{
  "schemaVersion": 1,
  "generatedAtUtc": "2026-10-15T01:30:00.0000000Z",
  "synthetic_mock": false,
  "syntheticLanguageExamples": true,
  "resultSource": "Saved Scala-graded reports and CSVs; no new grades or statistical estimates were computed for publication.",
  "measurement": "Explicitly supplied miniature grammars with eight fixed worked examples; not demonstrated absence from training or grammar inference from examples alone.",
  "common": {
    "model": "gpt-5.6-terra",
    "reasoningEffort": "low",
    "fewShotCount": 8,
    "styles": [
      "braces",
      "generic_end",
      "typed_end",
      "named_end",
      "padded_end"
    ],
    "lexicalRegimes": [
      "natural",
      "nonce"
    ],
    "samplingUnit": "structural_family"
  },
  "pilot": {
    "state": "complete_exploratory",
    "phase": "p1",
    "planned": 1800,
    "terminal": 1800,
    "modelEvaluable": 1800,
    "correct": 1622,
    "knownGenerationTokens": 6148886,
    "unresolvedReservedTokens": 0,
    "tasks": [
      "scope_lookup",
      "active_stack",
      "ast_to_source"
    ],
    "readingViews": [
      "before_close",
      "after_close"
    ],
    "generationView": "complete",
    "promptVersion": "1",
    "scorerVersion": "1",
    "protocolHash": "f7dd15dd3c44b507506e123517157926a35a53b52c1d4aab278664f512a1e182",
    "sourceHash": "f3e5a0a43c3ca7332be53f098c2b8d9064360312d6e7d5efc16cc4a74fd646ab",
    "report": "pilot/report.md"
  },
  "main": {
    "state": "provisional_credit_restopped_saved_checkpoint_20261007",
    "phase": "p1",
    "planned": 5120,
    "terminal": 2753,
    "modelEvaluable": 2688,
    "correct": 2331,
    "ambiguousOutcome": 1,
    "terminalNotDispatched": 5,
    "untouchedUnexecuted": 2365,
    "reportNotDispatchedTotal": 2372,
    "knownGenerationTokens": 9063965,
    "unresolvedReservedTokens": 707127,
    "task": "scope_lookup",
    "view": "after_close",
    "promptVersion": "1",
    "scorerVersion": "1",
    "protocolHash": "f57aadd6b497ff6d05715753dd59802be050dce8ec5706ff41216721515e6f4a",
    "sourceHash": "1d041fa5f4ae2a499b89151fb99958312a49b883a6315adad4d65e63307edca1",
    "inferentialCiComputed": false,
    "report": "main/report.md",
    "recoveryWitness": "supporting/main-zero-http-recovery.json",
    "reproductionWitness": "supporting/main-provisional-reproduction.json",
    "note": "On Oct7 UTC, the user reported charged and a fresh real capability probe passed. The unchanged frozen reading Main resumed without resending terminal trials. It saved 2,753 dispositions: 2,688 completed/evaluable, 59 API refusals, the original ambiguous outcome and five terminal-unexecuted outcomes. After another credit_balance_exhausted rejection, the exact native model process and guarded continuation chain stopped. The remaining 2,367 trials comprise two count-only pending and 2,365 untouched; all count-only/raw artifacts remain preserved. The old 2,410 accepted terminal bytes and 60 reservations were verified; the new 14,238-token hold raises 61 open holds to 707,127. Strict grades use released saved data only. Partial CI remains not_computed, and infrastructure missingness exceeds 1%. P1/main/P2/mock are separate.",
    "dispatched": 2748,
    "infrastructureMissing": 60,
    "snapshotAtUtc": "2026-10-07T14:08:56.2343959Z",
    "checkpointSource": "verification/main-credit-restop-checkpoint-20261007",
    "checkpointSnapshotSha256": "f6ac0abc5a436d34d9cc92327183eced78dfaf33b9d2b8f7897ed79e0a569749",
    "inferentialCiStatus": "not_computed",
    "heldReservations": [
      {
        "reservationId": "f92a43fa61e8d13dc4204692d22ffc4633917aaaec83d778457701acae458a64/1",
        "tokens": 14131,
        "window": "2026-10-05",
        "interpretation": "Original dispatched response unknown; retained"
      },
      {
        "reservationId": "dea5611a59249adf8adab67d8daf26e1d511bc7a8df7a04ab80c1844b06309b6/1",
        "tokens": 14838,
        "window": "2026-10-06",
        "interpretation": "Proven count-only no generation; frozen ledger conservatively retains pre-dispatch reservation"
      }
    ],
    "recoveryWitnessScope": "Historical original 2026-10-05 zero-HTTP recovery to 725 terminal dispositions; this witness is preserved unchanged and does not describe the current graded checkpoint.",
    "reproductionWitnessScope": "Historical original 725-terminal provisional report reproduction only; preserved unchanged, not the current graded checkpoint.",
    "currentScoreReportWitnessSource": "verification/main-credit-restop-checkpoint-20261007/snapshot.json",
    "currentScoreReportWitnessSha256": "f6ac0abc5a436d34d9cc92327183eced78dfaf33b9d2b8f7897ed79e0a569749",
    "countOnlyRecovery": {
      "sourceWitness": "verification/main-count-only-recovery-v2/recovery-witness.json",
      "sourceWitnessSha256": "f2cc43f10db9b6af67856dec05fe7df5521a43a9d32971e43a7dbf54e934cdcf",
      "capturedAt": "2026-10-06T10:17:44.935Z",
      "priorTerminalCount": 1024,
      "priorTerminalBytesUnchanged": true,
      "ledgerBytesUnchanged": true,
      "externalHttpAttempts": 0,
      "logicalCalls": 0,
      "recoveredTrialId": "dea5611a59249adf8adab67d8daf26e1d511bc7a8df7a04ab80c1844b06309b6",
      "terminalDisposition": "not_dispatched",
      "rawCountHttpDurationMs": 257,
      "tokenCountFinishedDurationMs": 263,
      "knownCurrentUtcWindowTokensAtRecovery": 1024996,
      "unresolvedReservedTokens": 28969
    },
    "countOnlyRecoveryWitness": "supporting/main-count-only-recovery-v2.json",
    "countOnlyRecoveryWitnessSha256": "1c6a01b4520811dc6061327670573d226853cb3d6d4b7ec8546569a8f7cb6fd9",
    "apiRejected": 59,
    "countOnlyNonTerminal": 2,
    "nonTerminal": 2367,
    "openReservationCount": 61,
    "heldReservationsScope": "Original accepted two-reservation subset remains listed for history; the full private ledger retains61 holds707127 including all old60/692889 and one new14238 credit refusal hold.",
    "creditRefusalReservedTokens": 678158,
    "newCreditRestopReservedTokens": 14238
  },
  "fullProgramExtension": {
    "state": "provisional_credit_restopped_saved_strict_reports_20261007",
    "phase": "p2",
    "tasks": [
      "source_to_ast",
      "ast_to_source"
    ],
    "view": "complete",
    "plannedStructuralFamilies": 48,
    "plannedGenerations": 960,
    "plannedPerDirectionStyle": 96,
    "liveExecuted": true,
    "liveGenerations": 519,
    "measuredSuccessRate": null,
    "completeMockWorkflowExecuted": true,
    "postChangeFullTestPassed": true,
    "postChangeFullTestCounts": {
      "core": 28,
      "bench": 19,
      "preflight": 12,
      "parsing": 13
    },
    "plannedPromptVersion": "2",
    "plannedScorerVersion": "2",
    "protocolHash": "2ac04cc60928800d0c1fdd37986546a6fee23bd1955ccc011c76099cbde60471",
    "depths": [
      2,
      4,
      8,
      16
    ],
    "fillers": [
      0,
      8,
      32
    ],
    "design": "docs/full-program-structures.md",
    "config": "supporting/full-program-pilot.json",
    "mockVerification": {
      "synthetic_mock": true,
      "planned": 960,
      "families": 48,
      "terminal": 960,
      "strictCorrect": 960,
      "simulatedAdapterAttempts": 1920,
      "externalHttpCalls": 0,
      "externalHttpEvidence": "Explicit --mock without --execute uses MockClient; code-path evidence, not packet capture.",
      "reportFilesByteIdentical": true,
      "resumeAdditionalLogicalTrials": 0,
      "resumeAdditionalAdapterAttempts": 0,
      "unresolvedReservedTokens": 0,
      "liveGenerations": 0,
      "immutableArtifactsUnchanged": true,
      "checksPassed": true,
      "protocolHash": "2ac04cc60928800d0c1fdd37986546a6fee23bd1955ccc011c76099cbde60471",
      "sourceHash": "41ffc2647b8f3c3838a17c2425b0be97007fe611f5d0a68d3dd9d1c79e1b0b96",
      "witness": "supporting/p2-mock-verification.json",
      "witnessSha256": "224e9091619a7ce85cce16c663fff2238b3694161fc4f5540734bfd383b5c4dc",
      "report": "mock-p2/report.md",
      "workflowRunUrl": "https://github.com/kmizu/syntax-locality-experiments/actions/runs/37371856651/attempts/3",
      "workflowJobId": 112109491633,
      "artifactId": 11390722284,
      "artifactDigestSha256": "a0058b2519122ca9bffe84e60fe1b0a1875b696416cbbbc69969e86b47392bc4"
    },
    "note": "The same P2 live plan resumed after a fresh real capability probe following the user-reported charge. A new shared credit rejection stopped both native runners. The saved partial report retains960 planned trials,521 terminal dispositions,519 dispatches,512 model-evaluable responses and348 correct. Generation is168/268 and whole-program parsing180/244, displayed separately. Seven credit-refusal terminals, two count-only expiry dispositions, one remaining count-only operation and438 untouched are preserved; no dispatched orphan was found. Old300 terminals and raw backing, all968 plan/request/mode artifacts and45 source/build inputs were verified. Two reused derived token-count cache durations became0 while exact input counts/request identities and raw HTTP timing bytes remain unchanged. Held usage166039 is not zero. All partial inferential CI stays not_computed. No terminal retry, fallback, best-of-N or automatic billing action was performed.",
    "sourceHash": "41ffc2647b8f3c3838a17c2425b0be97007fe611f5d0a68d3dd9d1c79e1b0b96",
    "liveCompletedResponses": 512,
    "liveTerminalDispositions": 521,
    "modelEvaluable": 512,
    "strictCorrect": 348,
    "report": "p2/report.md",
    "inferentialCiStatus": "not_computed",
    "measuredSuccessRateScope": "No single pooled rate is displayed; generation and parsing use their separate strict saved summary rows.",
    "liveSnapshotAt": "2026-10-07T13:57:32.7824474Z",
    "liveExecutionStatus": "supporting/p2-live-current-status.json",
    "liveCapabilities": "supporting/p2-live-capabilities.json",
    "unresolvedReservedTokens": 166039,
    "knownGenerationTokensAllWindows": 3293502,
    "checkpointSource": ".worktrees/grammar-structures/verification/p2-checkpoints-v3/terminal-521",
    "checkpointManifestSha256": "0d19355dca589179157836f7742ec6b82c17baea166e56e050032090c5a3f2ed"
  },
  "shortDependencyExtension": {
    "state": "complete_exploratory_single_replicate_not_frozen",
    "phase": "p3",
    "worktreeBranch": "p3-short-dependency",
    "worktreeCommits": [
      "6aed57575a56abcbd2dc111a535db1535a54a836",
      "d98fc14352ea11e25ed432b42f66649d7db61c1b",
      "39d69491f2fc8cd5bcd5086327aeed82ace34506",
      "14a166a97a50779307c8969cfa0583eb33b3a824"
    ],
    "sourceHash": "f57b680ff77f4a0b4190ede060a71f008bb1671a65d82f100fc9e6a5ffa4aff0",
    "sourceHashScope": "p3a, p3b-fq and p3b-short (worktree at d98fc14; earlier snapshot, recomputation matched)",
    "p3cSourceHash": "9d1d8aaf7ab703c39702f518386873b091a4d1f3f0b967ba13dc0b5f856b193c",
    "p3cSourceHashScope": "p3c (worktree at 14a166a; the published supporting/p3-source/ snapshot, recomputation matched)",
    "sourceHashRecomputedFromPublishedSnapshot": true,
    "sourceSnapshot": "supporting/p3-source/README.md",
    "publicRootSourcesContainP3Code": false,
    "model": "gpt-5.6-terra",
    "reasoningEffort": "low",
    "replicates": 1,
    "fewShotCount": 8,
    "depths": [
      4,
      8,
      16
    ],
    "fillers": [
      0,
      32,
      128
    ],
    "lexicalRegimes": [
      "natural",
      "nonce"
    ],
    "samplingUnit": "structural_family",
    "preflight": "p3-preflight/capabilities.json",
    "preflightP3c": "p3c-preflight/capabilities.json",
    "design": "docs/p3-short-dependency-design.md",
    "configs": [
      "supporting/p3a-pilot.json",
      "supporting/p3b-fq-pilot.json",
      "supporting/p3b-short-pilot.json",
      "supporting/p3c-pilot.json",
      "supporting/p3.md"
    ],
    "inferentialCiStatus": "not_computed",
    "runs": {
      "p3a": {
        "planned": 288,
        "terminal": 288,
        "modelEvaluable": 288,
        "correct": 191,
        "infrastructureMissing": 0,
        "families": 36,
        "tasks": [
          "ast_to_source"
        ],
        "styles": [
          "generic_end",
          "named_end",
          "depth_end",
          "indent"
        ],
        "strictCorrectOf72": {
          "generic_end": 15,
          "named_end": 62,
          "depth_end": 54,
          "indent": 60
        },
        "protocolHash": "b7fa756e73e53321ccebf9b0080aa5de75f23efcce70310503e8eaf49749e35a",
        "report": "p3a/report.md"
      },
      "p3b-fq": {
        "planned": 216,
        "terminal": 216,
        "modelEvaluable": 216,
        "correct": 187,
        "infrastructureMissing": 0,
        "families": 72,
        "tasks": [
          "scope_lookup",
          "ast_to_source"
        ],
        "styles": [
          "named_end"
        ],
        "importQualification": "fq",
        "strictCorrectOf72": {
          "ast_to_source/complete": 55,
          "scope_lookup/before_close": 72,
          "scope_lookup/after_close": 60
        },
        "protocolHash": "326f29ffbd4735878c67f6ebeb76183f732c22dbb27241fefb043e5ce99ab0e4",
        "report": "p3b-fq/report.md"
      },
      "p3b-short": {
        "planned": 216,
        "terminal": 216,
        "modelEvaluable": 216,
        "correct": 177,
        "infrastructureMissing": 0,
        "families": 72,
        "tasks": [
          "scope_lookup",
          "ast_to_source"
        ],
        "styles": [
          "named_end"
        ],
        "importQualification": "short",
        "strictCorrectOf72": {
          "ast_to_source/complete": 50,
          "scope_lookup/before_close": 72,
          "scope_lookup/after_close": 55
        },
        "protocolHash": "256e503ae744e165c61804b9f6cf92e13073bd66e8f66069f4799f0cd3d4c9d0",
        "report": "p3b-short/report.md"
      },
      "p3c": {
        "planned": 288,
        "terminal": 288,
        "modelEvaluable": 288,
        "correct": 255,
        "infrastructureMissing": 0,
        "families": 36,
        "familiesIdenticalToP3a": true,
        "tasks": [
          "ast_to_source"
        ],
        "styles": [
          "named_end",
          "indent",
          "indent_named_end",
          "indent_generic_end"
        ],
        "strictCorrectOf72": {
          "named_end": 67,
          "indent": 64,
          "indent_named_end": 64,
          "indent_generic_end": 60
        },
        "depth16Filler128CorrectOf8": {
          "named_end": 7,
          "indent": 7,
          "indent_named_end": 7,
          "indent_generic_end": 3
        },
        "failureBreakdown": {
          "valid_syntax_wrong_ast": 30,
          "invalid_generated_syntax": 3,
          "invalid_generated_syntax_parse_error_code": "invalid_nop",
          "indentation_or_closing_line_parse_errors_in_combined_styles": 0
        },
        "protocolHash": "a266e4cb45980b516d6d55b5df4f91fdadcf710c97de5b940ecaa1973ea9f01c",
        "sourceHash": "9d1d8aaf7ab703c39702f518386873b091a4d1f3f0b967ba13dc0b5f856b193c",
        "report": "p3c/report.md"
      }
    },
    "pairedDiscordanceP3c": {
      "pairing": "named_end versus each other style within p3c; familyId + lexicalRegime + depth + filler; model-evaluable pairs only",
      "named_end_vs_indent_named_end": {
        "pairs": 72,
        "bothCorrect": 61,
        "namedEndOnly": 6,
        "otherOnly": 3,
        "bothWrong": 2
      },
      "named_end_vs_indent": {
        "pairs": 72,
        "bothCorrect": 60,
        "namedEndOnly": 7,
        "otherOnly": 4,
        "bothWrong": 1
      },
      "named_end_vs_indent_generic_end": {
        "pairs": 72,
        "bothCorrect": 56,
        "namedEndOnly": 11,
        "otherOnly": 4,
        "bothWrong": 1
      }
    },
    "replicateAgreementNamedEnd": {
      "pairing": "p3a named_end versus p3c named_end; identical 72 requests matched by requestHash (same AST, rendering, examples and lexicon); separate generations",
      "pairs": 72,
      "p3aCorrect": 62,
      "p3cCorrect": 67,
      "bothCorrect": 59,
      "p3aOnlyCorrect": 3,
      "p3cOnlyCorrect": 8,
      "bothWrong": 2,
      "interpretation": "Single-replicate differences of roughly five per 72 are within run-to-run variation."
    },
    "pairedDiscordance": {
      "pairing": "familyId + task + view + lexicalRegime; model-evaluable pairs only",
      "ast_to_source/complete": {
        "pairs": 72,
        "bothCorrect": 41,
        "fqOnly": 14,
        "shortOnly": 9,
        "bothWrong": 8
      },
      "scope_lookup/after_close": {
        "pairs": 72,
        "bothCorrect": 46,
        "fqOnly": 14,
        "shortOnly": 9,
        "bothWrong": 3
      },
      "scope_lookup/before_close": {
        "pairs": 72,
        "bothCorrect": 72,
        "fqOnly": 0,
        "shortOnly": 0,
        "bothWrong": 0
      }
    },
    "note": "Three live exploratory P3 pilots executed on 2026-10-14 from the isolated p3-short-dependency worktree after preflight p3-preflight/capabilities.json passed, and a fourth (p3c, combined indentation + closing-line cues: indent_named_end H and indent_generic_end I with named_end D and indent F re-run on P3a's 36 families) on 2026-10-15 after preflight p3c-preflight/capabilities.json passed. Every planned trial is terminal with zero infrastructure missingness, zero refusal and zero incomplete; synthetic_mock=false. One replicate, no frozen confirmatory plan, no confidence intervals; differences are descriptive. Short-distance cues alone (G depth_end) do not reach D named_end; redundant explicit information (D, fq import paths) performs best in this pilot; filler 128 lowers D/G/F at depth >= 8. Combining indentation with named closers (H) did not improve on named closers alone in p3c (64 vs 67 of 72), and the identical named_end requests scored 62 in p3a and 67 in p3c, so single-replicate differences of about five per 72 are within run-to-run variation. Not evidence that the short-dependency hypothesis is supported. Raw trials/ bodies, budget and usage ledgers remain private; cases, scores, summaries, paired comparisons, errors, reports, failure inputs and protocols are published byte-identically. P3 runs are not pooled with P1, main, P2 or each other."
  },
  "promptExemplar": {
    "trialId": "024fb909d07dabef1ff984148793631a0780afdb01b69520a4154755b8637602",
    "selection": "Lexicographically smallest trial id among saved pilot ast_to_source / named_end / natural cases, chosen without inspecting outcomes.",
    "task": "ast_to_source",
    "style": "named_end",
    "lexicalRegime": "natural",
    "depth": 4,
    "filler": 32,
    "request": "supporting/pilot-named-end-ast-to-source-request.json",
    "prompt": "supporting/pilot-named-end-ast-to-source-prompt.txt",
    "note": "Saved canonical logical request is byte-preserved. Historical v1 transport object-key insertion order was not captured; no claim of exact wire-byte reconstruction."
  },
  "interruptionNote": "supporting/live-interruption.txt",
  "primaryHypothesis": "When generating an unfamiliar language, redundant closing syntax repeating kind or kind and name may help preserve structure with less reliance on distant context. Whole-program correctness is measured; internal dependency and attention are not measured.",
  "currentExecutionNote": "supporting/live-recovery-20261007.txt"
}