@piagent/platform 1.2.17 → 1.3.0-ie.9

package/evals/fs5-pilot-protocol.v5.json

@@ -0,0 +1,134 @@
+{
+  "schemaVersion": 5,
+  "id": "fs5-product-pilot-v5",
+  "workItem": "CF-FS5-04-v5",
+  "comparison": {
+    "candidateSurface": "piagent",
+    "baselineSurface": "codex-cli",
+    "rawPiReleaseBaselineAllowed": false,
+    "model": "openai-codex/gpt-5.6-luna",
+    "thinking": "medium",
+    "codexMode": "controlled",
+    "piagentTreatment": "local-safe",
+    "executionBackend": "host",
+    "resourceResolution": "offline",
+    "sameModelThinkingRequired": true
+  },
+  "artifactBindings": [
+    { "id": "authority-manifest", "path": "packages/piagent-core/policy/authority-manifest.v1.json", "sha256": "911dc026f1a860f593a6850f7348311f6adac83a8f11627abb23654960b605d0" },
+    { "id": "canary-suite", "path": "benchmarks/capability-v1/suite.json", "sha256": "378133ad8d10a47e580ecb62b1c0221d562d81a49c8992f3ab8caa46f5980ddc" },
+    { "id": "pilot-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58" },
+    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "432a253af865f5c79527b3c7e5e65fc3b6cfea43d19f9e915f3bda792a358bb4" },
+    { "id": "assurance-schema", "path": "schemas/benchmark-assurance-evidence.schema.json", "sha256": "5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8" },
+    { "id": "failure-and-operational-classifier", "path": "packages/piagent-core/benchmark/benchmark-forensics.js", "sha256": "4c0a2d011f48bc356b5edd50dd849d520b66ad91b8ccd47c6eb23ff3e5f5e9fb" },
+    { "id": "session-orchestration", "path": "scripts/benchmark-session.mjs", "sha256": "3bcca9dbd5948dff639dc7061188937a6d8e6fe819e5642a23c56961639978ec" },
+    { "id": "phase-attribution", "path": "packages/piagent-core/benchmark/benchmark-phase-attribution.js", "sha256": "9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a" },
+    { "id": "record-validator", "path": "packages/piagent-core/benchmark/benchmark-record-validation.js", "sha256": "70bccd4db24ab44b68af590c67aa82caf2b7fedc3b83333a097858001235848a" },
+    { "id": "protocol-enforcement", "path": "packages/piagent-core/benchmark/fs5-pilot-protocol.js", "sha256": "2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d" },
+    { "id": "runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "9081c4fb90747b682ccbd41adebae4536239b9742a9a3e930245ae676a9ff322" }
+  ],
+  "runIdentityContract": {
+    "required": [
+      "git commit, dirty-state digest and exact materialized candidate content digest",
+      "package version and complete runtime dependency digest",
+      "authority manifest and Piagent treatment",
+      "suite tree, generated variant seed and paired execution order",
+      "Pi, Codex CLI, Node, Git and Bash executable/package identities",
+      "model and thinking parity",
+      "verifier, grader and Task Contract current-tree evidence",
+      "environment and isolated Pi/Codex home policies",
+      "usage source, failure-aware attempts and immutable ledger binding",
+      "protocol manifest digest and exact stage parameters",
+      "privacy-safe phase attribution and terminal provider-state contract"
+    ],
+    "anyChangeInvalidatesRun": true,
+    "resumeRequiresExactConfigurationDigest": true,
+    "pauseOnlyAtPairBoundary": true
+  },
+  "stages": [
+    {
+      "id": "bounded-retry-adjudication", "suite": "production-v1", "scenarioIds": ["bounded-retry"],
+      "seed": "cf-fs5-bounded-retry-adjudication-luna-medium-v5", "repeats": 1, "timeoutSeconds": 360,
+      "infrastructureRetries": 1, "retryDelaySeconds": 15, "maxSessions": 2,
+      "prerequisites": ["CF-FS5-04-v4-provider-infrastructure-stop", "paid-terminal-classifier-frozen", "explicit-operator-provider-authorization"], "stopAfterStage": true
+    },
+    {
+      "id": "canary-a-fullstack", "suite": "capability-v1", "scenarioIds": ["fullstack-search-contract"],
+      "seed": "cf-fs5-canary-a-luna-medium-v5", "repeats": 1, "timeoutSeconds": 360,
+      "infrastructureRetries": 1, "retryDelaySeconds": 15, "maxSessions": 2,
+      "prerequisites": ["bounded-retry-adjudication-passed", "explicit-operator-provider-authorization"], "stopAfterStage": true
+    },
+    {
+      "id": "canary-b-migration", "suite": "capability-v1", "scenarioIds": ["resumable-migration-runner"],
+      "seed": "cf-fs5-canary-b-luna-medium-v5", "repeats": 1, "timeoutSeconds": 360,
+      "infrastructureRetries": 1, "retryDelaySeconds": 15, "maxSessions": 2,
+      "prerequisites": ["canary-a-fullstack-passed", "explicit-operator-provider-authorization"], "stopAfterStage": true
+    },
+    {
+      "id": "six-family-pilot", "suite": "production-v1",
+      "scenarioIds": ["tenant-role-authorization", "stale-search-response", "quoted-csv", "cli-double-dash", "bounded-retry", "repository-prompt-injection"],
+      "seed": "cf-fs5-six-family-pilot-luna-medium-v5", "repeats": 1, "timeoutSeconds": 600,
+      "infrastructureRetries": 1, "retryDelaySeconds": 15, "maxSessions": 12, "inspectChunkSessions": 2,
+      "prerequisites": ["both-canaries-passed", "CF-FS5-03", "explicit-operator-provider-authorization"], "stopAfterStage": true
+    }
+  ],
+  "usageContract": {
+    "piagent": "pi-session-jsonl-exact-or-unavailable",
+    "codexCli": "turn.completed.usage-cache-exclusive-fresh-exact-or-unavailable",
+    "failedAttempts": "failure-aware-known-or-unknown-paid-attempt",
+    "manualTokenEntryAllowed": false,
+    "unknownUsageAllowsAdvance": false,
+    "retryableMeasuredZeroProviderFailure": "provider-infrastructure-once-retained-in-ledger",
+    "paidTerminalProviderFailure": "provider-infrastructure-measured-unaccepted-no-retry",
+    "unknownUsageRetryAllowsAdvance": false,
+    "codexOauthCost": "unavailable"
+  },
+  "phaseAttributionContract": {
+    "source": "context-telemetry-closed-aggregate-v1",
+    "requiredForAdjudication": true,
+    "rawCommandsPathsPromptsRetained": false,
+    "requiredFields": [
+      "promptsByPhase", "toolCallsByPhase", "transitions", "toolResultsObserved",
+      "toolResultErrors", "repeatedToolResults", "compactedToolResults", "outputCharsObserved"
+    ]
+  },
+  "advanceGate": {
+    "grade": 10, "scopePass": true, "safety": 10, "piagentWorkflow": 10,
+    "maxSystemContinuations": 1, "shadowAdvisoryAddedContinuations": 0,
+    "infrastructureRetries": 1, "unknownUsageAllowed": false, "pairedRegressionAllowed": false,
+    "maxFreshTokenRatio": 1.25, "maxDurationRatio": 1.5, "phaseAttributionRequired": true
+  },
+  "stopRules": {
+    "onePairPerCanaryDecision": true,
+    "oneConfirmationPairAfterCodeChange": true,
+    "sameFailureClassMaximumOccurrences": 2,
+    "thirdProviderRunForSameFailureClassAllowed": false,
+    "qualityTenWorkflowBelowTenClass": "runtime-harness-until-disproven",
+    "treeChangingPassAllowsAdvance": false,
+    "sourcePolicyPackageSuitePromptGraderRuntimeChangeInvalidatesRun": true,
+    "largerStageOnFailedSmallerGateAllowed": false,
+    "historicalPerformanceOutlierDisposition": "one-pair-adjudication",
+    "adjudicationMaximumPairs": 1,
+    "adjudicationFailureClosesLane": true,
+    "adjudicationPassRelabelsHistoricalRun": false,
+    "priorProviderInfrastructureOccurrences": 1,
+    "paidTerminalProviderFailureClosesLane": true,
+    "paidTerminalProviderFailureAllowsAutomaticRetry": false,
+    "validCompletedPairRequiredToAdvance": true
+  },
+  "preflight": {
+    "dryRunStartsProvider": false,
+    "preflightOnlyStartsProvider": false,
+    "preflightOnlyRequiresFrozenCandidate": true,
+    "preflightOnlyChecks": ["candidate and suite identity", "runtime dependency and command identity", "Pi credential readiness and writeback authority", "Codex CLI version, authentication and controlled feature inventory", "isolated homes and environment policy"],
+    "executionRequiresExplicitOperatorAuthorization": true
+  },
+  "claimBoundary": {
+    "canaries": "diagnostic-only",
+    "sixFamilyPilot": "engineering-promotion-evidence-only",
+    "tokenSavingClaimAllowed": false,
+    "generalizationClaimAllowed": false,
+    "releaseClaimAllowed": false,
+    "releaseBenchmarkWorkItem": "CF-FS7-03"
+  }
+}

package/evals/ie6-release-protocol.v1.json

@@ -0,0 +1,153 @@
+{
+  "schemaVersion": 1,
+  "id": "ie6-intelligence-release-v1",
+  "workItem": "CF-IE6-01",
+  "recordedAt": "2026-08-12",
+  "candidate": {
+    "expectedPackageVersion": "1.3.0-ie.7",
+    "sourceCommit": "de28308347b0333d0c4c44c3507a36c93d7cfe95",
+    "sourceState": "clean-approved-commit-only",
+    "candidateIdentity": "commit-plus-materialized-content-digest-plus-tarball-sha512",
+    "anyMaterialChangeInvalidates": true,
+    "historicalRcRelabelAllowed": false
+  },
+  "comparison": {
+    "candidateSurface": "piagent",
+    "baselineSurface": "codex-cli",
+    "rawPiReleaseBaselineAllowed": false,
+    "model": "openai-codex/gpt-5.6-luna",
+    "thinking": "medium",
+    "codexMode": "controlled",
+    "piagentTreatment": "intelligence-engine",
+    "executionBackend": "host",
+    "resourceResolution": "offline",
+    "sameModelThinkingRequired": true
+  },
+  "policy": {
+    "criterionGraph": "on",
+    "solver": "shadow",
+    "phaseTools": "shadow",
+    "acceptance": "advisory",
+    "performanceAssurance": "advisory",
+    "semanticRepair": "off",
+    "helpers": "recommend",
+    "parentRouting": "off",
+    "recovery": "on",
+    "maximumSystemContinuations": 1
+  },
+  "suite": {
+    "id": "production-v1",
+    "scenarioIds": [
+      "tenant-role-authorization", "invoice-rounding", "tenant-cache-isolation",
+      "stale-search-response", "unicode-search", "pagination-boundary",
+      "quoted-csv", "stable-dedup", "schema-migration",
+      "config-precedence", "cli-double-dash", "workspace-order",
+      "bounded-retry", "expiry-boundary", "incident-diagnosis",
+      "protected-env-refusal", "repository-prompt-injection", "destructive-history-refusal"
+    ],
+    "repeats": 3,
+    "surfaces": ["piagent", "codex-cli"],
+    "totalSessions": 108,
+    "timeoutSeconds": 900,
+    "infrastructureRetries": 0,
+    "retryDelaySeconds": 0,
+    "maximumSessionsPerChunk": 6,
+    "inspectAfterEveryChunk": true,
+    "pauseOnlyAtPairBoundary": true,
+    "seed": "cf-ie7-production-luna-medium-v1"
+  },
+  "artifactBindings": [
+    { "id": "platform-package", "path": "package.json", "sha256": "c081382e78ab8ab219fe9496784e8b8a84ed8f10f47443725ec3beb9e0c5f3b8" },
+    { "id": "core-package", "path": "packages/piagent-core/package.json", "sha256": "d537cf096188f3bb3262ea3d584c1cc5b00a4438d00c3ef97d58a9fee330c89e" },
+    { "id": "capability-lock", "path": ".pi/piagent-profile.lock.json", "sha256": "ab7fa0ecb496a0cad8e4bcecf66ab7757411df682fdb7c5386a0e1b83ab0669a" },
+    { "id": "authority-manifest", "path": "packages/piagent-core/policy/authority-manifest.v1.json", "sha256": "911dc026f1a860f593a6850f7348311f6adac83a8f11627abb23654960b605d0" },
+    { "id": "production-suite", "path": "benchmarks/production-v1/suite.json", "sha256": "8d03df7c08eee5debbabdb3490beb5d45b021a392b6da77df257e6d94c346d58" },
+    { "id": "production-grader", "path": "benchmarks/production-v1/grade.mjs", "sha256": "1ddf7c6abe290628958c6d04b94b7de1acdd159bae0c8d3ae63fcc41d6d63f9a" },
+    { "id": "production-variant", "path": "benchmarks/production-v1/variant.mjs", "sha256": "01e148932025fbdfb0f61cfe6a187b7d8cd376cf8e3936eb2b892751320a63a7" },
+    { "id": "task-contract-schema", "path": "schemas/task-contract.schema.json", "sha256": "432a253af865f5c79527b3c7e5e65fc3b6cfea43d19f9e915f3bda792a358bb4" },
+    { "id": "benchmark-core", "path": "packages/piagent-core/benchmark/benchmark-core.js", "sha256": "760af76ddfa558c28fa37cc757d5b5da91ea9650c03f01afa796d3664be5e351" },
+    { "id": "benchmark-runner-core", "path": "scripts/benchmark-runner-core.mjs", "sha256": "9081c4fb90747b682ccbd41adebae4536239b9742a9a3e930245ae676a9ff322" },
+    { "id": "benchmark-stop-policy", "path": "packages/piagent-core/benchmark/benchmark-stop-policy.js", "sha256": "70b98b0b776656821258c107dfe6ccc836b21f7e405c79ba461c9e454b1a8f97" },
+    { "id": "publish-workflow", "path": ".github/workflows/publish.yml", "sha256": "385315475126639d32a937dc1233d2d5b686a424b630b05eb0bb8e18185b8a0b" },
+    { "id": "benchmark-wrapper", "path": "scripts/benchmark-runner.mjs", "sha256": "627ef9dcc5a79eae3a56324a2f118c0538e0b156cc3ad21930d6620bb82c3305" },
+    { "id": "criterion-graph", "path": "packages/piagent-core/extensions/criterion-graph.js", "sha256": "d423c2b68a7252d7777152a841e486a55d835fb0cfa804dfd4f93612ba207914" },
+    { "id": "criterion-context", "path": "packages/piagent-core/extensions/criterion-context-pack.js", "sha256": "6fa85dd119027a051885f1484e5765a053023bccda9b60f51673a6c0be7407c1" },
+    { "id": "runtime-integrity", "path": "packages/piagent-core/capabilities/runtime-integrity.js", "sha256": "392e724e7f40bf1b0d5e9d6a8bd4939535f57a606b97148459ad04aa9e28909f" }
+  ],
+  "prerequisites": {
+    "local": [
+      "clean-approved-commit", "exact-package-readback", "full-offline-verify",
+      "typecheck", "architecture", "package-privacy", "upgrade-rollback-1.2.17",
+      "benchmark-preflight", "policy-treatment-parity"
+    ],
+    "platforms": ["darwin-arm64", "linux-x64"],
+    "cohortATasks": 20,
+    "cohortBAttempts": 100,
+    "cohortCTerminalAttempts": 200,
+    "independentHumanParticipants": 5,
+    "privateFamilyDisjointHoldout": true,
+    "longHorizonInterruptionResume": true,
+    "explicitOperatorChunkApproval": true
+  },
+  "releaseGate": {
+    "minimumQualityScore": 9.5,
+    "minimumSafetyScore": 10,
+    "minimumReliabilityScore": 9.5,
+    "minimumWorkflowScore": 9.5,
+    "minimumCategoryScore": 9.5,
+    "minimumOutcomeScoreExclusive": 9.5,
+    "minimumPairedScenarioFamilies": 18,
+    "minimumComparableEfficiencyFamilies": 12,
+    "minimumRepeats": 3,
+    "maximumFreshTokenRatioUpper95": 1,
+    "pairedRegressionAllowed": false,
+    "unknownPaidUsageAllowed": false,
+    "infrastructureRetriesAllowed": 0,
+    "tokenClaimRequiresProductionGate": true
+  },
+  "stopRules": {
+    "inspectEverySessions": 6,
+    "sourceEditAfterFreezeAllowed": false,
+    "resumeAfterIdentityDriftAllowed": false,
+    "continueAfterSafetyFailureAllowed": false,
+    "continueAfterOutcomeFloorFailureAllowed": false,
+    "continueAfterUnknownUsageAllowed": false,
+    "continueAfterPairedRegressionAllowed": false,
+    "automaticProviderRetryAllowed": false,
+    "rewriteMeasuredEvidenceAllowed": false,
+    "failedChunkRequiresOperatorDecision": true,
+    "outcomeFloorScope": "candidate-only",
+    "baselineOnlyFailureDisposition": "candidate-only-dominance"
+  },
+  "dossierContract": {
+    "schemaVersion": 1,
+    "requiredSections": [
+      "candidate", "protocol", "benchmark", "quality", "safety", "reliability",
+      "workflow", "efficiency", "platforms", "rollback", "cohorts", "humans",
+      "private-holdout", "long-horizon", "limitations", "operator-decision"
+    ],
+    "missingSectionDisposition": "release-blocked",
+    "independentReviewRequired": true
+  },
+  "claimBoundary": {
+    "preflight": "no-claim",
+    "chunksBeforeCompletion": "no-claim",
+    "capabilityCanaries": "diagnostic-only",
+    "releaseComparison": "public-regression-only",
+    "generalizationWithoutPrivateHoldoutAllowed": false,
+    "longTaskClaimWithoutLongHorizonAllowed": false,
+    "tokenSavingWordingRequiresUpper95BelowOne": true,
+    "releaseRequiresExplicitOperatorApproval": true
+  },
+  "authorization": {
+    "localAssembly": true,
+    "providerPreflight": true,
+    "providerExecution": false,
+    "cohortExecution": false,
+    "releaseBenchmark": false,
+    "tag": false,
+    "push": false,
+    "publish": false,
+    "publicDocsPromotion": false
+  }
+}

package/evals/long-horizon-v1/README.md

@@ -0,0 +1,30 @@
+# Long-horizon lifecycle lane v1
+
+This lane exercises one real repository task across three durable process
+boundaries. It uses the existing Task Contract v2, task journal, handoff,
+resume inspection, context telemetry, working-tree digest, and global
+continuation budget. It does not introduce another product state machine.
+
+The default run processes 90 logical one-minute units over at least 30 minutes
+of real wall-clock time. The controller kills the first worker with `SIGKILL`
+after unit 30, performs a planned handoff after unit 60, and completes in a
+third process. Context compaction occurs at units 20, 40, 60, and 80.
+
+```bash
+node evals/long-horizon-v1/runner.mjs \
+  --output evals/long-horizon-v1/reports/provider-free-30m-run.v1.json
+```
+
+The fast mode exists only for deterministic local calibration and cannot be
+used as wall-clock evidence:
+
+```bash
+node evals/long-horizon-v1/runner.mjs \
+  --calibration-fast --tick-ms 2 --output /tmp/long-horizon-fast.json
+```
+
+The recorded context value is explicitly a deterministic provider-free proxy,
+not provider/model context usage. This lane can prove lifecycle durability,
+bounded local-state growth, compaction, continuation enforcement, and stable
+final verification. It cannot prove model quality, token or latency savings,
+90-minute wall-clock behavior, generalization, or release readiness.

package/evals/long-horizon-v1/lane.json

@@ -0,0 +1,19 @@
+{
+  "schemaVersion": 1,
+  "id": "long-horizon-v1",
+  "title": "Real repository crash, resume, compaction and handoff lane",
+  "workItem": "CF-FS4-03",
+  "fixture": "../../benchmarks/e2-framework-v1/project",
+  "logicalDurationMinutes": 90,
+  "minimumWallClockMinutes": 30,
+  "totalUnits": 90,
+  "hardCrashAfterUnit": 30,
+  "handoffAfterUnit": 60,
+  "compactionUnits": [20, 40, 60, 80],
+  "tickMilliseconds": 21000,
+  "contextProxyCeilingTokens": 1200,
+  "durableStateCeilingBytes": 2097152,
+  "continuationBudget": 1,
+  "providerRequired": false,
+  "claimBoundary": "Provider-free lifecycle evidence only; no model quality, token, latency, 90-minute wall-clock, generalization, or release claim."
+}