planner/policy.yaml: sources (see "Use every possible run…"), and the samples (all, lowBias) once more sources count.
planner/build_policy.py / core/atlas/model.py:
read through the hierarchical back-off, the empirical-Bayes alpha guard, gate_open / can_bond, and the first-attempt givens;
each should have a test with a small hand-made dataset.
planner/experiments.yaml / build_experiments.py: check that the experiment list still matches what the bot can run (src/experiments.js in DragonKind/DragonKind).
The bot's copy of the solver.push-policy now sends policy.json, solver.js and graph.js to DragonKind/DragonKind by pull request (DragonKind PR #1). Ken's vendored solver.js is much older. Add a smoke test to that pull request, or to this repo's CI, that runs the bot's AtlasPolicy.advise() against the new files.
The new PDF (2026-11-16). Its "Black Dragon Gates" (investigate, help, climb) match the model's gate. Its Black encounter is now "turn to face the threat → hold its gaze". Check core/spec/game.yaml against it.
This review is still open.
- [ ] **`planner/policy.yaml`:** sources (see "Use every possible run…"), and the samples (`all`, `lowBias`) once more sources count.
- [ ] **`planner/build_policy.py` / `core/atlas/model.py`:**
- read through the hierarchical back-off, the empirical-Bayes alpha guard, `gate_open` / `can_bond`, and the first-attempt `givens`;
- each should have a test with a small hand-made dataset.
- [ ] **`planner/experiments.yaml` / `build_experiments.py`:** check that the experiment list still matches what the bot can run (`src/experiments.js` in DragonKind/DragonKind).
- [ ] **The bot's copy of the solver.** `push-policy` now sends `policy.json`, `solver.js` and `graph.js` to DragonKind/DragonKind by pull request (DragonKind PR #1). Ken's vendored `solver.js` is much older. Add a smoke test to that pull request, or to this repo's CI, that runs the bot's `AtlasPolicy.advise()` against the new files.
- [ ] **The new PDF (2026-11-16).** Its "Black Dragon Gates" (investigate, help, climb) match the model's gate. Its Black encounter is now "turn to face the threat → hold its gaze". Check `core/spec/game.yaml` against it.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
This review is still open.
planner/policy.yaml: sources (see "Use every possible run…"), and the samples (all,lowBias) once more sources count.planner/build_policy.py/core/atlas/model.py:gate_open/can_bond, and the first-attemptgivens;planner/experiments.yaml/build_experiments.py: check that the experiment list still matches what the bot can run (src/experiments.jsin DragonKind/DragonKind).push-policynow sendspolicy.json,solver.jsandgraph.jsto DragonKind/DragonKind by pull request (DragonKind PR #1). Ken's vendoredsolver.jsis much older. Add a smoke test to that pull request, or to this repo's CI, that runs the bot'sAtlasPolicy.advise()against the new files.core/spec/game.yamlagainst it.