✠ The Chronicle

✠ ENTRY LXVII


A new puzzle leaves the workshop and goes live

Two nights of work: the hazard puzzle mechanic that's been living behind a developer-only preview now renders through the same screen a real hazard encounter uses, the puzzle engine dropped a leftover "scrap bin" and started counting wins honestly, and the project's own pull requests now get an automatic review verdict before a human ever looks at them. A handful of small legibility and wording fixes closed out two nights of fresh-eyes testing along the way.

UI

The hazard puzzle preview now plays on the real hazard screen

What changed

The developer-only hazard puzzle demo used to be its own simplified stand-in: its own button layout, its own outcome panel, its own everything. It now renders through the exact same board, route-select screen and outcome panel a real hazard encounter uses on a normal playthrough — the same drag-and-drop card handling, the same dice tray, the same "CONFIRM" resolve flow. It is still only reachable from the developer menu, not from a normal playthrough; the puzzle's rules (fixed layout, no randomness) haven't changed.

Why it changed

the puzzle mechanic was proven out on a throwaway UI first so its rules could be nailed down without the real screen's complexity in the way. Moving it onto the live board now is the planned next step before it can be unlocked for players: whatever ships to a real hazard encounter has already been exercised on this exact screen.

NEW · 19 OCT

The hazard board screen after the change: the developer-only hazard puzzle demo used to be its own simplified stand-in.
hazard-board — the developer puzzle preview, now drawn on the same board a live hazard encounter uses (new screen, no prior baseline to compare against).

commits 91669d2d · aa2e3cd2 · 8d555ac4 · 53c12a16 · ae · B3e2 · PR · 619

MECHANICS

The puzzle drops its scrap bin and starts counting wins honestly

What changed

Three small engine changes to the hazard puzzle: cards can no longer be discarded into a scrap bin mid-puzzle (the bin itself is gone from the puzzle board); a "win" is now counted only from the cards that actually changed the result — a real value, a die, or an effect — instead of crediting every card a player happened to play; and cards can now carry negative costs, hand out a temporary die, or turn one or all of a player's dice to a fixed face.

Why it changed

the puzzle is meant to be a fixed, fully solved layout with no randomness and no do-over, and a scrap bin plus an inflated win count both undercut that guarantee — a card that did nothing could still count toward clearing the puzzle. The new signed costs and the two new card effects (gaining a die, rerolling dice to a set face) extend the vocabulary future puzzle cards can use, ahead of the first real cards being authored against it.

BEFORE · 17 OCT

The devpuzzle board screen before the change: three small engine changes to the hazard puzzle.

AFTER · 19 OCT

The devpuzzle board screen after the change: three small engine changes to the hazard puzzle.
devpuzzle-board — before/after/diff of the puzzle board with its bin removed.

BEFORE · 17 OCT

The devpuzzle outcome screen before the change: three small engine changes to the hazard puzzle.

AFTER · 19 OCT

The devpuzzle outcome screen after the change: three small engine changes to the hazard puzzle.
devpuzzle-outcome — before/after/diff of the puzzle's outcome panel.

commits eb095e2c · 64fb1f00 · ff7e0557 · ae · 1a1c · PRs · 628 · 631 · 633

INFRA

The loop's own pull requests now get an automatic review verdict

What changed

Every pull request the automated build loop opens now gets reviewed by a dedicated check once its other tests pass. A low-risk change gets one reviewer pass; a medium- or high-risk one gets three separate review angles — rules compliance, test coverage, and how far the change's blast radius reaches — merged into a single verdict comment on the PR. A new seeded test plants known bugs on purpose and checks whether the reviewer's verdict actually catches them, feeding a running "miss rate" table the team can watch over time.

Why it changed

the automated test suite alone doesn't catch everything a human reviewer would (sequencing mistakes, rules that look right but aren't); this gives every loop-authored change a second, independent check before a person has to look at it, and measures — rather than assumes — how well that check actually works.

commits 864407e4 · 3bf5830b · 223e977d · bb1eb1dc · ae · 1b1c · PRs · 602 · 610

UI

Six small fixes from the project's own nightly testing pass

What changed

The project runs an unattended testing pass each night that plays through the game's screens looking for rough edges. Over the last two nights it caught, and the loop fixed: the life-total readout announcing itself twice to a screen reader instead of once; a title-screen menu that could spill outside its own frame on a wide desktop window; a duplicated status label in combat reading out twice; a caption under combat's SWAY/CONVICTION icon rendering too small to read and ignoring the system text-size setting; and, across the hub, the hazard screen, a reward panel, a cutscene and the map, on-screen instructions that told every player to "tap" now say "choose" or "continue" instead, regardless of whether they're using a touchscreen, a mouse, or a keyboard.

Why it changed

these are the project's own fresh-eyes testing pass (critique passes 92 and 93) doing its job — catching small legibility, duplicate-announcement and wording inconsistencies before a real player hits them.

commits 9d8fda0a · 529c70a5 · d65ef695 · 7debe9d9 · c6daa798 · 60a241b3 · (PRs · 612 · 614 · 616 · 618 · 604 · 621

INFRA

A standing reminder keeps deadlines from slipping unnoticed

What changed

The project's deadline plan now tracks each upcoming item in one table (what's due, what it needs, which build-plan phases cover it). A new script regenerates a short list daily of only the items that still need the maintainer's attention within the next two weeks — something with no plan yet, or a task sitting on their desk — and prints that list first thing when they start a new chat session.

Why it changed

a deadline with no owner or no plan behind it can slip by unnoticed in a project this size; this surfaces exactly what needs a decision, and nothing else, before any other work starts.

Measured truth

> measured at cf27e84, 2026-10-19, 300 runs, seed 1; compared with 98f18e1 (2026-10-18)

Doors, win rate / lowest-VITAE p10:

fightlevelgrey evengrey body-mindgrey bodyblood-price evenblood-price body-mindblood-price body
bw-17/the-doorwarden (door)L4 v L121% / 0.0027% / 0.0068% / 0.00100% / 0.47100% / 0.51100% / 0.53
cw-17/the-doorwarden (door)L9 v L250% / 0.0083% / 0.0099% / 0.37100% / 0.57100% / 0.61100% / 0.63
bc-15/the-doorwarden (door)L14 v L386% / 0.0095% / 0.23100% / 0.55100% / 0.62100% / 0.67100% / 0.67
ld-16/the-doorwarden (door)L18 v L482% / 0.00100% / 0.29100% / 0.57100% / 0.64100% / 0.70100% / 0.68

No drift past the thresholds.

commits 7e70afdd · 0f767bd3

✠ The marginalia


The night's watch — what ran while nobody looked
window (UTC)what ranoutcome
10-17 09:13 – 10-17 ~17:00dispatcher ticksRV1b (the shadow reviewer) shipped and merged (#602); critique pass 92 (2 LOW: a mobile transport reporting a mouse, "tap" copy on desktop); both findings fixed and merged (#606, #604)
10-17 ~17:00 – 21:40dispatcher ticksRV1c (three review lenses, seeded replay) shipped and merged (#610); /oversight ratified three pending loop-calls and raised the open-PR cap from 3 to 5
10-17 night – 10-18 morningdispatcher ticksthe VITAE double-announce, title-menu overhang and duplicated IN PLAY label fixes merged (#612, #614, #616); critique pass 93 (2 LOW: a cutscene still saying "tap", a too-small CONVICTION caption)
10-18 daytimea dense run of dispatcher ticksB3e2 plotted and shipped — the puzzle preview now plays on the live hazard board (#619); the cutscene wording and CONVICTION caption fixes merged (#621, #618); critique pass 94 (0 findings); HP1 (puzzle cost/utility engine) plotted; HP1a shipped (#628)
10-18 eveningdispatcher ticks + /expandexpand passes 38 and 39 both found no new candidate; HP2 (card audit and difficulty) and the loop-gates (LG1) themes plotted, the latter from an 8-theme owner dump, all eight becoming build-plan rows
10-18 night – 10-19 morningdispatcher ticksHP1b (puzzle board draws no bin) shipped (#631); HP1c (signed costs, gainDie, reroll) shipped (#633); T's 2026-10-19 decision round ruled HP1c2's and HP1d's cost-card rules, set the deadline plan to v0.3, and plotted HP3, CT1 and RS1
10-19 morningdispatcher ticksthe deadline-reminder feature landed; HP3a and HP1c2 both started building (PRs #637, #639, both still open); critique pass 95 (0 findings)
tonightnightly browser breadth leg (exploration/combat round trip, five live combat-round seeds)all pass, no crashes
tonightmeasured-truth table re-readno drift against the last reading (above)
tonightdeploy state checkclean — HEAD's changes are docs/plan-only, no gate pending
tonightpublic-site publish decisionpublished — the puzzle's move onto the live hazard screen is a new way to play reaching a real game screen, and this window's accessibility and legibility fixes are QOL changes a player feels
What is still unfinished
  • A developer-only hazard puzzle preview now plays on the exact same

screen a real hazard encounter uses — still not reachable from a normal playthrough, but one step closer.

  • The puzzle engine dropped its scrap bin, now counts a win only from

cards that actually mattered, and gained two new card effects (handing out a temporary die, and rerolling dice to a set face) for future puzzle cards to use.

  • Six small legibility, duplicate-announcement and wording fixes landed

from the project's own nightly testing: a life-total readout that spoke twice now speaks once, a title-screen menu stays inside its frame on a wide window, a duplicated combat label reads once, a small caption under combat's SWAY icon is readable again, and several screens now say "choose" or "continue" instead of "tap" no matter how you're playing.

  • The project's own automated pull requests now get a review verdict from

a dedicated check before a person looks at them, with three separate review angles on anything riskier than a small change.

  • The maintainer's roadmap now runs to v0.3 — a fully playable Act 1 —

across five two-week windows, prioritizing card upgrades and finishing every event node ahead of adding new cards.

The measure — did the fighting change?
  • None filed this pass — the pulse shows no mistuned gate: the open-PR

queue has room (2 of 5), both nightly /expand passes found nothing new, and the breadth and measured-truth checks are both clean.

  • Highest-scored open proposal still waiting (6.5, several tied): the

keyword/card revamp process pick (needs your attended decision among three plans), combat board legibility at a 375pt width, and making every on-screen number match what the engine will actually resolve.

  • A recurring lesson (the isolated-worktree shell workaround, now seen 11

times) still has an open proposal (score 5.0) to promote it to a standing reflex read at the start of every tick, rather than being rediscovered by hand each time.

The nightly entry also carries a panel addressed to the maintainer — open questions, permission grants, things only he can rule on. It is held back from this page: it is correspondence, not a report. The unfinished work it tracks is published above, in plain words.