experiments.json

{
  "schema_version": 1,
  "last_verified_at": "2026-09-11T04:18:44Z",
  "experiments": [
    {
      "id": "EXP-001",
      "title": "How much context can an AI coding agent safely not see?",
      "status": "PLANNED",
      "hypothesis": "Policy-driven reduction can remove substantial operational payload without reducing correctness when provenance and raw-evidence escalation remain available.",
      "participating_repositories": [
        "context-firewall",
        "decision-evidence-protocol",
        "agent-trajectory-profiler",
        "verifiable-agent-handoff"
      ],
      "roles": {
        "primary": "context-firewall",
        "expected_support": [
          "decision-evidence-protocol",
          "agent-trajectory-profiler"
        ],
        "potential_support": [
          "verifiable-agent-handoff"
        ]
      },
      "baseline": "Unreduced raw tool output supplied under the same task, model, prompt, fixture, and correctness gate.",
      "experimental_arms": [
        "failure-focused deterministic Context Firewall packet without a payload ceiling",
        "bounded 8,192-byte Context Firewall packet with provenance and an unfulfilled raw request",
        "bounded 8,192-byte Context Firewall packet with exact raw exposure when policy requires escalation"
      ],
      "primary_metric": "Correctness pass rate by visible-context fraction.",
      "secondary_metrics": [
        "input bytes and provider-recorded input tokens",
        "suppression ratio",
        "raw-evidence escalation rate",
        "missing-evidence defect rate",
        "latency, cost, and trajectory payload amplification"
      ],
      "correctness_gate": "Each arm must produce the fixture-defined correct repository result and pass the same deterministic tests and safety assertions before efficiency comparison.",
      "failure_classifications": [
        "system-under-test correctness failure",
        "unsafe evidence omission",
        "expected escalation",
        "unexpected escalation",
        "harness failure",
        "provider or model failure",
        "missing telemetry",
        "interrupted run"
      ],
      "dataset_fixture_identity": "FROZEN_OFFLINE: six public-safe Python repair tasks, evaluator-only accepted implementations, and the deterministic strict-TAP oracle are bound by corpus identity sha256:07f8ef2810267a02b3aa03a45b7de3512f6066b03bd82c833394121825f98d60 within freeze identity sha256:6ac668a2fd6b4390825a7f1470f1f8cdab05a6d2d01b41d4cfd4412e93df74d8 at opsle/research main 04234a65bf36192d63f1dd173c440d45a6604d2b. These are benchmark prerequisites, not experiment results.",
      "model_provider_configuration": "PREREGISTERED: OpenAI Responses API /v1/responses with public model identifier gpt-5.6-sol, medium reasoning effort, default service tier, no temperature/top_p/seed overrides, store=false, stream=false, parallel_tool_calls=false, and no automatic retry, fallback, or resume. Subject configuration identity: sha256:bc46a7d72ab776db966e84d9d473efbd0a7e2028ce993eecf4a09d22577ee6cf. No distinct dated model snapshot was publicly listed on 2026-08-29, so launch requires a catalogue-drift preflight and the identifier does not prove immutable weights.",
      "run_identities": [],
      "result_artifacts": [],
      "offline_benchmark_freeze": {
        "status": "OFFLINE_COMPONENTS_FROZEN",
        "research_release_sha": "04234a65bf36192d63f1dd173c440d45a6604d2b",
        "freeze_identity": "sha256:6ac668a2fd6b4390825a7f1470f1f8cdab05a6d2d01b41d4cfd4412e93df74d8",
        "corpus_identity": "sha256:07f8ef2810267a02b3aa03a45b7de3512f6066b03bd82c833394121825f98d60",
        "arm_manifest_identity": "sha256:7ce68d9c53232f5941140edb5e3fa2fff0ac0f5bba33bda14cdf1e09888b64f3",
        "allocation_method_identity": "sha256:6ef1e63dd51bdd967cd9ea472ebd70714bbd50a053ba1df7491f4dbe64595a57",
        "qualification_artifact": "program/evidence/exp-001-offline-freeze/qualification-report.json",
        "qualification_artifact_sha256": "sha256:fe3ce2d215cdd1eebb24185f78a68d54ab717437c30a0a8cf96baf3f4b6ad076",
        "qualification": "PASS",
        "task_count": 6,
        "oracle_case_count": 252,
        "arm_rendering_count": 48,
        "context_firewall_invocation_count": 36,
        "decision_evidence_validation_count": 36,
        "trajectory_profile_count": 48,
        "experiment_runs_added": 0,
        "provider_model_runs_added": 0,
        "allocation_status": "METHOD_FROZEN_SEED_UNSET",
        "lifecycle_impact": "NO_PROMOTION: offline components are frozen, but launch configuration and instantiated allocation remain unreproducible.",
        "limitations": [
          "The corpus is Python-only and supports no cross-language claim.",
          "Offline qualification does not establish preserved model correctness, token or cost savings, causal benefit, or a safe context-reduction frontier.",
          "No provider/model configuration, repetition count, seed commitment, sealed arm mapping, or subject adapter is registered."
        ]
      },
      "launch_preregistration": {
        "status": "PREREGISTERED_AWAITING_EXACT_BUDGETED_LAUNCH_AUTHORIZATION",
        "research_release_sha": "31848c3f25ff9371055932657e8e2f8ad54cc8c7",
        "preregistration_identity": "sha256:e8a0ed7a303e5f4b1f5df089046b6e2fe5ab281b60f9aaeba64ddd2fe972f8c7",
        "subject_configuration_id": "sha256:bc46a7d72ab776db966e84d9d473efbd0a7e2028ce993eecf4a09d22577ee6cf",
        "allocation_index_identity": "sha256:e32f0a4c94842f44b713d40e89c81ea85c98db13927834792620629a5ed95600",
        "seed_commitment": "sha256:a10f7985f6fc0268ab42bc25270c3ebe434ad5655f92156ee4a5471c46a51bd3",
        "planned_subject_count": 240,
        "block_count": 60,
        "subjects_per_arm": 60,
        "repetition_count": 10,
        "provider_model_runs_added": 0,
        "experiment_runs_added": 0,
        "verification_artifact": "program/evidence/exp-001-preregistration/verification-report.json",
        "verification_artifact_sha256": "sha256:b20a7c216e8bc87a589993eb131fdcecf81206823b51c4ddfd85c2e5ba33fc2e",
        "qualification": "PASS",
    "lifecycle_impact": "NO_PROMOTION: launch controls are frozen and provider-free verified, but no authorized subject, measured correctness result, or replication exists.",
        "limitations": [
          "Preregistration is not provider/model authorization; all run and result ledgers remain empty.",
          "The public model identifier does not prove immutable weights because no distinct dated snapshot was listed.",
          "The allocation remains blinded only while the coordinator seed stays outside Git and every subject context.",
      "The provider-free subject adapter and offline replay establish contract behavior, not model correctness or comparative benefit."
    ]
  },
  "block_coordinator_qualification": {
    "status": "PROVIDER_FREE_QUALIFIED_NO_SUBJECT_RUN",
    "research_release_sha": "9ee43197880c18d4e185cf7e29e02a151d22a12e",
    "contract_identity": "sha256:873d28781f9f7fed3ffabb8e95b73cdf5d88bdb5f5b04dd7189b8e55f3466a89",
    "coordinator_revision": "sha256:d7a2e5709fdc3b36b0fedd464bd7ccb791b067da5f327d17f7a76314f3d0af90",
    "preregistration_identity": "sha256:e8a0ed7a303e5f4b1f5df089046b6e2fe5ab281b60f9aaeba64ddd2fe972f8c7",
    "allocation_index_identity": "sha256:e32f0a4c94842f44b713d40e89c81ea85c98db13927834792620629a5ed95600",
    "qualification_artifact": "program/evidence/exp-001-block-coordinator/qualification-report.json",
    "qualification_artifact_sha256": "sha256:a274f7c1912968c8501d43f7758d8ca393ab8204e2b4c317e05b69619c1da426",
    "value_receipt_artifact": "program/evidence/exp-001-block-coordinator/value-receipt.json",
    "value_receipt_artifact_sha256": "sha256:9981b81f5670ee4330172c11c42fffc0ff55bf48fb6c68d8fb87a69d17d6ac92",
    "qualification": "PASS",
    "block_count": 1,
    "fixture_authorization_count": 4,
    "live_authorization_count": 0,
    "arm_rendering_count": 8,
    "result_envelope_count": 8,
    "context_firewall_invocation_count": 6,
    "decision_evidence_validation_count": 6,
    "trajectory_profile_count": 8,
    "subject_visible_arm_identity_count": 0,
    "provider_model_runs_added": 0,
    "experiment_runs_added": 0,
    "experiment_results_added": 0,
    "authorization_consumed": false,
    "private_artifacts_persisted": false,
    "lifecycle_impact": "NO_PROMOTION: one secret-backed block was prepared twice with fixture-only authorizations, but no live authorization was consumed and no provider/model subject or correctness result exists.",
    "limitations": [
      "Fixture-only authorizations cannot pass the live authorization-class gate and do not authorize a provider request.",
      "Secret-backed deterministic replay proves coordinator mechanics, not model correctness or comparative benefit.",
      "The catalogue/pricing preflight and exact live four-label authorization set remain external launch prerequisites.",
      "No token, cost, latency, correctness, or causal savings claim is made."
    ]
  },
  "live_authorization_preflight": {
    "status": "PROVIDER_FREE_VALIDATED_UNRELEASED",
    "implementation_base_sha": "9056b35703d94d85055868ed6778d7d79804485d",
    "authorization_set_id": "exp001-live-authset-03703f1af7db48c4eab9fe33b8f55073",
    "authorization_set_identity": "sha256:8259f5b664235e5914ad4a53966a1155f8f90ad2014dc704f50ea394af32d0ec",
    "authorization_class": "LIVE_PROVIDER_RUN",
    "live_authorization_label_count": 4,
    "authorization_validation_count": 13,
    "model_catalogue_identity": "sha256:334a48850dacf3d9ad5d5c9088287fecc78925e64e10f974cb63575b74592b3c",
    "pricing_preflight_identity": "sha256:a19e793c2c2453d949687f614bee65be84b9abfa6218ea25c5302af7608a38db",
    "model_catalogue_artifact": "program/evidence/exp-001-live-preflight/model-catalogue.json",
    "model_catalogue_artifact_sha256": "sha256:5f7d2f26c70fedd67d71f9146da87d0181700ecb0d73a11f6db9c969644de05f",
    "pricing_preflight_artifact": "program/evidence/exp-001-live-preflight/pricing-preflight.json",
    "pricing_preflight_artifact_sha256": "sha256:070cc44476e26b23a7c8bfa1d94a5c64ed9e62b6dbe81c0d1ce52a136b89cc6f",
    "qualification_artifact": "program/evidence/exp-001-live-preflight/qualification-report.json",
    "qualification_artifact_sha256": "sha256:bb2c6df8e6dcd3293af9034be58e159503ba2e083d30ef98eeb45922e4f9588f",
    "value_receipt_artifact": "program/evidence/exp-001-live-preflight/value-receipt.json",
    "value_receipt_artifact_sha256": "sha256:0d5d00262a660fc09590dc9a427435aefe7bdf5eebe917abe91b2587a8e0167d",
    "qualification": "PASS",
    "deterministic_replay_count": 2,
    "deterministic_payloads_byte_identical": true,
    "fixture_rejected_from_live_gate": true,
    "subject_rendering_count": 0,
    "subject_visible_canonical_arm_identifier_count": 0,
    "provider_model_runs_added": 0,
    "authorization_consumptions_added": 0,
    "experiment_runs_added": 0,
    "experiment_results_added": 0,
    "result_envelopes_added": 0,
    "lifecycle_status": "PLANNED",
    "lifecycle_impact": "NONE_NO_ADVANCEMENT: the external live set and current pricing preflight are provider-free validated but unconsumed and unreleased; no subject, run, or result exists.",
    "limitations": [
      "The external authorization set remains outside Git and every subject context.",
      "Account-specific model access is unverified because no provider call was made.",
      "No distinct dated gpt-5.6-sol snapshot is documented, so immutable weights cannot be claimed.",
      "No token, cost, latency, correctness, savings, or causal-benefit claim is made."
    ]
  },
  "prerequisite_progress": [
        {
          "repository": "context-firewall",
          "head_sha": "953c48f1cfd154d6b7ed10b51b87fe54e4df45f2",
          "status": "COMPLETED",
          "evidence": [
            "https://github.com/opsle/context-firewall/blob/953c48f1cfd154d6b7ed10b51b87fe54e4df45f2/src/reducer.js",
            "https://github.com/opsle/context-firewall/blob/953c48f1cfd154d6b7ed10b51b87fe54e4df45f2/src/value-receipt.js",
            "https://github.com/opsle/context-firewall/blob/953c48f1cfd154d6b7ed10b51b87fe54e4df45f2/fixtures/corpus.js"
          ],
          "limitations": "Prototype conformance and observational Visible Value telemetry do not establish model correctness, benchmark readiness, causal benefit, or a safe context-reduction frontier."
        },
        {
          "repository": "decision-evidence-protocol",
          "head_sha": "b17ae3b41cea7cb0b9e0befe43e885b5aa0e4a09",
          "status": "COMPLETED",
          "evidence": [
            "https://github.com/opsle/decision-evidence-protocol/blob/b17ae3b41cea7cb0b9e0befe43e885b5aa0e4a09/src/context-firewall-v1.js",
            "https://github.com/opsle/decision-evidence-protocol/blob/b17ae3b41cea7cb0b9e0befe43e885b5aa0e4a09/src/context-firewall-value.js",
            "https://github.com/opsle/decision-evidence-protocol/blob/b17ae3b41cea7cb0b9e0befe43e885b5aa0e4a09/fixtures/context-firewall-packet-v1/vectors.json",
            "https://github.com/opsle/decision-evidence-protocol/pull/2"
          ],
          "limitations": "Receipt and source conformance plus validation telemetry establish deterministic protocol behavior, not model correctness, benchmark readiness, comparative benefit, or a safe context-reduction frontier."
        },
        {
          "repository": "agent-trajectory-profiler",
          "head_sha": "0a89661640721d6a39f127514b993d29bd728d47",
          "status": "COMPLETED",
          "evidence": [
            "https://github.com/opsle/agent-trajectory-profiler/blob/0a89661640721d6a39f127514b993d29bd728d47/src/context-evidence.js",
            "https://github.com/opsle/agent-trajectory-profiler/blob/0a89661640721d6a39f127514b993d29bd728d47/src/value-summary.js",
            "https://github.com/opsle/agent-trajectory-profiler/blob/0a89661640721d6a39f127514b993d29bd728d47/fixtures/context-firewall-manifest.json",
            "https://github.com/opsle/agent-trajectory-profiler/pull/2"
          ],
          "limitations": "Measurement and receipt-summary conformance establish deterministic accounting and class preservation, not preserved model correctness, benchmark readiness, comparative benefit, or a safe context-reduction frontier."
        },
        {
          "repository": "research",
      "head_sha": "9ee43197880c18d4e185cf7e29e02a151d22a12e",
      "status": "COORDINATOR_PROVIDER_FREE_QUALIFIED",
          "evidence": [
            "https://github.com/opsle/research/blob/04234a65bf36192d63f1dd173c440d45a6604d2b/experiments/exp-001/benchmark.json",
            "https://github.com/opsle/research/blob/04234a65bf36192d63f1dd173c440d45a6604d2b/experiments/exp-001/corpus/manifest.json",
            "https://github.com/opsle/research/blob/04234a65bf36192d63f1dd173c440d45a6604d2b/experiments/exp-001/harness.py",
            "https://github.com/opsle/research/blob/04234a65bf36192d63f1dd173c440d45a6604d2b/program/evidence/exp-001-offline-freeze/qualification-report.json",
            "https://github.com/opsle/research/pull/9",
            "https://github.com/opsle/research/blob/31848c3f25ff9371055932657e8e2f8ad54cc8c7/experiments/exp-001/preregistration-v1/preregistration.json",
        "https://github.com/opsle/research/blob/31848c3f25ff9371055932657e8e2f8ad54cc8c7/program/evidence/exp-001-preregistration/verification-report.json",
        "https://github.com/opsle/research/pull/11",
        "https://github.com/opsle/research/blob/9ee43197880c18d4e185cf7e29e02a151d22a12e/experiments/exp-001/coordinator-v1/coordinator.py",
        "https://github.com/opsle/research/blob/9ee43197880c18d4e185cf7e29e02a151d22a12e/program/evidence/exp-001-block-coordinator/qualification-report.json",
        "https://github.com/opsle/research/pull/13"
      ],
      "limitations": "The corpus, oracle, arms, allocation, launch configuration, subject adapter, and one-block coordinator are frozen and provider-free qualified; live authorization, catalogue/pricing admission, model correctness, experimental benefit, and replication remain unverified."
        }
      ],
      "replication_status": "NOT_STARTED",
      "verdict": "PENDING",
      "blockers": [
        "The validated live authorization set remains unconsumed; separate execution authority is required before any provider/model subject launch.",
        "Account-specific API entitlement remains unverified under the zero-provider-call policy and requires a fresh fail-closed launch-time catalogue check.",
        "No distinct dated gpt-5.6-sol snapshot was publicly listed on 2026-08-30, so immutable model weights cannot be claimed."
      ],
      "theory_reconciliation": {
        "verified_at": "2026-08-29T11:24:05Z",
        "context_firewall_definition": "The deterministic multi-adapter evidence boundary is the concept; the current TAP-subset reducer is one adapter prototype.",
        "experimental_hypothesis": "The experiment continues to test correctness under reduced model-visible context, not the definition or existence of Context Firewall.",
        "gearbox_dependency": "NONE",
    "sequence": "The offline benchmark, launch controls, one-block coordinator, exact live authorization set, and current catalogue/pricing preflight are provider-free validated; zero authorizations are consumed and zero model/provider subjects or experiment runs/results exist.",
        "gearbox_publication_status": "COMPLETED_INDEPENDENT_PREREQUISITE",
        "gearbox_repository_head_sha": "f3fab9f292cf4eabd7200615d444f98881f57d55",
        "model_provider_runs_added": 0
      },
  "next_task": "Provider-free live-preflight publication is already merged through PR #15 and hardening PR #16. Keep EXP-001 PLANNED; any subject run requires new durable execution authority and a fresh fail-closed preflight, not repetition of the published preparation.",
      "current_evidence_review": {
        "observed_at": "2026-09-11T04:18:44Z",
        "evidence": "program/registry.json#/reconciliation",
        "decision": "PLANNED retained: no authorized subject run or result found in current Research source or inspected durable Tasks history; provider-free qualification is not experiment consumption.",
        "superseded_recommendation": "Independently review and release the provider-free live-authorization preflight branch; do not consume authorization or launch a provider/model subject.",
        "supporting_evidence": [
          "https://github.com/opsle/research/commit/2ca23904f122583f2808f729b6ea9642f0ddba0b",
          "https://github.com/opsle/research/commit/9b9b5ca0fffdbf18713da7ffd339f8fe8f77818c"
        ],
        "qualification_status_scope": "live_authorization_preflight.status is the frozen pre-release qualification label required by the existing validator. PR #15/#16 establish subsequent publication, not experiment consumption."
      }
    },
    {
      "id": "AV-EXP-001",
      "title": "Minimum Defensible Verification — Real Repository Shadow Calibration",
      "status": "RECORDED",
      "hypothesis": "On one pinned real public repository, Affected Verification can propose less than the full verification workload while selecting every check that the frozen full-catalog oracle demonstrates was relevant.",
      "participating_repositories": [
        "affected-verification",
        "research"
      ],
      "roles": {
        "primary": "affected-verification",
        "expected_support": [
          "research"
        ],
        "potential_support": []
      },
      "baseline": "The complete frozen 17-check Zustand verification catalog at b57db4f86ef179285da216eeb291266da82c361c, executed for every scenario and authoritative over all selector predictions.",
      "experimental_arms": [
        "FULL frozen verification catalog",
        "NATIVE Vitest 4.1.10 related selector",
        "AV_CORE with normalized Git, catalog, source-graph, and policy evidence",
        "AV_WITH_NATIVE_EVIDENCE with native output as one normalized evidence source"
      ],
      "primary_metric": "Selection misses reported individually against checks whose full-catalog outcome changed because of a frozen scenario.",
      "secondary_metrics": [
        "relevant-check recall and scenario-level misses",
        "exact selected and skipped test files, test executions, and non-test checks as separate units",
        "uncertainty broadening and full-verification escalation",
        "observed wall-clock telemetry without a causal time-saved claim",
        "native-versus-AV selection differences"
      ],
      "correctness_gate": "Every frozen catalog check runs for every scenario; selector predictions never accept a scenario, and a miss is any omitted oracle-relevant failing check.",
      "failure_classifications": [
        "selection miss",
        "selected relevant failure",
        "irrelevant full-catalog failure",
        "incomplete or indeterminate full run",
        "baseline instability",
        "harness defect requiring a versioned amendment",
        "evidence or identity drift",
        "conservative broadening",
        "insufficient evidence"
      ],
      "dataset_fixture_identity": "Zustand b57db4f86ef179285da216eeb291266da82c361c; preregistration commit 0544362d7659093b7f0b4f89ee8f68023fd269c3; catalog sha256:8c5b224deaa7077690341248a18a2155310e2b16072e1607a9b4cd546e3a0914; ten scenario identities and patches frozen in the preregistration.",
      "model_provider_configuration": "NONE: no model/provider benchmark subject or external provider workload was used.",
      "run_identities": [
        "sha256:68b8582a9ce7b86bfa5431d89d2dea07f8c34b88d1d0350bab25c99fa5b236df"
      ],
      "result_artifacts": [
        "https://github.com/opsle/affected-verification/blob/641aee9d29a89e2a8819f00817ccee8e5d234dcb/benchmark/av-exp-001/REPORT.md",
        "https://github.com/opsle/affected-verification/blob/641aee9d29a89e2a8819f00817ccee8e5d234dcb/benchmark/av-exp-001/results-v2/summary.json",
        "https://github.com/opsle/affected-verification/blob/641aee9d29a89e2a8819f00817ccee8e5d234dcb/benchmark/av-exp-001/results-v2/analysis.json",
        "https://github.com/opsle/affected-verification/blob/641aee9d29a89e2a8819f00817ccee8e5d234dcb/benchmark/av-exp-001/results-v2/evidence-manifest.json"
      ],
      "target": {
        "repository": "https://github.com/pmndrs/zustand.git",
        "sha": "b57db4f86ef179285da216eeb291266da82c361c",
        "license": "MIT"
      },
      "preregistration": {
        "commit_sha": "0544362d7659093b7f0b4f89ee8f68023fd269c3",
        "amendment_count": 3,
        "comparative_outcomes_observed_before_commit": false
      },
      "benchmark_result": {
        "affected_verification_main_sha": "641aee9d29a89e2a8819f00817ccee8e5d234dcb",
        "results_commit_sha": "97f0301c28e9840e18c5aa35a6cbf95b95f7c6cf",
        "summary_identity": "sha256:68b8582a9ce7b86bfa5431d89d2dea07f8c34b88d1d0350bab25c99fa5b236df",
        "analysis_identity": "sha256:c431d8849edce79d6121a290f49288ed2710e406600d8b66d3588e6b82c73a1d",
        "evidence_bundle_identity": "sha256:1e176b7a40b5f16451797d87784f560f932b686f2fe261731526709331ff1172",
        "native_selector_identity": "sha256:f9f8f41244923a6daa6f86b0818889c955b6b26362432b76ae4ee361f10171d6",
        "scenario_count": 10,
        "synthetic_fault_count": 6,
        "synthetic_benign_change_shape_count": 4,
        "relevant_check_count": 8,
        "native_selected_relevant_check_count": 6,
        "native_missed_relevant_check_count": 2,
        "av_core_selected_relevant_check_count": 8,
        "av_core_missed_relevant_check_count": 0,
        "av_with_native_selected_relevant_check_count": 8,
        "av_with_native_missed_relevant_check_count": 0,
        "av_full_escalation_count": 3
      },
      "replication_status": "SAME_HOST_DETERMINISTIC_REPLAY_ONLY",
      "verdict": "PASS for completing the preregistered shadow calibration: no AV selection miss was observed in the frozen corpus. This does not establish general safety, correctness equivalence, causal savings, or bounded trust.",
      "blockers": [
        "Only one repository and one TypeScript/Vitest ecosystem were calibrated.",
        "The corpus uses six synthetic faults and four synthetic benign change shapes rather than a historical real-change replay.",
        "The adapter is benchmark-only and no independent qualifying replication exists.",
        "Affected Verification remains OBSERVE/SHADOW; no TRUSTED_BOUNDED class is authorized."
      ],
      "lifecycle_impact": "PROMOTE_TO_VERIFIED_ONLY: the narrow scoped correctness and safety claims pass meaningful automated and revision-bound benchmark checks, but this run is explicitly capped below BENCHMARK_READY and does not authorize trusted execution.",
      "next_task": "The proposed second calibration was completed as AV-EXP-002. Preserve this first calibration and the subsequent miss; admit further work only through current Tasks authority and evidence-backed program priorities.",
      "current_evidence_review": {
        "observed_at": "2026-09-11T04:18:44Z",
        "evidence": "program/registry.json#/reconciliation",
        "decision": "Existing experiment status, immutable identities, artifacts and verdict retained; package/product integration does not satisfy an experiment or lifecycle gate.",
        "superseded_recommendation": "Preregister and run a second public-repository shadow calibration in a different ecosystem with a meaningful native selector and the same full-catalog oracle discipline.",
        "supporting_evidence": [
          "https://github.com/opsle/research/commit/f1dca5ed07ec1e2d558d90387c750743b3dec962"
        ]
      }
    },
    {
      "id": "AV-EXP-002",
      "title": "Cross-Ecosystem Minimum Defensible Verification Shadow Calibration",
      "status": "RECORDED",
      "hypothesis": "On a pinned Python repository with an established ecosystem affected-test selector, Affected Verification preserves every oracle-relevant frozen catalog check while proposing less than FULL and broadening when evidence is incomplete.",
      "participating_repositories": [
        "affected-verification",
        "research"
      ],
      "roles": {
        "primary": "affected-verification",
        "expected_support": [
          "research"
        ],
        "potential_support": []
      },
      "baseline": "The complete frozen 2,024-check Click verification catalog at 36baa15ff831b939a22bc527cd76ce653ef6f66d, containing 2,016 pytest nodes and eight non-test checks, executed for every scenario and authoritative over all selector predictions.",
      "experimental_arms": [
        "FULL frozen verification catalog",
        "ECOSYSTEM_SELECTOR using pytest-testmon 2.2.0 under its test-selection contract",
        "AV_CORE with normalized Git, Python import graph, pytest catalog, verification catalog, and policy evidence",
        "AV_WITH_SELECTOR_EVIDENCE with pytest-testmon output as an additional normalized evidence source"
      ],
      "primary_metric": "Selection misses reported individually against checks whose full-catalog outcome changed because of a frozen scenario.",
      "secondary_metrics": [
        "relevant-check recall and scenario-level misses",
        "exact selected and skipped pytest nodes, test files, and non-test checks by compatible class",
        "uncertainty broadening and full-verification escalation",
        "observed wall-clock telemetry without a causal time-saved claim",
        "static-graph versus runtime selector compensation",
        "normalized comparison to AV-EXP-001 without aggregating incompatible units"
      ],
      "correctness_gate": "Every frozen catalog check runs for every scenario after selector and AV proposals are frozen; FULL remains authoritative, and a miss is any omitted oracle-relevant failing check.",
      "failure_classifications": [
        "outside selector contract",
        "dependency evidence miss",
        "verification-class omission",
        "policy omission",
        "adapter defect",
        "planner defect",
        "oracle or harness defect",
        "unresolved",
        "conservative broadening",
        "insufficient evidence"
      ],
      "dataset_fixture_identity": "Click 36baa15ff831b939a22bc527cd76ce653ef6f66d; preregistration commit f8a183c460535f3352fad2fb4990b0c54818d623; catalog sha256:28ed20abf60e7c785052308298dc6ed647b7a20a737513e8fb3c76aa62d9094c; corpus sha256:d5bc43405a5ab0ac34feef6d5fd5df111eace7f1a355400f963f8a7d4399640b; eleven frozen scenarios and patches.",
      "model_provider_configuration": "NONE: no model/provider benchmark subject or external provider workload was used; one interactive Codex session used native shell and patch facilities without child agents.",
      "run_identities": [
        "sha256:5b3f99bfbebd3a0d061651d66adfb5a6aaef899475c6267cb20e4040e6ed5768"
      ],
      "result_artifacts": [
        "https://github.com/opsle/affected-verification/blob/3ff41688dded6e96e65da7cc44fe2608cf86d073/benchmark/av-exp-002/REPORT.md",
        "https://github.com/opsle/affected-verification/blob/3ff41688dded6e96e65da7cc44fe2608cf86d073/benchmark/av-exp-002/results-v1/summary.json",
        "https://github.com/opsle/affected-verification/blob/3ff41688dded6e96e65da7cc44fe2608cf86d073/benchmark/av-exp-002/results-v1/cross-experiment.json",
        "https://github.com/opsle/affected-verification/blob/3ff41688dded6e96e65da7cc44fe2608cf86d073/benchmark/av-exp-002/results-v1/evidence-manifest.json"
      ],
      "target": {
        "repository": "https://github.com/pallets/click.git",
        "sha": "36baa15ff831b939a22bc527cd76ce653ef6f66d",
        "license": "BSD-3-Clause"
      },
      "preregistration": {
        "commit_sha": "f8a183c460535f3352fad2fb4990b0c54818d623",
        "identity": "sha256:93743e15caee647de1807ee27d35518cd9520394983803e1b93a58e9a2841db6",
        "amendment_count": 5,
        "comparative_outcomes_observed_before_commit": false
      },
      "benchmark_result": {
        "affected_verification_main_sha": "3ff41688dded6e96e65da7cc44fe2608cf86d073",
        "results_commit_sha": "5d126e0ae17557065b55ab84a46f6a3577a49989",
        "summary_identity": "sha256:5b3f99bfbebd3a0d061651d66adfb5a6aaef899475c6267cb20e4040e6ed5768",
        "evidence_bundle_identity": "sha256:435d8e5356ed6868edfc1747523ff75d1b327389bf14cc2675e867e45f4de705",
        "cross_experiment_identity": "sha256:6ac2695ee80c9cb711e756846dad4c7138a3177c11d733456636259c36948da0",
        "selector_baseline_identity": "sha256:9a3eb5630a868869179585210e9e6034e0c16039d97874921a09f63ef61aafd8",
        "scenario_count": 11,
        "synthetic_fault_count": 7,
        "synthetic_benign_change_count": 1,
        "uncertainty_scenario_count": 3,
        "relevant_check_count": 87,
        "ecosystem_selector_selected_relevant_check_count": 77,
        "ecosystem_selector_missed_relevant_check_count": 10,
        "ecosystem_selector_scenario_miss_count": 7,
        "av_core_selected_relevant_check_count": 86,
        "av_core_missed_relevant_check_count": 1,
        "av_core_scenario_miss_count": 1,
        "av_core_full_broadening_count": 6,
        "av_with_selector_selected_relevant_check_count": 86,
        "av_with_selector_missed_relevant_check_count": 1,
        "av_with_selector_scenario_miss_count": 1,
        "av_with_selector_full_broadening_count": 5,
        "av_miss": {
          "scenario_id": "AV2-006",
          "check_id": "pytest:tests/test_imports.py::test_light_imports",
          "classification": "PLANNER_OR_ADAPTER_MISS",
          "finding": "A subprocess instrumented runtime imports through the public click package; the static Python graph declared completeness and pytest-testmon also omitted the relevant node."
        }
      },
      "major_findings": [
        "Catalog, policy escalation, explicit uncertainty, skip evidence, FULL fallback, shadow classification, and Visible Value generalized without changing the core plan schema.",
        "Python dependency evidence, pytest node identity, conftest coupling, selector database lifecycle, and dynamic or subprocess imports required material ecosystem-specific logic.",
        "Runtime selector evidence compensated for static dynamic-plugin uncertainty in AV2-009 but could not repair AV2-006 because the selector also omitted the relevant runtime-import test.",
        "Both deliberately degraded cases denied aggressive skipping and required FULL."
      ],
      "replication_status": "SAME_HOST_RELEASE_AND_BUNDLE_VERIFICATION_ONLY",
      "verdict": "FAIL for the safety hypothesis: AV_CORE and AV_WITH_SELECTOR_EVIDENCE each selected 86/87 oracle-relevant checks and omitted the same runtime/subprocess import test in AV2-006. The controlled shadow calibration itself completed and FULL exposed the miss.",
      "blockers": [
        "The AV2-006 runtime/subprocess import miss invalidates a positive targeted-sufficiency claim for the frozen corpus.",
        "The Python and pytest-testmon adapters are benchmark-only and no historical real-change replay or independent qualifying replication exists.",
        "Affected Verification remains OBSERVE/SHADOW; no TRUSTED_BOUNDED class is authorized."
      ],
      "lifecycle_impact": "REMAIN_VERIFIED: the run adds revision-bound falsification evidence and failure modes, but the observed selection miss and explicit run cap do not establish BENCHMARK_READY or EXPERIMENTED lifecycle promotion for the repository.",
      "next_task": "The proposed AV2-006 repair was completed as bounded AV-EXP-003 evidence. Preserve this historical FAIL; do not repeat the superseded repair recommendation or infer general selector safety.",
      "current_evidence_review": {
        "observed_at": "2026-09-11T04:18:44Z",
        "evidence": "program/registry.json#/reconciliation",
        "decision": "Existing experiment status, immutable identities, artifacts and verdict retained; package/product integration does not satisfy an experiment or lifecycle gate.",
        "superseded_recommendation": "Preregister and execute a selection-miss repair for AV2-006 that represents runtime/subprocess import uncertainty without changing the preserved AV-EXP-002 result.",
        "supporting_evidence": [
          "https://github.com/opsle/research/commit/e8a2c36678ac4d4f72a79a56e03e9f896811be02"
        ]
      }
    },
    {
      "id": "AV-EXP-003",
      "title": "Opaque Dependency Boundary Repair",
      "status": "RECORDED",
      "hypothesis": "Affected Verification may skip a check only when available evidence defends completeness for every declared dependency mechanism capable of connecting the change to that check; an unmodeled, incomplete, unknown, or opaque boundary forces selection of that check unless identified evidence closes it.",
      "participating_repositories": [
        "affected-verification",
        "research"
      ],
      "roles": {
        "primary": "affected-verification",
        "expected_support": [
          "research"
        ],
        "potential_support": []
      },
      "baseline": "The permanently preserved AV-EXP-002 FAIL and exact frozen AV2-006 Click scenario, plus frozen AV-EXP-001/002 selector/full-oracle evidence and ten preregistered generalized repair cases.",
      "experimental_arms": [
        "historical AV-EXP-001 and AV-EXP-002 AV selections preserved as the pre-repair baseline",
        "Affected Verification plan v2 with check-level dependency completeness and deterministic boundary evidence",
        "FULL frozen oracle for every repair case and preserved complete FULL evidence for every prior-corpus replay"
      ],
      "primary_metric": "Zero repaired selection misses, including selection of the exact known AV2-006 check for a generalized dependency-completeness reason.",
      "secondary_metrics": [
        "prior and repaired test checks or executions selected in compatible exact units",
        "additional verification introduced by dependency-safety policy",
        "tests still skipped, scenarios remaining targeted, scenarios broadened, and FULL escalations",
        "check-level boundary provenance and forced-selection explanations",
        "non-test check differences by class"
      ],
      "correctness_gate": "FULL remains authoritative in SHADOW; every adversarial catalog is completely executed, every frozen prior relevant set is replayed, and any newly missed relevant check fails the experiment.",
      "failure_classifications": [
        "known regression remains omitted",
        "new regression miss",
        "target-specific special case",
        "malformed or unsupported boundary evidence accepted",
        "incomplete full oracle",
        "unmeasured or hidden precision cost",
        "historical result mutation",
        "unsafe trust promotion"
      ],
      "dataset_fixture_identity": "Affected Verification final 97f490a67337552fee25757266f3dc034660dca0; AV-EXP-003 preregistration 7aa4d13e42d6a547973d7f2a6b330821145cedc2; Click 36baa15ff831b939a22bc527cd76ce653ef6f66d; result sha256:03b2f7d6a380c84f6a1749531067cf8b87404c879f42380de8f07cce48251519; regression matrix sha256:7260c2d3476a6e78323e75d36c54c8409ea4cb18fa3a8f9a76b5533e1df08615.",
      "model_provider_configuration": "NONE: one interactive Codex session used native shell, patch, Git, and Graphify facilities; no Codex children, external model/provider workloads, or production systems were used.",
      "run_identities": [
        "sha256:03b2f7d6a380c84f6a1749531067cf8b87404c879f42380de8f07cce48251519",
        "sha256:7260c2d3476a6e78323e75d36c54c8409ea4cb18fa3a8f9a76b5533e1df08615"
      ],
      "result_artifacts": [
        "https://github.com/opsle/affected-verification/blob/97f490a67337552fee25757266f3dc034660dca0/benchmark/av-exp-003/REPORT.md",
        "https://github.com/opsle/affected-verification/blob/97f490a67337552fee25757266f3dc034660dca0/benchmark/av-exp-003/results-v1/summary.json",
        "https://github.com/opsle/affected-verification/blob/97f490a67337552fee25757266f3dc034660dca0/benchmark/av-exp-003/results-v1/repair-regression-matrix.json",
        "https://github.com/opsle/affected-verification/blob/97f490a67337552fee25757266f3dc034660dca0/benchmark/av-exp-003/results-v1/boundary-evidence.json",
        "https://github.com/opsle/affected-verification/blob/3ff41688dded6e96e65da7cc44fe2608cf86d073/benchmark/av-exp-002/REPORT.md"
      ],
      "target": {
        "repository": "https://github.com/pallets/click.git",
        "sha": "36baa15ff831b939a22bc527cd76ce653ef6f66d",
        "license": "BSD-3-Clause"
      },
      "preregistration": {
        "commit_sha": "7aa4d13e42d6a547973d7f2a6b330821145cedc2",
        "known_failure_outcome_disclosed": true,
        "repaired_outcomes_observed_before_commit": false
      },
      "benchmark_result": {
        "affected_verification_main_sha": "97f490a67337552fee25757266f3dc034660dca0",
        "result_identity": "sha256:03b2f7d6a380c84f6a1749531067cf8b87404c879f42380de8f07cce48251519",
        "regression_matrix_identity": "sha256:7260c2d3476a6e78323e75d36c54c8409ea4cb18fa3a8f9a76b5533e1df08615",
        "evidence_manifest_identity": "sha256:932292a4f47372cab963db85772bb3d6c1e4fa536edd7b99f9686b72afd7e93c",
        "known_replay": {
          "scenario_id": "AV2-006",
          "check_id": "pytest:tests/test_imports.py::test_light_imports",
          "historical_outcome": "MISS",
          "repaired_outcome": "SELECTED",
          "additional_test_executions_per_av_arm": 1
        },
        "adversarial_scenario_count": 10,
        "adversarial_repaired_miss_count": 0,
        "adversarial_prior_tests_selected": 10,
        "adversarial_repaired_tests_selected": 17,
        "adversarial_additional_tests_due_to_repair": 7,
        "adversarial_tests_still_skipped": 13,
        "adversarial_targeted_scenarios_retained": 10,
        "adversarial_broadened_scenarios": 7,
        "adversarial_full_escalations": 0,
        "av_exp_001_new_misses": 0,
        "av_exp_001_additional_test_executions": 0,
        "av_exp_002_av_core_new_misses": 0,
        "av_exp_002_av_core_additional_test_executions": 6,
        "av_exp_002_with_selector_new_misses": 0,
        "av_exp_002_with_selector_additional_test_executions": 7,
        "non_test_check_differences": 0
      },
      "major_findings": [
        "Evidence-source agreement and evidence completeness are independent; static and native selector omission cannot prove irrelevance across an unmodeled boundary.",
        "Check-local fail-closed selection repaired the known replay without global FULL broadening; all ten repair scenarios remained targeted.",
        "The bounded Python inspector found two open subprocess/child-interpreter checks among 2,016 Click pytest checks and retained provenance without claiming it closes those boundaries.",
        "AV-EXP-002 remains a permanent FAIL and no trust promotion follows from repairing its known miss."
      ],
      "replication_status": "SAME_HOST_DETERMINISTIC_REPLAY_ONLY",
      "verdict": "PASS only for the preregistered defect-repair claim: the known AV2-006 check is selected, ten generalized cases have zero misses, no frozen prior AV-relevant check becomes newly missed, and exact precision cost is measured. This does not solve dynamic dependencies, prove general safety, or authorize trusted execution.",
      "blockers": [
        "The static boundary inspector identifies but does not close child-process, runtime import, arbitrary plugin, or reflection boundaries.",
        "Historical real-change replay, a production-quality evidence adapter, and independent qualifying replication remain missing.",
        "Affected Verification remains OBSERVE/SHADOW; no TRUSTED_BOUNDED class is authorized."
      ],
      "lifecycle_impact": "REMAIN_VERIFIED: the run repairs and measures one defect under SHADOW but is explicitly capped below lifecycle or trust promotion.",
      "next_task": "Child-process import tracing remains an evidence-gated research proposal, not an authorized workload item. Require current Tasks admission and preregistration before any new bounded OBSERVE/SHADOW experiment.",
      "current_evidence_review": {
        "observed_at": "2026-09-11T04:18:44Z",
        "evidence": "program/registry.json#/reconciliation",
        "decision": "Existing experiment status, immutable identities, artifacts and verdict retained; package/product integration does not satisfy an experiment or lifecycle gate.",
        "superseded_recommendation": "Preregister a bounded child-process import-tracing evidence-provider experiment to test whether selected opaque checks can regain precision without weakening the fail-closed rule.",
        "supporting_evidence": [
          "program/registry.json#/program_control"
        ]
      }
    },
    {
      "id": "LEGACY-001",
      "title": "Graphify plus Antigravity semantic adapter integration observation",
      "status": "RECORDED",
      "hypothesis": "A bounded read-only subscription-CLI adapter can add schema-valid semantic graph fragments without an API-key semantic backend.",
      "participating_repositories": [
        "research"
      ],
      "roles": {
        "primary": "research",
        "expected_support": [],
        "potential_support": []
      },
      "baseline": "Graphify deterministic AST extraction alone.",
      "experimental_arms": [
        "Two bounded read-only Antigravity subscription CLI semantic passes"
      ],
      "primary_metric": "Schema-valid semantic nodes and edges admitted after source verification.",
      "secondary_metrics": [
        "provider-recorded input tokens",
        "provider-recorded output tokens",
        "graph health warnings"
      ],
      "correctness_gate": "Fragments must match the strict JSON schema and remain subordinate to direct source verification.",
      "failure_classifications": [
        "invalid fragment",
        "unsupported inference",
        "dangling extraction edge",
        "same-endpoint collapse",
        "provider failure"
      ],
      "dataset_fixture_identity": "Retired predecessor read-only snapshot at e2b061dd4ee1404ef59b27a9a76bf97a8fbbde1c.",
      "model_provider_configuration": "Antigravity subscription CLI; exact model identity and configuration were not recorded and remain UNVERIFIED.",
      "run_identities": [],
      "result_artifacts": ["experiments/graphify-antigravity/README.md"],
      "replication_status": "NOT_REPLICATED",
      "verdict": "An integration path was observed; completeness and comparative benefit were not established.",
      "blockers": [
        "Exact run identities, model configuration, fixtures, and independent replication were not recorded."
      ],
      "next_task": "Retain as legacy evidence; do not promote a concept stage from this observation alone.",
      "current_evidence_review": {
        "observed_at": "2026-09-11T04:18:44Z",
        "evidence": "program/registry.json#/reconciliation",
        "decision": "Existing experiment status, immutable identities, artifacts and verdict retained; package/product integration does not satisfy an experiment or lifecycle gate."
      }
    }
  ]
}