zorya-classifier 0.0.0-stage → 0.2.0

package/docs/evidence/v02/wesnoth-proof-root-review-20261020-v1.json

@@ -0,0 +1,76 @@
+{
+  "schema": "independent-wesnoth-excluded-proof-review-v1",
+  "sourceMapSha256": "41d2b65d3098c77c68d560832938c9e38bf2a549496e338cb158835ebbdff417",
+  "imageSha256": {
+    "a": "560f6f6fa5b69e0ff30f5416457f537ede44c2d023cd9096785ad98865d19c86",
+    "b": "cc7763407c1084802c3ab5e78c538ce7da6c7c1186d396c4daae6304bfabd569",
+    "a2": "0a89337c48966476368131328402d056940da7fefc67071e6883b825a5b27917"
+  },
+  "roiRawSha256": {
+    "a": "49416ea53f982154205b2a0e6c04deea74c5a34db6d3eddfa6e706793e14927a",
+    "b": "440809bfeff0badb16815394f64ee40b62d1c3ee904c3670e97bf927d40546e1",
+    "a2": "49416ea53f982154205b2a0e6c04deea74c5a34db6d3eddfa6e706793e14927a"
+  },
+  "captureSize": {
+    "width": 1024,
+    "height": 768
+  },
+  "contentRect": {
+    "left": 856,
+    "top": 11,
+    "width": 160,
+    "height": 212
+  },
+  "bitmapRoi": {
+    "left": 856,
+    "top": 11,
+    "width": 159,
+    "height": 212
+  },
+  "sameStateRoiByteIdentical": true,
+  "markerCount": 51,
+  "changedPixels": 1297,
+  "float32RepeatedSampler": {
+    "projectedPixels": 1363,
+    "deltaInside": 1297,
+    "outside": [],
+    "projectedUnchanged": 66,
+    "allMarkerProjectionsIntersectDelta": true,
+    "sampledSourceYAtRow53": 215,
+    "cursorAtRow53": 215.9999237060547
+  },
+  "idealRationalSampler": {
+    "projectedPixels": 1363,
+    "deltaInside": 1294,
+    "outside": [
+      [923, 64],
+      [924, 64],
+      [925, 64]
+    ],
+    "projectedUnchanged": 69,
+    "allMarkerProjectionsIntersectDelta": true,
+    "sampledSourceYAtRow53": 216
+  },
+  "providerCalls": 0,
+  "freshCohortImagesRead": false,
+  "freshCohortTruthRead": false,
+  "scope": "Excluded proof only. Projection is an upper bound, not an exact foreground mask; painter overdraw leaves projected pixels unchanged. Full-frame A/A2 stability is not claimed.",
+  "operatorSha256": "b37bfea4abeeb2b25c4fb630129bd1da483b798f5df8bba2846ca4e15c171a1f",
+  "rawReviewSha256": "c0e359dd2811ee90d1a356b7ac8357fd50d3bf2182f68380f249c40230c4928d",
+  "sourceTransform": {
+    "map": "Wesnoth 1.16.11 official 4p_Siege_Castles.map",
+    "rawGrid": [38, 38],
+    "defaultBorderSize": 1,
+    "unscaledBitmap": [648, 864],
+    "markerRect": "L=18*x-6; T=24*y+(odd(x)?0:-12); width=18; height=24; half-open",
+    "sampling": "float32(source/destination); float32 accumulator initialized0; truncation before each float32 addition",
+    "sourceAudit": "wesnoth-1.16.11-minimap-projection-source-audit.md"
+  },
+  "limits": [
+    "This proof is excluded from the new blind cohort.",
+    "Same-state stability is established only for the minimap bitmap ROI.",
+    "Whole-source projection covers markers, including pixels hidden by later terrain painting. It is not an exact foreground mask.",
+    "Controlled Wesnoth rendering is one source family; this is not a diverse natural-game accuracy benchmark.",
+    "No model request or localization accuracy was measured by this proof."
+  ]
+}

package/docs/evidence/vision-cost-reconciliation.json

@@ -0,0 +1,37 @@
+{
+  "priorReportSha256": "25c463dc9b005c2702dd94e78b12cee861e5f27827c5093cda857771837b19c7",
+  "reconciled": [
+    {
+      "path": ".run/vision-selection-v1/qwen/photo/photo-dev-01/wire-response-1.json",
+      "sha256": "2fa3b79d85a81fc392c13a68b3174e28e40f7c1c8502f89c7e534b218a8cc686",
+      "costUsd": 0.00004498
+    },
+    {
+      "path": ".run/vision-selection-v1/qwen/photo/photo-dev-17/wire-response-1.json",
+      "sha256": "483a9270a3f91cddc71bd819ddd2b146dce3d5176b516fe651984d444033d94c",
+      "costUsd": 0.00004114
+    },
+    {
+      "path": ".run/vision-selection-v1/qwen/photo/photo-heldout-09/wire-response-1.json",
+      "sha256": "f4cf9a5021bafd083d0620b91ccf95a0f77a31317333610510b394af71c4dbb2",
+      "costUsd": 0.00003826
+    },
+    {
+      "path": ".run/vision-selection-v2/haiku/photo/photo-dev-16/wire-response-1.json",
+      "sha256": "a27afbda79dbfa8bcaab0a2cedb958ffe848292fb1792e89c3f32476352d0d85",
+      "costUsd": 0.0001922
+    },
+    {
+      "path": ".run/vision-selection-v2/haiku/text/support-heldout-016/wire-response-1.json",
+      "sha256": "a51ad18129c35d4d788a69a933b0fa25edb64d47979c6d5cf0a29f513f8606be",
+      "costUsd": 0.0001622
+    }
+  ],
+  "seed": {
+    "admittedProviderCalls": 288,
+    "knownCostUsd": 0.029322660000000056,
+    "unknownCostCalls": 3,
+    "inFlightCalls": 0
+  },
+  "note": "Billing metadata only; failed classifications remain failures. Two auth errors and one provider error still retain unknown-cost reservations."
+}

package/docs/evidence/vision-model-summary.json

@@ -0,0 +1,562 @@
+{
+  "protocols": {
+    "vision-selection-v1": "v1 requires min4 ranks for every categorical question",
+    "vision-selection-v2": "v2 requires one classification/final rank, four beam ranks",
+    "decisions-selection-v1": "native typed Choice/Noul",
+    "decisions-completion-v1": "same native runtime, missing development photos/heldout texts only"
+  },
+  "summary": {
+    "nex": {
+      "photo": {
+        "count": 10,
+        "correct": 6,
+        "errors": 0,
+        "p50Ms": 1599.773122999999,
+        "p95Ms": 3901.195785,
+        "knownValidatedCostUsd": 0.00030315,
+        "cases": [
+          "photo-dev-01",
+          "photo-dev-02",
+          "photo-dev-03",
+          "photo-dev-04",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-07",
+          "photo-dev-08",
+          "photo-dev-09",
+          "photo-dev-10"
+        ],
+        "mistakes": ["photo-dev-05", "photo-dev-06", "photo-dev-07", "photo-dev-08"]
+      },
+      "text": {
+        "count": 0,
+        "correct": 0,
+        "errors": 0,
+        "p50Ms": null,
+        "p95Ms": null,
+        "knownValidatedCostUsd": 0,
+        "cases": [],
+        "mistakes": []
+      },
+      "heldout": {
+        "count": 0,
+        "correct": 0
+      }
+    },
+    "qwen": {
+      "photo": {
+        "count": 60,
+        "correct": 53,
+        "errors": 3,
+        "p50Ms": 3939.7940679999883,
+        "p95Ms": 5315.089853000012,
+        "knownValidatedCostUsd": 0.00230825,
+        "cases": [
+          "photo-dev-01",
+          "photo-dev-02",
+          "photo-dev-03",
+          "photo-dev-04",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-07",
+          "photo-dev-08",
+          "photo-dev-09",
+          "photo-dev-10",
+          "photo-dev-11",
+          "photo-dev-12",
+          "photo-dev-13",
+          "photo-dev-14",
+          "photo-dev-15",
+          "photo-dev-16",
+          "photo-dev-17",
+          "photo-dev-18",
+          "photo-dev-19",
+          "photo-dev-20",
+          "photo-heldout-01",
+          "photo-heldout-02",
+          "photo-heldout-03",
+          "photo-heldout-04",
+          "photo-heldout-05",
+          "photo-heldout-06",
+          "photo-heldout-07",
+          "photo-heldout-08",
+          "photo-heldout-09",
+          "photo-heldout-10",
+          "photo-heldout-11",
+          "photo-heldout-12",
+          "photo-heldout-13",
+          "photo-heldout-14",
+          "photo-heldout-15",
+          "photo-heldout-16",
+          "photo-heldout-17",
+          "photo-heldout-18",
+          "photo-heldout-19",
+          "photo-heldout-20",
+          "photo-heldout-21",
+          "photo-heldout-22",
+          "photo-heldout-23",
+          "photo-heldout-24",
+          "photo-heldout-25",
+          "photo-heldout-26",
+          "photo-heldout-27",
+          "photo-heldout-28",
+          "photo-heldout-29",
+          "photo-heldout-30",
+          "photo-heldout-31",
+          "photo-heldout-32",
+          "photo-heldout-33",
+          "photo-heldout-34",
+          "photo-heldout-35",
+          "photo-heldout-36",
+          "photo-heldout-37",
+          "photo-heldout-38",
+          "photo-heldout-39",
+          "photo-heldout-40"
+        ],
+        "mistakes": [
+          "photo-dev-01",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-17",
+          "photo-heldout-07",
+          "photo-heldout-09",
+          "photo-heldout-11"
+        ]
+      },
+      "text": {
+        "count": 30,
+        "correct": 26,
+        "errors": 0,
+        "p50Ms": 1101.8768629999831,
+        "p95Ms": 1246.0037309999461,
+        "knownValidatedCostUsd": 0.00099663,
+        "cases": [
+          "support-dev-001",
+          "support-dev-002",
+          "support-dev-003",
+          "support-dev-004",
+          "support-dev-005",
+          "support-dev-006",
+          "support-dev-007",
+          "support-dev-008",
+          "support-dev-009",
+          "support-dev-010",
+          "support-heldout-001",
+          "support-heldout-002",
+          "support-heldout-003",
+          "support-heldout-004",
+          "support-heldout-005",
+          "support-heldout-006",
+          "support-heldout-007",
+          "support-heldout-008",
+          "support-heldout-009",
+          "support-heldout-010",
+          "support-heldout-011",
+          "support-heldout-012",
+          "support-heldout-013",
+          "support-heldout-014",
+          "support-heldout-015",
+          "support-heldout-016",
+          "support-heldout-017",
+          "support-heldout-018",
+          "support-heldout-019",
+          "support-heldout-020"
+        ],
+        "mistakes": [
+          "support-dev-009",
+          "support-heldout-009",
+          "support-heldout-014",
+          "support-heldout-019"
+        ]
+      },
+      "heldout": {
+        "count": 40,
+        "correct": 37
+      }
+    },
+    "luna": {
+      "photo": {
+        "count": 60,
+        "correct": 54,
+        "errors": 1,
+        "p50Ms": 3202.183490000025,
+        "p95Ms": 4788.130577999982,
+        "knownValidatedCostUsd": 0.01018445,
+        "cases": [
+          "photo-dev-01",
+          "photo-dev-02",
+          "photo-dev-03",
+          "photo-dev-04",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-07",
+          "photo-dev-08",
+          "photo-dev-09",
+          "photo-dev-10",
+          "photo-dev-11",
+          "photo-dev-12",
+          "photo-dev-13",
+          "photo-dev-14",
+          "photo-dev-15",
+          "photo-dev-16",
+          "photo-dev-17",
+          "photo-dev-18",
+          "photo-dev-19",
+          "photo-dev-20",
+          "photo-heldout-01",
+          "photo-heldout-02",
+          "photo-heldout-03",
+          "photo-heldout-04",
+          "photo-heldout-05",
+          "photo-heldout-06",
+          "photo-heldout-07",
+          "photo-heldout-08",
+          "photo-heldout-09",
+          "photo-heldout-10",
+          "photo-heldout-11",
+          "photo-heldout-12",
+          "photo-heldout-13",
+          "photo-heldout-14",
+          "photo-heldout-15",
+          "photo-heldout-16",
+          "photo-heldout-17",
+          "photo-heldout-18",
+          "photo-heldout-19",
+          "photo-heldout-20",
+          "photo-heldout-21",
+          "photo-heldout-22",
+          "photo-heldout-23",
+          "photo-heldout-24",
+          "photo-heldout-25",
+          "photo-heldout-26",
+          "photo-heldout-27",
+          "photo-heldout-28",
+          "photo-heldout-29",
+          "photo-heldout-30",
+          "photo-heldout-31",
+          "photo-heldout-32",
+          "photo-heldout-33",
+          "photo-heldout-34",
+          "photo-heldout-35",
+          "photo-heldout-36",
+          "photo-heldout-37",
+          "photo-heldout-38",
+          "photo-heldout-39",
+          "photo-heldout-40"
+        ],
+        "mistakes": [
+          "photo-dev-01",
+          "photo-dev-05",
+          "photo-dev-15",
+          "photo-dev-17",
+          "photo-heldout-07",
+          "photo-heldout-27"
+        ]
+      },
+      "text": {
+        "count": 30,
+        "correct": 30,
+        "errors": 0,
+        "p50Ms": 962.3370350000332,
+        "p95Ms": 1260.0757910000393,
+        "knownValidatedCostUsd": 0.0029629,
+        "cases": [
+          "support-dev-001",
+          "support-dev-002",
+          "support-dev-003",
+          "support-dev-004",
+          "support-dev-005",
+          "support-dev-006",
+          "support-dev-007",
+          "support-dev-008",
+          "support-dev-009",
+          "support-dev-010",
+          "support-heldout-001",
+          "support-heldout-002",
+          "support-heldout-003",
+          "support-heldout-004",
+          "support-heldout-005",
+          "support-heldout-006",
+          "support-heldout-007",
+          "support-heldout-008",
+          "support-heldout-009",
+          "support-heldout-010",
+          "support-heldout-011",
+          "support-heldout-012",
+          "support-heldout-013",
+          "support-heldout-014",
+          "support-heldout-015",
+          "support-heldout-016",
+          "support-heldout-017",
+          "support-heldout-018",
+          "support-heldout-019",
+          "support-heldout-020"
+        ],
+        "mistakes": []
+      },
+      "heldout": {
+        "count": 40,
+        "correct": 38
+      }
+    },
+    "haiku": {
+      "photo": {
+        "count": 16,
+        "correct": 11,
+        "errors": 2,
+        "p50Ms": 3094.087862999999,
+        "p95Ms": 4186.938432999999,
+        "knownValidatedCostUsd": 0.0025682,
+        "cases": [
+          "photo-dev-01",
+          "photo-dev-02",
+          "photo-dev-03",
+          "photo-dev-04",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-07",
+          "photo-dev-08",
+          "photo-dev-09",
+          "photo-dev-10",
+          "photo-dev-11",
+          "photo-dev-12",
+          "photo-dev-13",
+          "photo-dev-14",
+          "photo-dev-15",
+          "photo-dev-16"
+        ],
+        "mistakes": ["photo-dev-01", "photo-dev-05", "photo-dev-07", "photo-dev-15", "photo-dev-16"]
+      },
+      "text": {
+        "count": 30,
+        "correct": 29,
+        "errors": 1,
+        "p50Ms": 757.0049549999967,
+        "p95Ms": 1003.6543539999984,
+        "knownValidatedCostUsd": 0.0046254,
+        "cases": [
+          "support-dev-001",
+          "support-dev-002",
+          "support-dev-003",
+          "support-dev-004",
+          "support-dev-005",
+          "support-dev-006",
+          "support-dev-007",
+          "support-dev-008",
+          "support-dev-009",
+          "support-dev-010",
+          "support-heldout-001",
+          "support-heldout-002",
+          "support-heldout-003",
+          "support-heldout-004",
+          "support-heldout-005",
+          "support-heldout-006",
+          "support-heldout-007",
+          "support-heldout-008",
+          "support-heldout-009",
+          "support-heldout-010",
+          "support-heldout-011",
+          "support-heldout-012",
+          "support-heldout-013",
+          "support-heldout-014",
+          "support-heldout-015",
+          "support-heldout-016",
+          "support-heldout-017",
+          "support-heldout-018",
+          "support-heldout-019",
+          "support-heldout-020"
+        ],
+        "mistakes": ["support-heldout-016"]
+      },
+      "heldout": {
+        "count": 0,
+        "correct": 0
+      }
+    },
+    "luna-decisions": {
+      "photo": {
+        "count": 60,
+        "correct": 55,
+        "errors": 0,
+        "p50Ms": 1434.9253419999986,
+        "p95Ms": 3436.2048670000004,
+        "knownValidatedCostUsd": 0.0059048,
+        "cases": [
+          "photo-heldout-01",
+          "photo-heldout-02",
+          "photo-heldout-03",
+          "photo-heldout-04",
+          "photo-heldout-05",
+          "photo-heldout-06",
+          "photo-heldout-07",
+          "photo-heldout-08",
+          "photo-heldout-09",
+          "photo-heldout-10",
+          "photo-heldout-11",
+          "photo-heldout-12",
+          "photo-heldout-13",
+          "photo-heldout-14",
+          "photo-heldout-15",
+          "photo-heldout-16",
+          "photo-heldout-17",
+          "photo-heldout-18",
+          "photo-heldout-19",
+          "photo-heldout-20",
+          "photo-heldout-21",
+          "photo-heldout-22",
+          "photo-heldout-23",
+          "photo-heldout-24",
+          "photo-heldout-25",
+          "photo-heldout-26",
+          "photo-heldout-27",
+          "photo-heldout-28",
+          "photo-heldout-29",
+          "photo-heldout-30",
+          "photo-heldout-31",
+          "photo-heldout-32",
+          "photo-heldout-33",
+          "photo-heldout-34",
+          "photo-heldout-35",
+          "photo-heldout-36",
+          "photo-heldout-37",
+          "photo-heldout-38",
+          "photo-heldout-39",
+          "photo-heldout-40",
+          "photo-dev-01",
+          "photo-dev-02",
+          "photo-dev-03",
+          "photo-dev-04",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-07",
+          "photo-dev-08",
+          "photo-dev-09",
+          "photo-dev-10",
+          "photo-dev-11",
+          "photo-dev-12",
+          "photo-dev-13",
+          "photo-dev-14",
+          "photo-dev-15",
+          "photo-dev-16",
+          "photo-dev-17",
+          "photo-dev-18",
+          "photo-dev-19",
+          "photo-dev-20"
+        ],
+        "mistakes": [
+          "photo-heldout-07",
+          "photo-dev-01",
+          "photo-dev-05",
+          "photo-dev-06",
+          "photo-dev-17"
+        ]
+      },
+      "text": {
+        "count": 30,
+        "correct": 30,
+        "errors": 0,
+        "p50Ms": 216.72845600000073,
+        "p95Ms": 407.2245180000027,
+        "knownValidatedCostUsd": 0.0020347,
+        "cases": [
+          "support-dev-001",
+          "support-dev-002",
+          "support-dev-003",
+          "support-dev-004",
+          "support-dev-005",
+          "support-dev-006",
+          "support-dev-007",
+          "support-dev-008",
+          "support-dev-009",
+          "support-dev-010",
+          "support-heldout-001",
+          "support-heldout-002",
+          "support-heldout-003",
+          "support-heldout-004",
+          "support-heldout-005",
+          "support-heldout-006",
+          "support-heldout-007",
+          "support-heldout-008",
+          "support-heldout-009",
+          "support-heldout-010",
+          "support-heldout-011",
+          "support-heldout-012",
+          "support-heldout-013",
+          "support-heldout-014",
+          "support-heldout-015",
+          "support-heldout-016",
+          "support-heldout-017",
+          "support-heldout-018",
+          "support-heldout-019",
+          "support-heldout-020"
+        ],
+        "mistakes": []
+      },
+      "heldout": {
+        "count": 40,
+        "correct": 39
+      }
+    }
+  },
+  "evidencePins": [
+    {
+      "path": "docs/evidence/vision-selection-v1-report.json",
+      "sha256": "9ca1fb956c2d937f14e55ca264b09c33d26707693fcf1095243c5e4fb5adc1b5",
+      "originalLocalRunSha256": "a340e426671b59dd5725ff4bd52f8535f559688ccabb4e270dfe22ddba0fc80f"
+    },
+    {
+      "path": "docs/evidence/vision-selection-v1-manifest.json",
+      "sha256": "1818066f005f116d61721ed31a82c3c816ecdfe019fc8d2e498904656c121d58",
+      "originalLocalRunSha256": "1818066f005f116d61721ed31a82c3c816ecdfe019fc8d2e498904656c121d58"
+    },
+    {
+      "path": "docs/evidence/vision-selection-v2-report.json",
+      "sha256": "cc5f454157872ebf96c9297960d994d813a3fb95dce74ea38c1232c7dd31634c",
+      "originalLocalRunSha256": "7281a7a98b09be42397daea563a3f81eabb0ef8e315a65ff413c3c2ee5524fce"
+    },
+    {
+      "path": "docs/evidence/vision-selection-v2-manifest.json",
+      "sha256": "8ac42170f6afbd14aa7c79b8709e7e24413eb9003231d920f99e24386a565ca9",
+      "originalLocalRunSha256": "8ac42170f6afbd14aa7c79b8709e7e24413eb9003231d920f99e24386a565ca9"
+    },
+    {
+      "path": "docs/evidence/decisions-selection-v1-report.json",
+      "sha256": "45b4b2a22e42cae9e879d95f557a9c0fb03f291862d51a044c6337ec61ad093e",
+      "originalLocalRunSha256": "25c463dc9b005c2702dd94e78b12cee861e5f27827c5093cda857771837b19c7"
+    },
+    {
+      "path": "docs/evidence/decisions-selection-v1-manifest.json",
+      "sha256": "4c9288d1142971bbbe3313bd25ca449bf6554051a2fe7d39fbe3d9f5cb8edec0",
+      "originalLocalRunSha256": "4c9288d1142971bbbe3313bd25ca449bf6554051a2fe7d39fbe3d9f5cb8edec0"
+    },
+    {
+      "path": "docs/evidence/decisions-completion-v1-report.json",
+      "sha256": "60475bf7148fa3392a844cb263f5cb4f1754d9b446a0b46e7498da7b9ae0f444",
+      "originalLocalRunSha256": "15c728c3614d34f21a30c0de5bb62253d097ed316b4d80c5a8d1d485d257f76b"
+    },
+    {
+      "path": "docs/evidence/decisions-completion-v1-manifest.json",
+      "sha256": "c2c8dbabf1799fe365cf959a6fae7eba120598264dcbc8d5c1b65b05b152337a",
+      "originalLocalRunSha256": "c2c8dbabf1799fe365cf959a6fae7eba120598264dcbc8d5c1b65b05b152337a"
+    }
+  ],
+  "costs": "Rejected responses with cost metadata are reconciled for budget only; their classifications remain failures. Non-numeric/missing costs retain reservations. No invoice verification.",
+  "publication": "Public JSON is formatted for repository style; both original local bytes and published file hashes are pinned. No recorded model outcomes are changed.",
+  "finalComparisonBudget": {
+    "admittedProviderCalls": 332,
+    "maxCalls": 1000,
+    "maxCostUsd": 0.5,
+    "knownCostUsd": 0.03456876000000009,
+    "unknownCostCalls": 4,
+    "inFlightCalls": 0,
+    "reservationPerUnknownUsd": 0.05,
+    "accountedExposureUsd": 0.2345687600000001,
+    "exceededByReportedCost": false
+  },
+  "localizationSmoke": {
+    "caseId": "photo-dev-02",
+    "status": "located",
+    "physicalCalls": 3,
+    "report": "luna-decisions-locator-smoke-report.json"
+  }
+}