{
  "campaignId": "fc-3194f803055c",
  "cells": {
    "cli/claude-sonnet|public.swe.martinblech-xmltodict-issue-257": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 9,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 1,
        "withheld": 0
      },
      "evalId": "fr-20260902-6616b188",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "cli/claude-sonnet|vvdex.knowledge.grounded-rag-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 9,
        "withheld": 0
      },
      "evalId": "fr-20260902-2e7f4171",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "cli/claude-sonnet|vvdex.knowledge.memory-fact-update-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 10,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "evalId": "fr-20260901-46eb9548",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "cli/codex|public.swe.martinblech-xmltodict-issue-257": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 10,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "evalId": "fr-20260902-6616b188",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "cli/codex|vvdex.knowledge.grounded-rag-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 10,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "evalId": "fr-20260902-2e7f4171",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "cli/codex|vvdex.knowledge.memory-fact-update-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 10,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "evalId": "fr-20260901-46eb9548",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "codestral-latest|public.swe.martinblech-xmltodict-issue-257": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 10,
        "withheld": 0
      },
      "evalId": "fr-20260902-6616b188",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "codestral-latest|vvdex.knowledge.grounded-rag-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 1,
        "model_fail": 9,
        "withheld": 0
      },
      "evalId": "fr-20260902-2e7f4171",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    },
    "codestral-latest|vvdex.knowledge.memory-fact-update-1": {
      "certified": null,
      "certifiedCounts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 8,
        "withheld": 0
      },
      "evalId": "fr-20260901-46eb9548",
      "grader": null,
      "integrity": "verified_contained",
      "withheldBy": null
    }
  },
  "certificationReceipts": {
    "public.swe.martinblech-xmltodict-issue-257": "87ba9833637f004939e1a7b2d3a69b2d0e0e37b1040ef5d403eb9c392695bee0",
    "vvdex.knowledge.grounded-rag-1": "95deabd2469ddea08d716e0cc5c2c966b874be73743798020aeedaa2fd99d70f",
    "vvdex.knowledge.memory-fact-update-1": "0e21831617f6d3b3680450adc93ec0aa0367a79dfc6ed07538d8a69a05a104ee"
  },
  "counts": {
    "campaignId": "fc-3194f803055c",
    "cells": 9,
    "certifiedPasses": 52,
    "lanes": {
      "cli/claude-sonnet": {
        "certified_pass": 20,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 10,
        "withheld": 0
      },
      "cli/codex": {
        "certified_pass": 30,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "codestral-latest": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 1,
        "model_fail": 27,
        "withheld": 0
      }
    },
    "recordCount": 3,
    "recordDigests": [
      "1127e8356d42d31767a3ddc4d6c6cd0a515f5bbf7239ef0a82129e00fc2e6900",
      "138425da59aabc18aa208c339aa41919f387e1ecb80fb725cfed8f7d6ba78713",
      "157ec324a925c0b10fc8769d01f628b6143c40a00fa2dc70ea309a70e55a37a8"
    ],
    "refused": 0,
    "suspectRecords": [],
    "withheldByHarness": 0
  },
  "disclosure": {
    "public.swe.martinblech-xmltodict-issue-257": {
      "class": "public_source",
      "rule": "public-source disclosure (ownership public_source declared by engine registry, upstream https://github.com/martinblech/xmltodict under MIT; family public_oss_swe): the submitted change is a patch against a public repository, so a redacted excerpt of it is published."
    },
    "vvdex.knowledge.grounded-rag-1": {
      "class": "answer_key",
      "rule": "answer-key disclosure (ownership vvdex_proprietary \u2014 engine registry; family rag_grounding): the exam is VVDex evaluation IP and the graded output is an answer or a keyed value, so no submission content, diff, model claim or hidden-assertion name is published."
    },
    "vvdex.knowledge.memory-fact-update-1": {
      "class": "answer_key",
      "rule": "answer-key disclosure (ownership vvdex_proprietary \u2014 engine registry; family long_term_memory): the exam is VVDex evaluation IP and the graded output is an answer or a keyed value, so no submission content, diff, model claim or hidden-assertion name is published."
    }
  },
  "lanes": [
    {
      "certifiedPasses": 30,
      "countable": 30,
      "counts": {
        "certified_pass": 30,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "cli/codex",
      "rate": 1.0,
      "rateInterval95": [
        0.8865,
        1.0
      ]
    },
    {
      "certifiedPasses": 20,
      "countable": 30,
      "counts": {
        "certified_pass": 20,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 10,
        "withheld": 0
      },
      "lane": "cli/claude-sonnet",
      "rate": 0.6666666666666666,
      "rateInterval95": [
        0.4878,
        0.8077
      ]
    },
    {
      "certifiedPasses": 2,
      "countable": 29,
      "counts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 1,
        "model_fail": 27,
        "withheld": 0
      },
      "lane": "codestral-latest",
      "rate": 0.06896551724137931,
      "rateInterval95": [
        0.0191,
        0.2196
      ]
    }
  ],
  "refused": [],
  "retired_for_future_evaluation": {
    "vvdex.knowledge.grounded-rag-1": {
      "environmentId": "vvdex.knowledge.grounded-rag-1",
      "historicalResults": "Historical sealed results on this revision are unaffected: the disclosure happened after those runs, so it could not have influenced them.",
      "reason": "post-run public disclosure of answer-bearing submission content",
      "retiredOn": "2026-09-02",
      "version": "0.1.0"
    },
    "vvdex.knowledge.memory-fact-update-1": {
      "environmentId": "vvdex.knowledge.memory-fact-update-1",
      "historicalResults": "Historical sealed results on this revision are unaffected: the disclosure happened after those runs, so it could not have influenced them.",
      "reason": "post-run public disclosure of answer-bearing submission content",
      "retiredOn": "2026-09-02",
      "version": "0.1.0"
    }
  },
  "reviews": {
    "public.swe.martinblech-xmltodict-issue-257": {
      "classification": [],
      "law": "a suspect examiner may not blame the model: verdict-bearing suspicion not cleared by in-record evidence withholds every model_fail in the run",
      "suspect": false,
      "withheld": [],
      "withheldCount": 0
    },
    "vvdex.knowledge.grounded-rag-1": {
      "classification": [],
      "law": "a suspect examiner may not blame the model: verdict-bearing suspicion not cleared by in-record evidence withholds every model_fail in the run",
      "suspect": false,
      "withheld": [],
      "withheldCount": 0
    },
    "vvdex.knowledge.memory-fact-update-1": {
      "classification": [],
      "law": "a suspect examiner may not blame the model: verdict-bearing suspicion not cleared by in-record evidence withholds every model_fail in the run",
      "suspect": false,
      "withheld": [],
      "withheldCount": 0
    }
  },
  "schema": "vvdex.forge.campaign-counts/v1"
}