Every push to main builds the static site (npm run build -> dist/) and force-pushes it as the only commit of
the `pages` branch. Ken's web server serves that branch (a git-sync sidecar or a pull on a timer): nothing
on his server needs a secret, since the repository is public, and the site updates within minutes of every
merge, data PRs included.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- bot-cycle.yml: no schedule. It needs a runner on the bot's machine that no one has added, so every 4 hours
a run queued forever. The bot's data already arrives through the daily source check (it reads
DragonKind/DragonKind) and the model goes back through push-policy. Still runnable by hand.
- update-sources.yml: a concurrency group, so two dispatches never build and push at once.
- .gitignore: drop `/site` (a Python template's mkdocs line). site/ is the Atlas page; new files there were
silently ignored.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
analysis/known/build_known.py writes data/public/known_answers.json from pdf_encounters.json and tree.json;
solver.js knownAnswer(known, offered, { target, sceneColour }) matches a screen's choices against it (the guide's
paraphrases by their words, the bot's texts exactly) and returns a pick when one choice is clearly on the guide's
recommended path to the target, plus the evidence for every choice to hand to Clef. Rebuilt daily, checked in CI,
and sent to the bot with the route model.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
The daily check keeps the community route PDF only in its private raw store (never published here). When that
copy differs from "Threshing Options.pdf" in the bot's private repo, `python -m collection send-pdf` opens (or
updates) a pull request there with the new PDF and the lines its text gained and lost, so Ken can update the
hand-written routes in src/pdfRoutes.js (the bot does not read the PDF itself).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- update-sources: download DragonKind/DragonKind (read with ATLAS_BOT_TOKEN, or DRAGONKIND_TOKEN if set) and
import it as the bot source. Needs DK_ACCOUNT_SALT (the same salt as earlier imports, so accounts keep their keys);
without it the bot source is skipped as before.
- The bot load keeps one file, data/bot/trajectories-bot.jsonl: the bot's repo holds every account's full history,
so each import is merged by attempt id (attempts already published stay; older per-import files are folded in).
One copy of each attempt also keeps build-graph.js from counting an attempt twice.
- New workflow push-policy: when policy.json, solver.js or graph.js change on main, open (or update) a pull request
on DragonKind/DragonKind that replaces vendor/threshing-atlas/ with them. Ken merges it.
- forge open-pr: --repo/--base for another repository, and reuse the open pull request for the same branch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
The evening source check stopped (correctly) on two answers the coding rules did not know:
- first sighting "2 Greens": a new form option. Coded as Green, with a data flag that keeps the
original answer (first_sighting_alias in core/spec/form.yaml).
- row label "Unknown" on a bulk-pasted row: added to the entry values in data/dictionary.yaml.
Workflow: when `collection update` reports the build failed, skip the rebuild (it would only fail
again) and always open the "Source check failed" issue, so new answers are reported, not lost.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- python -m collection rebuild: graph, route model, experiment plan, the Black evidence, the screen
tree, statistics, build. refresh, cycle and both update-sources workflows use it, so the morning
update also rebuilds the reports the tests check (it rebuilt only graph, model and stats before)
- graph.json had not been rebuilt since the played-steps fix (no Node on the Mac)
- docs: the branch is pushed; README mentions rebuild
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- game.yaml: `givens` (first or later attempt) and `distinct` levels that are not pooled; the shadow roll
depends on the attempt first. Form submissions show the shadow on 9% of first attempts and 0.3% of later ones.
- Reporting bias: a roll whose bias cannot be measured can borrow another roll's (`borrow: {shadow: meadow}`),
so form submissions inform the shadow again. An assumption, stated in planner/policy.yaml.
- policy.json: targets.<colour>.firstAttempt (this attempt, eventually). Black: about 1.4% on a first attempt,
0.07% on a later one.
- solver.js: advisor options `attempt`, `fix`, sampling "uniform"; assign() and armFor() for the plan.
- planner/experiments.yaml + build_experiments.py -> data/public/experiments.json (progress and status per
experiment, rebuilt from the bot's runs).
- python -m collection cycle: import bot runs, refit, re-plan, deliver the kit; .gitea/workflows/bot-cycle.yml.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014jqCC2LZ4x1WHQ1V57nTsv
- forge.yaml: the repository's host (Gitea at git.kenschiano.net, limited visibility);
the "Report a run" collector, the page's links (data/public/links.json, filled in by
site/build.js) and the dataset metadata read it
- collection/etl/forge.py: issues, pull requests (merge when checks succeed) and labels
for Gitea and GitHub; `python -m collection forge ...`
- .gitea/workflows: CI and the daily update (pushes a branch, opens a pull request that
merges itself, opens an issue when a source breaks); .github/workflows kept for GitHub
- docs: Gitea setup (REPO_SETTINGS), the bot reading atlas files from the repository
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ