Step 4a of the Python-first core.
- analysis/stats/measures.yaml (declarative): outcomes and denominators, route targets and
comparison groups, encounter contexts, the route map's stages, six filter presets and the
route thresholds (5, 10, 20, 30).
- analysis/stats/site_stats.py computes every number the page shows, for every preset, and
writes data/public/stats.json: factor tables, models, chance events, encounters, the route map
for every outcome, route rankings with q-values at every threshold, the valley map's place and
step counts, and The Odds' summary cards.
- The page now only displays: app.js, ui.js and explorer.js read stats.json; engine.js is no
longer loaded. The three filter controls become one "Runs" menu; minimum runs become fixed
thresholds. Removed: "Load a newer route export", click-to-select and "Follow" on the Deep
Dive route map, and the unused old valley renderer. The coded runs are a CSV download.
- tests/parity/compare_stats.py: every number matches what engine.js computed, for all six
presets (factor tables, models, encounters, route rankings at every threshold, route map).
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
1. Letting the Red pass. Only climbing down quickly puts you above the Red, so a tree run that
passed on the Red and then met the Greens let it pass. Those 16 runs were coded as slow
descents; they are now "Descend quickly: let the Red pass" and count for the Greens'
encounter, so the route model's "let it pass" option rests on its own runs. A tree run with
both the Red and the Greens attempted (impossible in one run) is now flagged.
2. Leaving your squadmates. The form only offered "> Left them" from 2026-10-27 (0 of 504 squad
runs before chose it, 40 of 144 after). Earlier and undated squad runs are now unknown rather
than "stayed together" (form.yaml: left_squadmates_asked_from).
3. Bot text. The wounded cadet's other two choices map onto the form; a test checks every choice
the bot has logged maps onto a form variable (encounter answers excepted, not modelled yet).
Each rule has a test in tests/core/test_coding.py. Data rebuilt (runs, graph, policy, report);
the headline odds move by at most 0.02 points. The Methods tab describes both rules.
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Step 2 of the Python-first, declarative core.
- core/spec/game.yaml: the game as data: every decision and its options, every roll and the
earlier steps its odds may depend on, every bonding encounter, and the flow of a run from the
entrance to the dragons (set / keep / drop / go / end / encounter). Also how the form's coded
answers map onto the game.
- core/atlas/model.py: what each coded run reveals about each step, count tables, empirical-Bayes
shrinkage, and the posterior draws (same random numbers as before, reproducible from the seed).
- core/atlas/solver.py: one interpreter of game.yaml used three ways (solve, evaluate a fixed
policy, export the published tree) instead of the tree written out five times. Vectorised
over the 200 posterior draws: the build takes about 2 minutes instead of 4, with no memory flag.
- planner/build_policy.py and policy_report.py replace the JS build and report.
Checked against the JavaScript model: every number in policy.json agrees to within 4e-12
(last-digit differences in log/cos). Three "still best" shares differ by one draw in 200, at
near-ties under the "eventually" objective. POLICY.md is otherwise identical.
tests/parity/compare_solver.py keeps checking JS against Python in CI until the site's JS advisor
replaces solver.js; tests/core/test_game.py checks game.yaml is complete and consistent.
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Moves only; no behaviour changes. Every file now lives under the part of the project it
belongs to, each section has a README, and docs/PRIVACY.md collects the rules all parts follow.
core/ engine.js, graph.js, solver.js, python/clean.py
collection/ sources/ (form, register, PDF check, daily update), bot/ (importer)
analysis/ stats/ (Python statistics), network/ (build-graph.js), reports/ (POLICY.md)
planner/ build-policy.js, policy-report.js
site/ atlas/ (the page), build.js, qa/
tests/ core/ (unit), parity/ (JS vs Python), fixtures/
Checked against the previous layout: tests pass; coded runs, test outputs, graph.json and
policy.json are byte-identical; the built site differs only in comments that name file paths.
The published file names Ken's bot downloads (solver.js, graph.js, data/policy.json) are unchanged.
Also removes the undefined/ folder of QA screenshots committed by mistake (the QA scripts now
default to a temp folder) and test.txt, and drops the unused npm workspaces.
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Solver: adds leaving your squadmates, your reaction to the first dragon (options offered per
dragon, from the form's categories) and, below the tree, jumping onto the Red or letting it pass
to the two Greens. Bot crosswalk and option patterns map the game's text for each.
Map: a choice you make now holds for every route to that dragon, together with the earlier steps
of the path you were looking at, so a later choice never sends the route back through another
place. Routes your choices rule out are hidden; choosing an option closes its card.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
After helping your squadmate the game asks your first move (pause to assess, charge the
smaller cadet, or get between your squadmate and the bigger one) and, after pausing, what to
do about the smaller cadet's dagger. Both are now decisions in the solver, and the chance of
dying in the fight depends on them (about 3 in 10 riders who charged in died; almost none who
paused). With that split, helping now beats walking away.
On the map, several choices made at one place share one diamond, which opens a card listing
each choice in the game's order. Policy rebuilt; the policy build gets a larger heap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
The solver treated the cadet encounter as one seven-way choice made at the
first dragon, so the map put a diamond there with no matching choices. It is
now a sequence, as in the game: at the scream, investigate or keep walking;
in the clearing, help your squadmate or walk away; then the bullies' fate
(after surviving the fight) or what to do for the wounded cadet. Each
choice point sits at its own place on the map and opens only its own
options; an option that stays in the same place (helping, in the clearing)
appears as a clickable label there. Bot histories that only know the final
resolution still resolve correctly; the bot advisor maps each step onto the
screen text. Policy rebuilt; tests added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- engine.js / analyze.py: one MISSING set ("Unknown", "Unsure",
"Unsure/unknown", "Multiple", "(not recorded)", "(not asked)", ...).
Factor tables compare only runs that answered each question and report how
many were left out; models are complete-case for every choice variable;
route rankings group only fully known routes; the route flow skips
missing steps.
- "I've lost count" now counts as a later attempt.
- The chase goal keeps its "not asked" stratum as an adjustment only; it is
no longer displayed as a finding.
- Page: no unknown levels in charts, tables or filters; factor explorer notes
how many runs were left out; Methods explains the rule.
JS/Python parity still exact.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- packages/core/src/solver.js: decision tree with perfect recall over the
community form's choices; chance events (first dragon, deaths, shadow,
meadow dragon, encounter outcome) depend on earlier choices through
hierarchical Dirichlet back-off with empirical-Bayes concentration per
event; backward induction for P(colour this attempt) and Dinkelbach for
P(colour) / (P(colour) + P(other colour)); 200 posterior draws for
credible intervals and "still best" rates; cross-fitted held-out check.
- Only genuine attempts inform encounter outcomes (goal / Malek intent).
- data/public/policy.json (all runs and a low-bias sample) and the readable
docs/POLICY.md; published on Pages with solver.js for the bot.
- Bot advisor: re-solves from any history, maps advice onto screen text,
reports margins so ties can be randomised. docs/BOT_INTEGRATION.md.
- Tests on a synthetic world with known answers; daily source update now
rebuilds the policy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
The Threshing Routes form was redesigned (5,651 responses). Both implementations
(engine.js and clean.py) now read either version:
- one direction question + one colour question, with the older per-direction
columns used only to fill gaps and cross-check
- cadet encounter as an action path (scream, first and second combat move)
plus a separate outcome
- new fields: sighting_action, combat_second, rare_reported, malek_intent,
left_squadmates, tail; "Why are you doing this run?" mapped onto the goal
("Blue or Black" for rare chasers)
- the rare-dragon question feeds saw_blue_meadow / saw_shadow and the branch
The old export still codes identically. JS/Python parity: all match.
Data: form 2,573 -> 5,651 rows, Dragon Register 216 -> 273, new route PDF
fingerprint. Supersedes the automated pull request #2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
- apps/atlas: the web page (valley map, route flow, Dragon Register)
- packages/core: coding rules and statistics (engine.js) and the shared route
graph (graph.js): scene graph, decision-variable graph, bot-to-form crosswalk,
Beta credible intervals
- packages/ingest: DragonKind bot importer to anonymised trajectories (no prose,
no account identifiers); data/bot holds Ken's sample
- research/python: independent reference implementation + update_sources.py
- tests: JS/Python parity checks and graph/importer unit tests
- CI, GitHub Pages deploy, daily check of the community sheets and route PDF
- Email addresses in free-text form answers are redacted; the raw spreadsheet
is not committed
- MIT code licence, CC BY 4.0 data licence
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ