zorya-classifier 0.0.0-stage → 0.2.0

package/docs/evidence/luna-decisions-fine-probe.json

@@ -0,0 +1,19 @@
+{
+  "httpStatus": 502,
+  "transportError": false,
+  "durationMs": 1977,
+  "requestSha256": "e11285a5d5d6914645ebc210d7bb5d36afd402ac3ab9e4e5fce4d3a28ccc2290",
+  "responseSha256": "082902bef07a9357f83d3ba4f88b500337975a3127159ef86a4968d12abca170",
+  "reportedCostUsd": null,
+  "aggregatePhysicalCalls": 329,
+  "aggregateKnownCostUsd": 0.03266816000000008,
+  "aggregateUnknownCostCalls": 4,
+  "aggregateAccountedExposureUsd": 0.2326681600000001,
+  "requestEvidence": "Recorded public-corpus fine request; original image-bearing request retained locally, SHA256 above",
+  "response": {
+    "error": {
+      "message": "OpenAI refused to answer question \"selection\"",
+      "code": 502
+    }
+  }
+}

package/docs/evidence/luna-decisions-locator-smoke-manifest.json

@@ -0,0 +1,39 @@
+{
+  "caseId": "photo-dev-02",
+  "model": "openai/gpt-6-luna-decisions",
+  "provider": "openai",
+  "maxNewCalls": 3,
+  "priorBudget": {
+    "admittedProviderCalls": 329,
+    "maxCalls": 1000,
+    "maxCostUsd": 0.5,
+    "knownCostUsd": 0.03266816000000008,
+    "unknownCostCalls": 4,
+    "inFlightCalls": 0,
+    "reservationPerUnknownUsd": 0.05,
+    "accountedExposureUsd": 0.2326681600000001,
+    "exceededByReportedCost": false
+  },
+  "imageSha256": "68d92b810be3205702e9ea527654ffbbb2eab3f48fd73f66770a0a0df4f484a1",
+  "corpus": {
+    "corpus-manifest.json": "00f4459996db0f61a38235adaf512e1bfc52e02e1d599bb590b62a98217e4cdb",
+    "photo-manifest.json": "a99f47d4e1c064581c6cc9123b58a5c81b196d15227963bf015041544f713aa3",
+    "text/text-manifest.json": "5e74c8c13cd1f1f8a2e2ecc22e90e6c0c83ebf5f9407dfed5b87204d3b8831ac",
+    "text/inputs/policy.json": "9db3010c97fa056fbb4a43ea51a5c973bd75b9d21f8d07f2c15d6d176e712b8a",
+    "inputs/dev.json": "0c443231956e9578afa43d4bf884cc9741594ae69ac5f957657b9c5a24422bb6",
+    "ground-truth/dev.json": "c618e1a950ae89e0c9ced0c2b1ca249ea7adc891f61c0579468c479613ea4b26",
+    "text/inputs/dev.json": "e7d2bc07559cba70494d6e4672cc3e1113c070aba8b2694ee45e76ebcbc04fcc",
+    "text/ground-truth/dev.json": "4827650b560f20b32b0366dea14947681d29ef9fefb5a059301ba5abf5c031f0",
+    "inputs/heldout.json": "2ce42723824900de9a8eca60b58220567b938a88855ec41a5a882bb410d3b1e4",
+    "ground-truth/heldout.json": "737ee884b8bbf557191d5848b56f17ac056fd01793cd8fb738c5624ea119c7d2",
+    "text/inputs/heldout.json": "e33beeeb011c55992d955ce66397717875c0084bcc21be29ecd123bd29ccce4a",
+    "text/ground-truth/heldout.json": "d29cf2fb7b980d138fd7c8e712708fa9730aa6637f350468ae15651bf9f5c4b0"
+  },
+  "runtime": {
+    "index": "69d648d8339d9f2bd5faf1e6db05b23fec163490603a8e72b44750726df8c3ba",
+    "nativeAdapter": "1b8950207342ff463f0b2e5c8552cace9a80db9896637601b924d50625134c3f",
+    "locator": "8d768151c486d48ef221b9764a34233379999bae3359b667ef543fda734ff798"
+  },
+  "harnessSha256": "c0e3d98e54158546bca7a395984e97161e8ce6bf8191686ae1f6c206fb3270d7",
+  "note": "One explicit fixed positive smoke after a preserved failed diagnostic; no automatic retry, ground truth used only after model result"
+}

package/docs/evidence/luna-decisions-locator-smoke-report.json

@@ -0,0 +1,149 @@
+{
+  "caseId": "photo-dev-02",
+  "result": {
+    "status": "located",
+    "region": {
+      "left": 384,
+      "top": 426,
+      "width": 64,
+      "height": 57
+    },
+    "normalizedRegion": {
+      "left": 0.375,
+      "top": 0.623718887262079,
+      "width": 0.0625,
+      "height": 0.08345534407027819
+    },
+    "selection": {
+      "choice": "A",
+      "nativeChoice": "A",
+      "probabilities": {
+        "A": 0.69,
+        "B": 0.12,
+        "C": 0.05,
+        "D": 0.05,
+        "none": 0.09
+      },
+      "tied": ["A"],
+      "confidence": 0.61
+    },
+    "method": "three-choice-grid-v1",
+    "metadata": {
+      "image": {
+        "width": 1024,
+        "height": 683,
+        "coordinateSpace": "oriented-source-pixels"
+      },
+      "stages": [
+        {
+          "model": "openai/gpt-6-luna-decisions-20261006",
+          "provider": "OpenAI",
+          "requestId": "gen-dec-1792278183-chHSntFM0E1U3fsGWzEi",
+          "durationMs": 1636.31997,
+          "costUsd": 0.0003104,
+          "attempts": 1
+        },
+        {
+          "model": "openai/gpt-6-luna-decisions-20261006",
+          "provider": "OpenAI",
+          "requestId": "gen-dec-1792278185-1n5kxDEToDokqs0PB4Ig",
+          "durationMs": 1316.7197260000003,
+          "costUsd": 0.0013121,
+          "attempts": 1
+        },
+        {
+          "model": "openai/gpt-6-luna-decisions-20261006",
+          "provider": "OpenAI",
+          "requestId": "gen-dec-1792278187-NG6D4CV54lamboCuNEA1",
+          "durationMs": 986.0284590000001,
+          "costUsd": 0.0002781,
+          "attempts": 1
+        }
+      ],
+      "durationMs": 4751.135625,
+      "knownCostUsd": 0.0019006000000000001,
+      "unknownCostCount": 0
+    }
+  },
+  "metrics": {
+    "attempted": 1,
+    "located": 1,
+    "failed": 0,
+    "unknown": 0,
+    "notFound": 0,
+    "successful": 1,
+    "overallSuccessRate": 1,
+    "present": {
+      "attempted": 1,
+      "located": 1,
+      "hits": 1,
+      "misses": 0,
+      "failed": 0,
+      "unknown": 0,
+      "notFound": 0,
+      "hitRate": 1
+    },
+    "negative": {
+      "attempted": 0,
+      "falsePositives": 0,
+      "correctRejections": 0,
+      "failed": 0,
+      "unknown": 0,
+      "falsePositiveRate": null,
+      "correctRejectionRate": null
+    },
+    "locatedAreaFraction": {
+      "count": 1,
+      "min": 0.005215959004392387,
+      "max": 0.005215959004392387,
+      "mean": 0.005215959004392387,
+      "p50": 0.005215959004392387,
+      "p95": 0.005215959004392387
+    },
+    "presentLocatedDistance": {
+      "count": 1,
+      "min": 0,
+      "max": 0,
+      "mean": 0,
+      "p50": 0,
+      "p95": 0
+    },
+    "geometryCoverage": 1,
+    "distanceUnit": "normalized-xy-euclidean-divided-by-sqrt2",
+    "latency": {
+      "count": 1,
+      "minMs": 4753.023501000001,
+      "maxMs": 4753.023501000001,
+      "meanMs": 4753.023501000001,
+      "p50Ms": 4753.023501000001,
+      "p95Ms": 4753.023501000001,
+      "quantileMethod": "nearest-rank"
+    },
+    "cases": [
+      {
+        "presence": "present",
+        "status": "located",
+        "center": {
+          "x": 0.40625,
+          "y": 0.6654465592972181
+        },
+        "areaFraction": 0.005215959004392387,
+        "distanceToNearestBox": 0,
+        "pointInsideAnyBox": true
+      }
+    ]
+  },
+  "physicalCalls": 3,
+  "budget": {
+    "admittedProviderCalls": 332,
+    "maxCalls": 1000,
+    "maxCostUsd": 0.5,
+    "knownCostUsd": 0.03456876000000009,
+    "unknownCostCalls": 4,
+    "inFlightCalls": 0,
+    "reservationPerUnknownUsd": 0.05,
+    "accountedExposureUsd": 0.2345687600000001,
+    "exceededByReportedCost": false
+  },
+  "durationMs": 4753.023501000001
+}

package/docs/evidence/pack-check.json

@@ -0,0 +1,26 @@
+{
+  "status": "PASS",
+  "command": "npm run pack:check",
+  "consumer": {
+    "independentConsumer": "passed",
+    "declarations": "compiled",
+    "sharp": "0.34.5",
+    "vips": "8.17.3",
+    "classification": "yes",
+    "localization": "located",
+    "controlledDecisionCalls": 4,
+    "providerHttpCalls": 0
+  },
+  "testedSnapshot": {
+    "files": 60,
+    "unpackedSize": 648482,
+    "sha256": "a68e64d52b6120635b1f9945586bbe679846f90945b109970639c18180556c8a",
+    "runtimeDistFiles": 34,
+    "runtimeDistSha256": "62d5a0075e5ccca6627ba81f8ca8be427f499ded7508570171c00520bcf369ca",
+    "consumerDeclarationFiles": 184,
+    "npmTransport": "node npm-cli.js; shell:false"
+  },
+  "sourceLogSha256": "2b2654fa14f71ff0861c51282acf047f1b07cfe1eeda9fd44c94884bdd1e5311",
+  "scope": "Independent Linux archive installation; installed source-byte equality and secret scan; fixture classification/localization with native Sharp",
+  "note": "Docs/evidence added afterward change final archive SHA. Final release SHA and source commit are recorded in the external release manifest; tested dist must stay equal."
+}

package/docs/evidence/protocol-probes.json

@@ -0,0 +1,61 @@
+{
+  "purpose": "Two independent diagnostic replays of invalid distributions; not scored into the primary benchmark",
+  "model": "typesafe/jev-1.14",
+  "provider": "typesafe",
+  "maxCalls": 2,
+  "maxReservedUsd": 0.04,
+  "retries": 0,
+  "startedAt": "2026-10-17T21:37:19.055Z",
+  "calls": [
+    {
+      "input": "dev/photo-dev-02/locate/call-2/request.json",
+      "status": 200,
+      "durationMs": 2546.6695290000002,
+      "error": {
+        "code": "INVALID_RESPONSE",
+        "message": "See preserved wire and adapter validation"
+      },
+      "choices": {
+        "selection": {
+          "choice": "C000",
+          "options": 153,
+          "sum": 0.9900000000000004,
+          "max": 0.06999999999999999,
+          "selectedProbability": 0.06999999999999999
+        }
+      },
+      "usage": {
+        "input_tokens": 27042,
+        "output_tokens": 1567,
+        "cost": 0.001135764
+      },
+      "model": "typesafe/jev-1.14-20261014",
+      "provider": "TypeSafe"
+    },
+    {
+      "input": "dev/photo-dev-05/locate/call-3/request.json",
+      "status": 200,
+      "durationMs": 1236.846798,
+      "choices": {
+        "selection": {
+          "choice": "A",
+          "options": 5,
+          "sum": 1,
+          "max": 0.5599999999999999,
+          "selectedProbability": 0.5599999999999999
+        }
+      },
+      "usage": {
+        "input_tokens": 6785,
+        "output_tokens": 69,
+        "cost": 0.00028497
+      },
+      "model": "typesafe/jev-1.14-20261014",
+      "provider": "TypeSafe"
+    }
+  ],
+  "rawManifestSha256": "bb81b6f26a61b2a320968f7aa3a850da6b8f4ffd3950bfea7f40d3a0d351eb8f",
+  "knownCostUsd": 0.001420734,
+  "physicalHttpCalls": 2,
+  "limitations": "These are new diagnostic replays; their outcomes/costs do not replace historical benchmark failures or unknown costs."
+}

package/docs/evidence/rejected-vision-selection-v1-qwen-photo-dev-01.json

@@ -0,0 +1,42 @@
+{
+  "id": "gen-1792276536-jAVcZakU5K4hZH0PEDu1",
+  "object": "chat.completion",
+  "created": 1792276536,
+  "model": "qwen/qwen3.7-flash",
+  "provider": "Alibaba",
+  "system_fingerprint": null,
+  "service_tier": null,
+  "choices": [
+    {
+      "index": 0,
+      "logprobs": null,
+      "finish_reason": "stop",
+      "native_finish_reason": "stop",
+      "message": {
+        "role": "assistant",
+        "content": "{\n  \"answers\": {\n    \"selection\": {\n      \"ranking\": [\n        \"absent\"\n      ]\n    }\n  }\n}",
+        "refusal": null,
+        "reasoning": null
+      }
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 1339,
+    "completion_tokens": 37,
+    "total_tokens": 1376,
+    "cost": 0.00004498,
+    "is_byok": false,
+    "prompt_tokens_details": {
+      "cached_tokens": 0,
+      "cache_write_tokens": 0,
+      "audio_tokens": 0,
+      "video_tokens": 0
+    },
+    "cost_details": {
+      "upstream_inference_cost": 0.00004498,
+      "upstream_inference_prompt_cost": 0.00004017,
+      "upstream_inference_completions_cost": 0.00000481
+    },
+    "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 }
+  }
+}

package/docs/evidence/rejected-vision-selection-v1-qwen-photo-dev-17.json

@@ -0,0 +1,42 @@
+{
+  "id": "gen-1792276636-K22KM54vMUs5iAFdpLQ1",
+  "object": "chat.completion",
+  "created": 1792276636,
+  "model": "qwen/qwen3.7-flash",
+  "provider": "Alibaba",
+  "system_fingerprint": null,
+  "service_tier": null,
+  "choices": [
+    {
+      "index": 0,
+      "logprobs": null,
+      "finish_reason": "stop",
+      "native_finish_reason": "stop",
+      "message": {
+        "role": "assistant",
+        "content": "{\n  \"answers\": {\n    \"selection\": {\n      \"ranking\": [\n        \"absent\"\n      ]\n    }\n  }\n}",
+        "refusal": null,
+        "reasoning": null
+      }
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 1211,
+    "completion_tokens": 37,
+    "total_tokens": 1248,
+    "cost": 0.00004114,
+    "is_byok": false,
+    "prompt_tokens_details": {
+      "cached_tokens": 0,
+      "cache_write_tokens": 0,
+      "audio_tokens": 0,
+      "video_tokens": 0
+    },
+    "cost_details": {
+      "upstream_inference_cost": 0.00004114,
+      "upstream_inference_prompt_cost": 0.00003633,
+      "upstream_inference_completions_cost": 0.00000481
+    },
+    "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 }
+  }
+}

package/docs/evidence/rejected-vision-selection-v1-qwen-photo-heldout-09.json

@@ -0,0 +1,42 @@
+{
+  "id": "gen-1792276690-iPbRn7jjry4MlWuoTX2L",
+  "object": "chat.completion",
+  "created": 1792276690,
+  "model": "qwen/qwen3.7-flash",
+  "provider": "Alibaba",
+  "system_fingerprint": null,
+  "service_tier": null,
+  "choices": [
+    {
+      "index": 0,
+      "logprobs": null,
+      "finish_reason": "stop",
+      "native_finish_reason": "stop",
+      "message": {
+        "role": "assistant",
+        "content": "{\n  \"answers\": {\n    \"selection\": {\n      \"ranking\": [\n        \"absent\"\n      ]\n    }\n  }\n}",
+        "refusal": null,
+        "reasoning": null
+      }
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 1115,
+    "completion_tokens": 37,
+    "total_tokens": 1152,
+    "cost": 0.00003826,
+    "is_byok": false,
+    "prompt_tokens_details": {
+      "cached_tokens": 0,
+      "cache_write_tokens": 0,
+      "audio_tokens": 0,
+      "video_tokens": 0
+    },
+    "cost_details": {
+      "upstream_inference_cost": 0.00003826,
+      "upstream_inference_prompt_cost": 0.00003345,
+      "upstream_inference_completions_cost": 0.00000481
+    },
+    "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 }
+  }
+}

package/docs/evidence/rejected-vision-selection-v2-haiku-photo-dev-16.json

@@ -0,0 +1,42 @@
+{
+  "id": "gen-1792277576-zHlKznkIQTisZQIC56q0",
+  "object": "chat.completion",
+  "created": 1792277576,
+  "model": "anthropic/claude-haiku-5.5",
+  "provider": "Anthropic",
+  "system_fingerprint": null,
+  "service_tier": "default",
+  "choices": [
+    {
+      "index": 0,
+      "logprobs": null,
+      "finish_reason": "stop",
+      "native_finish_reason": "end_turn",
+      "message": {
+        "role": "assistant",
+        "content": "```json\n{\"answers\":{\"selection\":{\"ranking\":[\"present\"]}}}\n```",
+        "refusal": null,
+        "reasoning": null
+      }
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 1772,
+    "completion_tokens": 30,
+    "total_tokens": 1802,
+    "cost": 0.0001922,
+    "is_byok": false,
+    "prompt_tokens_details": {
+      "cached_tokens": 0,
+      "cache_write_tokens": 0,
+      "audio_tokens": 0,
+      "video_tokens": 0
+    },
+    "cost_details": {
+      "upstream_inference_cost": 0.0001922,
+      "upstream_inference_prompt_cost": 0.0001772,
+      "upstream_inference_completions_cost": 0.000015
+    },
+    "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 }
+  }
+}

package/docs/evidence/rejected-vision-selection-v2-haiku-support-heldout-016.json

@@ -0,0 +1,42 @@
+{
+  "id": "gen-1792277601-duFBXdHe6XXSngyZZOxH",
+  "object": "chat.completion",
+  "created": 1792277601,
+  "model": "anthropic/claude-haiku-5.5",
+  "provider": "Anthropic",
+  "system_fingerprint": null,
+  "service_tier": "default",
+  "choices": [
+    {
+      "index": 0,
+      "logprobs": null,
+      "finish_reason": "stop",
+      "native_finish_reason": "end_turn",
+      "message": {
+        "role": "assistant",
+        "content": "```json\n{\"answers\":{\"selection\":{\"ranking\":[\"technical_support\"]}}}\n```",
+        "refusal": null,
+        "reasoning": null
+      }
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 1457,
+    "completion_tokens": 33,
+    "total_tokens": 1490,
+    "cost": 0.0001622,
+    "is_byok": false,
+    "prompt_tokens_details": {
+      "cached_tokens": 0,
+      "cache_write_tokens": 0,
+      "audio_tokens": 0,
+      "video_tokens": 0
+    },
+    "cost_details": {
+      "upstream_inference_cost": 0.0001622,
+      "upstream_inference_prompt_cost": 0.0001457,
+      "upstream_inference_completions_cost": 0.0000165
+    },
+    "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 }
+  }
+}