zorya-classifier 0.1.0 → 0.2.0

package/docs/evidence/v02/categorical-heldout-admission-20261020-v4.json

@@ -0,0 +1,270 @@
+{
+  "schema": "categorical-heldout-admission-v1",
+  "createdAt": "2026-10-20T18:28:30.207Z",
+  "freeze": {
+    "file": "categorical-heldout-run-freeze-20261020-v4.json",
+    "sha256": "97aa1173d0569c02522789a00ffd469e1a49792e564c91b104ba719ff6c176c1"
+  },
+  "methodSelectionSha256": "fabb05c2c472c30a8cb20a5f9b6d697b932d94aedc43976359746463ef8708b9",
+  "producerSource": {
+    "candidate": {
+      "files": 67,
+      "sha256": "0a93a965faf8f6ffbae630d03d00d0e68e2fdc69ac31e3d2ba171253e5515ac4"
+    },
+    "noImage": {
+      "files": 73,
+      "sha256": "69c85bacace07e09356186d1a72252402bf10c28e114697f35a121be97887619"
+    },
+    "sourceMapFile": "categorical-heldout-producer-source-map-v4.json",
+    "onlySourceChangeFromV3": "fixed photo input manifest digest",
+    "runtimeSourceUnchanged": true
+  },
+  "noImagePrompt": {
+    "variant": "grammar-plus-examples",
+    "sha256": "e442e9dc2bb920db4bf91b11e6b1e7f507f78f3a778af19fcdb79e98eb87cf5d"
+  },
+  "preDispatchGates": {
+    "localTests": 451,
+    "failed": 0,
+    "skipped": 0,
+    "check": "passed",
+    "format": "passed",
+    "ci": {
+      "receipt": "ci-e551253.json",
+      "jobs": 4,
+      "passed": 4
+    },
+    "installedArchive": {
+      "revision": "e551253376ff4658696a58a1f78ec7605d8ef670",
+      "sha256": "a667f43544e327eda17c5b1f7ab7a0e97da9ae0ab5e944d89813b5da84c5f6c3",
+      "files": 273,
+      "unpackedBytes": 3771154,
+      "consumer": "passed",
+      "httpCalls": 0,
+      "archiveRetained": false
+    },
+    "resource": "proposal-resource-soak-261dc23.json",
+    "runtimeParity": {
+      "unchangedSourceFiles": 53,
+      "sinceRevision": "0c365cb13173e290365a5aa2a1aec122c42b0b4d",
+      "distFiles": 106,
+      "distManifestSha256": "87f6a18d46107ca08b37170db1e9204514dfc72a1dfc3181f59e8ffe76d85c95"
+    },
+    "gitleaksHistory": {
+      "version": "8.30.1",
+      "revision": "e551253376ff4658696a58a1f78ec7605d8ef670",
+      "commits": 91,
+      "findings": 0
+    },
+    "scope": "Verified source baseline and unchanged runtime. The new pin-only source/freeze commit must pass its own local source checks and hosted CI before root GO.",
+    "latestBaselineCi": {
+      "receipt": "ci-d2042e0.json",
+      "jobs": 4,
+      "passed": 4
+    }
+  },
+  "exposure": {
+    "rootAndDispatcherInspectedFreshPixels": false,
+    "rootAndDispatcherInspectedPerCaseTruth": false,
+    "rootMetadataOnly": true,
+    "curatorRoleOwnsLabels": true,
+    "admissionBeforeAnyHeldoutProviderDispatch": true,
+    "priorFailedBlindCohortExcluded": true
+  },
+  "sourceAdmission": {
+    "originalPrivateReceiptSha256": "8952e3ec254f69a72620d550c831929fb3a704b8c9da3f860f9488919634c57f",
+    "metadata": {
+      "aggregateCounts": {
+        "cases": 80,
+        "negative": 20,
+        "photoCasesWithAtLeastOneReferenceMask": 21,
+        "photoReferenceMaskInstances": 32,
+        "positive": 60
+      },
+      "candidateOrder": ["photo", "ui", "map"],
+      "caseCounts": {
+        "map": {
+          "cases": 20,
+          "freshTruthSha256": "9710edd246a9c2253172bfbd6205b9b2ded393dbd54e0d7a73d301c8f69b6f78",
+          "inputSha256": "2c18d2c1bae481d0a1239802f32d4ac985bc0d33a028eaafb490fd5995de7a3b",
+          "negative": 5,
+          "positive": 15,
+          "sourceLedgerSha256": "8dc35d0afd4918c80d96fe263729454bf096b470779bd4c05c4b8e6b0cc215ef",
+          "sourceTruthSha256": "ee279f6568613c0255984f89d636df23fff02a1c98f041fcb266dbd2201a03c3"
+        },
+        "photo": {
+          "cases": 40,
+          "freshTruthSha256": "e0b6b8b977ac31e41e74d1f73044e8638bbfeb8c76fae5732a1207112c7b7caa",
+          "inputSha256": "93cf40732075ab63fd26b77806a7b118382e816c6ab691bf3e5ee68a2acbf013",
+          "negative": 10,
+          "positive": 30,
+          "sourceLedgerSha256": "19507c1c53c7eec76791ae0cd2997db839867afc564d15fc92829b43436d79fe",
+          "sourceTruthSha256": "e0b6b8b977ac31e41e74d1f73044e8638bbfeb8c76fae5732a1207112c7b7caa"
+        },
+        "ui": {
+          "cases": 20,
+          "freshTruthSha256": "047011b2e15c1f43f6d0ab3aff413ffc7cb24fe9c2c08e9ce8be2c12ee3e176c",
+          "inputSha256": "1bf7940fd92fdc1f7fa295296dfc5f84c5f4c32ccc41b7a2e80a0059bff918c3",
+          "negative": 5,
+          "positive": 15,
+          "sourceLedgerSha256": "6ee2d979470127330460f1c95bf4d11b6bd54addd43a6c9ee242f5770dfd7fc6",
+          "sourceTruthSha256": "41fe8ac9769aa7d1c0d8dd4d11a62cdd7ff47ee3e9eac0e046a00f9140b0f680"
+        }
+      },
+      "combinedPins": {
+        "inputs": {
+          "bytes": 45933,
+          "sha256": "6b6cbcf6347fd906f06c78d2b4a36e131344fadb7a06327d7e41ecec8b73f822"
+        },
+        "source-ledger": {
+          "bytes": 176215,
+          "sha256": "0382399d8c4184e3ea411f645b0df948efacad820c5982c1119c7f1e67f54133"
+        },
+        "truth": {
+          "bytes": 188218,
+          "sha256": "1d14963f8743604b5a78181744e56abe1cee53a9e8d259c529279824742912e3"
+        }
+      },
+      "commonImageRoot": "private staging; 112 hash-verified files (80 images and 32 masks); no out-of-root links",
+      "individualInputs": {
+        "map": {
+          "path": ".run/v03-heldout-curation/minetest/heldout-v3/map-oracle-v6/inputs.json",
+          "sha256": "2c18d2c1bae481d0a1239802f32d4ac985bc0d33a028eaafb490fd5995de7a3b"
+        },
+        "photo": {
+          "path": ".run/v03-heldout-curation/heldout80-photo-v3/inputs.json",
+          "sha256": "93cf40732075ab63fd26b77806a7b118382e816c6ab691bf3e5ee68a2acbf013"
+        },
+        "ui": {
+          "path": ".run/v03-heldout-curation/mousepad/heldout-v3/ui-oracle-v2/inputs.json",
+          "sha256": "1bf7940fd92fdc1f7fa295296dfc5f84c5f4c32ccc41b7a2e80a0059bff918c3"
+        }
+      },
+      "limitations": [
+        "Controlled UI and map families do not claim natural-capture diversity.",
+        "Photo mask coverage is incomplete; all available official masks are retained and scored separately.",
+        "Map capture uses one controlled Minetest minimap family with stable source-verified marker support."
+      ],
+      "mapAdmission": {
+        "all20WorldmodInitFilesMatchGenericOracle": true,
+        "captureHistory": {
+          "allSourceAttemptsPreserved": true,
+          "baseV5Cases": 20,
+          "baseV5SourceStabilityRejects": 2,
+          "baseV5StrictPass": 18,
+          "boundedRetryCaptures": 3,
+          "excludedOriginalCaseAfterMaxAttempts": 1,
+          "preservedV4MismatchRun": true,
+          "replacementAttempts": {
+            "setupIncomplete": 1,
+            "strictSourcePass": 1
+          },
+          "replacementCaseAccepted": true,
+          "retryPasses": 1,
+          "setupIncompleteSeparateFromSourceStability": true,
+          "sourceStabilityRejectsAcrossBaseAndRetries": 5
+        },
+        "captureSourceSha256": "67aad0d41786d033f99ef63a521f269404df38368dcdfde8fe4376a0aa0cd45c",
+        "cases": 20,
+        "controlledMapFamilyLimit": "single Minetest 5.6.1 engine/game family; 150x150 minimap crop",
+        "derivationSourceSha256": "ff4720c3256f31e39f8f2207dbffd626695313b181acec6bb731abcbf09366a7",
+        "genericWorldmodOracleSha256": "cab8f3f62179fafaf0b4083790179c7d7a5e3d285e4a5fc9c501b9bc433ed79c",
+        "officialMarkerAssetSha256": "17386c72db28b5ed7c99a60fcbfa5b3fa7286cab7c1228d895bb54f0a5985ad5",
+        "oracleSourceSha256": "cab8f3f62179fafaf0b4083790179c7d7a5e3d285e4a5fc9c501b9bc433ed79c",
+        "rendererSourceSha256": "ded29762ae3c723dd923a89a01c5b8a197bb30443a0155bf0cc22b3bff13f855",
+        "sourceGroup": "controlled-map:minetest-5.6.1-minetest-game",
+        "stratum": "controlled-real-2d-minimap"
+      },
+      "photoAdmission": {
+        "exposureExclusion": "source ledger records exclusion of old v1/v2 image and photographer groups",
+        "license": "CC-BY-2.0 Flickr sources, per-image provenance recorded",
+        "negativeContradictions": 0,
+        "negativeRule": "explicit absence; exclude any negative with target-class box or segmentation row",
+        "source": "Open Images V5 validation",
+        "uniqueImages": 40,
+        "uniquePhotographerSourceGroups": 40
+      },
+      "schemaVersion": 1,
+      "status": "ready-for-root-freeze",
+      "uiAdmission": {
+        "cases": 20,
+        "controlledCssPaintOracle": "AT-SPI control identity/allocation plus exact four painted magenta edges and opaque fill",
+        "naturalCaptureClaim": false,
+        "provenance": {
+          "appLicenseEvidenceSha256": "026d2237c9b625a5de60dcc5f794c8b760788df4137d731a4aa391068a4dd30c",
+          "captureAndOracleSourceSha256": {
+            "capture.py": "6482195c3039a73e006406ef8f3e76f43b3ab7304bafbfc007e468e11206860a",
+            "derive.py": "0de05c65bbba52773094c6658554640edcd7347c0b76b3a77c87d6c273630fe9",
+            "paint-provider.c": "d94138e461226ac09d3de0f9f808d24b3194236ec7e6e685405fc6fcd0c8a9db",
+            "prepare_inputs.py": "737719c8dbe2e332cc81b7a146cc7d0b389e2989cc1a5381a9734f225ee30dea"
+          }
+        },
+        "sourceGroup": "controlled-ui:mousepad-0.6.1",
+        "stratum": "controlled-real-application"
+      }
+    }
+  },
+  "dispatchAuthorization": {
+    "status": "pending-root-go-after-new-source-checks",
+    "producer": "A5 sole paid dispatcher",
+    "order": ["candidate 80 cases", "photo-only no-image control 40 cases"],
+    "retry": 0,
+    "fallback": false,
+    "ordinaryFailurePolicy": "preserve all errors and charges; no restart/reselection"
+  },
+  "limitations": [
+    "One controlled Mousepad UI family and one controlled Minetest minimap family do not establish natural game/UI diversity.",
+    "Native vision candidate and no-image control use different models; the comparison is between recipes, not a causal image ablation.",
+    "Provider timing under concurrent source/CI work is unranked.",
+    "This admission receipt authorizes measurement; it does not qualify accuracy or authorize a release without the unchanged quality gates."
+  ],
+  "baseRootSourceRevision": "e551253376ff4658696a58a1f78ec7605d8ef670",
+  "executableSourceCommit": "assigned by the dispatch receipt after this source/freeze commit; closure below is authoritative",
+  "previousZeroHttpAdmissionFailure": {
+    "receipt": "categorical-heldout-predispatch-20261020-v3.json",
+    "physicalProviderCalls": 0,
+    "preserved": true
+  },
+  "producerOnlyInputDerivative": {
+    "receiptSha256": "0b948b0f2b072da4c230259b68d46fe1a77ed9129e82f439e1be2e229a49e8fc",
+    "metadata": {
+      "schema": "categorical-photo-neutral-instruction-derivative-v1",
+      "createdAt": "2026-10-20T18:26:48.405Z",
+      "sourceInputSha256": "93cf40732075ab63fd26b77806a7b118382e816c6ab691bf3e5ee68a2acbf013",
+      "derivedInputSha256": "5d125e853ed93558ab634e406ca90cf931569953279c1630d49dcf239b9baa79",
+      "cases": 40,
+      "changedFields": 40,
+      "onlyChangedField": "locate.instruction",
+      "originalInstruction": "Locate each visible instance of the named physical object class. Return unknown if unsupported; do not infer from absent labels.",
+      "approvedInstruction": "Locate the visible physical instance of the named object; return unknown when unsupported.",
+      "allOtherJsonValuesIdentical": true,
+      "rawOriginalPreserved": true,
+      "imagesTargetsIdsDimensionsClassifyUnchanged": true,
+      "truthAndMasksUnchanged": true,
+      "all80TargetPolicyAndPrivacyChecksPass": true,
+      "physicalProviderCalls": 0,
+      "pixelsDecoded": false,
+      "truthRead": false
+    },
+    "publicCorpusScopeUnchanged": true,
+    "actualManifests": [
+      {
+        "group": "photo",
+        "sha256": "5d125e853ed93558ab634e406ca90cf931569953279c1630d49dcf239b9baa79",
+        "cases": 40
+      },
+      {
+        "group": "ui",
+        "sha256": "1bf7940fd92fdc1f7fa295296dfc5f84c5f4c32ccc41b7a2e80a0059bff918c3",
+        "cases": 20
+      },
+      {
+        "group": "map",
+        "sha256": "2c18d2c1bae481d0a1239802f32d4ac985bc0d33a028eaafb490fd5995de7a3b",
+        "cases": 20
+      }
+    ],
+    "acquisitionReceiptInputPinsAreHistorical": true,
+    "curatorIndependentlyConfirmedSameReplacementRule": true
+  }
+}

package/docs/evidence/v02/categorical-heldout-dispatch-authorization-20261020-v4.json

@@ -0,0 +1,79 @@
+{
+  "schema": "categorical-heldout-dispatch-authorization-v1",
+  "recordedAt": "2026-10-20T18:34:19.252000+00:00",
+  "executableSourceRevision": "c107ddfc8a129070e4bf7232f2c713b93fe650d4",
+  "phase": "heldout",
+  "freeze": {
+    "file": "categorical-heldout-run-freeze-20261020-v4.json",
+    "sha256": "97aa1173d0569c02522789a00ffd469e1a49792e564c91b104ba719ff6c176c1"
+  },
+  "producerSource": {
+    "candidateFiles": 67,
+    "candidateSha256": "0a93a965faf8f6ffbae630d03d00d0e68e2fdc69ac31e3d2ba171253e5515ac4",
+    "noImageFiles": 73,
+    "noImageSha256": "69c85bacace07e09356186d1a72252402bf10c28e114697f35a121be97887619"
+  },
+  "offlineAdmission": {
+    "cases": 80,
+    "photo": 40,
+    "ui": 20,
+    "map": 20,
+    "targetPolicyAndTaskPrivacyChecks": "passed",
+    "pixelsOrTruthInspectedByRootOrDispatcher": false,
+    "physicalCalls": 0
+  },
+  "sourceGates": {
+    "localTests": 451,
+    "failures": 0,
+    "skipped": 0,
+    "strictCheck": "passed",
+    "format": "passed",
+    "hostedCiReceipt": "ci-c107ddf.json",
+    "hostedCiRun": 39475196959,
+    "hostedWindowsLinuxNodeJobs": 4,
+    "jobsPassed": 4,
+    "gitleaksHistory": {
+      "version": "8.30.1",
+      "commits": 93,
+      "findings": 0
+    }
+  },
+  "installedArchive": {
+    "sha256": "866ba6262d9bae307f044ab3174785910c670afe1e17a6c2c4963175fe1b5335",
+    "files": 281,
+    "unpackedBytes": 3840895,
+    "consumer": "passed",
+    "providerCalls": 0,
+    "runtimeDistFiles": 106,
+    "runtimeDistSha256": "87f6a18d46107ca08b37170db1e9204514dfc72a1dfc3181f59e8ffe76d85c95",
+    "retainedByPackCheck": false
+  },
+  "nativeAndResourceScope": {
+    "sourceRuntimeFiles": 53,
+    "unchangedSince": "0c365cb13173e290365a5aa2a1aec122c42b0b4d",
+    "nativeWindowsReceipt": "native-proposal-windows-0c365cb.json",
+    "twoHourResourceReceipt": "proposal-resource-soak-261dc23.json",
+    "gameSessionClaim": false
+  },
+  "authorization": {
+    "rootGoIssuedBeforeFirstDispatch": true,
+    "ownerAuthorization": "existing session authorizes public model tests and eventual publication after gates",
+    "soleDispatcher": "A5",
+    "order": ["candidate80", "photo-no-image-control40"],
+    "retries": 0,
+    "fallback": false,
+    "freshTruthAccessForScoringOnlyAfterBothProducersTerminate": true
+  },
+  "observedFirstDispatch": {
+    "dispatcherRunId": "v02-heldout-dispatch-2026-10-20T183257373Z-72857a56",
+    "profile": "candidate",
+    "startedAt": "2026-10-20T18:32:57.373Z",
+    "runId": "v02-categorical-heldout-native-luna-categorical-presence-sam-heldout-target-plus-neutral-v1-2026-10-20T183257519Z-d6cfb60e"
+  },
+  "status": "authorized-measurement-in-progress",
+  "limitations": [
+    "This receipt records dispatch readiness, not an accuracy result or release qualification.",
+    "The earlier v3 zero-HTTP setup failure and source-oracle capture rejections remain preserved.",
+    "Final publication still requires independent quality evaluation and checks for the final release source/archive."
+  ]
+}

package/docs/evidence/v02/categorical-heldout-predispatch-20261020-v3.json

@@ -0,0 +1,57 @@
+{
+  "schema": "categorical-heldout-zero-http-admission-failure-v1",
+  "freezeFile": "categorical-heldout-run-freeze-20261020-v3.json",
+  "freezeSha256": "d621530a0bea286d7ae3df83af7d74f7ecb2b94b7873f1cd3397ef085e695a9a",
+  "sealedProducerRevision": "6e398aecdbe8e34ade71ed83a7d64b7ab85a02d3",
+  "physicalProviderCalls": 0,
+  "logicalProviderCalls": 0,
+  "producerCaseOutputs": 0,
+  "noImageControlStarted": false,
+  "status": "prefetch-admission-failed",
+  "diagnosis": {
+    "validator": "validateTargetPolicyTasks",
+    "source": "benchmarks/v02/categorical-presence-fresh-runner.ts",
+    "errorCode": "INVALID_INPUT",
+    "message": "Fresh case has an unapproved locate instruction",
+    "photoInstructionsRejected": 40,
+    "uiInstructionsRejected": 0,
+    "mapInstructionsRejected": 0,
+    "manifestHashesMatchFreeze": true,
+    "caseCountsPass": true,
+    "taskIdentityPrivacyGuardPasses": true,
+    "originalInstruction": "Locate each visible instance of the named physical object class. Return unknown if unsupported; do not infer from absent labels.",
+    "requiredInstruction": "Locate the visible physical instance of the named object; return unknown when unsupported."
+  },
+  "preservedDispatchArtifacts": {
+    "start": {
+      "sha256": "959e1a9ca90a532c974c2d53766cfac79d911b0014100f21d17aae3d1845c760",
+      "record": {
+        "schema": "categorical-heldout-dispatch-v1",
+        "dispatcherRunId": "v02-heldout-dispatch-2026-10-20T181953455Z-1a4c23ae",
+        "phase": "heldout",
+        "profile": "candidate",
+        "freezeSha256": "d621530a0bea286d7ae3df83af7d74f7ecb2b94b7873f1cd3397ef085e695a9a",
+        "startedAt": "2026-10-20T18:19:53.455Z",
+        "state": "dispatching"
+      }
+    },
+    "incomplete": {
+      "sha256": "d096dd0d30d48faeb14b7298b5ba0c3e2004609ab7b98e701055377016abbfec",
+      "record": {
+        "schema": "categorical-heldout-dispatch-incomplete-v1",
+        "dispatcherRunId": "v02-heldout-dispatch-2026-10-20T181953455Z-1a4c23ae",
+        "profile": "candidate",
+        "finishedAt": "2026-10-20T18:19:53.488Z",
+        "errorCode": "INVALID_INPUT"
+      }
+    }
+  },
+  "repairPolicy": "Create a separate immutable photo-input derivative replacing only the common instruction with the selected DEV neutral suffix; preserve all targets, IDs, images, references, labels and original receipts. Bind new executable source and input pins before first provider dispatch. No method or scoring gate changes.",
+  "blindness": {
+    "providerSawAnyFreshImage": false,
+    "rootOrDispatcherInspectedFreshPixels": false,
+    "rootOrDispatcherInspectedPerCaseTruth": false,
+    "rootInspectedOnlyUniqueGenericInstructionCounts": true
+  },
+  "accuracyInterpretation": "This setup failure produced no classification/localization outcomes; it cannot qualify or fail accuracy and must not be rewritten as a successful run."
+}