAK,SAAS,MAAM,OAAO;EAC5C,MAAM,eACJ,IAAI,iBAAiB,KAAA,KAAa,IAAI,iBAAiB,YACnD,IAAI,eACJ;EACN,IAAI,iBAAiB,IAAI,iBAAiB,KAAK;CACjD;CAEF,KAAK,MAAM,SAAS,QAClB,IAAI,iBAAiB,OAAO,SAAS,MAAM,OAAO;EAChD,MAAM,eACJ,MAAM,iBAAiB,KAAA,KAAa,MAAM,iBAAiB,YACvD,MAAM,eACN;EACN,IAAI,iBAAiB,IAAI,iBAAiB,KAAK;CACjD;CAEF,OAAO;AACT;AAEA,SAAS,yBAAyB,QAAiE;CACjG,MAAM,MAA8B,CAAC;CACrC,KAAK,MAAM,SAAS,QAClB,KAAK,MAAM,OAAO,MAAM,OAAO,CAAC,GAAG;EACjC,MAAM,UAAU,IAAI,sBAAsB;EAC1C,IAAI,YAAY,IAAI,YAAY,KAAK;CACvC;CAEF,OAAO;AACT;AAEA,SAAS,WACP,MACA,QACA,WAC4B;CAC5B,MAAM,MAAkC,CAAC;CACzC,KAAK,MAAM,OAAO,MAChB,IAAI,eAAe,KAAK,SAAS,MAAM,OAAO;EAC5C,MAAM,YAAY,IAAI,QAAQ,IAAI;EAClC,IAAI,KAAK,EAAE,QAAQ,OAAO,cAAc,YAAY,YAAY,EAAE,CAAC;CACrE;CAEF,KAAK,MAAM,SAAS,QAClB,IAAI,iBAAiB,OAAO,SAAS,MAAM,OACzC,IAAI,KAAK,EAAE,SAAS,MAAM,KAAK,UAAU,KAAK,EAAE,CAAC;CAGrD,OAAO;AACT;AAEA,SAAS,gBAAgB,UAAsD;CAC7E,MAAM,aAAa,SAAS,QAAQ,YAAgC,YAAY,IAAI;CACpF,IAAI,WAAW,WAAW,GAAG,OAAO;CACpC,OAAO,WAAW,OAAO,OAAO,CAAC,CAAC,SAAS,WAAW;AACxD;AAEA,SAAS,eAAe,KAAgB,WAAmC;CACzE,IAAI,uBAAuB,GAAG,GAAG,OAAO;CACxC,MAAM,QAAQ,cAAc,GAAG;CAC/B,OAAO,UAAU,KAAA,IAAY,OAAO,SAAS;AAC/C;AAEA,SAAS,uBAAuB,KAAyB;CACvD,OAAO,IAAI,iBAAiB,KAAA,KAAa,IAAI,iBAAiB;AAChE;AAEA,SAAS,iBAAiB,OAA6B,WAAmC;CACxF,IAAI,MAAM,iBAAiB,KAAA,KAAa,MAAM,iBAAiB,WAAW,OAAO;CACjF,IAAI,MAAM,OAAO,OAAO,OAAO;CAC/B,IAAI,eAAe,MAAM,KAAK,GAAG,OAAO,MAAM,SAAS;CACvD,OAAO,MAAM,OAAO,OAAO,OAAO;AACpC;AAEA,SAAS,wBACP,SACkD;CAClD,OAAO,YAAY,YAAY,YAAY,eAAe,YAAY;AACxE;AAEA,SAAS,gBAAgB,SAA4D;CACnF,IAAI,YAAY,aAAa,OAAO;CACpC,IAAI,wBAAwB,OAAO,GAAG,OAAO;AAE/C;AAEA,SAAS,UAAU,MAA4B,OAA8B;CAC3E,OAAO,KACJ,QAAQ,QAAQ,IAAI,aAAa,KAAK,CAAC,CACvC,IAAI,aAAa,CAAC,CAClB,OAAO,cAAc;AAC1B;;;;;;;AAQA,SAAS,cAAc,KAAoC;CACzD,MAAM,QAAQ,mBAAmB,KAAK,IAAI,aAAa,YAAY,YAAY,QAAQ;CACvF,OAAO,eAAe,KAAK,IAAI,QAAQ,KAAA;AACzC;AAEA,SAAS,WAAW,IAAsC;CACxD,IAAI,GAAG,WAAW,GAAG,OAAO;CAC5B,OAAO,GAAG,QAAQ,KAAK,MAAM,MAAM,GAAG,CAAC,IAAI,GAAG;AAChD;AAEA,SAAS,iBAAiB,IAAuB,GAA0B;CACzE,IAAI,GAAG,WAAW,GAAG,OAAO;CAC5B,MAAM,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;CAC3C,OAAO,OAAO,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,OAAO,MAAM,IAAI,CAAC,CAAC;AACzF;AAEA,SAAS,eAAe,OAAiC;CACvD,OAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK;AAC3D;AAEA,SAAS,WAAW,GAAkB,GAAiC;CACrE,IAAI,MAAM,QAAQ,MAAM,MAAM,OAAO;CACrC,OAAO,IAAI;AACb;AAEA,SAAS,SAAS,KAAoB,QAA+B;CACnE,IAAI,QAAQ,MAAM,OAAO;CACzB,IAAI,UAAU,GAAG,OAAO,OAAO,IAAI,IAAI;CACvC,OAAO,QAAQ,IAAI,KAAK,IAAI,GAAG,GAAG,IAAI,MAAM;AAC9C;AAEA,SAAS,gBACP,SACA,YACe;CACf,MAAM,OAAO,OAAO,SAAS,WAAW,cAAc,IAClD,QAAQ,gBAAgB,OACtB,OACA,QAAQ,WAAW,iBAAiB,KAAK,IAAI,QAAQ,aAAa,KAAK,CAAC,IAC1E;CACJ,MAAM,UAAU,OAAO,SAAS,WAAW,YAAY,IACnD,QAAQ,cAAc,OACpB,OACA,QAAQ,WAAW,eAAe,KAAK,IAAI,QAAQ,WAAW,KAAK,CAAC,IACtE;CACJ,IAAI,SAAS,QAAQ,YAAY,MAAM,OAAO;CAC9C,OAAO,KAAK,IAAI,MAAM,OAAO;AAC/B;AAEA,SAAS,QAAQ,GAAmB;CAClC,IAAI,CAAC,OAAO,SAAS,CAAC,GAAG,OAAO;CAChC,OAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AACnC;AAEA,SAAS,cACP,QACA,QACA,SACA,QACQ;CACR,MAAM,SAAS,sBAAsB,OAAO,IAAI;CAChD,MAAM,aAAa,aAAa,QAAQ,cAAc,cAAc,QAAQ,WAAW,eAAe,QAAQ,YAAY,YAAY,IAAI,QAAQ,QAAQ,EAAE,aAAa,IAAI,QAAQ,SAAS;CAC9L,IAAI,OAAO,WAAW,GAAG,OAAO,GAAG,OAAO,IAAI;CAC9C,OAAO,GAAG,OAAO,IAAI,WAAW,WAAW,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,GAAG;AAC/E;AAEA,SAAS,IAAI,GAA0B;CACrC,IAAI,MAAM,MAAM,OAAO;CACvB,OAAO,EAAE,QAAQ,CAAC;AACpB"}
|