@tangle-network/agent-eval 0.203.1 → 0.204.0

package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/failure-receipt.attempt-1.json

@@ -0,0 +1,58 @@
+{
+  "schemaVersion": 1,
+  "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+  "recordedAt": "2026-10-01T16:47:43.515Z",
+  "failure": {
+    "stage": "dispatch",
+    "error": {
+      "name": "Error",
+      "message": "Operational interruption after native artifact capture",
+      "stack": "Error: Operational interruption after native artifact capture\n    at Object.dispatch (file:///home/drew/webb/_wt/lab-autoroute-campaign-20261001/[eval1]:16:22)\n    at async executeCell (file:///home/drew/webb/_wt/lab-evidence-unified/node_modules/.pnpm/@tangle-network+agent-eval@0.203.0/node_modules/@tangle-network/agent-eval/dist/llm-judge-C-0P1WiT.js:459:19)\n    at async file:///home/drew/webb/_wt/lab-evidence-unified/node_modules/.pnpm/@tangle-network+agent-eval@0.203.0/node_modules/@tangle-network/agent-eval/dist/llm-judge-C-0P1WiT.js:888:15\n    at async Promise.allSettled (index 0)\n    at async runCampaign (file:///home/drew/webb/_wt/lab-evidence-unified/node_modules/.pnpm/@tangle-network+agent-eval@0.203.0/node_modules/@tangle-network/agent-eval/dist/llm-judge-C-0P1WiT.js:935:22)\n    at async file:///home/drew/webb/_wt/lab-autoroute-campaign-20261001/[eval1]:19:17"
+    }
+  },
+  "cell": {
+    "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+    "cellId": "capture-retention:0",
+    "scenarioId": "capture-retention",
+    "rep": 0,
+    "artifact": null,
+    "judgeScores": {},
+    "costUsd": 0,
+    "costProvenance": {
+      "kind": "observed",
+      "usd": 0
+    },
+    "costCallIds": [],
+    "tokenUsage": {
+      "input": 0,
+      "output": 0
+    },
+    "durationMs": 1,
+    "seed": 42,
+    "cached": false,
+    "errorStage": "dispatch",
+    "error": "Operational interruption after native artifact capture"
+  },
+  "cost": {
+    "totalCalls": 0,
+    "pendingCalls": 0,
+    "unresolvedCalls": 0,
+    "reservedCostUsd": 0,
+    "inputTokens": 0,
+    "outputTokens": 0,
+    "reasoningTokens": 0,
+    "cachedTokens": 0,
+    "cacheWriteTokens": 0,
+    "totalCostUsd": 0,
+    "costProvenance": {
+      "kind": "observed",
+      "usd": 0
+    },
+    "byChannel": [],
+    "unpricedModels": [],
+    "fullyPriced": true,
+    "usageComplete": true,
+    "accountingComplete": true,
+    "incompleteReasons": []
+  }
+}

package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/spans.jsonl

@@ -0,0 +1 @@
+{"name":"capture-invocation","cellId":"capture-retention:0","startMs":1790873263516,"invocation":2,"durationMs":0}

package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/observation.json

@@ -0,0 +1,121 @@
+{
+  "invocations": 2,
+  "observed": [
+    {
+      "invocation": 1,
+      "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+      "sha256": "b5485b0030654d9ca660401cc07d42e16decdbe5bdc4d8a059eb20045202656b",
+      "value": {
+        "cellId": "capture-retention:0",
+        "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+        "invocation": 1,
+        "content": "Actual filesystem capture from invocation 1"
+      }
+    },
+    {
+      "invocation": 2,
+      "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+      "sha256": "6b0ce0b4eac76fbfdbdf88be21f5d73939d6949b23a7a5579253d74541232ebc",
+      "value": {
+        "cellId": "capture-retention:0",
+        "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+        "invocation": 2,
+        "content": "Actual filesystem capture from invocation 2"
+      }
+    }
+  ],
+  "firstContentStillPresent": false,
+  "finalBytes": {
+    "cellId": "capture-retention:0",
+    "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+    "invocation": 2,
+    "content": "Actual filesystem capture from invocation 2"
+  },
+  "reports": [
+    {
+      "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+      "splitDigest": "sha256:4a729491b198148335bdd5b8f6429d48611d3b44480f1982e77f3a91e4cf2a5f",
+      "seed": 42,
+      "reps": 1,
+      "startedAt": "2026-10-01T16:47:43.512Z",
+      "endedAt": "2026-10-01T16:47:43.721Z",
+      "durationMs": 209,
+      "cells": [
+        {
+          "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+          "cellId": "capture-retention:0",
+          "scenarioId": "capture-retention",
+          "rep": 0,
+          "artifact": {
+            "cellId": "capture-retention:0",
+            "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+            "invocation": 2,
+            "content": "Actual filesystem capture from invocation 2"
+          },
+          "judgeScores": {},
+          "costUsd": 0,
+          "costProvenance": {
+            "kind": "observed",
+            "usd": 0
+          },
+          "costCallIds": [],
+          "tokenUsage": {
+            "input": 0,
+            "output": 0
+          },
+          "durationMs": 0,
+          "seed": 42,
+          "cached": false,
+          "retryAttempts": 1
+        }
+      ],
+      "aggregates": {
+        "byJudge": {},
+        "byScenario": {},
+        "cost": {
+          "totalCalls": 0,
+          "pendingCalls": 0,
+          "unresolvedCalls": 0,
+          "reservedCostUsd": 0,
+          "inputTokens": 0,
+          "outputTokens": 0,
+          "reasoningTokens": 0,
+          "cachedTokens": 0,
+          "cacheWriteTokens": 0,
+          "totalCostUsd": 0,
+          "costProvenance": {
+            "kind": "observed",
+            "usd": 0
+          },
+          "byChannel": [],
+          "unpricedModels": [],
+          "fullyPriced": true,
+          "usageComplete": true,
+          "accountingComplete": true,
+          "incompleteReasons": []
+        },
+        "cellsExecuted": 1,
+        "cellsSkipped": 0,
+        "cellsCached": 0,
+        "cellsFailed": 0,
+        "cellsDispatchFailed": 0,
+        "cellsJudgeFailed": 0,
+        "cellsUnclassifiedFailed": 0
+      },
+      "runDir": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry",
+      "artifactsByPath": {
+        "capture-retention:0/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+        "capture-retention:0/failure-receipt.attempt-1.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/failure-receipt.attempt-1.json"
+      },
+      "scenarios": [
+        {
+          "id": "capture-retention",
+          "kind": "operational",
+          "scenarioDigest": "sha256:955e060f603d0f2e9cd84b8426b05de1faab5b9dbbefdc5d494834afdcb56fc9"
+        }
+      ]
+    }
+  ],
+  "modelInference": false,
+  "researchUnits": 0
+}

package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-summary.json

@@ -0,0 +1,321 @@
+{
+  "retry": {
+    "invocations": 2,
+    "observed": [
+      {
+        "invocation": 1,
+        "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+        "sha256": "b5485b0030654d9ca660401cc07d42e16decdbe5bdc4d8a059eb20045202656b",
+        "value": {
+          "cellId": "capture-retention:0",
+          "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+          "invocation": 1,
+          "content": "Actual filesystem capture from invocation 1"
+        }
+      },
+      {
+        "invocation": 2,
+        "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+        "sha256": "6b0ce0b4eac76fbfdbdf88be21f5d73939d6949b23a7a5579253d74541232ebc",
+        "value": {
+          "cellId": "capture-retention:0",
+          "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+          "invocation": 2,
+          "content": "Actual filesystem capture from invocation 2"
+        }
+      }
+    ],
+    "firstContentStillPresent": false,
+    "finalBytes": {
+      "cellId": "capture-retention:0",
+      "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+      "invocation": 2,
+      "content": "Actual filesystem capture from invocation 2"
+    },
+    "reports": [
+      {
+        "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+        "splitDigest": "sha256:4a729491b198148335bdd5b8f6429d48611d3b44480f1982e77f3a91e4cf2a5f",
+        "seed": 42,
+        "reps": 1,
+        "startedAt": "2026-10-01T16:47:43.512Z",
+        "endedAt": "2026-10-01T16:47:43.721Z",
+        "durationMs": 209,
+        "cells": [
+          {
+            "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+            "cellId": "capture-retention:0",
+            "scenarioId": "capture-retention",
+            "rep": 0,
+            "artifact": {
+              "cellId": "capture-retention:0",
+              "runAttemptId": "64928c35-e5f2-4ae6-a4c1-266088c5ecf9",
+              "invocation": 2,
+              "content": "Actual filesystem capture from invocation 2"
+            },
+            "judgeScores": {},
+            "costUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "costCallIds": [],
+            "tokenUsage": {
+              "input": 0,
+              "output": 0
+            },
+            "durationMs": 0,
+            "seed": 42,
+            "cached": false,
+            "retryAttempts": 1
+          }
+        ],
+        "aggregates": {
+          "byJudge": {},
+          "byScenario": {},
+          "cost": {
+            "totalCalls": 0,
+            "pendingCalls": 0,
+            "unresolvedCalls": 0,
+            "reservedCostUsd": 0,
+            "inputTokens": 0,
+            "outputTokens": 0,
+            "reasoningTokens": 0,
+            "cachedTokens": 0,
+            "cacheWriteTokens": 0,
+            "totalCostUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "byChannel": [],
+            "unpricedModels": [],
+            "fullyPriced": true,
+            "usageComplete": true,
+            "accountingComplete": true,
+            "incompleteReasons": []
+          },
+          "cellsExecuted": 1,
+          "cellsSkipped": 0,
+          "cellsCached": 0,
+          "cellsFailed": 0,
+          "cellsDispatchFailed": 0,
+          "cellsJudgeFailed": 0,
+          "cellsUnclassifiedFailed": 0
+        },
+        "runDir": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry",
+        "artifactsByPath": {
+          "capture-retention:0/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/captured-response.json",
+          "capture-retention:0/failure-receipt.attempt-1.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-retry/capture-retention_0/failure-receipt.attempt-1.json"
+        },
+        "scenarios": [
+          {
+            "id": "capture-retention",
+            "kind": "operational",
+            "scenarioDigest": "sha256:955e060f603d0f2e9cd84b8426b05de1faab5b9dbbefdc5d494834afdcb56fc9"
+          }
+        ]
+      }
+    ],
+    "modelInference": false,
+    "researchUnits": 0
+  },
+  "resume": {
+    "invocations": 2,
+    "observed": [
+      {
+        "invocation": 1,
+        "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0/captured-response.json",
+        "sha256": "36f08e9a29651e4506af2c01fa13df81fff0711f1c53f3f4633f79397b90c569",
+        "value": {
+          "cellId": "capture-retention:0",
+          "runAttemptId": "08005dbd-35ff-488f-8942-22732fc5f036",
+          "invocation": 1,
+          "content": "Actual filesystem capture from invocation 1"
+        }
+      },
+      {
+        "invocation": 2,
+        "path": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0/captured-response.json",
+        "sha256": "bd867bf7291384d653ab799dc2eebf768878ca3a8681fa869684d6da0808cfd7",
+        "value": {
+          "cellId": "capture-retention:0",
+          "runAttemptId": "cc7546fa-7b18-4a50-8444-204439c84bae",
+          "invocation": 2,
+          "content": "Actual filesystem capture from invocation 2"
+        }
+      }
+    ],
+    "firstContentStillPresent": false,
+    "finalBytes": {
+      "cellId": "capture-retention:0",
+      "runAttemptId": "cc7546fa-7b18-4a50-8444-204439c84bae",
+      "invocation": 2,
+      "content": "Actual filesystem capture from invocation 2"
+    },
+    "reports": [
+      {
+        "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+        "splitDigest": "sha256:4a729491b198148335bdd5b8f6429d48611d3b44480f1982e77f3a91e4cf2a5f",
+        "seed": 42,
+        "reps": 1,
+        "startedAt": "2026-10-01T16:47:43.722Z",
+        "endedAt": "2026-10-01T16:47:43.723Z",
+        "durationMs": 1,
+        "cells": [
+          {
+            "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+            "cellId": "capture-retention:0",
+            "scenarioId": "capture-retention",
+            "rep": 0,
+            "artifact": null,
+            "judgeScores": {},
+            "costUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "costCallIds": [],
+            "tokenUsage": {
+              "input": 0,
+              "output": 0
+            },
+            "durationMs": 1,
+            "seed": 42,
+            "cached": false,
+            "errorStage": "dispatch",
+            "error": "Operational interruption after native artifact capture"
+          }
+        ],
+        "aggregates": {
+          "byJudge": {},
+          "byScenario": {},
+          "cost": {
+            "totalCalls": 0,
+            "pendingCalls": 0,
+            "unresolvedCalls": 0,
+            "reservedCostUsd": 0,
+            "inputTokens": 0,
+            "outputTokens": 0,
+            "reasoningTokens": 0,
+            "cachedTokens": 0,
+            "cacheWriteTokens": 0,
+            "totalCostUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "byChannel": [],
+            "unpricedModels": [],
+            "fullyPriced": true,
+            "usageComplete": true,
+            "accountingComplete": true,
+            "incompleteReasons": []
+          },
+          "cellsExecuted": 0,
+          "cellsSkipped": 0,
+          "cellsCached": 0,
+          "cellsFailed": 1,
+          "cellsDispatchFailed": 1,
+          "cellsJudgeFailed": 0,
+          "cellsUnclassifiedFailed": 0
+        },
+        "runDir": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume",
+        "artifactsByPath": {
+          "capture-retention:0/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0/captured-response.json",
+          "capture-retention:0/failure-receipt.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0/failure-receipt.json"
+        },
+        "scenarios": [
+          {
+            "id": "capture-retention",
+            "kind": "operational",
+            "scenarioDigest": "sha256:955e060f603d0f2e9cd84b8426b05de1faab5b9dbbefdc5d494834afdcb56fc9"
+          }
+        ]
+      },
+      {
+        "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+        "splitDigest": "sha256:4a729491b198148335bdd5b8f6429d48611d3b44480f1982e77f3a91e4cf2a5f",
+        "seed": 42,
+        "reps": 1,
+        "startedAt": "2026-10-01T16:47:43.723Z",
+        "endedAt": "2026-10-01T16:47:43.979Z",
+        "durationMs": 256,
+        "cells": [
+          {
+            "manifestHash": "6e492a4dc01bf06d2f7e169cdde455d0b0416e34bd4396c1ec7f72f795d57fa7",
+            "cellId": "capture-retention:0",
+            "scenarioId": "capture-retention",
+            "rep": 0,
+            "artifact": {
+              "cellId": "capture-retention:0",
+              "runAttemptId": "cc7546fa-7b18-4a50-8444-204439c84bae",
+              "invocation": 2,
+              "content": "Actual filesystem capture from invocation 2"
+            },
+            "judgeScores": {},
+            "costUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "costCallIds": [],
+            "tokenUsage": {
+              "input": 0,
+              "output": 0
+            },
+            "durationMs": 256,
+            "seed": 42,
+            "cached": false
+          }
+        ],
+        "aggregates": {
+          "byJudge": {},
+          "byScenario": {},
+          "cost": {
+            "totalCalls": 0,
+            "pendingCalls": 0,
+            "unresolvedCalls": 0,
+            "reservedCostUsd": 0,
+            "inputTokens": 0,
+            "outputTokens": 0,
+            "reasoningTokens": 0,
+            "cachedTokens": 0,
+            "cacheWriteTokens": 0,
+            "totalCostUsd": 0,
+            "costProvenance": {
+              "kind": "observed",
+              "usd": 0
+            },
+            "byChannel": [],
+            "unpricedModels": [],
+            "fullyPriced": true,
+            "usageComplete": true,
+            "accountingComplete": true,
+            "incompleteReasons": []
+          },
+          "cellsExecuted": 1,
+          "cellsSkipped": 0,
+          "cellsCached": 0,
+          "cellsFailed": 0,
+          "cellsDispatchFailed": 0,
+          "cellsJudgeFailed": 0,
+          "cellsUnclassifiedFailed": 0
+        },
+        "runDir": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume",
+        "artifactsByPath": {
+          "capture-retention:0/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0/captured-response.json"
+        },
+        "scenarios": [
+          {
+            "id": "capture-retention",
+            "kind": "operational",
+            "scenarioDigest": "sha256:955e060f603d0f2e9cd84b8426b05de1faab5b9dbbefdc5d494834afdcb56fc9"
+          }
+        ]
+      }
+    ],
+    "modelInference": false,
+    "researchUnits": 0
+  }
+}