CI / test (pull_request) Successful in 40s
- The odds card for the Black: "Low evidence", "about 1 in N" first attempts, the 95% range written as "between 1 in X and 1 in Y", later attempts effectively never, why it is uncertain (every logged shadow is a rider who chose to post), and the steps (shadow -> reached the Black -> bonded). Dashed border. - Valley map: the Black's routes are solved for a first attempt (they showed ~0% per attempt), labelled "on a first attempt", and "Why this route?" gives the same range and caveat. - "The Black is not dice" panel: why the odds are uncertain, and what replaces the assumption (fresh-account bot runs to the tree). - analysis/black/evidence.py wrote "odds_ratio": Infinity (no later-attempt shadow at all), which is not JSON: the page's two Black panels showed "could not be loaded". Now null, and the report says "unbounded". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
analysis
Making sense of the data.
| Folder | What it does |
|---|---|
stats/measures.yaml |
Declarative. What the site measures: outcomes and their denominators, route targets and comparison groups, the bonding-encounter contexts, the route map's stages, the filter presets and the route thresholds. |
stats/site_stats.py |
Computes every number the Atlas page shows for every preset and writes data/public/stats.json; the page computes nothing itself. |
stats/stats_core.py |
The statistics themselves: rates with Wilson intervals, odds ratios with Fisher's test and Benjamini–Hochberg, Mantel–Haenszel, ridge logistic models that adjust for data source and chase goal. Tested against known answers (tests/core/test_stats.py). |
network/ |
build-graph.js combines every source into data/public/graph.json (the decision-variable graph and the bot's scene graph). CI fails if the committed graph is out of date. |
reports/ |
Generated reports. POLICY.md is written by planner/policy_report.py; do not edit it by hand. |