cmd/

decidelab asks whether inference changes what the bot plays, and whether the changes win games.

exfit tabulates, for every logged exchange (or, with -kind placement, tile placement) that was inferred, the static leave value of each leave the exchanger could have kept against its hypergeometric prior.

fetch_lexica downloads the lexicon data files the test suite needs into a data directory, using the same code path (and the same source) as the shell's `set lexicon`.

inferdiag — Tier 1 + Tier 2 Bayesian inference diagnostic harness.

inferlab replays a finished autoplay run's inference positions and reports, for each one, what the engine believed and what was actually true.

leaveval prints the engine's static leave value for each leave read from stdin, one per line -- a way to ask, over a corpus, whether the leaves the imputation misjudges have anything in common the leave table already knows.

make_all_wmps builds a .wmp word map for every lexicon a game can be created with on woogles, ready to be uploaded to the data volume.

make_wmp builds a .wmp file from a .kwg file.

mlreads finds the positions in an autoplay match where the ML bot chose a different move from the static-equity (HastyBot) move for the same position, and records both with their equities and the game's outcome, so the interesting disagreements can be adjudicated by a sim and shown.

playability estimates how much knowing each word is worth.

shadow_test is a command-line tool that runs shadow agreement tests.

simlabel labels positions with a Monte Carlo sim's verdict on every candidate play, to measure how often a bot's choice matches the sim's (check).

taufit — maximum-likelihood estimation of the rangefinder softmax temperature (tau) from real BestBot-vs-human games on Woogles.

wmp_test is a command-line tool that runs WMP agreement tests.