Files
threshing-atlas/docs/ROADMAP.md
T
ekcSchianoandClaude Opus 5.5 d2ffb84b5e Solver: best decisions per dragon colour, and the bot advisor
- 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
2026-11-01 22:34:38 +00:00

2.0 KiB

Roadmap

The aim is to move from summary statistics to path finding: for each dragon colour, the decision policy that maximises the chance of bonding it, with honest uncertainty.

  1. Repo and packaging (done): monorepo, CI, GitHub Pages, JS/Python parity tests.
  2. Graph format (packages/core, first version done): typed scene, chance and terminal nodes; history-context keys; per-source counts (bot, form, PDF, Discord); fingerprints and short labels only, no game prose. Importers in packages/ingest for the bot's knowledge.json, runs.jsonl and encounters.jsonl (done), the form (done, via the crosswalk in graph.js), and a PDF scene crosswalk (to do). Daily source checks for the form, the register and the PDF (done).
  3. Context learning (research/): tests of history dependence (G-tests, BH); variable-order Markov / context tree with BIC; temporal-tiered DAG structure learning (BDeu/BIC); latent "character" model vs specific triggers vs memoryless baseline, compared on held-out likelihood; hierarchical Dirichlet back-off pooling.
  4. Solver (first version done: packages/core/src/solver.js, docs/POLICY.md; still to do: richer history models from step 3): backward induction on the augmented state (scene, history summary), i.e. an influence diagram with perfect recall. Two objectives side by side per colour:
    • P(target colour on this attempt);
    • eventual success when deliberate deaths are free: maximise p_X / (p_X + p_other bond), solved by Dinkelbach iteration. Monte Carlo credible intervals, top-k routes, required steps and bottlenecks.
  5. Valley graph UI: every option at every point, with P(survive) and P(bond) per colour, the policy overlay for a chosen colour, and untested branches marked.
  6. Bot loop (advisor and sync published, see docs/BOT_INTEGRATION.md; experiments.json to do): the atlas publishes policy.json and experiments.json (ranked by value of information); the private DragonKind bot consumes them and returns anonymised runs.