Commit Graph
2 Commits
Author SHA1 Message Date
slaguru666andClaude Opus 5 4e28de2a64 Play PAPER WINDOW with the shipped pregens; fix four silent defects
Ran the starter end to end. It closed, and it found four defects in what I shipped an
hour earlier — all in the scenario data, all silent.

R-247: the Borrowed Authority did not work. I wrote scope "operate", which is not one
of the five scope tiers; an unrecognised scope ranks 0 and every action came back
"impossible" with no error. The first scene of the starter is the Borrowed Authority
tutorial. Fixed, guarded in check-kits, and covered by a test that documents the
failure mode.

R-248: the contact subsystem was unreachable in the scenario built around it. The case
file carried a standing for the Returned Clerks and no GROUP ITEM; resolveContact
returns null without one, so READ, SIGNAL and OFFER did nothing. Groups are data now,
and a test asserts every standing on a shipped case file has a group behind it.

R-249: legs wrote `kindId` where resolveLegDialog reads `leg.kind`. It worked only
because the terrain text happens to begin with "Extraction".

Not a defect: phase four went Difficult because Resources hit 0 and destitution makes
every remaining phase Difficult. Best beat in the session.

All four phases failed and the case still closed — Containment 8 to 2, Resources 8 to
0 — which is what the scenario is built to survive. Two of six closing requirements
succeeded and the settlement still completed. The Cancellation Man opened with a
critical against a raised riot shield and went through 10 points of armour, teaching
the shield and graded-defence lessons in one roll.

Also documented: nobody on the starter team can aim the return, because only the Anchor
Officer trains Transposition. The extraction page now says to narrate it rather than
roll six dice that will all fail.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-11 23:39:43 +01:00
slaguru666andClaude Opus 5 43508259ac Consolidation pass after a Codex review
Four real defects, each verified against the source before acting on it.

Trade expertise was coupled to agency rank. A probationary burglar came out at 35-50%
and a veteran at 65-80%, as though fifteen years of picking locks were something the
department conferred. TRADE_BANDS is now fixed at 50-65/30-50 at every rank, and
INDUCTION scales with service instead (20-32 probationary to 45-60 veteran) because
that half genuinely is the department's. The two axes were the wrong way round. This
is the thing I flagged myself after PAPER HARBOUR and then left alone — flagging a
defect is not the same as fixing it.

Bonus points were sprayed across every skill in the game. grantFullSkillList runs
immediately before the loop and puts all fifty-nine skills on the sheet, and the loop
picked from actor.items, so a veteran's ninety points landed anywhere. The comment two
hundred lines above has always said "over the core skills"; the code never did.

"Random species" passed "baseline" into generators that support random perfectly well.

No behavioural tests. The strongest point in the review: the three guards are static
and none of them can tell whether a rule is READ, which is the only kind of defect this
project has ever shipped. tools/check-behaviour.mjs now runs 28 deterministic tests,
gated into the build, covering R-231/232/243/245/246, graded defences and the lamp
rule. Proved it bites by re-breaking R-246.

The register moved to postings.mjs, a Foundry-neutral module the engine, rulebook,
guards and tests all import normally — they were previously pulled out of the engine
with a regex and eval, which worked and was a trap. My first version of that used
`export … from`, which re-exports without local bindings, so ROLES was undefined and
the system threw on init: node --check passes that, loading it in Foundry does not.

Also: the contact doc comment still said "only OFFER moves standing"; check-kits
accepted mas/app, which are the DISPLAY names of siz/cha, so a posting declaring one
would have passed and produced an undefined characteristic; package.json disagreed with
the manifest; README counts were three revisions stale and are now generated at build.

Deferred with reasons: splitting the 6,700-line engine into modules. Right, and a
multi-session refactor whose risk is exactly the silent breakage demonstrated above.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-11 22:57:20 +01:00