> For the complete documentation index, see [llms.txt](https://docs.thecolliery.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.thecolliery.org/benchmarks/coalmine/readme.md).

# CoalMine—canary detection

Antivirus-style detection-rate measurement for the canaries—the same idea as AV-Comparatives: a corpus of **planted, labeled defects** plus **clean decoys**, scored mechanically.

```
benchmarks/CoalMine/
  fixtures/<suite>/                 ← 7 suites: rot-canary (16 fixtures) +
    f01-… f08/f12-…                    scale-canary · resilience-audit ·
                                       telemetry-canary · testability-canary ·
                                       drift-canary · supply-chain-audit
                                       (8 fault fixtures + 3 clean decoys each),
                                       every fixture with expected.json
                                       (ground truth: file, line, category, severity)
    d01-…-clean-decoy/              ← zero-defect files; any finding = false positive
  results/<date>[-<suite>]-<model>-r<N>.json ← one agent run (rep N): structured findings
                                       + a "suite" field (absent = rot-canary)
  score.mjs                         ← mechanical matcher (suite-aware) → report
  RESULTS.md                        ← K-rep aggregate across suites + engines
  AG-RUN-PROMPT.md                  ← the Antigravity arm's copy-paste protocol
```

## Running an eval

1. Have your agent run the **rot-canary** skill (DEPTH=QUICK) over each fixture directory, recording findings as `{ fixture, file, line, category, severity }`. Categories: `dead-code · bug-prone · resource-leak · concurrency · silent-failure · input-boundary · doc-rot`.
2. Save the run as `results/<YYYY-MM-DD>-<model>.json`: `{ "model": "...", "date": "...", "skillVersion": "...", "findings": [...] }`
3. Score it: `node score.mjs` (newest run by default)—prints the report. Add `--write` to (re)generate `RESULTS.md`: `node score.mjs --write`.

A match = same fixture + file + category, line within ±3. Severity is scored separately. Findings on decoys are false positives.

## Why

"Antivirus-grade quality" needs a number, not an adjective. Results are model-dependent—like AV detection rates are engine-dependent—so every run records the model and skill version, and re-running after a skill or model change shows regressions mechanically.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.thecolliery.org/benchmarks/coalmine/readme.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.