THROUGH TRAIN Act Three publishes a single figure for its standoff - "the party is
wiped in 24% of runs" - and tells the GM not to soften it. Measured against every party
the duty roster can field, that fight runs from 0% to 100%.
It is not that the published number is wrong. It is that no single number can be right
for this fight, because the answer was settled at character selection: four armed
postings walk it, four trades are massacred, and the GM reading one figure is reading
somebody else's session.
--spread measures the encounter against every C(16,4) party - 1820 of them, exhaustive
rather than sampled, about a minute - and reports the floor, the median and the ceiling
with the parties that produce them. Exhaustive on purpose: a GM planning a session
wants the actual worst case, not an estimate of it.
It also prints each agent's effect on the wipe rate averaged over every party they
appear in, which is the line that gets used at the table. For the Act Three standoff:
okonkwo -43.2 points ashcroft +16.1
sandoval -17.7 nkemdirim +13.5
holloway -15.9 ferriby +12.4
Okonkwo is worth forty-three points of wipe rate on his own. That is a scene-shaping
fact about the encounter that no amount of re-measuring the median would surface.
The published tables are NOT rewritten here. What to do about a 100-point spread is a
design decision - constrain the party, print a range, or rebalance the fight - and it
belongs to whoever wrote the scenario.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The harness measured a game in which every blow went to a general pool. That was
tolerable while every creature was a baseline human and merely wrong once the bestiary
had a Cadence in it: a creature whose entire mechanical identity is being whittled down
body by body was being measured as a person with an odd portrait, and the number it
printed was about nobody.
locationFor and resolveLocationHit move from ringbrp.mjs to rules.mjs, and the
category mapping inside woundPenaltyFor becomes woundPenaltyFrom beside them, for the
reason applyDifficulty and resolveBands moved before them: node cannot load the engine,
and the harness may not own a second copy of a rule. All three arrive with the
spot-checks they never had - 103 rules now, 346 formulas verified.
What the loop does now: 1d20 against the DEFENDER's own species table, melee finding
limbs and shooting finding centre of mass; damage to that location and to the general
pool, so the two agree; disabled at the location maximum and destroyed at twice it,
cumulatively. Then the consequences actually apply - a destroyed head is unconscious
immediately (conditionFor has always accepted that flag and was never passed it, so a
headshot used to leave the target swinging), a ruined arm costs -30 to every attack
including shooting, a ruined leg costs -30 to dodge, and each Cadence body lost costs
the whole creature -10 to everything.
Verified as mechanics rather than as numbers that moved: no species can return a
location it does not have across all 40 rolls; vesh and cadence have no head at any
roll and cadence has no vital either; a vesh ridge is hit 3/20 where a human head is
1/20, which is the "long target rather than a small one" the tables were designed for.
A built Cadence has five bodies at 4 points each and no vital; a built Vesh has six
locations with the ridge the largest.
Measured consequences, seed 11, 400 runs, four duty-roster agents:
keepers x6 31.0% -> 36.5% wiped
the_choir x11 0.3% -> 0.0% wiped, and now degrading as bodies go rather than
being a person who cannot be shot in the head
long_walker 42.0% wiped, which nothing had ever measured
Still no bleeding, panic, Coherence, cover, range bands or fire modes. Still a floor.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
THROUGH TRAIN publishes "MEASURED, 400 runs each" and a 24% wipe rate; the starter
publishes its own table. Nothing in this repository could produce either number, so both
were claims rather than results and neither could be re-checked after a rule moved.
The harness knows no rules of its own. Everything comes out of rules.mjs, because a
harness with its own copy of the combat loop measures a game nobody is playing and
drifts silently the first time a rule changes. Two things had to move for that to be
possible:
applyDifficulty, resolveBands and gradeRoll -> rules.mjs. The whole of d100 resolution
lived in ringbrp.mjs, which node cannot import. All three are pure, so they moved and
the engine re-exports them; no caller changes. check-rules then refused them for having
no spot-check, so they now have fifteen, covering the 1% floor, the 96-99 rule and
00-always-fumbles. They had none in their entire life inside the engine.
expandFromRegister -> tools/expand-spec.mjs. A roster agent is written as a job and a
rank and carries no skills or kit; everything it can do is derived at build time by a
function locked inside build-packs.mjs, which cannot be imported because importing it
runs the build. A simulated agent expanded by a second copy of that logic is not the
agent that gets packed. Moved verbatim; both sides import one copy.
Deterministic throughout: same creature, party and seed, same numbers, which is what
lets a published table be verified rather than remembered.
What it finds: for the Act Three standoff the seed barely matters (0.5-2.8% wipe across
five seeds, eight rounds throughout) and party composition dominates everything. Four
support agents against the six miners wipe 92.8% of the time; swap in the heavy and the
warden and it is 0.3%. The published 24% is not a property of the encounter, and no
single figure can be.
Models Reaction re-rolled each round, banding, graded defences, the stacking defence
penalty, damage by band, major wounds and the dying clock. Does NOT yet model hit
locations, bleeding, panic, cover, range bands or fire modes, all of which make a losing
fight worse. Read the output as a floor, never a ceiling.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
