deepgrain-foundry/docs/agent-log.md

# Agent log

What agents have done, newest first. Add an entry for every piece of work: the date, which
note or request it was for, what changed, the files touched, the tests run, and anything left
open or not verified. Keep each entry short; the git history has the details.

## 2026-10-27: t66 dev console (Claude Code + Codex)

- Backtick opens a dev console in any build. The first time in a world it warns that the save
  will be marked; accepting saves `dev: true` and a `devLog` (last 200 commands) in the world,
  and the worlds list shows a "Dev console" badge. Commands: help, money, debt, give (display or
  internal name, past the bag limit), bit, bag, unlock, tp (cursor or x y), free on|off, speed,
  pause, step, weather clear|rain. Command logic is sim-side (`devCommand` in
  `src/sim/devconsole.js`); the panel is `src/ui/devconsole.js`.
- Not built: `time` (no clean way to set the time of day) and undo (commands are logged instead).
  Free building also skips `payCost`, so ghosts built later are free too; removing a piece built
  for free still refunds its normal cost.
- Also fixed a flaky build-card check: the wall-material chip markup (`data-wall-picker=""`)
  made the selection card rebuild its button every refresh.
- Files: `src/sim/devconsole.js`, `src/ui/devconsole.js`, `src/styles/devconsole.css`,
  `src/shell.html`, `src/manifest.json`, `src/sim/items.js`, `src/sim/save.js`, `src/sim/world.js`,
  `src/ui/input.js`, `src/ui/loop.js`, `src/ui/saves.js`, `src/ui/session.js`,
  `src/ui/selection.js`, `src/ui/notebook.js`, `src/ui/build-menu.js`, tests
  `t_dev_console.js`, `t_dev_console.py`, `test/run_all.py`, `docs/PROJECT_MAP.md`.
- Tests: `python3 test/run_all.py --browser` all passed. Built in a parallel worktree; Codex
  hit its usage limit near the end and Claude finished it (the payCost fix).

## 2026-10-27: t67 belt drop drawn as two capped ends (Claude Code)

- A flat belt feeding an uphill belt's low end on the same row (and two uphill slopes side by
  side) stays joined, but grains really drop about 3 grains onto the face there, so it now
  draws as two capped ends instead of the bridge phase 2b drew (it overlapped and stepped
  badly). `beltDrop` in `src/render/slope-belts.js` decides it for `slopeBeltConnections` and
  `flatBeltCaps`; the shader and `world.js` packing went back to their pre-2b versions.
- Tests: `t_joins.js` and `t_smooth.py` drop checks updated; the existing sand test confirms
  the same-row feed carries sand up. `python3 test/run_all.py --browser` all passed.

## 2026-10-27: t67 phase 2b, joins follow the travel direction (Claude Code + Codex)

- The owner tested phase 2: horizontal → uphill and vertical-up → diagonal didn't join like the
  rest. New rule (DESIGN.md): a piece's outlet points at its block plus its travel vector, and
  the next piece joins when it sits there travelling the same general way. Belts bend it where
  grains can't follow: an uphill belt's top feeds a flat belt on the same row, and a flat belt
  one block below an uphill low end still joins.
- `src/sim/joins.js` reworked around inlets and outlets; screw drawing, belt drawing, screw
  grain flow and bars follow it. Straight-screw mouths no longer reach a diagonal across a
  corner; straight-to-straight transfers (lift side-feeding) unchanged.
- Files: `src/sim/joins.js`, `src/sim/conveyors.js`, `src/sim/items.js`,
  `src/render/diagonal-screws.js`, `src/render/slope-belts.js`, `src/render/slope-belt-shader.js`,
  `src/render/world.js`, `src/ui/previews.js`, `tools/scenes/joins.js`, tests `t_joins.js`,
  `t_diag_screws.js`, `t_slope_belt_art.js`, `t_smooth.py`, `docs/PROJECT_MAP.md`.
- Tests: `python3 test/run_all.py --browser` all passed.
- Open: Codex hit its usage limit before finishing; the belt drop drawing was finished in the
  next entry. Not playtested.

## 2026-10-27: t67 phase 2, screw conveyors follow the join rule (Claude Code + Codex)

- Screw drawing, grain flow and bars now all go through `endJoin`. New joins: a horizontal in
  E at a diagonal's low end, and a vertical in E at either end; both draw an elbow that bends
  inside the straight piece's end block. Same-row high-end elbows are unchanged.
- No longer joined (moves nothing): a same-row horizontal at a low end, a vertical directly
  above or below a diagonal's last block, and any other side contact. Joined belts and filters
  take discharge on their carrying face at either end. A horizontal screw feeding the side of a
  vertical lift still works (`straightScrewNext` keeps straight-to-straight transfers).
- The opposite-direction check is one helper, `screwAccepts`.
- Files: `src/render/diagonal-screws.js`, `src/render/smooth-shaders.js`, `src/render/world.js`,
  `src/ui/previews.js`, `src/sim/conveyors.js`, `src/sim/machine-tick.js`, `src/sim/items.js`,
  `src/sim/cells.js`, `tools/scenes/joins.js` (replaces `t62-joins.js`), `test/t_diag_screws.js`,
  `test/t_smooth.py`, `docs/PROJECT_MAP.md`.
- Tests: `python3 test/run_all.py --browser` all passed; determinism diff changed only screw
  checks. Screenshots: `shots/joins-*.png`.
- Open: inside the new elbows the flight stripes of both pieces overlap, which looks a little
  busy in the smooth look. Layouts built on the removed joins need moving.

## 2026-10-27: t63 vacuum picks up grains in mid-air (Claude Code + Codex)

- `toolVacuum` now takes flyers (grains a screw launches or that were blown out) inside its
  brush and reach before loose cells, with the same pick filter, flask rule, bag-full stop and
  48-grain budget. `vacuumTargetAt` also considers them. Rain is skipped: it belongs to the
  clouds, and taking it would make water.
- Files: `src/sim/vacuum.js`, `test/t_basics.js`. Built on a separate worktree branch in
  parallel with t67 and fast-forwarded.
- Tests: `python3 test/run_all.py --browser` all passed; new checks only in the determinism diff.

## 2026-10-27: t67 phase 1, shared join module (Claude Code + Codex)

- Owner decision (DESIGN.md, replaces the t62 same-row rule): one join rule for every transport
  piece. New DOM-free `src/sim/joins.js`: `pieceEnds`, `endJoin`, `pieceJoins`; rule tables are
  compiled once and joins must be mutual. Belt drawing now uses it: `slopeBeltConnections`,
  `flatBeltCaps` (now also for ghosts), and the pixel filter frame (`beltLike(b, end)`). Sloped
  belt drag previews now pass a planned map.
- Deliberate drawing changes where the old belt rules disagreed: flat belt caps no longer
  treat any same-row diagonal screw as joined, and filter frames open only at a real join.
- Files: `src/sim/joins.js`, `src/manifest.json`, `src/render/slope-belts.js`,
  `src/render/setup.js`, `src/render/world.js`, `src/ui/previews.js`, `test/t_joins.js`,
  `test/t_slope_belt_art.js`, `docs/PROJECT_MAP.md`, `docs/DESIGN.md`.
- Tests: `python3 test/run_all.py --browser` all passed; t_joins checks 4,232 placements,
  symmetry, planned maps, and that belt sand flow agrees with the rule. Sim output unchanged.
- Open: screws still use their old rules until phase 2.

## 2026-10-27: t62 corner placements stay separate (Claude Code + Codex)

- Owner tried the corner join and found the jog odd; decision (recorded in DESIGN.md): only
  same-row joins connect. Removed join kinds 5/6 and `straightScrewNext`; `diagonalScrewNext`
  and `entDiagonalScrewTick` now continue across a corner only into a diagonal running the same
  way, so a corner straight piece no longer receives grains or bars. Kept the shared geometry
  file and the smooth-look elbow seam fix.
- Files: `src/render/diagonal-screws.js`, `src/render/smooth-shaders.js`, `src/sim/conveyors.js`,
  `src/sim/items.js`, `src/sim/machine-tick.js`, `test/t_diag_screws.js`, `test/t_smooth.py`,
  `tools/scenes/t62-joins.js`, `docs/PROJECT_MAP.md`, `docs/DESIGN.md`.
- Tests: `python3 test/run_all.py --browser` all passed; determinism diff changed only the
  replaced corner checks.

## 2026-10-27: t62 screw conveyor joins (Claude Code + Codex)

- Straight screws touching a diagonal's end only at the corner now join, in both looks and in
  ghosts/previews, and grains and bars cross the joint. Same-row joins were already elbows.
  Also removed the seam across the top of connected elbows in the smooth look.
- Files: new `src/render/diagonal-screws.js` (join and casing geometry moved out of
  `src/render/world.js`), `src/render/smooth-shaders.js`, `src/ui/previews.js`,
  `src/sim/conveyors.js` (`straightScrewNext`), `src/sim/machine-tick.js`, `src/sim/items.js`,
  `src/manifest.json`, `docs/PROJECT_MAP.md`, `tools/scenes/t62-joins.js`, tests
  `test/t_diag_screws.js`, `test/t_smooth.py`.
- Tests: `python3 test/run_all.py --browser` all passed; determinism diff showed only timing
  lines and the new checks.
- Open: the corner join doesn't check that the straight piece ends there, so a straight run
  passing over the diagonal's corner also draws a jog. Not playtested by hand.

## 2026-10-27: t66 dev console in every build (Claude Code)

- Owner decision: the console ships in production too, with a warning when it's turned on that
  the save gets marked (and future achievements won't unlock). Recorded in `docs/DESIGN.md`;
  t66 gained subtasks for the warning, the save mark, and game speed and fun controls. No code
  changed.

## 2026-10-27: t66 dev console added to Todo (Claude Code)

- Owner request: a dev console to add money, items and upgrades quickly for testing. Subtasks
  cover how it opens (and hiding it from real players), money and debt, bag items, bit and bag
  tiers, other shortcuts, and marking changed worlds. No code changed.

## 2026-10-27: Save compatibility rule dropped from AGENTS.md (Claude Code)

- Owner: old saves don't matter; only they play, for testing, and can start a new game.
  `AGENTS.md` still said to keep saves loading (never renumber ids, add migrations), which
  contradicted the DESIGN decision "No backward compatibility for now". It now matches. The t65
  subtask no longer asks for saves to load. `AGENTS.md` is local (gitignored). No code changed.

## 2026-10-27: t65 Rust conversion added to Todo (Claude Code)

- Owner request: a task to convert the game to Rust for performance, after the base game is more
  playable (sooner if performance becomes a problem). Subtasks: settle the scope (whole game or the
  hot sim code; WebAssembly or native), profile for a baseline, keep tests and saves working. No
  code changed.

## 2026-10-27: Shift-drag draws a line for transport pieces (Claude Code)

- Owner request: Shift shouldn't fill a rectangle for belts, conveyors and similar. New
  `lineOnly(code)` in `src/sim/pieces.js` (belts, filters, screw conveyors, shafts, pipe pieces);
  `primaryRelease` (`src/ui/actions.js`) and the drag preview (`src/ui/previews.js`) ignore Shift
  for them, and `pieceCode` (`src/ui/selection.js`) no longer flattens a diagonal belt drag on
  Shift. Build-menu "How to place" and the notebook say so; the notebook's stale "cleats up or
  down" line now says uphill only.
- Tests: play.py checks a Shift-drag with a belt lays a 3-piece sloped line. Full
  `run_all --browser` passes.

## 2026-10-27: t61 sticky materials slide freely on built surfaces (Claude Code)

- The owner's video showed clay lumps creeping down a sloped wall; removing a piece was
  unrelated. Cause: STICK gates every diagonal step, so cohesive materials crawl on 45°. In
  `powder()` (`src/sim/cells.js`) a grain resting on a built solid (not F_NAT) now always tries
  the diagonal; the random draw is unchanged, so generated worlds are identical.
- Measured: a clay grain down a 20-piece built slope took 77 ticks, same as sand (before: never
  arrived in 2000 ticks). Heaps on a built floor and on natural ground are unchanged.
- Test: `t_slopes.js` clay vs sand transit. `python3 test/run_all.py --browser`: ALL PASSED.

## 2026-10-27: Unremovable one-cell belt left after cutting slopes (Claude Code)

- The owner's exported save had a lone belt cell no block owned, so nothing could build there or
  remove it. It was a sloped belt's seal: a seal may sit in a cell its neighbour leaves open (a
  slope's open corner, or under a grate, screw or screen), but removal only let it go when the
  neighbour was a sloped wall. Now `blockOwnsCell` decides (`src/sim/blocks.js`), and loading a
  world clears belt and screen cells no block or seal owns (`clearStraySeals`, `save.js`).
- Found with a fuzz of random slope/belt/filter/wall/grate builds removed in random orders and by
  cut (0 strays in 4000 runs after the fix). The owner's save loads clean and the spot is
  buildable. Tests in `t_slope_belts.js` fail on the old code. `run_all --browser`: ALL PASSED.

## 2026-10-27: t42 feedback: downhill belts drop the sidewall too; new tasks (Claude Code)

- Downhill sloped belts now draw as plain belts: no far wall and no ribs, placed or as ghosts or
  drag previews (`slopeWallAt`, ghost drawing in `render/world.js`, `previewBlock` now gets the
  belt direction from `previews.js` and `preview-selection.js`). `slopeUphill` takes the
  direction bits instead of a block.
- Test: `t_slope_belt_art.js` checks a downhill run draws no wall or ribs. Drag preview checked
  headlessly with a counting canvas (downhill draws the band only); placed belts and ghosts
  checked in screenshots. `python3 test/run_all.py --browser`: ALL PASSED. Rebuilt dist.
- Filed t59 (world longevity audit), t60 (renewable resources, discussion), t61 (sloped walls
  sticky after removing a piece mid-flow: not reproduced headless, see its report), t62 (screw
  conveyor horizontal/diagonal joins), t63 (vacuum falling grains).

## 2026-10-26: t42 feedback: uphill stall at a seam, no cleats downhill, slope pick (Claude Code)

- The owner's long uphill run stalled with a few grains, with a flashing seam between pieces 9
  and 10. Cause: each block kept its own cleat phase, so a piece added to a running belt was out
  of step; at the seam a rib started one or two cells behind another and pushed its pocket into
  that rib's back, stalling the run for good. Every block of a run now takes the phase of the
  run's lower-left block (`slopeCleatsTick`, memoised `slopeRunKey`).
- Owner decision: downhill belts have no cleats. They carry like a flat belt (`carryColumn`) and
  draw only the sidewall; ribs, friction drag and pushes are uphill only (`slopeUphill`).
  `pushRow`'s downhill levelling is gone. Downhill now 90 grains/s (was 38).
- Middle click on a sloped belt's band, where it is drawn into the empty block below, now picks
  the belt (`pickUnderCursor` falls back to `slopeBeltAt`).
- Tests: `t_slope_belts.js` adds pieces added 3/7/11/13 ticks later (13 stalled before) and the
  band pick (fails without the fix); downhill floor back to 60; cleat checks uphill only.
  `t_slope_belt_art.js` draws ribs on uphill runs. `python3 test/run_all.py --browser`: ALL
  PASSED. Screenshots of an up-and-down run in both looks opened. Rebuilt dist.

## 2026-10-26: t42 feedback: cleats that really hold and push their load (Claude Code)

- The owner saw ribs pass through heavy loads and the load climb as before. Rewrote the slope
  physics in `src/sim/slope-conveyors.js`: no column carry on slopes; each step every rib moves
  one face cell and pushes the rows of grains in its way (`pushRow`, front first, never between
  two rib cells or into a machine); a contact-layer friction drag that gravity beats on 45°;
  a boot-section scoop at the foot of uphill runs; a run stalls (all its moves undone) if a rib
  can't push; grains under a newly placed rib are lifted out. Ribs stop short of built cells.
  Slopes now step before flat belts (`machine-tick.js`), so a feeding belt finds the room.
- Drawing: ribs drawn by the smooth wall shader at the simulation phase (rubber bars with a
  light cap); pixel ribs lighter with a cap; far wall a little lighter (`render/world.js`,
  `slope-belt-shader.js`, `slope-belts.js`).
- Rates: uphill 78.7 (suite) to 86 grains/s (heavy probe), downhill 38, flat 90. Perf: 576
  loaded slopes 1.03 ms/tick; 50 slopes in a generated world 2.19 vs 1.83 ms/tick without.
- Tests: `t_slope_belts.js` adds "no grain ever inside a rib under heavy feed", "a grain on the
  bare face slides back to a rib, then rides up with it"; downhill rate floor lowered to 30.
  `python3 test/run_all.py --browser`: ALL PASSED. Open: downhill rate; t58 (low-end joins).

## 2026-10-26: t42/t43 drawing and performance follow-up (Codex)

- Filled the far sidewall down to the belt, behind grains, with two-tone pixel corrugation
  and a continuous smooth 45-degree band. Rubber ribs use the physics mask in all eight
  phases. Fixed smooth wall seams; matched ghosts, cursor previews and menu icons.
  Smooth screens retain tinted mesh, now between solid rails without the pixel overlay;
  loaded-only shake and transport are unchanged.
- Empty cleats only wake nearby load; per-chunk flags bypass mask reads for distant powder.
  Empty eight-block slope: 3 -> 0 moving areas after 120 ticks. Uphill sand/gravel/copper ore
  71.3 grains/s, coal 71.2; exact sorting retained. Isolated 1f8a89d -> current, median five
  generated falling-world samples: no slopes 2.006 -> 2.063 ms/tick; 50 slopes 1.952 -> 2.076.
  Powder probe with cached zero flag: 45.02 -> 50.56 ms (2.14M calls); residual 12 percent
  microbenchmark overhead, about 3 percent in the generated no-slope world. Existing loaded
  576-block stress: 0.515 -> 1.064 ms/tick. slope-cost.js records the reproducible methodology;
  raw results /tmp/t42-perf-{pre,after}-isolated.txt. Pre-t42 loaded slopes use old grip physics.
- Files: sim cells/slope-conveyors/world; render slope-belts/slope-belt-shader/world;
  UI previews/build-menu; t_slope_belt_art.js, t_smooth.py, slope-cost.js; PROJECT_MAP,
  DESIGN and t42/t43 reports. Reused enterFree for the ordinary entry rules.
- Final python3 test/run_all.py --browser: ALL PASSED (26 headless suites, 360 play,
  16 smooth checks; art suite 57). Before/after suite outputs differ only by five new
  checks (/tmp/t42-output.diff). Opened shots/t{42,43}-follow-{belt,screen}-{pixel,smooth}-z{4,8}.png;
  rechecked smooth belt after fixing seams; opened t42-follow-ghosts-{pixel,smooth}-z8.png
  with both updated menu icons visible. Added sleep, border/load wake, wall geometry,
  grain colour and smooth grain-class checks. Rebuilt dist, regenerated map; both tasks Review.
  Owner playtest and hardware GPU appearance remain unverified. No commit.

## 2026-10-26: t42: sidewall belts with moving cleats (Codex)

- Continued the uncommitted prototype. `sim/slope-conveyors.js` supplies moving barrier
  lines (8 face cells apart, 6 grains tall), finite periodic pocket capacity and saved phase.
  Removed slope grip; natural powder movement runs between steps and surplus rolls back.
  Fixed the shared slope-to-screw mouth boundary so grains cannot fall under its casing.
  Pixel/smooth drawing uses the actual mask, continuous dark far walls, matching previews/icons.
- Files: manifest; sim cells/conveyors/slope-conveyors/machine-tick/items/pieces/blocks/world/save;
  render slope-belts/slope-belt-shader/world; UI actions/selection/selection-pickers/previews/
  build-menu/notebook; slope test suites and play.py; slope-flow/slope-cost experiments,
  slope-conveyors scene, PROJECT_MAP, DESIGN and t42/t43. Removed both requested scratch files.
- Reproducible `node tools/experiments/slope-flow.js`: sand, gravel, copper ore 71.3 and
  coal 71.2 grains/s uphill, each 90.0 flat; 20-second window after 220 warm-up ticks.
  Measured settled pile angles 33.7 degrees for all four. Final HEAD -> current stress measurement: 0.486 -> 1.038 ms/tick for 576 loaded
  slope blocks (1024x512); no-slope powder microbenchmark, 2.14M calls: 45.07 -> 65.55 ms.
  Median of five samples after warm-up; scripts use DG_HARNESS for the HEAD snapshot in
  /tmp/t42-baseline. Barriers skip mask reads when none exist, but still have a CPU cost.
- Opened shots/t42-belt-{pixel,smooth}-z{4,8}.png. Headless checks pass, including pocket
  overflow, collision masks, unstamped bare-face sliding, joins, screw mouths, water, bars,
  save phase, all eight drawing phases and unchanged seed hashes 777/77/4242.
  Full python3 test/run_all.py --browser: ALL PASSED (26 headless suites, 360 play
  checks, 15 smooth checks). Rebuilt dist and regenerated PROJECT_MAP. Both tasks moved to
  Review. The owner still decides whether the lower rate feels
  right. Pocket movement is a discrete envelope, not a calculation of individual forces.
  No commit.

## 2026-10-26: t43: inclined vibrating screens (Codex)

- Diagonal filters always run downhill through all placement/settings/undo/paste paths;
  R leaves a picked slope unchanged. The picker shows fixed downhill travel rather than two
  misleading direction buttons. Flat filters retain their original carry. Slope grains get
  gravity plus one mesh-contact nudge every 4 ticks (15 steps/s); unselected bars use the
  same cadence. No whole-column carry or between-step grip on slopes.
- Pixel and smooth mesh frames shake a little while loaded, with stable idle drawing.
  Updated ghosts, icons, build text, notebook and the relevant sim/UI/render files listed
  in t42 above. Added real browser drag/R/picker checks and art shake checks.
- Both shapes and both sorting modes deliver the complete 64 sand + 64 gravel mix exactly.
  Generous-feed discharge is 60.0 grains/s down-left and 59.3 down-right; rest passes through.
  Opened shots/t43-screen-{pixel,smooth}-z{4,8}.png. Headless and smooth suites pass; the
  first browser run found a hover fixture above the mesh and a later timeout. Corrected the
  coordinate and reset picked slope state between fixtures. All new browser checks pass.
  A subsequent run caught an existing 200 ms silo HUD timing race; wait for the expected
  HUD text before the same assertion. Final full browser run: ALL PASSED, 360 play
  checks plus 15 smooth checks; all 26 headless suites pass. t43 moved to Review.
  No uphill-save migration. No commit; owner playtest remains.

## 2026-10-26: t44: rename screw tubes to screw conveyors (Codex)

- Player text uses Screw conveyor; code uses screw/SCREW, including the piece id, part
  constants, direction tables, transport helpers, shader variable, build icons and settings
  access. Numeric ids and save/blueprint versions unchanged; no aliases or migrations added.
- Updated affected files across src/sim, src/render, src/ui, test and tools, plus README,
  AGENTS and current DESIGN/TESTING/system docs. Regenerated PROJECT_MAP and built dist.
  Renamed test/t_diag_tubes.js to test/t_diag_screws.js with git mv; no scene filename needed
  renaming. Past log entries, docs/history and task title/Context preserved.
- All 26 suite outputs saved in /tmp/t44-before and /tmp/t44-after. Raw diff:
  /tmp/t44-output.diff; name/timing-normalized diff: /tmp/t44-output-normalized.diff (empty).
  No numeric output differences except ms timings; numeric literals unchanged in all 39
  edited JavaScript/Python files. Only tube matches in scoped source/docs are the physical
  casing description in pieces.js and casing/flight rendering comment in world.js.
- python3 test/run_all.py --browser: ALL PASSED (26 headless suites, 352 play checks,
  15 smooth checks). Focused menu, tooltip, hover, notebook, screw blueprint-code round-trip
  and both starter blueprint-code checks pass in pixel and smooth; no page errors.
  Opened shots/t44-{menu,hover,notebook}-{pixel,smooth}.png and the play diagonal screenshot.
  Temporary UI fixture initially lacked SX/SY; corrected to match shot.py before these checks.
- Task t44 moved to Review. Owner playtest remains. No commit.

## 2026-10-26: t38: show silo contents at the cursor and in a wide window (Codex)

- `src/ui/info.js`: contents were confined to the bottom info plate (hidden below 720 px),
  omitted from cursor labels, and cone gaps lacked mach ownership. Hover now resolves the
  full 12x24 outline and shows contents beside the cursor; physical footprints stay unchanged.
- `src/render/machines.js`: shared framed three-grain window in both looks, fixed pixel grain
  noise, visible near-empty row, lifted dark fuel colours, and stencils fitted beside the glass.
  `tools/scenes/silo.js` adds snow and one grain of coal. All 31 accepted powders checked.
- `test/t_silo_contents.js`: 5 checks confirm snow intake, identity, water accounting, warm
  storage and release. `test/play.py` checks hover/label at five outline points empty/filled,
  narrow HUD, fill pixels at zooms 2-12 and stable drawing for all powders in both looks.
- Opened and inspected `shots/t38-silos-{pixel,smooth}-z{2,4,8}.png`. Updated t38 and DESIGN;
  regenerated PROJECT_MAP. Final `python3 test/run_all.py --browser`: ALL PASSED (26 headless
  suites, 352 play checks, 15 smooth checks). Task moved to Review; owner playtest remains.

## 2026-10-26: t41: distant resonance-drill aim is normal (Codex)

- `src/ui/preview-drill.js`, `previews.js`: keep tuned fan and light crosshair, add a solid far
  edge in pulse colour, remove the drill-only red reach circle. `info.js` omits drill distance
  warnings; other tools retain their warning, including beside silo contents.
- `src/sim/drill.js`: removed only the distance status; too-hard/cannot-be-drilled wording and
  `toolDig` are unchanged. Notebook drill text remains accurate; gamepad code untouched.
- `test/t_drill.js`: new distance regression failed before fix, now passes; two former warning
  expectations updated intentionally. `test/play.py`: held mouse aim makes no toast/warning,
  hard granite still toasts, no red reach circle, gamepad aim stays within reach.
- Opened and inspected `shots/t41-drill-beyond-reach.png`. Focused browser checks pass.
  Initial full browser run exposed test timing/paused-input fixture mistakes, corrected without
  changing gameplay. Final `python3 test/run_all.py --browser`: ALL PASSED (26 headless
  suites, 352 play checks, 15 smooth checks). Updated t41 and DESIGN; moved to Review. No commit.

## 2026-10-26: t39: seal touching square-block corners (Codex)

- `cells.js`: six inline checks reject diagonal moves between two PH_SOLID neighbours, with
  no new random calls. `worldgen.js` saves/disables/restores CORNER_SEAL alongside LSTEP/LFILM
  while settling. No material ids, save fields or piece geometry changed.
- Reproduced two further leaks: free flyers (`vacuum.js`) and diagonal tube mouths pulling sand
  between external walls (`conveyors.js`); blocked those steps only. Horizontal liquid scans
  already check each intervening cell; vacuum collection removes grains into the bag. Bars
  resolve ordinary movement by axis. Screw lane transport remains explicit and unchanged.
- `test/t_corner_seal.js`: 23 checks, including both staircase orientations for sand/water/steam,
  all four diagonal intakes, a flyer, world hashes/flag restoration, and an open pile that remains
  57x17 with identical material layout and random state when sealing is disabled.
- All 25 headless suites pass. Before/after outputs: `t_slopes` square-funnel outflow 379 -> 339
  (more sand retained on steps); `t_drill` ore outside the wall 56 -> 55 (solid corners hold ore);
  `t_world` awake chunks after 900 gameplay ticks following reload 148 -> 149 (different settling
  after corner blocking). Disabling the checks restores drill/world baseline output exactly.
  Other existing output is identical except timings. Evidence: /tmp/t39-before, /tmp/t39-after.
- Full mat/temp/flags hashes match original code for seeds 777/77/4242 at 512x384 and 2048x1024,
  plus 424242 at 1024x640 (/tmp/t39-world-hashes-{before,after}.txt).
- Factory benchmark at 1024x640: idle 0.69 -> 0.68 ms/tick; 1/4/8 modules
  2.05/2.36/2.65 -> 2.13/2.44/2.70 ms/tick. Timing is indicative: settling changes, and this
  fixture reported zero machine-moved grains, so it does not prove loaded transport performance.
- Rebuilt dist and regenerated PROJECT_MAP. Inspected `shots/t39-corners.png` in pixel look:
  sand rests along the upper steps and pours through the central outlet.
- Final `python3 test/run_all.py --browser`: ALL PASSED (25 headless suites, 323 play checks,
  15 smooth checks). Updated t39 subtasks/report for Review and recorded the design decision.
  No commit made; owner playtest remains.

## 2026-10-25: t40 art pass: sloped belts and filters drawn as one belt (Claude Code, Codex)

- Sloped runs draw as the flat belt (rollers, studs) or filter (tinted mesh) band turned to 45°,
  continuous, with elbows into flat belts, filters and tubes and rounded free ends, in both looks
  (`src/render/slope-belts.js`, new `slope-belt-shader.js`, `world.js`, `smooth-shaders.js`);
  ghosts and icons match. Drawing only.
- `test/t_slope_belt_art.js`; `run_all.py --browser` ALL PASSED.

## 2026-10-25: t40: sloped belts and filters (Claude Code, Codex)

- Block codes 19-22 (`BP_BELT_UP/DOWN`, `BP_FILTER_UP/DOWN`, `src/sim/pieces.js`): slope-shaped
  belts and filters whose 45° face runs, laid by a diagonal Belt or Filter drag. `conveyorStep`,
  `holdSlopeColumn` and a shared front-first order for belts and filters in `src/sim/conveyors.js`;
  bars in `items.js`; saves, blueprints, undo, hover, previews; `src/render/slope-belts.js`.
- 90 grains/s flat, up and down for belts and filters; a sloped filter sorted 64/64 with no loss.
  Seeds 777, 77 and 4242 generate identically.
- `test/t_slope_belts.js` (32 checks); `run_all.py --browser` ALL PASSED. Art pass follows.

## 2026-10-25: t35 follow-up: diagonal tube elbows (Claude Code, Codex)

- Diagonal tubes are as thick as straight ones, and an end block joining a straight tube, belt or
  filter draws an elbow at its centre, so the pipes meet centred and edge to edge; free ends get a
  collar; ghosts match (`src/render/world.js`, `smooth-shaders.js`, `src/ui/previews.js`,
  `preview-selection.js`). Scene `tools/scenes/t35-joins.js`.
- `test/t_diag_tubes.js`; `run_all.py --browser` ALL PASSED.

## 2026-10-25: t35/t36 follow-ups: wall drags make slopes, diagonal tube art (Claude Code, Codex)

- Wall piece: a diagonal drag lays slopes (`pieceCode`/`turnSlope` in `src/ui/selection.js`), T flips
  a drag or cycles square and four corners for one block; the Sloped wall piece is gone from the
  menu, slope codes map to the wall piece (`src/sim/pieces.js`); old slope saves load
  (`test/fixtures/slopes-v2.json.gz`).
- Diagonal tubes draw as one continuous 45° pipe in both looks, overhanging empty cells at joints
  (`src/render/world.js`, `smooth-shaders.js`). Drag ghosts lost the long side arrow; each ghost
  block keeps its own chevron (`src/ui/previews.js`). The owner found the diagonal still meets
  straight tubes off-centre; an elbow fix follows.
- `test/t_slopes.js`, `t_diag_tubes.js`, `play.py`; `run_all.py --browser` ALL PASSED.

## 2026-10-25: t37 follow-up: grains on built things cast no shadows (Claude Code, Codex)

- The owner saw a flickering shadow under a grinder fed by a stream: grains resting on its
  housing for a moment shaded. `looseForLight` (`src/render/light.js`) now treats a loose stack
  that rests on any `builtForLight` cell, or ends in air or gas, as see-through. Stacks on natural
  ground or deeper than 12 cells still shade.
- `test/t_light.js` (a grinder stream checked for 180 ticks in both lighting modes);
  `run_all.py --browser` ALL PASSED.

## 2026-10-25: t37 follow-up: falling grains cast no shadows (Claude Code, Codex)

- The owner saw a single falling grain darken a column of sky. `onBeltForLight` became
  `looseForLight` (`src/render/light.js`): a loose stack that ends in air or gas within 12 cells
  (falling) is see-through, as are belt and filter loads; stacks resting on any solid or a tube
  part still shade.
- `test/t_light.js`; `run_all.py --browser` ALL PASSED.

## 2026-10-25: t37 follow-up: belt loads cast no shadows (Claude Code, Codex)

- The owner saw grains riding a belt overhead still shading the ground. `onBeltForLight(i)` in
  `src/render/light.js` walks down through up to 12 loose cells to a belt or filter; such cells
  are skipped by the light roof and pass light like air. `builtForLight` is unchanged, so the t33
  loose-grain shading still applies to belt loads. Heaps on rock, silos and wall floors still shade.
- `test/t_light.js` (56 checks); `run_all.py --browser` ALL PASSED.

## 2026-10-25: t35: diagonal screw tubes (Claude Code, Codex)

- Parts `P_TUR`..`P_TUL` (11-14), `TUBE_PARTS`, and an `IS_TUBE` table replacing every
  `P_TU..P_TR` range check (`src/sim/material-tables.js` and its users); tube variants 4-7;
  diagonal drags for tubes; grains in `src/sim/conveyors.js`/`cells.js`, bars in `items.js`;
  both looks, previews, hover, copy/paste, undo, blueprints, saves.
- `test/t_diag_tubes.js` (85 checks): 120 grains/s straight, up and down over ten blocks; joins,
  no spills, water passes, bars both ways, controls, persistence. Existing suites unchanged and
  passing; seeds unchanged. `run_all.py --browser` ALL PASSED. Screenshots
  `shots/t35-diagonal-*.png`, scene `tools/scenes/t35-diagonal.js`.

## 2026-10-25: t36: sloped walls and diagonal wall lines (Claude Code, Codex)

- `BP_SLOPE_BL`..`BP_SLOPE_TR` (15-18) in `src/sim/pieces.js` (`isSlope`, `isWall`, `slopeCell`,
  `slopeCode`); `src/sim/blocks.js`: half cost and refund, `slopeSeal` corner cell and
  `slopeSealOwner`, `diagonalDrag` and `blockLine(.., diagonal)`. Build menu, T key
  (`turnSlope`), previews, copy/paste, undo, blueprints, hover, both looks.
- `test/t_slopes.js`: funnel 400/400 sand (square steps 379/400), water 400/400, no leaks in all
  four orientations from both sides, bars slide, every wall material; play.py menu and diagonal
  drag. Seeds unchanged. `run_all.py --browser` ALL PASSED. Screenshots
  `shots/t36-funnel-*.png`, scene `tools/scenes/t36-funnel.js`.

## 2026-10-25: t33: loose grains and solid ground look different (Claude Code, Codex)

- Smooth look (`src/render/smooth-shaders.js`): natural ground gets a soft seam where it meets
  loose grains as well as air; packed grains keep the pebble shading and are slightly lighter.
  Pixel look and thermal lens (`drawWorld()` in `src/render/world.js`): loose grains a touch
  lighter and speckled, solid cells touching loose grains darker.
- Smooth frame 1.08 -> 1.10 ms. `t_smooth.py` checks the seam and the lighter grains;
  `tools/scenes/t33-ground.js` for screenshots (`shots/t33/`).
- play.py: 'F on an automatic ingot mold stays quiet' counted toasts on screen, and failed
  now and then when an older toast expired or a new one arrived in the same 150 ms; it now
  records the toasts the key press makes (none). `run_all.py --browser` ALL PASSED apart from
  that check, which passes in the new form.

## 2026-10-25: t37: buildings cast no shadows (Claude Code, Codex)

- `computeLight()` (`src/render/light.js`): a per-view-column light roof that skips player-built
  cells (`builtForLight`: unnatural solids, block, machine or part cells), and built cells pass
  light like air. `drawWorld()` stops marking built cells as shading terrain for the smooth look.
  The simulation's `roofY` is untouched (built roofs still condense steam).
- `test/t_light.js`; screenshots `shots/t37-before-*.png`, `shots/t37-after-*.png`.
  `run_all.py --browser` ALL PASSED.

## 2026-10-25: t34: the nozzle blows no further than it sucks (Claude Code, Codex)

- `toolBlow()` (`src/sim/vacuum.js`) aims at the capped point from `nozzle()` when the cursor is
  out of reach (thrown bars already did); `drawPreviews()` draws the blow brush there, in the blow
  colour rather than the out-of-reach red.
- `test/t_nozzle.js`: aiming at 3x and 5x the reach lands sand within reach + 16 grains (brush and
  settling, measured 15); inside reach it still lands around the cursor. play.py: the preview
  sits at the capped point. `run_all.py --browser` ALL PASSED. Screenshot
  `shots/t34-blow-out-of-reach.png`.

## 2026-10-25: t21: a loaded world starts around the player first (Claude Code, Codex)

- Owner: "lets speed up the first load speed. load in areas around the player first".
- `src/sim/save.js`: `deserializeWorld` no longer wakes every area; `startLoadWake()` keeps the
  load-time heat scan's result per area, wakes those within 150 grains of the bot, and queues the
  rest by distance; `loadWakeTick()` (from `simTick`) wakes 12 a tick and, every 8 ticks, any still
  waiting within 150 grains of the bot. The queue is not saved.
- Huge world, ms/tick before -> after: ticks 0-600 average 15.2 -> 5.5, max 79.7 -> 16.2, 95th
  percentile 21.0 -> 11.0; ticks 600-1800 average 7.3 -> 10.0 (the same work, later). Every area
  has started by tick 1024 (17 s).
- `test/t_load.js` (19 checks). `run_all.py --browser` ALL PASSED.

## 2026-10-24: t21 phase 4: duplicates merged, largest files split (Claude Code, Codex)

- Merged seven real duplicates: wall refunds (`removeWallCell`, walls and molds), tube edges and
  lanes (`entTubeEdge`, `entTubeLane`), pipe text, piece dimensions, font lookup, reach circles,
  surface averaging (`meanSurface`). Look-alikes left alone: data tables, sprite rows, GLSL, and
  the hot movement and heat code.
- Split by topic: `ui/previews.js` -> `preview-drill.js`, `preview-selection.js`;
  `render/smooth.js` -> `smooth-shaders.js`; `ui/notebook.js` -> `notebook-market.js`,
  `notebook-blueprints.js`; `sim/materials.js` -> `material-tables.js`; `sim/worldgen.js` ->
  `worldgen-helpers.js`; `ui/selection.js` -> `selection-pickers.js`, `selection-panels.js`.
  `render/world.js` left whole (one large function).
- Every suite's output identical apart from timings; seeds generate the same worlds.
  `run_all.py --browser` ALL PASSED.

## 2026-10-24: t21 phase 3: speed-ups that change nothing (Claude Code, Codex)

- Profiled a Huge world (6144x2048) headless and in the browser. After a load every area is
  awake: 8-15 ms/tick for the first 600 ticks, settling to about 0.7 ms/tick after 3 minutes. In
  the browser, drawing is cheap; the full roof scan every 20 ticks was the largest single item.
- Kept: `updateDirtyRoof()` (`src/sim/ledger.js`) rescans only area columns marked by
  `wakeXY`/`wakeHeatXY`/`simMove`/`heatChunk` (`roofDirty`); `updateRoof()` is still the full scan
  for setup code. Quiet Huge world 4.60 -> 1.88 ms per scan, digging 4.67 -> 3.48.
  `heatChunk` skips `react()` for solids and powders below their first reaction threshold
  (`REACT_MIN`). Huge, ms/tick: digging 2.54 -> 2.34, quiet 1.39 -> 1.18.
- Reverted (no consistent gain): inlining `condPair`, one shared part rule for `enter`/`passable`.
- `test/t_sim_optimizations.js`: the dirty roof matches a full scan; `heatChunk` matches the
  reference loop in `test/fixtures/heat-chunk.js` byte for byte, random state included (catches a
  reaction added to `react()` but not to `REACT_MIN`).
- Every suite's output identical before and after apart from timings; seeds 777 and 4242
  generate the same worlds. `run_all.py --browser` ALL PASSED.

## 2026-10-24: t21: piles near the bot sort quickly (Claude Code, Codex)

- Owner: "a pile should sort when the player is nearby. the sorting shouldnt take long anyway."
- `src/sim/cells.js`: `SORTP` 0.003 -> 0.06, `SORTP_FUEL` 0.003 for sinking through burning fuel
  (at 0.06 dirt sank through a buried burning wood pile and `t_world`'s charcoal check failed);
  `wakeBotPowder()` every 8 ticks from `simTick` wakes areas with loose grains (`powderArea`
  hint, kept by `simMove`) within 100 grains of the bot. `worldSteps` settles at 0.003: seeds
  unchanged.
- Cost on a Medium dig site: 0.108 -> 0.134 ms/tick. `t_physics` sorting checks: near the bot ore
  sorts in 432 ticks then sleeps; far away it stays; sand on dirt never sorts.
- Tests: `run_all.py --browser` ALL PASSED.

## 2026-10-24: t21 phase 2: physics test pass for sand, molten metal and gases (Claude Code, Codex)

- `test/t_physics.js` (41 checks): heaps, digging under a heap, ore sorting through ash, sand
  sinking in a pond, world edges; molten copper levelling (slower than water), setting, sinking in
  a pond, slag over copper, filling a cavity; smoke rising and fading (longer under a roof), steam
  condensing, fire going out, gas in sealed pockets, gas leaving at the top row. Every scenario
  checks that its areas go back to sleep.
- Found and fixed: gas stuck in a pocket never finished fading or condensing because its area
  slept (`gas()` keeps its area awake: steady cost 1.34 -> 1.37 ms/tick); a rain drop or blown
  grain landing on steam deleted that steam's water (`landFlyer`/`depositNear` in
  `src/sim/vacuum.js` now send it to the clouds).
- Found, owner decided: a resting pile never sorts. To follow: piles near the bot keep sorting,
  and sorting gets quick.
- Tests: `run_all.py --browser` ALL PASSED; seeds 777 and 77 generate the same worlds.

## 2026-10-24: t21 phase 1: water test pass, faster pouring, rain spreads (Claude Code, Codex)

- `test/t_water.js` (20 checks): conservation, levelling and sleep, pools pouring off ledges,
  world edges and corners, pumps, freezing and thawing, boiling and condensing, a pool draining
  into a hot shaft.
- `liquid()`/`move()` in `src/sim/cells.js`: multi-cell steps (`LSTEP`, scaled by `FLUID`),
  a longer reach for top-layer grains (`LFILM`), and waking the row ahead only when a top-layer
  grain steps sideways (waking on every liquid move kept 207 of 2048 areas busy after a reload).
  `worldSteps` settles with both at 1: seeds 777, 77 and 4242 hash the same as before.
- Found: cooled steam drifting under a roof could stay steam for ever, because its heat area
  slept. `gas()` now keeps the heat pass on it.
- `weatherTick()`: raining clouds drift at half pace; rain falls over the band and one either
  side. `t_weather`: 'close to where it gathered' now samples drops in the air (landed water
  spreads); a new wind check.
- Test fixes, not weakening: `t_conveyors` pit check counts steam on its way up (the water total
  was exact); `t_molds` burning-bed check seeds `RS` (its result hung on earlier tests' draws:
  0.28-1.04 depending on the sequence, identical on the old and new code).
- Steady cost after a reload (2048x1024): 1.26 -> 1.24 ms/tick. Tests: `run_all.py --browser`
  ALL PASSED. Screenshots `shots/t21-pour-120.png`, `shots/t21-pour-900.png`.

## 2026-10-24: t1: bars stack neatly, Rack blocks store them (Claude Code, Codex)

- `232f5af`: `entTick` snap-and-tip (`entSupportCount`, `src/sim/items.js`);
  `test/t_bar_stacking.js` (12 bars fill a pit's bottom six rows; fails on the old code).
- Rack blocks: `src/sim/racks.js` (`rackState` per block, FIFO runs, `rackIntake` from `entTick`,
  `rackTick` release, `rackRefill`), `BP_RACK`/`RACKB`, F toggle in `interactAt`, removal returns
  bars with `giveOrDrop`, saved in meta `racks` and validated on load, shelves overlay, hover,
  Machines tab, notebook. `test/t_racks.js` (29 checks), play.py menu check.
- Tests: `run_all.py --browser` ALL PASSED. Screenshot `shots/t1-racks.png`.

## 2026-10-24: Backlog combined and rewritten (Claude Code)

- On the owner's request: t12 (exotic ores) and t31 (choosing claims) merged into t13, now
  "Distinct worlds: generation options, planet presets, exotic ores and a choice of claims";
  t24 (water fixes) merged into t21, now "Code health: refactor, speed up and test the physics".
  t2 and t14 rewritten with subtasks. t12, t24 and t31 archived (`tasks/archive/`), not deleted.

## 2026-10-24: t32: minimap and full map (Claude Code, Codex)

- `src/ui/map.js`: colour cache (`resetMap`, `mapArea`, `updateMap` from the frame loop),
  minimap (`#minimap` in `#hud .br`, `SET.minimap`, `SET.minimapZoom`), full map modal `m-map`
  (`openMap`, `zoomMap`, drag pan). `input.js`: M map, Shift+M mute. `#cards` moved to `.bc`.
- Frame time on Medium: 2.18 ms with the minimap off, 2.29 ms on.
- Tests: play.py map checks (buttons, zoom setting, cache updates after digging, fog black, M/Esc,
  Shift+M, cards in .bc, setting hides it, frame time). `run_all.py --browser` ALL PASSED.
  Screenshots `shots/t32-hud.png`, `shots/t32-map.png`, `shots/t32-map-fog.png`.

## 2026-10-24: New worlds get camp names too (Claude Code, Codex)

- The new-world dialog prefills `claimName(seed, existing names)` instead of "World N", and
  follows seed edits until the player types their own name; a blank name falls back to it too
  (`src/ui/loop.js`, `session.js`). play.py checks updated. All tests pass.

## 2026-10-24: t30: progress while generating, loading and saving (Claude Code, Codex)

- `worldgen.js`: `worldSteps` generator yields `[fraction, label]`; `generateWorld` (sync, tests)
  and `generateWorldAsync` (yields every 30 ms) run it. Small and Medium mat/temp/flags hashes
  match the old generator exactly.
- `session.js` `showLoading(on, p, label, title)` with `#loadstep`; `GAME.loading` stops the
  frame loop's sim and drawing while worlds swap. `loadSavedWorld` for Continue and the world
  list; `worldNewest`/`unpackBuffers`/`startFromSave` report progress; travel and new claims too.
  `#saving` chip after 150 ms of saving.
- Tests: t_world progress labels and identical worlds; play.py overlay, responsiveness during
  async generation, Continue labels, saving chip. All pass except t_smooth half-resolution
  (fails under load, as before). Screenshots `shots/t30-loading.png`, `shots/t30-saving.png`.

## 2026-10-24: t3 follow-up: claim names; t30 added (Claude Code, Codex)

- New claims get a camp name from their seed ("Cinder Gulch"; `claimName` in `src/ui/claims.js`,
  skipping names already in the world list) instead of "Claim N". The Claims tab shows "Claim N"
  beside the name and has a Rename button on every row (reuses `renameWorld`).
- Owner's request added to Todo as t30: progress meters while generating, loading and saving.
- Tests: play.py claims checks extended; t_smooth half-resolution still fails under load.

## 2026-10-24: t3 fix: deep ore burned and melted after a reload (Claude Code)

- Owner: a Huge world ran at about 10 fps after travelling between claims, even after a refresh.
  Cause: reloading wakes every area, and t3's deep coal (ignites at 430°C, natural rock reaches
  that at about 70% depth) and gold next to the magma burned and melted forever: about 2,100 of
  12,288 areas stayed busy, 65 ms per tick. A fresh world hid it because nothing deep was awake.
- `worldgen.js`: an ore is not placed where `AMB[y]` is within 120°C of its ignition or 200°C of
  its melting point; a cooler coal band in stone and granite at 45%+ depth replaces most of the
  lost coal. Huge after a reload: 335 areas awake, about 5 ms per tick; Medium 104 (126 before
  t3). Medium seed 424242 ore is now about 4.2x the original (coal 3.2x, gold 2.8x).
- `test/t_world.js`: no ore in too-hot rock; a reloaded world settles with no fires and under
  10% of areas moving; ore balance ranges updated. Existing saves made with the bug keep it.

## 2026-10-24: t3: a world that lasts (Claude Code, Codex)

- `26d17c2` deeper is richer (`worldgen.js` ore thresholds drop with depth below fy 0.25).
  Medium seed 424242 ore grains before -> after: coal 33,493 -> 170,437; copper 26,508 ->
  127,298; iron 77,679 -> 361,683; zinc 11,797 -> 53,527; tin 7,723 -> 38,551; silver 3,638 ->
  12,166; gold 1,681 -> 5,858; total 162,519 -> 769,520 (4.73x). Shallow ore unchanged. Codex's
  first try stretched noise by depth (streaks) and gave gold 26x; redone.
- `4a88254` waste rock: `REFINE` granite/basalt about 2x; granite and basalt bulk orders; prices.
- `56c93ae` Huge 6144x2048 and gzip-packed save buffers (`packBuffers`/`unpackBuffers`, old raw
  saves still load). Medium 29.0 MB -> 0.44 MB, Huge 92.8 MB -> 1.36 MB; Huge generation 5.9 s,
  save 0.5 s, 0.39 ms per idle tick.
- New claims: `src/ui/claims.js` (career record `career:<id>` in IndexedDB, applied over each
  loaded claim; `openNewClaim`, `travel`, Claims tab), `session.js` `setupNewWorld`, `saves.js`.
  I switched claims.js to the repo's single quotes. Screenshot `shots/t3-claims.png`.
- Tests: play.py 268 checks and all headless suites pass. t_smooth's half-resolution check
  fails under load on this machine; the build before t3 failed it 3 of 3 times under the same
  load, so it isn't caused by this work.

## 2026-10-24: t4 follow-ups and t28 Brick press (Claude Code, Codex)

- `9739ca0`: Refiner and Stone cutter code removed; `save.js` skips unknown machine types on load
  and refunds a refiner ($120) or cutter ($90). Their old t_pipes/t_stone checks now use blocks.
- `f5e5591`: `drawProcBlocks` (`render/machines.js`) draws Grinder wheels and the Saw blade over a
  plain housing texture; turns while busy, no `rnd()`.
- `08d2d65`: `PLANK` item; the Saw cuts `LWOOD` too (`SAW_CUTS`, per-material `cut` counts).
  "Stone brick" is "Stone block" in player text. Saw text explains the pile-up under it.
- t28: `BP_PRESS`/`PRESSB`, `GREENB` and `BARBR` items, `FIRETO` (items.js fires a hot green
  brick into a Brick). Machines tab, hover, notebook, milestone hint.
- Tests: `run_all.py --browser` passed after each commit (t_smooth's flake rerun alone).
  Screenshots `shots/t4-wheels{,-smooth}.png`, `shots/t28-press.png`.

## 2026-10-24: t4: Grinder and Saw hopper blocks (Claude Code, Codex)

- New block pieces `grinder`/`saw` (`BP_GRIND`/`BP_SAW`, materials `GRINDER`/`SAWB`, NMAT 97).
  `procBlocksTick` (`src/sim/processors.js`) works the row above each stack top and passes the
  result down the stack (`procFeed`), reusing `REFINE` and `CUT_PER_BRICK`; per-block state in
  `procState`, rebuilt on place/remove/load. Refiner and Stone cutter are `hidden`; their ticks are
  unchanged. Texture in `render/world.js`, hover text in `info.js`, Machines tab, notebook.
- Codex hit its output limit before finishing the tests; I fixed the test setup (a leftover brick
  blocked the outlet), kept the old machine icons, stopped "blocked" flickering on idle ticks,
  and added the play.py Machines-tab and drag-a-row checks.
- Tests: t_stone all pass, play.py 243 pass; t_smooth's half-resolution check fails inside
  run_all at load ~9 and passes alone (known flake). Screenshots `shots/t4-hoppers{,-smooth}.png`.

## 2026-10-24: t25: air streaks for the Pneumatic nozzle (Claude Code, Codex)

- `drawAirStreaks` in `src/ui/previews.js`: 12 teal streaks spiral into the ring while sucking, 10
  amber streaks run from the nozzle to the cursor in a cone while blowing. `performance.now()`
  only, no `rnd()`; skipped with particles off or in fog. `test/play.py`: drawing both doesn't
  change the sim RNG. Also commits the `docs/PROJECT_MAP.md` refresh missed in t27/t26.
- Tests: `run_all.py --browser` passed (play.py 240; t_smooth's half-resolution flake at load 9
  passed on rerun). Screenshots `shots/t25-{suck,blow}.png`.

## 2026-10-24: t26: the Vacuum is renamed the Pneumatic nozzle (Claude Code, Codex)

- `TOOLS[1].name`, a title row in the tool card ("Pneumatic nozzle" plus a `width N` chip, moved
  from the sub line), and player text in `session.js`, `selection.js`, `notebook.js` and
  `progress.js`. Ids, function names and comments still say vacuum.
- Tests: `run_all.py --browser` all passed (play.py 239). Screenshots `shots/t26-card{,-900}.png`.

## 2026-10-24: t27: tool messages as toasts (Claude Code, Codex)

- `setStatus` still records `GAME.status` and now also shows a toast (`err` or `info` class), unless
  told to stay quiet (the "New drill bit/bag" F messages, which have their own card). `toast`
  restarts the timer of a showing toast with the same text instead of adding another, so per-tick
  messages make one toast. The card's status line and its CSS are gone; `doInteract` no longer
  toasts twice. Files: `src/ui/{state,selection,actions}.js`, `style.css`, `test/play.py`.
- Tests: `run_all.py --browser` all passed (play.py 237). Screenshot `shots/t27-toasts.png`.

## 2026-10-24: t20 follow-up: bag hint and Clear stay visible (Claude Code)

- The hint and Clear were inside the scrolling `#baglist`, so a long Ctrl list scrolled them away.
  They now render into a new `#bagfoot` under the list (`src/shell.html`, `renderBag`); Clear is a
  full-width, highlighted "Clear selection" button (`style.css`).
- `test/play.py`: a long Ctrl list scrolls while the hint and a full-width Clear stay visible.
  Screenshots `shots/t20c-ctrl{,-900}.png`. Tests: `run_all.py --browser` all passed (play.py 232).

## 2026-10-24: t20 follow-ups: deselect when blown out, Ctrl + wheel, Ctrl+click (Claude Code)

- Blowing the last of a selected material removes it from the selection (`toolTick`); materials
  selected before you have any stay. Window `wheel` listener: Ctrl + wheel never zooms the page
  and scrolls the scrollable panel under the mouse (the Ctrl bag list). Bag clicks add when
  `GAME.ctrl` is set even if the click has no `ctrlKey` (owner had to hold Ctrl+Shift; not
  reproduced in headless Chromium, which passes ctrlKey).
- Files: `src/ui/{actions,input,selection,notebook}.js`, `test/play.py` (229 checks).
- Tests: `run_all.py --browser` all passed.

## 2026-10-24: t20 follow-up: one selection, full list while Ctrl is held (Claude Code, Codex)

- `GAME.blow` is now the selection for picking up and blowing out (`vacFilter` -2 = the selection).
  `blowList()` returns only selected materials in the bag; no pruning, no auto-select. `GAME.ctrl`
  (input.js) makes `renderBag` add a "Not in bag" section of every `seen` material the vacuum can
  take; selected but empty materials always show there. Clear button, click-the-only-row, gamepad B.
- Files: `src/ui/{actions,selection,hud,input,state,gamepad,previews,notebook,session}.js`,
  `src/sim/{vacuum,progress}.js`, `src/styles/style.css`, `test/play.py`, `test/t_basics.js`.
- Tests: `run_all.py --browser` passed (play.py 226) except `t_smooth` half-resolution at load 8.4,
  which passed on rerun. Screenshots `shots/t20b-{noctrl,ctrl,ctrl-900}.png`.

## 2026-10-24: t20: pick up what you click, blow out what's in hand (Claude Code, Codex)

- After discussing options with the owner: one Vacuum, a drag locks onto the material it starts on
  (Shift: everything), and `GAME.blow` is "the hand" (click/Q/E, Ctrl+click mix). Empty materials
  leave the hand; an empty hand blows nothing; a pickup fills an empty hand. Removed `GAME.collect`,
  R marks, Ctrl+right-click, `openVacuumMenu`, `ensureBlow` and the `.vpick` styles.
- Files: `src/ui/{actions,selection,hud,state,previews,build-menu,info,notebook,session}.js`,
  `src/sim/{vacuum,progress}.js` (`toolVacuum` returns the first material picked up),
  `src/styles/style.css`, `test/play.py`, `test/t_basics.js`.
- Tests: `run_all.py --browser` all passed (play.py 214 checks). Screenshots `shots/t20-*.png`
  (empty hand, one item, mix, 900px).

## 2026-10-24: t17 follow-up: no jagged rim, unexplored is black (Claude Code)

- The owner still saw a jagged grey edge. Causes found by sampling pixels: the fade went to 0
  (black) while unexplored was drawn (12, 10, 10), leaving a dark stepped rim; the ±40 noise lit
  the faintest cells; rays that reached rock already dim dropped to 0 in 2–3 grains.
- `src/render/world.js`: unexplored and off-map are pure black (`OUT`, smooth-look buffers);
  partial fog eases into black along a smoothstep, with noise scaled by s·(1−s). `src/sim/fog.js`:
  sight into rock fades over 8 grains as a fraction of its brightness.
- `test/play.py`: the three unexplored-colour checks expect black (owner's decision).
  `docs/systems/fog.md` updated. Screenshots `shots/fog-dig-black-{pixel,smooth,nolight}.png`.
- Tests: `run_all.py --browser` passed except `t_smooth` once at load 7 (passed on rerun).

## 2026-10-24: t17 follow-up: softer fog edges (Claude Code, Codex)

- The owner found the edges sharp and stepped. `src/sim/fog.js`: sense ring 24 grains with a
  12-grain fade (`FOG_SENSE_FADE`), rays fade 8 grains into rock and over the last 16 of range,
  surface fade 16, `fogRevealAround` gets a 6-grain soft edge. `src/render/world.js`: partly seen
  grains get ±40 of per-grain noise (`noiseB`), so edges look grainy, not stepped.
- `test/t_fog.js`: a point in the fade band is between 0 and 128. `docs/systems/fog.md` updated.
  Screenshots `shots/fog-edge-{before,after}{,-nolight}.png` compared.
- Tests: all pass except `t_smooth` half-resolution, which failed 1 of 3 runs with the change and
  passed 2 of 2 without, at load 6–12; that check runs with fog off, so the drawing change does
  not run there. Fog drawing 0.43 → 0.45 ms.

## 2026-10-24: t17 Fog of war reveals the ground around the bot (Claude Code, Codex)

- `src/sim/fog.js`: `FOG_SENSE = 20`; `fogLook` also reveals every fog cell within 20 grains,
  through rock (full to 16, fading to 20). Claude rewrote Codex's per-grain `Math.hypot` loop to
  work per fog cell. `fogLook` cost: 0.042 → 0.043 ms (every 4 ticks).
- `test/t_fog.js`: rock 12 grains away is seen, near the edge partly, 25 away still hidden.
- `docs/systems/fog.md`, DESIGN.md decision. Also: `.kandown/kandown_work.md` rule to write
  tasks from chat in our own words (no owner labels, dates or quotes); t16/t19 descriptions
  rewritten that way; `docs/kandown.md` notes that web-board descriptions were lost.
- Tests: `run_all.py --browser` all pass (220 `play.py` checks). `t_smooth` half-resolution failed
  on two runs while the machine was loaded (load 6–10) and passed on the final run; timing
  `fogLook` showed the change is not the cause. Screenshot `shots/fog-sense.png` checked.

## 2026-10-24: t19 Vacuum blow all visual (Claude Code, Codex)

- The owner's description was lost in kandown; chosen in chat: bag rows light up and the card
  shows it while Shift is held. t18 was deleted at the owner's request (to decide later: what
  blow does when the chosen material runs out).
- `hud.js` `blowingAll()` (Shift held and not flying, or blowing everything); `selection.js` card
  ("Everything · mix of N", no swatches so it stays one line) and bag `.on` rows; `previews.js`
  cursor swatches.
- `test/play.py`: Shift preview on and off, flying doesn't preview (the bot is put back after,
  or a later gamepad dig check fails). Tests: `run_all.py --browser` all pass (220 `play.py`
  checks).

## 2026-10-24: t16 follow-up: Shift blows all, Ctrl+click in the bag, compact menu (Claude Code)

- The owner's subtask "shift to blow all" and two chat requests. Codex was out of credits, so
  Claude made the edits.
- `input.js`/`actions.js`: `secondaryPress(shift)` sets `GAME.blowAll`; `hud.js` `blowAllList()`
  (loose bag materials, no bars); `previews.js` shows it. `selection.js`: Ctrl+click a bag name
  toggles blow, Ctrl+right-click toggles pick-up; the vacuum menu is one `.vrow` per material
  (CSS in `style.css`). Text in the card, bag hint and `notebook.js`.
- `test/play.py`: Shift blow-all (replaces the old "Shift still blows only coal" check, per the
  owner's change), Ctrl+click and Ctrl+right-click, plain right click does nothing, one line per
  menu row at 1280 and 900.
- Tests: `run_all.py --browser` all pass (215 `play.py` checks). Screenshots checked.

## 2026-10-24: t16 Vacuum tool rework, again (Claude Code, Codex)

- The owner: too many modes; chose an F menu with two pick-up states.
- `src/ui/selection.js`: `openVacuumMenu()` (a tile per material with ⬇ Pick up / ⬆ Blow, plus
  Pick up everything and Blow one at a time); `GAME.pickMode`, `setPickMode`, `cyclePickMode`,
  `GAME.vacLocked` and the 'first' locking removed. Bag rows lose ⬇/⬆ buttons (a ⬇ tag shows
  marks). `actions.js`: pick-up is everything unless marked; F with the vacuum opens the menu
  when nothing in reach can be used. Card chips open the menu (`build-menu.js`). T and gamepad B
  no longer change vacuum modes. Text in `notebook.js`, `progress.js`, `session.js` (Welcome),
  `previews.js`; CSS `.vpick`.
- `test/play.py`: mode checks replaced (no marks takes everything, marks, Shift, menu toggles,
  F order, card chip, no bag buttons, menu screenshots at 1280 and 900); `test/t_basics.js` name.
- Tests: `run_all.py --browser` all pass (208 `play.py` checks). Codex ran out of credits before
  its final test run; Claude fixed one check (it mutated a mold's kind) and the suction line.

## 2026-10-24: t6 Rework the vacuum/blower tool, with t15 (Claude Code, Codex)

- The owner folded t15 (a material menu for the vacuum) into t6's chosen design.
- `src/ui/state.js`, `hud.js`, `actions.js`, `selection.js`: `GAME.pickMode` ('first', 'marked',
  'all', saved in localStorage) and `GAME.blow` (a set) replace `dropMat`, `blowMix`,
  `matChosen` and the Ticked mix. Pick-up marks stay in `GAME.collect` (same storage key).
- Bag rows with the vacuum: ⬇ pick-up toggle, name (blow only this), ⬆ blow toggle. Tool card:
  a pick-up mode chip (T or click cycles) and the blow choice (Q/E). Gamepad button 1 cycles the
  mode when nothing else uses it. Text updated in `info.js`, `notebook.js`, `progress.js` (the
  copper hint), `src/sim/vacuum.js` (a status line), `docs/DESIGN.md`.
- `test/play.py`: modes, automatic switching, independent marks, mixes, persistence, toggles only
  with the vacuum; `test/t_basics.js`: the new status. Screenshots checked at 1280 and 900.
- Tests: `run_all.py --browser` all pass (214 `play.py` checks).

## 2026-10-24: decisions for t1, t4, t6 (Claude Code)

- Options given in chat; the owner chose: t1 neat bar piles plus a Bar rack; t4 Grinder and Saw
  blocks replace the Refiner and Stone cutter (old ones keep working); t6 the bag as the vacuum's
  control panel with separate pick-up and blow marks. Recorded in `docs/DESIGN.md` and as
  subtasks on each task. Nothing built yet.

## 2026-10-24: t8 Fog of war for unexplored areas (Claude Code, Codex)

- Decided with the owner: revealed by line of sight; unseen is black with a soft edge.
- New `src/sim/fog.js` (2×2-grain cells, 0-255, only ever rises): `fogInit` (sky and surface
  seen), `fogLook` (256 rays, 48 grains, through air, gas and liquid, 4 grains into rock), called
  every 4 ticks from `simTick`. Never read by the simulation.
- `src/sim/save.js`: fog saved; saves without it rebuild it (surface, 12 grains around built
  things, the bot's view). `src/sim/world.js`, `worldgen.js`: allocate and initialize.
- `src/render/world.js`: grains scaled by a bilinear fog value; fully unseen grains drawn like
  out-of-world grains; no glow from fog; thermal lens fogged too. `overlays.js`: rain and
  particles hidden in fog. `src/ui/info.js`: "Unexplored". Setting `fog` (Game tab).
- Browser tests and tools turn fog off (`play.py`, `t_smooth.py`, `shot.py`, `perf.py`,
  `playthrough.py`) except the fog checks. New `test/t_fog.js`; `docs/systems/fog.md`.
- Tests: all 12 headless suites, `play.py` (185 checks), `t_smooth` pass. Screenshots checked in
  both looks. `tools/perf.py`: drawWorld 0.34 → 0.43 ms with fog on (no change with light off).

## 2026-10-24: t5 Material selection menu (Claude Code, Codex)

- Decided with the owner: F opens it for walls and molds, R keeps cycling.
- `src/ui/selection.js`: `openWallPicker` (a dialog of every wall material: what you have, the
  cost for this piece, the heat it stands; owned first, unaffordable dimmed; glass says whether
  it is from shards or bars); `pieceCostNow(P, m)` takes a material.
- `src/ui/actions.js`: F with a wall or mold piece opens it. `src/ui/build-menu.js`: the material
  chip reads "· R · F" and opens it on click; placement help mentions F. `src/ui/notebook.js`:
  Controls and Blueprints text. `src/styles/style.css`: `.wpick`.
- `test/play.py`: F and the chip open it, Brick picks RBRICK, R still cycles, molds have their own
  menu and cost, it fits at 1280×800 and 900×700 (screenshots checked).
- Tests: `run_all.py --browser` pass, apart from the `t_smooth` half-resolution flake.

## 2026-10-24: t10 World name box for a new world (Claude Code, Codex)

- `src/shell.html`, `src/styles/style.css`: a Name box across the top of the New world panel.
- `src/ui/loop.js`: opening the panel fills it with the next `World N` and selects it; Enter in
  Name or Seed generates. `src/ui/session.js`: `startNewWorld(seed, w, h, name)`.
- `src/ui/saves.js`: `cleanWorldName` (trim, 40 characters, else a fallback), shared with
  `renameWorld`.
- `test/play.py`: a typed name reaches the Worlds list and the pause menu; the default and an
  all-spaces name both give `World N`. Screenshot `shots/play/new-world.png` checked.
- Tests: `run_all.py --browser` all pass.

## 2026-10-24: t7 Option to turn off smooth pan following the mouse (Claude Code, Codex)

- New setting `cameraLead` (Display: "Camera looks toward the pointer", on by default):
  `src/ui/state.js`, `src/ui/settings.js`; `src/ui/loop.js` gives the camera no look-ahead when
  it is off (mouse and gamepad).
- `test/play.py`: the toggle saves; with it off the camera stays on the bot with the mouse in a
  corner; with it on the camera moves toward the corner.
- Tests: headless suites and `play.py` pass; `t_smooth` half-resolution flake as before.

## 2026-10-24: t9 Script to run the local server for testing (Claude Code, Codex)

- New `tools/serve.py`: builds, serves `dist/` on 127.0.0.1 (port 8000, or the next free one up
  to 8010), sends `Cache-Control: no-store`, and rebuilds when `src/` or `build.py` changes; a
  failed build is printed and the old page keeps being served.
- `README.md`, `AGENTS.md`, `build.py` docstring, `docs/PROJECT_MAP.md`: point at it.
- Checked by hand: page served with no-store; touching a source file rebuilt it; with the owner's
  old server on 8000 it moved to 8001; Ctrl+C stops it. Headless suites and map check pass.

## 2026-10-24: t11 Welcome and first milestone shown again on an older world (Claude Code, Codex)

- Cause: nothing cleared the discovery cards (`#cards`) when a world was entered, so a new
  world's Welcome and milestone cards stayed up after loading an older world.
- `src/ui/session.js`: `enterGame` clears the cards, floaters, particles and the status line;
  a new world still shows its Welcome card afterwards.
- `test/play.py`: a new world shows Welcome; loading a world from the list shows no cards.
- Tests: `run_all.py --browser` passes (153 `play.py` checks). `t_smooth`'s half-resolution
  check flaked (passed and failed on reruns, with and without this change; the machine was busy).

## 2026-10-24: the to-do list moved from NOTES.md to a kandown board (Claude Code)

Request: initialize kandown (https://kandown.dev, 0.52.0) for the project, move all the notes
into it, and document how it works.

- Practised first in a scratch folder: `init --no-open` starts no daemon and makes no sample tasks
  when `tasks/` exists; `create` prints only the id and names the file `<id>_<slug>.md`; a
  multi-line `report: |` block is read (checked in kandown's parser) and `doctor` accepts it;
  `list --json` doesn't include `report`; project instructions in `.kandown/kandown_work.md`
  appear in `kandown work` (~1,100 tokens).
- A mistake on the way: `kandown init --help` ran `init` in the vault (it ignores `--help`) and
  started a daemon there. Stopped it, and removed the vault's earlier kandown test files as the
  owner asked. `docs/kandown.md` warns about it.
- `kandown init --no-open` in the repo. `.kandown/kandown.json`: default columns, Review's
  instructions say only the owner moves work to Done, tags and priority shown. `.gitignore`:
  `.kandown/kandown.html` (6.5 MB, regenerated).
- 14 notes became tasks t1-t14 (Todo: 6 tasks; Backlog: 5 tagged `discussion`, 3 tagged
  `idea`), each with the owner's note word for word under `## Context` (checked by script: 14 of
  14). `kandown doctor` is clean.
- `.kandown/kandown_work.md` holds the board rules (from AGENTS.md's old NOTES section);
  `docs/kandown.md` explains how kandown works here. `AGENTS.md` (session start, a new "Working
  from the task board" section linking both), `CLAUDE.md`, `README.md`, `docs/PROJECT_MAP.md`,
  `docs/DESIGN.md` (decision). `NOTES.md` removed.
- `kandown list` and `kandown doctor` checked inside Codex's sandbox: they work.
- Open: the web board (`kandown`) wasn't opened here; the owner uses it.

## 2026-10-24: several worlds, with a Worlds list (Claude Code, with Codex)

Note: "support multiple saves". The owner chose a world list.

- `src/ui/saves.js`: each world has an id and its own `<id>:main` / `<id>:auto` saves, plus an
  index record `worlds` (name, seed, size, created, saved, playtime). A save writes the world
  and its index entry in one transaction. `GAME.worldId` is the open world.
- Migration: an old single `main` / `auto` save is copied to "World 1", every byte compared,
  and only then are the old keys deleted (in the same transaction that writes the index). If
  anything fails the old save is kept and a toast says so.
- Title screen: Continue loads the most recently saved world; a Worlds button opens a list
  with Load, Rename (trimmed, at most 40 characters, escaped when shown) and Delete (with a
  confirmation). New and imported worlds get new ids. Settings' "Delete saves" became "Delete
  all worlds" with a confirmation. `shell.html`, `session.js`, `loop.js`, `settings.js`,
  `state.js`, `style.css`, `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- `test/play.py` (149 checks): migration of an old save, two worlds listed, load switches,
  rename survives a reload, HTML in a name shows as text, long names cut to 40, delete and
  cancel, Continue picks the newest, import adds a world. All 11 headless suites pass;
  `t_smooth.py` passed in Codex's run and failed its known flaky check in mine.
- Shots: `shots/worlds-{pixel,smooth}.png` and `-900` versions.
- Not committed: a line in `AGENTS.md` about running `kandown work` appeared in the working
  tree during this task; it was not part of the task and is left for the owner.
- Open: nothing.

## 2026-10-24: screw tubes carry bars (Claude Code, with Codex)

Note: "how do you transport bars? belts work but tubes do not, should they?" The owner chose
tubes over a separate bar lift.

- `src/sim/items.js`: a bar whose eight cells are all tube cells is in the tube (derived from
  its position, so saves need nothing new). On the tube beat it steps along the tube, snaps
  into the next block's lane at corners, and leaves a mouth with a launch like grains. A bar
  pushed by a belt into a tube's side or back, falling into a down tube or sitting on an up
  tube's intake enters when the aligned cells are free. Grains and bars wait for each other.
  Gas and liquid don't block a bar (Claude added this after review: smoke near furnaces stalled
  bars in Codex's version).
- Text: `pieces.js` tube description, `notebook.js`, `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- Tests: `t_conveyors` (55 checks): belt into a lift and out the top, right tube, up-to-right
  and right-to-down corners, blocked mouth waits without overlap, grains and a bar share a
  lift with sand conserved, save mid-tube, smoke and steam in the tube, melting in a tube,
  bars still ride and stop on belts and rest on top of sideways tubes. All 11 headless suites
  and `play.py` (131) pass; `t_smooth.py`'s half-resolution check failed on the slow fallback
  (known flaky).
- Shots: `tools/scenes/bar-lift.js`, `shots/bar-lift-{300,900}.png`: mold, belt, lift, corner,
  belt; the bars arrive at the top.
- Open: nothing.

## 2026-10-23: screw tubes use all their lanes (Claude Code, with Codex)

Note: "maybe the belts can move faster/carry more than the tubes… the tubes only carry one line of
materials on either side." The owner chose to fill every lane; belt speed is unchanged.

- Measured first (`test/t_conveyors.js`, a belt with a sand heap into the side of a 6-block up
  tube, 3,000 ticks): 1,459 grains out, lanes `[2940, 0, 0, 10]`; the edge lane was full (one
  lane carries at most ~1,500). Spreading only at intake changed little (1,486): belts push
  straight into the edge lane without going through `tubeIntake`.
- `src/sim/machine-tick.js`: on each tube beat a grain in a vertical tube may step sideways
  within its block to a lane with clearly fewer grains ahead (a strict gap stops shuffling).
  `src/sim/conveyors.js`: `tubeSpread`, shared by side intake (far lanes first) and corners.
- After: 2,071 out (92% of the 2,243 the same belt delivers to an open drop), lanes
  `[1937, 1116, 486, 872]`, sand conserved. Checks: all four lanes, conservation, at least 90%
  of the belt's delivery, a side-fed sideways tube fills more than one row. All 11 headless
  suites and `play.py` pass. `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- Found on the way: Codex's sandbox (`workspace-write`) blocks network sockets, and headless
  Chrome crashes at startup without one (SIGTRAP core dumps in pairs). Codex runs now pass
  `-c sandbox_workspace_write.network_access=true`.
- Open: nothing.

## 2026-10-23: the build card describes the build tool, not the piece (Claude Code)

Note: "instead of the current info box for the build tool showing the current building selected,
it should just show a description about the build tool in general." The owner kept the R/F chips.

- `src/ui/selection.js`: the build card reads "Build", then the R/F chips, Replace · G and
  ghosts, a line on placing, the build keys and the status line. No piece name or cost.
- `test/play.py`: the card shows "Build" and the placing text, not the piece's name, cost or
  description; the other card checks are unchanged. `play.py` and all 11 headless suites pass.
- Shots: `shots/build-card-general-{pixel,smooth}.png`. `docs/DESIGN.md`.
- Open: nothing.

## 2026-10-23: world sounds are heard around the bot, not the view (Claude Code)

Note: "the sound effect is still heard when the player's zoom level is farther out… sounds
shouldnt be heard based on the viewport, but rather in a radius from the player."

- `src/ui/audio.js` `sfxPlace`: gain from the distance to the bot's centre (full within 24
  grains, silent at 96), pan from the horizontal offset; zoom no longer matters.
- `test/play.py`: the audio checks use points relative to the bot; a new check zooms out to 2
  and plays a visible sizzle 200 grains away: silent (it played before). `play.py` and all 11
  headless suites pass. `docs/DESIGN.md`.
- Open: nothing.

## 2026-10-23: F on an ingot mold does nothing and says nothing (Claude Code)

Note: "now that ingot molds empty themselves, F still makes a message…"

- `src/sim/actions.js` `interactAt` passes ingot molds by, so F reaches anything else nearby;
  `src/ui/actions.js` shows no "Nothing to use there." when the cursor is on an ingot mold.
- Tests: `t_molds` (F on an ingot mold in every state returns nothing, changes nothing, no
  sound) and `play.py` (F over an ingot mold leaves no status and no toast). All 11 headless
  suites and `play.py` pass.
- Open: nothing.

## 2026-10-23: pixel-art silo for the pixel look (Claude Code, with Codex)

Note: "the silo art looks the same for both pixelated graphiocs and smooth."

- `src/render/machines.js`: `drawSilo` is the smooth (illustrated) version as before; in the
  pixel look `drawSiloPixel` draws it grain by grain: a drum shaded by column with rib rows and
  ladder rungs, a cone to the slide gate, legs, the roof hatch with its lid, and the sight glass
  filling in the contents colour. Codex's first version read poorly (dark side columns made
  the drum look narrow); Claude redrew the grain layout.
- `tools/scenes/silo.js` (run after `clear-sky.js`): three silos, empty, half sand, nearly full
  copper ore with the gate open. Shots `shots/silo-*.png` at zoom 6 and 12; smooth shots match
  the before shots pixel for pixel. `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- Tests: all 11 headless suites and `play.py` (129) pass. `t_smooth.py`'s half-resolution check
  failed twice on the slow fallback at load ~5 (known flaky; it passed in the previous task).
- Open: nothing.

## 2026-10-23: world sounds are local (Claude Code, with Codex)

Note: "there is a sound effect that seems to happen randomly but constantly… maybe the sound
effects are not local?" The owner then confirmed a new world is quiet.

- Measured headlessly over 5 game minutes: a fresh world emits no sounds; a lava and water
  contact 720+ grains from the bot emitted 2,889 sizzles (about 580 a minute), and `events.js`
  played every one at full volume. The sizzle (`heat.js` boiling and quenching) is the only sim
  sound with a position.
- `src/ui/audio.js`: `sfxPlace(x, y)` gives gain (1 in view, 0 one view width beyond it) and
  pan; `playSfx(name, arg, x, y)` skips silent ones before the rate limit and plays the rest
  through a gain and stereo panner (`AU.bus`). `src/ui/events.js` passes positions.
  `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- `test/play.py`: far off-screen is silent, on screen plays, just off-screen is quieter and
  panned. All 11 headless suites, `play.py` and `t_smooth.py` pass.
- Checked afterwards with the owner's exported world (seed 622525): over one minute, 48 sizzles
  and no other sounds, all from water boiling on naturally hot deep rock (a pool on 580°C basalt
  and drips in a shaft and a cave, about 500 grains below the bot). Nothing built there. With
  the fix every one of them is silent from the surface.

## 2026-10-23: fewer, smaller messages (Claude Code, with Codex)

Note: "lets change the engot mold and other status messages. they are way too large and just too
many messages…". The owner chose molds, machine warnings and shorter status lines.

- `src/render/overlays.js`: every mold always shows a small percentage (`moldLabel`), gold
  `100% · F` for a set bit or tank, red when spoiled or when an ingot mold is blocked
  (`mo.blocked`, set in `ejectIngot`, not saved). Details are in the hover panel (`info.js`),
  which also carries the furnace-loading advice that F messages used to append.
- `src/ui/events.js`: no cards or toasts for bars dropping, blocked molds or set castings (the
  sounds stay). Fan heat is a red ! marker on the fan (`mm.hot`, hover shows °C); melted tubes
  and machines destroyed by heat get a short red floating label (`warn` event, coloured
  `spawnFloater`) instead of a toast.
- Status lines shortened to one sentence across `src/sim` (molds, actions, blocks, blueprints,
  drill, items, machines, tools, undo, vacuum) and `src/ui` (actions, selection).
- Tests: `play.py` records every status line and checks none is over 90 characters (longest
  now 55); mold label and no-toast checks. `t_molds` (message lengths, `mo.blocked`), `t_build`
  (hot fan: `mm.hot`, no toast), `t_world`. All 11 headless suites and `play.py` pass.
- Shots: `shots/messages-pixel.png`, `shots/messages-smooth.png` (filling, blocked, set bit,
  hot fan).
- Open: the first-ever bar drop no longer shows an explaining card; the notebook still explains.

## 2026-10-23: ingot molds need no F; removing a mold returns its metal (Claude Code, with Codex)

Note: "now that the ingot molds drop the ingots, what is the purpose of F…". The owner chose no F
on ingot molds.

- `src/sim/molds.js` `ejectIngot`: when nothing over or in the mold is molten, junk and mixed
  metal (as scrap) and plugs over the trough or frozen in the grate are written as loose grains
  into the 4 rows under the mold; what doesn't fit waits. Slag and ash are thrown away. After a
  drop or a clean-out the grate above is raked. Bars and the short stub work as before.
- `collectMold` on an ingot mold only says "Ingot molds empty themselves." Bit and tank molds
  keep F. `removeMold` (every kind) gives the set metal to the bag: whole bars as bars, the rest
  as scrap; molten metal is left alone.
- Text: `troughState` flaws, `info.js`, `overlays.js` (no "· F" on ingots), `pieces.js`,
  `progress.js`, `notebook.js`, `docs/systems/stone-and-liquids.md`, `docs/DESIGN.md`.
- Tests: `t_molds` 60 checks (F does nothing on ingots; plate, slag and rake cases now
  automatic; junk out, mixed metal out as scrap, junk waits when blocked and while molten;
  removal returns bars and scrap, leaves molten metal). `furnace_play.js` adapted. All 11
  headless suites and `play.py` (122) pass.
- Open: undoing a mold removal brings the mold back empty (its metal stays in the bag).

## 2026-10-23: compact build card top-left (Claude Code, with Codex)

Note: "build tool has no info box like the other tools." The owner chose a compact card.

- `src/ui/selection.js`: the build tool has a card again: Q, piece name, cost (red if short), E;
  variant chips, Replace · G and ghosts waiting; one how-to line; the status line. No long
  description; that stays in the hover popup.
- `src/ui/build-menu.js`: the `.bb-status` row is gone (its contents are on the card); the
  variant chips are shared through `buildVariantChips(P)`. `src/styles/style.css`,
  `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- `test/play.py`: the card is visible while building with name and cost and without the
  description; status and Replace · G are read from the card; no `.bb-status` (122 checks pass).
  All 11 headless suites pass; `t_smooth.py`'s half-resolution check is still flaky under load.
- Shots: `shots/build-card-*` (belt and wall, both looks, 1280×800 and 900×700) look right.
- Open: at 900px wide the bottom-left hover panel still sits partly under the build bar.

## 2026-10-23: Replace is held G, not Ctrl (Claude Code, with Codex)

Note: "i see the issue was ctrl+w closing the tab… it happens for the bookmark hotkey too. maybe
change the replace key?" The owner chose holding G.

- `src/ui/input.js`: `GAME.ctrl` became `GAME.replace`, set while G is held. Ctrl no longer
  does anything on its own, so WASD works while replacing. Ctrl+S, Ctrl+Z, Ctrl+wheel and the
  leave-page prompt are unchanged.
- Flag users and text: `previews.js`, `actions.js` ("Hold G to replace it."), `build-menu.js`
  (Replace · G chip, key hints), `loop.js`, `state.js`, `notebook.js` controls, `docs/DESIGN.md`.
- `test/play.py`: the replace checks hold G; a new check flies right with D while G is held
  (121 checks pass). All 11 headless suites pass. `t_smooth.py`'s half-resolution check failed
  on the slow fallback again (known, flaky under load).
- Open: nothing.

## 2026-10-23: Ctrl+W no longer closes the game without asking (Claude Code)

Note: "crash: hoplding crtl and pressing W crashes the whole browser."

- Cause: Ctrl+W is the browser's close-tab shortcut (the whole window if it is the last tab);
  holding Ctrl for replace while pressing W to fly hits it. Chrome and Brave reserve it, so the
  page never sees the key and cannot cancel it.
- `src/ui/input.js`: a `beforeunload` handler makes the browser ask before leaving while
  `GAME.running`. The `visibilitychange` autosave no longer depends on the audio context
  existing (it used to return early before any sound had played).
- `test/play.py`: 2 checks (leaving asks while a world is open, not on the title screen); 120
  checks pass. All 11 headless suites pass; `t_smooth.py` passed on its rerun (the known flaky
  half-resolution check failed once at load ~5).
- Open: the real Ctrl+W prompt can only be seen in a headed browser; the owner checks it.

## 2026-10-23: build menu hover popup, top-left card hidden while building (Claude Code, with Codex)

Notes: "mouse hover reviels a new info popup…" and "dont need the short description in the build
menu…". The owner chose to hide the top-left card entirely while building.

- `src/ui/build-menu.js`: `#bb-pop` above the hovered tile (icon, name, cost, size, power, full
  description, how to place, filter help, build keys), clamped to the window; on a gamepad it
  shows for the selected tile. The `.bb-tip` row became `.bb-status`: variant (material,
  direction, filter) chips, Replace · Ctrl, ghosts waiting, and status messages. The bar now
  updates its parts separately so hovering doesn't rebuild the tiles.
- `src/ui/selection.js`: `#selcard` is hidden for the build tool unless copying, cutting or
  pasting; the build branch moved to the bar. `src/styles/style.css`: popup and status styles;
  on narrow screens the bottom-right panel moves up while building.
- `src/ui/notebook.js`, `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- Tests: `play.py` 118 checks (popup on hover, switching tiles, hidden on leave, gamepad,
  no `.bb-tip`, card hidden while building and back for tools 1 and 2, status in the bar; the
  Ctrl replace check now reads the bar). All 11 headless suites pass. `t_smooth.py`'s
  half-resolution check is still flaky under load (~4.8): 1 pass in 3 runs, as before.
- Screenshots in `shots/` (build-popup-*) at 1280×800 and 900×700 in both looks.
- Open: at 900px wide the bottom-left hover-info panel sits partly under the build bar.

## 2026-10-23: a backed-up heap on a belt no longer shears (Claude Code, with Codex)

Note: "belts with a big pile on top only move the grains on the top of the stack…". The owner's
setup: a belt ending in a screw tube going up; material mixed into the bottom layers never moved.

- Cause: `carryColumn` (`src/sim/conveyors.js`) skipped a blocked grain and carried on up the
  column, so with the front backed up the top slid over a stuck bottom. Reproduced headlessly
  first: over 4,000 ticks, 8,815 shear steps at the tube (1,415 against a wall) and only 57 of
  192 bottom sand grains left the belt.
- Fix: on a belt, the walk stops at the first blocked powder grain, so the load above stands
  still. On a filter the walk goes on only so chosen materials can still drop through; nothing
  above a blocked grain is carried. After: no shear, 119 of 192 sand grains out through the tube.
- `test/t_conveyors.js`: 3 new checks (no shear into an up tube, none against a wall, bottom
  layer leaves through the tube). `tools/scenes/belt-heap.js` sets up the owner's case for
  `tools/shot.py`; shots at 30 and 900 ticks look right. `docs/DESIGN.md`, `docs/PROJECT_MAP.md`.
- Tests: all 11 headless suites and `play.py` (109) pass. `t_smooth.py` failed its
  half-resolution check once at load 6.5 and passed on the rerun.
- Open: `t_smooth.py`'s half-resolution check is flaky on a busy machine (the game drops to
  pixel when frames are slow). It is worth making that test immune to the slow fallback.

## 2026-10-23: ingots drop out of the bottom of the mold (Claude Code, with Codex)

Note: "copper bars go into the furnace. also they shouldnt require the player to press F…".
The owner confirmed the first part was a bug: the knock-out threw bars up into the furnace.

- `src/sim/molds.js`: `ejectIngot` runs from `checkMolds` (every 15 ticks). Bars spawn directly
  under the mold, only where `entFreeAt` says there is room; the rest wait in the trough. Metal
  is taken from the top so the short stub stays on the floor. `ingot_drop` and `ingot_blocked`
  events (blocked is said once per spell via `mo.blockedSaid`, not saved). F on a ready ingot
  never launches bars upward; F on an empty, short or spoiled mold still cleans and rakes.
- `src/ui/events.js`: first drop gets a card, later drops a toast; blocked gets a toast. Ingots
  no longer emit `cast_ready`.
- Text: `pieces.js`, `progress.js`, `notebook.js`, `info.js`, `docs/DESIGN.md` (with the
  decision), `docs/systems/stone-and-liquids.md`, `docs/PROJECT_MAP.md`.
- Tests: `t_molds` (38 checks: downward drops, stub kept, top-up, blocked and warned once,
  partial drops, full entity pool, slag), `t_basics`, `t_stone`, `furnace_play.js` and
  `play.py` section 6 changed to the new behaviour. All 11 headless suites, `play.py` (109) and
  `t_smooth.py` (8) pass. A shot of a mold with a belt one block below shows the bars riding off.
- Open: a belt touching the mold's underside counts as blocked; bars need one open block. The
  owner may want a belt directly under the mold to work.

## 2026-10-23: Shift+Q/E build tabs, G pick key removed (Claude Code, with Codex)

Notes: "add shift + q/e to move tabs" and "having the "g pick" there is confusing".

- `src/ui/build-menu.js`: `selectBuildCat(c)` (shared with tab clicks) and `cycleBuildCat(d)`;
  the menu hint now reads Q/E cycle · Shift+Q/E tab.
- `src/ui/input.js`: Shift+Q/E switch tabs with the build tool; G is unbound.
- Text: build card (`selection.js`), notebook controls (`notebook.js`), `actions.js` comment,
  `docs/DESIGN.md`, `docs/PROJECT_MAP.md` (regenerated, `--check` passes).
- `test/play.py`: checks for Shift+Q/E, wrapping, the remembered piece per tab, plain Q/E,
  and that G no longer calls pick while middle mouse still does (109 checks).
- Tests: all 11 headless suites and `play.py` pass; a `tools/shot.py` shot of the Moving tab
  shows the new hint. `t_smooth.py` passes, but its "half-resolution smooth still draws" check
  failed on several runs with the 'slow' pixel fallback while the machine was busy (load ~14).
  The previous commit fails the same way under the same load, so this change didn't cause it.
- Open: that check is timing-sensitive on a loaded machine; rerun it when the machine is idle.

## 2026-10-23: browser tests run locally (Claude Code, with Codex)

Request: get the browser tests, Playwright and Chromium working on the owner's Arch machine
(Python 3.14 refuses a system-wide pip install).

- Playwright 1.64.1 and its Chromium 156 (revision 1248) are in an in-repo `.venv/` made with uv.

- Added `tools/use_venv.py` and called it before Playwright imports in `tools/shot.py`,
  `tools/perf.py`, `tools/playthrough.py`, `test/play.py` and `test/t_smooth.py`.
  The `sys.prefix` guard prevents an exec loop if the venv lacks Playwright.
- Updated `requirements.txt`, `.gitignore`, `README.md`, `AGENTS.md`, `docs/TESTING.md`
  and `docs/PROJECT_MAP.md` for Playwright 1.64.1 and venv setup.
- Checks: regenerated the project map and ran `tools/map.py --check`.
- Tests run with plain `python3`: `test/run_all.py --browser` all passed (11 headless suites,
  `play.py` 101 checks, `t_smooth.py` 8 checks) on the newer Chromium; `tools/shot.py` shots
  of the stone-uses scene look right in both looks. With `.venv` moved away, `play.py` exits
  with the setup message instead of a traceback.
- Open: nothing.

## 2026-10-23: export from claude.ai to a local repository (Claude, in claude.ai)

Request: split the source into small files, deduplicate and tidy a little, test fully, and
write the notes, docs and agent instructions for local development.

- **Source split.**
  - `core.js` (3,688 lines), `main.js` (2,012) and `render.js` (938) became 52 files in
    `src/sim/`, `src/render/` and `src/ui/`, each with a header line. The build order lives in
    `src/manifest.json`.
  - Checked that the joined code was identical to before, apart from comments and blank lines.
  - Every headless suite gave the same output, line for line, before and after.
  - Screenshots of two scenes, in both looks, were pixel-identical before and after
    (`tools/shot.py --freeze`).
- **Tidying, behaviour unchanged.**
  - One scan loop in `simMove` instead of two mirrored copies.
  - A `drawHazardRims` helper replaced 3 copies.
  - The belt and filter shader code shares `beltLineSD`, `bandRidge` and `bandColor`.
  - Removed unused `refundCost`, `wakeArea`, `brushFor`, and the `PARTS` / `PART_BY_ID` aliases.
  - The suites share `suite()` (check / done) from the harness, and failures now set the exit
    code.
- **Tests.**
  - Six broken old scripts were deleted; they used functions removed long ago.
  - Experiments and reports moved to `tools/experiments/`.
  - `t_v2` / `t_v4` / `t_v9` were renamed `t_basics` / `t_world` / `t_conveyors`.
  - Added `test/run_all.py`.
- **Tools.**
  - `tools/shot.py`: repeatable screenshots of a scene set up in JavaScript.
  - `tools/where.py`: maps built-page line numbers back to source files.
  - `tools/map.py`: the project map's index.
  - `tools/scenes/`: example scenes.
  - The game skips cursor previews when `GAME.hidePreviews` is set, for clean screenshots.
- **World files outside claude.ai.**
  - Pause → Export world file only worked inside claude.ai, through its downloads hook.
  - It now falls back to an ordinary browser download (`exportWorld` in `src/ui/saves.js`).
  - `test/play.py` checks that an exported file downloads and imports back on the title screen
    as the same world.
- **Build.** `build.py` writes `dist/index.html` (a full page), `dist/deepgrain-foundry.html`
  (the same without the document wrapper, as the claude.ai artifact used) and
  `dist/bundle-lines.json`.
- **Docs.** `AGENTS.md`, `CLAUDE.md`, `README.md`, `NOTES.md`, `docs/PROJECT_MAP.md`,
  `docs/DESIGN.md`, `docs/TESTING.md`, `docs/systems/*`, `docs/history/requests.md`.
- **Tests run.** All 11 headless suites pass, as do `test/play.py` (101 checks) and
  `test/t_smooth.py` (8 checks).
- **Open.**
  - The owner reports that big piles on belts still only move at the top (NOTES.md); not yet
    reproduced.
  - The game's last claude.ai artifact was version 24.