an endless marble-race generator in the browser; the url is the seed, so every track can be shared https://src.rest/cat/marbles/ 2026-09-30T20:46:48-06:00 2178fd3d2b5ca8ab549e2609f0eadc5153710bb9 Dami 2026-09-30T20:46:48-06:00 Remove the unsupported account_id field so Wrangler can deploy the complete site to the existing authenticated Pages project. b20b16518311f9eb4d573b3653af0c904aba2647 Dami 2026-09-30T20:42:01-06:00 Build browser JS and native physics together, then package static files and asset packs into ignored output/site. Save the existing Pages target in Wrangler configuration and document the deployment workflow and controls. Record completed UX, collision and camera verification: 230 tests passed, full type checks and build passed, and browser checks reported no errors. cfae4700bcf09cb8c7fd5928ce2cf8a37b5a5611 Dami 2026-09-30T20:41:32-06:00 Keep chosen wheel zoom through idle time, respawns and retargeting. Pair obstruction hits on the same mesh/instance and distinguish open foreground shells from volumes containing the lens. Wide shots tolerate foreground occlusion and ease through bends while retaining lens clearance. Detect teleports from consecutive subject positions rather than camera lag. Add eight geometry regressions; verify continuous motion at 30/60/120 fps. On e0b75o4y7s, an 80-second wide-camera comparison reduced moves over two units per frame from 21 to zero. bb28b82cc536eda960692c8eef76c9c6943f34e8 Dami 2026-09-30T20:41:24-06:00 Use a dedicated illustrated title screen with racer count and optional seed. Keep Classic loop, steering and diagnostics in Simulation options; name the Modern and Legacy steering choices explicitly. Remove redundant subtitles and the duplicate top-bar wordmark. Add visible camera, pause, fullscreen, share and settings controls with native keyboard behavior, responsive layouts and clear loading failure recovery. Hide hero/leader selection and its shortcut in solo or decorative-peer runs. 9784c75b23512b64a832ae2b4ec9e7d92619dbd3 Dami 2026-09-30T20:41:06-06:00 Split discontinuous braid ribbons and pack independent collision strips without joining triangles. Restrict static contacts to each racer checkpoint and adjacent checkpoints so rotated older segments cannot block the track. Preserve peer contacts and ghost behavior; export the rebuilt shim API. Regression checks cover disconnected topology, ramp/helix overlaps and both icy-braid lanes. Native browser soaks cleared the recorded failures; two drop-cup recoveries remain documented. 6e38293cbab70130b367cac05a0adce902505bd1 Dami 2026-09-30T20:40:33-06:00 Attach source window, family, zone and lane metadata to dense rail stations. Remap placement indices after filtered jump/reorientation meshes and carry provenance through concatenation. Record bounded recovery and LOD telemetry with generated-window exposure counts for diagnosing stalled pieces. Add provenance regressions and preserve the earlier recovery sample notes. a8237c65e1c2260f12cdc98937f338c2f53a6756 Dami 2026-09-30T20:40:16-06:00 Keep a separate committed sampling anchor so a moving ribbon head does not prevent slow motion from adding trail points. Reset the anchor on warps, fades and explicit clears. Covered by a slow-motion regression. 7e7a85f65097b1cb7e73fc3b5b18448127938bd1 Dami 2026-08-26T15:57:14-06:00 Keep endless orbit attached to the streamed subject, fade tunnel shells around the lens, and scope camera occlusion to authored track pieces. Make the title menu and gallery usable on small screens, preserve exact seeds and native input behavior, and hide inactive HUD chrome. Add constant-cost boost trails, dispose marble/WASM resources across regeneration, and cover trail and physics teardown regressions. Verified with lint, type-check, build, and 214 passing tests. 62919dbc41c478b3801086a082584cd42b8503a4 Dami 2026-08-10T04:10:51-06:00 The reported bug ("the outer shell of marbles sometimes gets a blue tint at some locations or angles") was the BOOST GLOW, not the glass. On ignition the racer wrote emissive 0x2255ff @0.9 to every material it owned — including the transmission SHELL. A transmission material adds emissive on top of the refracted core rather than blending with it, so that flooded the whole sphere with one flat colour; the same saturated blue on the CORE at 0.9 then swamped each racer's identity. Result: every boosting marble rendered the same periwinkle, which is why the tint came and went with location (boost corridors) and took the whole pack at once. Now: the shell never emits, and the glow is the racer's OWN colour pushed 30% toward white at 0.55 — the marble lights up from inside its glass and stays itself. Same law at pack scale (instance colour brightens instead of lerping 3/4 to a shared blue). Also, found while measuring: the shell carried two stacked reflection layers (RoomEnvironment PMREM at 1.2 plus clearcoat 1), spraying ~8% of an environment brighter than this world over every marble pixel, which the zone's mood lighting tinted. Measured on a red-cored marble, the wash lifted G/B from ~4 to ~47. One layer at 0.5 now; core saturation at race distance is up ~40% and the refractive rim and highlights are unchanged. 210 tests pass. fbf553171d5e53703fc34093e7c2c8485ec2955a Dami 2026-08-09T21:51:09-06:00 bedb2e352b57bcfa78ab89fab4243c18275117a6 Dami 2026-08-09T21:49:44-06:00 Client-side accumulating history — generation never knows (streaming law untouched; every mark parents to a segment group and sinks out with its ExitAnim): - SCARS: hard marble-static knocks (contact events ≥ 4.5 u/s) leave dark scuff marks at the hit point, sized by impact speed, capped 30 per segment, shared geometry/material, excluded from the fit box. - ICE SCUFFS: hard braking on icy stations drops pale translucent streak marks (rate-limited 0.5s per racer, deterministic — sim state only). - TIP-OVERS: a hard knock within 0.8u of trackside furniture (fans, banners, bumper mushrooms) tips it ~74 degrees away from the shove, eased over ~0.5s — and it STAYS fallen until the segment sinks. A fallen fan keeps spinning: the wreck is alive. - Telemetry: race JSON gains standing-mark count; scar bookkeeping cleans with segment disposal. NO audio (user rule). 210 green, check clean, goldens untouched (pure presentation). 11ae0ef5c39a001d0a8f858fc6050b1687f6b006 Dami 2026-08-09T21:46:49-06:00 New wasm surface, no audio (user rule — audio stays parked): - shim.c: hit events enabled on every marble shape (userData = idx+1; statics stay silent and resolve as −1), world hitEventThreshold 1.5 u/s (real knocks, not rolling contact); ph_hits flattens the step's b3ContactHitEvents to [idxA, idxB, approachSpeed, px, py, pz] × 6 floats — pull-based and deterministic. Rebuilt via build_physics.sh. - engine.ts hits(): reusable heap-view readback (HITS_CAP 256). - endless.ts consumes per step INSIDE the step callback (events clear next step): marble-marble hits arm the TEMPER trigger (real contacts now; the 2.5-radii proximity proxy stays as the sub-threshold fallback), and hard marble-static knocks (≥4.5 u/s) queue pendingScars — the M25 world-memory feed. - Headless test: head-on pair reports exactly the expected pair above threshold; hard drop reports a static knock. 210 green, check clean, goldens untouched (client+shim only). 6c1fa38aedfe3b3ef1a0cfb861598040fcd2aa9c Dami 2026-08-09T21:41:32-06:00 StyleField v2 — the under-exploited seam gets its identity draw: - style.ts TEMPERAMENTS (appended draw — finite streams byte-identical, genVersion untouched): balanced (common), treacherous (icy/gravity/ spur + spooky/barren/winter islands), soaring (jump/sprint/helix + tropical/meadow), tangled (braids/spurs/reorients + abstract/grove), rolling (slopes/corner-drops + rocky/autumn), serene (hazards damped + meadow/forest/grove). - The endless planner's clause draw now composes climate × temperament (world.ts composeClauseBias; path.ts takes one merged record); the island biome draw leans with the temperament too — a seed's sky, hazards and islands cohere into one character. - Corpus law test: treacherous seeds run measurably icier/heavier than serene ones; ≥4 temperaments emerge over 60 seeds. - Pin cycle #3 of the round: jump→grav-scan-2, propeller→grav-scan-0, spur@1→grav-scan-22, 180 reorient back to grav-scan-33, 3-chain → grav-scan-32@0-2 (one seed now pins open fork AND full chain), gust bank→wind-scan-9@4 (bank run-gap relaxed ≤4: climate biases spread gusts thinner than the M19a adjacency law assumed — an island's ledge row is the bank unit). 209 green, check clean, goldens re-pinned. 39a9911f5fa1c5a6d58fb7425b8d2d76c10d420f Dami 2026-08-09T21:34:08-06:00 Two braids per segment + 3-segment open chains — sub-packs now live on separate lanes for up to 60 cells: - path.ts: braid gate braids.length < 2 (an open fork is necessarily last); continuations may STAY OPEN (0.35) when the cap affords the full middle segment + a minimal merge — the alt lane rides an ENTIRE segment boundary-to-boundary (L2 = SEGMENT_TARGET, no merge, open state re-emitted with accumulated cells). OPEN_BRAID_CAP 34 → 60: fork segment + one full open middle + merge segment is the longest legal chain, still comfortably co-live under KEEP_SEGMENTS. - world.ts: altPath concatenates EVERY lane's loop (was last-wins — a two-braid segment lost its first lane's channel proxies). - proxies.ts: chains and merge windows pair IN PATH ORDER (find-first broke the moment a cont chain and a fresh fork shared a segment — the closed braid lost both junction composites; caught by the contract suite on braid-scan-0). - Client lane carry is transitive through braidContOf (0→1→2), so a rider's fork-drawn lane survives BOTH boundaries. - Tests: chain-walking V3 contract (transitive resolution, ≤3 segments, cap law, port continuity at every hop, stay-open middles emerge); altPath containment law (concatenated lanes); pinned 3-segment chain grav-scan-15 with both-lane drive soak through the whole network. 208 green, check clean, endless goldens re-pinned. 291336ca2fd71bcafcaa71b5e65417321aa5dc8c Dami 2026-08-09T21:25:22-06:00 One rule, five systems couple: riding a wake REFILLS the boost tank (DRAFT_REFILL 1/14 at full depth on top of base regen 1/12 — a committed wake rider roughly doubles its recharge; burns never sip the tow). - boostStep gains draftDepth (appended param, decision-table rows); decide() feeds (draft−1)/DRAFT_GAIN; riders keep their kinematic fake (no fuel at LOD distance). - Emergence: pilots with wake appetite line up for the economy, hoard, and detonate at attack range — cascading overtakes nobody scripted. - Telemetry: race JSON gains maxTrain + trains (s-consecutive drafting chains ≥3) so the trains are measurable in soaks. - All client-side: goldens byte-identical, scenario pins untouched. 206 green, check clean. 7e734b734e74ea8d5796e68ad0c2f71e5c42ac5f Dami 2026-08-09T21:22:14-06:00 drawZone was always seed-pure — so the generator legally reads the SAME climate the sky presents. Systems that never felt each other now couple: weather → surface → pilot brake styles → boost economy. - src/gen/climate.ts: MOODS + BIOME_MOOD_BIAS + drawZone moved verbatim from client atmosphere.ts (draw SHAPE unchanged — skies are part of every shared seed); segmentClimate(seed, i) = the BASE zone draw (the biome lean stays presentation-only: islands are planned inside the zone, so the base draw is the only self-consistent generation input). ZONE_SEGMENTS lives here now; atmosphere.ts re-exports for the client. - path.ts MOOD_CLAUSE_BIAS: snowdrift grows icy patches (x2.5) and damps sprints; goldhour soars (jump x1.6, sprint x1.4, icy x0.5); ember sprints + corner-drops; abyss pulls gravity x1.8; midnight forks (braid x1.5, spur x1.4); violet spirals (helix x1.7, reorient x1.3). One weighted draw either way — stream shape identical, outcomes lean. - furniture.ts: ember storms gust harder and rise (gust x1.4, updraft x1.4); snow muffles gusts (x0.7). Threshold multipliers only. - Corpus law test: snowdrift icy-rate > 1.5x goldhour; goldhour outjumps snowdrift — the coupling is measurable, not cosmetic. - THE pin cycle: the climate bias moved nearly every endless pin. Re-scanned + re-pinned: reorients grav-scan-20(+90)/grav-scan-13(180), helix jump-scan-14@0, spurs grav-scan-3@0/grav-scan-36@1 (grav-scan-18 rejected: a 180 twist upstream of its spur made a hostile scenario), gust bank wide-scan-30@8, tagged OPEN braid grav-scan-17@1 (boost-alt vs icy-main across a segment boundary), bump/vortex corpora widened. 206 green, check clean, endless goldens re-pinned, genVersion unchanged (finite untouched). 416d6ab4e8c96bec8d9705859ff7da66797301dd Dami 2026-08-09T21:07:18-06:00 'Islands are mostly just one type with a color phenotype plus decor' — now the island BODY itself is a vocabulary: - MASS_FAMILIES in islands.ts: 'cliff' (nature voxels, full shaping vocabulary — untouched code path) vs 'block' families (Platformer Kit voxels, measured: clean 1.0 full / 0.5 low courses, bottom-center, cosmetic 0.082 lip overhang, INTEGRATED tops — no cap tile). - BiomeSpec.massFamily: winter is BUILT OF SNOW now (block-snow body, tinted-cliff phenotype retired); new 'grove' biome (block-grass body, forest decor). Ten biomes total. - proxies blockHeight knows platformer courses (1x1 boxes per course, lip is cosmetic); islandRender treats block masses as occluders with value-jitter-only coloring (pre-textured tops must not be hue-tinted). - Platformer Kit imported + joined DECOR_PACKS / GOLDEN_PACKS / test manifests. - MEASURED RULINGS on the record: Space Kit terrain_* is a SURFACE kit (flat top + 0.5 skirts, hollow underside, off-origin authoring) — wrong for void islands seen from below, deferred again; block corner rim-rounding awaits a mesh-orientation measure. - Corpus: grove+winter block islands emerge across biome scan; 21.9k block voxels / 30 seeds. 205 green, endless goldens re-pinned, genVersion unchanged. 3b979a648b28c0ad022a69e44977f0d0fb89898e Dami 2026-08-09T20:41:43-06:00 1abfbcfa2ee94e3884c5b119d2560237cdb8e1a3 Dami 2026-08-09T20:41:17-06:00 'Forking currently exists as a singular generative pass only. It should allow for unresolved forks to resolve in later generation up to a point.' — the braid clause may now leave its fork OPEN at the segment boundary; the successor segment MUST open with the continuation and merge within OPEN_BRAID_CAP total cells (bounded irresolution). - path.ts: HeadState.openBraid carries {side, offset, cells, laneMods, wanderer} across the boundary; tryBraidSide draws an open coin (new child stream — zero draw perturbation; the pinned braids simply drew closed) and runs the braid flush to the boundary (its straight tail IS the seam margin; profile keeps no return budget); planSegment resolves an inherited fork FIRST via tryBraidContinuation (longest-fits-first, profile walks the carried offset home, V1 merge at a pinned straight). The alt lane's own boundary corridor is checked with the hasFlatContinuation CONTRACT pose (the next cell past the seam — the bug that silently killed every open attempt in the first cut), and the successor-probe gate is what makes leaving a fork open safe: it only commits when the continuation actually plans. - world.ts: fork mutation only on non-cont braids, merge only on non-open; alt chains build headless/tailless per case and hang contBraid/openBraid flags; EndlessSegment.braidContOf names the fork segment; lane-tag lip margins apply only at ends that ARE junctions. - proxies.ts: a continuation chain's carrier window keeps its plain channel RIBBON (no fork mouth cut into an ordinary wall) and no fork composite; the merge composite emits as ever. - Client: railFor + braidLaneFor + the rider hook carry the FORK segment's per-racer lane draw across the boundary (one draw per BRAID — no re-draw, no teleport-flip mid-braid); rival mirroring keys on the same label. - Tests: V3 contract (open resolves next segment exactly once, braidContOf, side/laneMods carry, cap law, alt-lane PORT CONTINUITY across the seam <1e-6); closed contract scoped to closed braids; lip law knows boundary ends; drive() soak on grav-scan-32 crosses the boundary on BOTH lanes and merges home. Corpus: 136 braids / 14 open / 0 planner failures over 160 seeds x 8 segments. 205 green, check clean, endless goldens re-pinned. b82e133b3c8b151f300cbb23c2ff29c4de94422d Dami 2026-08-09T20:22:25-06:00 Floating ground props read as set-dressing that forgot gravity. The void vocabulary is now things that plausibly float: - GENERIC: floating island SHARDS — torn-off islets (cliff course(s) + grass cap + ragged quarter underside), every piece sharing ONE float so the islet bobs and drifts as a body; plus space debris (meteors, crystals). - Biome tables purged of ground props (pirate sand rocks, holiday rock piles, autumn sand rocks OUT); ships, the ghost, the crypt and the snowflakes stay. - Shard clearance covers the whole islet (cap and underside), not just its base. - Test: floating-ground-prop ban (nature ids in the void must be shard components), capless-grass ban, shared-float/scale shard coherence. 203 green; endless goldens re-pinned. 2657265edb30939aae7e5cfd81bc7205072f4dbf Dami 2026-08-09T20:18:56-06:00 Finite-mode set-piece breadth, one batched genVersion shift (g491fc4c7a685 → g555803244402), endless goldens' segment hashes byte-identical (only the embedded __version moved — the supports gate held): - features.ts: a SECOND split-merge diamond on another run (own child stream, 0.5) — two lane-choice events per lap. The chicane pair from the design was ruled out by MEASUREMENT (mutated rings top out at ~9-cell straight runs; the pair needs 11) and adapted honestly. - stubs.ts: STUB GARDEN — 2-3 clustered sidings sharing one cap style plus a trackside fan facing the ring beyond the middle cap; with the M19a loop wind backport the garden literally breathes on passing marbles. Appended child stream. - scenery.ts: TWO-TIER ROTUNDA — measured: open and solid curve-wide forms are both exactly 0.5 tall with identical footprints, so a solid ring is a plinth the open ring stacks on flush. Appended child draw. - supports.ts: planStack gains an extended pattern pool (three-band taper four→two→single, pedestal) GATED to loop-mode call sites (assemble !open, takeable falsework) — the endless viaduct hash space is untouched and the endless golden test is the tripwire. Both new patterns obey the containment law (narrower always on wider). - Tests: scenery-ring law extended to rotunda tiers; pinned scenarios loop-scan-981 / loop-scan-20 / loop-scan-33. golden_tracks.json regenerated once. 202 green, check clean. eb4271346ab5c2112dfc25c36041e7611c8fb6b9 Dami 2026-08-09T20:07:46-06:00 Lane mods + wanderer character — one intentional reshuffle, tag-only law: - path.ts: Braid.laneMods drawn from a new child stream (0.5 none / 0.25 icy-alt+boost-main / 0.25 inverse); wanderer character (own child, 0.35) leans the SAME profile range outward via weighted picks (pick and weighted both consume one draw — non-wanderer braids stay byte-identical; the wanderer bias IS the reshuffle) and always carries the alt-boost compensation, so the long way around is never a strict loser. Breadth stays capped at B (no new twist-window veto pressure). - TAG-ONLY LAW: lane mods never join plan.mods (realize's exclude rules must not see them — geometry is sacred), never pin windows. world.ts stamps main-lane windowMods over the braid INTERIOR only (>=2 cells from fork and merge — the junction-lip law in friction form) and hangs per-part zone tags on the alt chain (LoopSegment.alt.zones, chain ends untagged); tagged windows' placements carry zone for the recolor. - world.ts realize generalized over plan.braids (bare braid-wfc label for braid 0 — existing seeds byte-identical; >1 braid is now a one-line planner change away, with its own pin cycle). - rail.ts: alt-chain stations OVERLAY the tag on the fork's centering ({center, icy}/{center, speed}) — untagged chains (diamonds, spurs, V1 braids) byte-identical. - Presentation: BOOST_PIPE amber joins ICE_PIPE/GRAVITY_PIPE — boost zones (sprints AND boost lanes) now recolor: zones must read. - Tests: corpus lip-law scan (plan.mods zones exempt — a jump's boost approach may brush the merge), wanderer-compensation law, alt-zones alignment, rail overlay signature on the pinned grav-scan-6@4 braid; drive() soak clears the tagged braid on BOTH lanes. New pins: grav-scan-6@4 icy/wanderer, grav-scan-10@3 boost lane; braid pins grav-scan-17/38 held. 199 green, check clean, endless goldens re-pinned, genVersion g491fc4c7a685 unchanged. fb13e4f3dbbd8f5858a573c0801eae576979a7ff Dami 2026-08-09T19:55:40-06:00 Directed seed-pure relationships expressed ONLY through existing actuators — no new physics surface, zero gen-side movement (goldens and scenario pins untouched; the A0 lock proves the trait appends clean). - peersCore rivalFor(seed, i, n): own label rival/${i}, fixed draw order (has 0.5 → offset 1-16 → valence 70/30 grudge → intensity 0.5-1). DIRECTED — i hunts its target, the target is oblivious (a symmetric pair would need a shared hash, banned by the rng law). Resolved once at pack build. - Traits appended AFTER M20a's (milestone order IS the draw order): brakeStyle 0.8-1.35 scales the PLANNING decel in the brake sweep (6·1.35 < A_BRAKE_MAX 10 — a late-braker can always shed what it planned; the band inequality is test-pinned), temper 0-1 opens a ~3s hot window on contact (temperStep mirrors boostStep; the contact proxy is a neighbor inside 2.5 radii — the shim has no contact events). - decide() expression, levers only: grudge target ahead reads closer (gap/(1+intensity) into an unchanged boostStep) and ally never triggers an attack; separation asymmetry via the REPORTED neighbor bravery (−0.1·intensity, autopilot compare untouched); defensive blocking stands in the chasing rival's line (offset ≤0.25 < CORRIDOR_DEAD, engage/release hysteresis + slew, strict priority overtake > block > wander on the one offset actuator); braid lane mirroring adopts the rival's own label-pure draw (grudge follows, ally opposes) behind a 10u proximity gate. All transient state resets on demote/promote/respawn; rivalry expresses in the live band only. - HUD: PlaceHud gains one quiet token `#3/128 · R#7`, flag-gated (?rival=1), Kenney-frame styling inherited — default decided at review. Telemetry: hero rival + grudge-ignition counter, spread/burst from M20a. - Tests: temper decision table, brakeStyle envelope behavior + band pins, rivalFor determinism/independence/coverage laws. 196 green, check clean, goldens byte-identical. 5c4fd0d74aee4e3302f11d16fd601f1e4cfda09d Dami 2026-08-09T19:49:38-06:00 The pack interacts through the air and the line, not just contact/fuel — all inside the existing envelope law, zero gen-side movement (both golden files byte-identical; scenario pins untouched). - A0 FIRST: tests/personality_snapshot.json + the appended-draw lock — pilotFor/racerIdentity/braidLaneAlt fixture captured PRE-change; the test walks only fixture keys, so appends stay green and any in-stream insert goes red. The fixture regenerates only from a known-good commit. - Traits appended after efficiency: wake 0.6-1.4 (draft appetite), lanePref 0.35-0.65 (braid threshold), chir ±1 (passing side). - draftFactor(gap, wake): wake window 1.2-5.0u behind the same-lane leader, ≤ +10%, threaded into autopilotForce as targetMult exactly like boostMult — wishspeed stays min(target, vSafe): drafting into a corner brakes identically (envelope test proves the cap). - decide(): captures the gap-ahead racer; same-lane gate on braids; r.draft at decision cadence (1-step-lag contract); overtake commitment holds pilotEff.offset = chir·0.27 (< CORRIDOR_DEAD 0.3 — the corridor spring never fights it) while a burn is an attack, engage <3.5u / release >5u hysteresis + 0.4s slew on the single offset actuator. State resets on demote/promote/respawn. - Riders: rank stamped in the sort pass; riderUpdate applies the same draft law CAPPED at +8% (chained riders must never outrun the hero-anchored stream) and draws the same per-racer braid lane at segment entry, so the fork split ratio holds at LOD distance. - braidLaneAlt(seed,idx,seg,pref): SAME label, SAME single draw, moving threshold only — historical coin == pref 0.5 (test-pinned). - Telemetry: race JSON gains respawnBurst (2s window), spreadP10P90 (the liveliness tripwire), per-racer draft. - Tests: draft window/falloff/cap tables, envelope safety, lane monotonicity + per-racer independence, trait bands, fuzz harness runs the draft law with client parity + asserts a wake window opened; determinism run-twice still byte-equal. 193 green, check clean. 74731b242648c8051e0cd701703bffcc99804114 Dami 2026-08-09T19:40:37-06:00 New generative system, zero perturbation of existing streams (new labeled family "void-decor"): - gen/endless/voidDecor.ts: per-segment clusters (0-2, expected ~1) of 1-5 pieces hung off the segment's own path samples — mid band 8-16u (slightly below the line), far band 18-40u scaled into fog silhouettes; biome-flavored tables (ships off tropical coasts, a ghost + crypt in spooky air, snowflakes in winter, sand shoals in autumn) + generic rock/crystal shoals; rare far-band EVENTS ~0.08 (meteor trains, a passing craft/satellite) with pure-in-index quiet-after suppression. Sites keep >=6u of every path sample (main+alt), island cells, and any occupancy claim at their column. Space Kit imported (153 pieces). - EndlessSegment.voidDecor is a SEPARATE field: the proxy path structurally cannot see it — non-collision by construction (contract test: proxies byte-count-identical with the field stripped). - Placement.float {amp, period, phase, drift} — the wind model: canon hashes it; genVersion untouched (finite never emits it). - client/voidRender.ts: instanced flat in the frame-correct segment group (spawn/exit waves see direct children; twist halves split per anchor), castShadow off, no occluder flag; float pass in finishFrame gated on the spawn wave (writer-overlap-free) with a 2s bob envelope + capped segment-local drift; fit box skips voidDecor-flagged meshes (far shoals must not inflate shadow/focus). - Tests: same-seed determinism, 30x8 structural scan (budget, clearance, resolution, float-on-every-piece, multi-pack emergence), proxy blindness contract, event rarity + never-consecutive law. 188 green, check clean. Endless goldens re-pinned (canon + content); genVersion g491fc4c7a685 unchanged. ad5a11148863dca8b4d2d345465fe3982e7cbdd9 Dami 2026-08-09T19:31:55-06:00 - Import pipeline: Holiday/Graveyard/Pirate/Survival kits packed (one colormap each); DEFAULT_PACKS grows to six. - decor.ts: SETTINGS rows carry pack + finite flags; pickSetting draws ONLY the historical nature trio (the genVersion guard — finite stream byte-identical, golden_tracks untouched); settingByName reaches the four new regex tables (patterns + scale ranges measured per pack: pirate is ~2.5-3x nature scale, survival props tiny). - islands.ts: BiomeSpec capId fully-qualified, per-biome landmark lists + landmarkScale (default = the historical nature statue list, draws byte-identical); four appended rows (w 1.2/0.8/1.2/1.0 — nature keeps ~70%, a themed island is an event). Caps reuse the 1x1 nature ground tile; measured: no themed pack ships a clean 1x1 cap. - Client: assets.ts loads a DECOR_PACKS list + manifests map; catalogs (client + goldenCatalog + all gen-side tests) carry all five packs — settings resolve against the catalog, so pack presence shapes output. - Presentation: islandRender biome tints (snow-creep courses, graveyard gloom, sandstone strata, ochre mottling; snow/sand cap repaints; pinned foliage leans); atmosphere drawZone takes a biome bias (zone's dominant island biome multiplies mood weights — winter leans snowdrift etc., same rng draw shape, presentation-only). - Tests: finite-trio lock, per-row manifest resolution, 40-seed themed scan (placement resolution + cap 1x1 law). 184 green, check clean. - Endless goldens re-pinned (biome reshuffle, island-local); scenario pins all held; genVersion g491fc4c7a685 unchanged. 1c0d8ec2b0e44a5bf85de6f00ef329b835a5f442 Dami 2026-08-09T19:20:26-06:00 Wind & hazard vocabulary, all appended draws (zero rng-stream reshuffle): - Placement.wind gains pulse {period,duty,phase} + swirl; canon hashes both. - fanField: pulseGain (cosine-eased duty windows on SIM time), swirl volumes (tangential force around a vertical column), FanField.setTime. - furniture: gust walls upgrade cross fans in place (strength 14, duty .45, period 2.4-3.6s); vortex = table fan at the lip, eddy over the channel; coordinateGusts post-pass (rng-free) phase-locks banks into travelling waves and sets same-host L/R pairs anti-phase. - endless Session: windT sim clock drives the field (advanceTime-pure). - Loop backport: FanField over loop fan placements; kinematic marble takes tangential wind into speed and lateral wind into a clamped channel sway. - Tests: pulse/swirl/coordination laws, corpus emergence (16x10), gust scenario rules + pinned bank (wide-scan-13), windy drive() soak through the bank with bit-identical replay. 181 green, check clean. - Endless goldens re-pinned (content adds); genVersion unchanged. 38d562aafc37964742db8c997283d7e10eff4e7d Dami 2026-08-09T18:40:55-06:00 Full remediation of docs/collision-mesh-audit.md (119/119 pieces, re-audited clean same day — docs/collision-mesh-reaudit.md), built on headless GLB measurement (gltf-transform over marble-kit.glb, pattern tools/measure_ports.ts). Measured truths, grid units: wall inner face 0.287, rim floor+0.25, floor bore ROUND r 0.25 in straight-hole and BOTH end-hole flavors, wide inner wall 0.575, split-double walls at branch ±0.287. - PARITY (T7): segmentProxies() is the ONE channel builder for production, the ?gallery=1 bench, and the sims (the bench had been omitting wide scales — a diagnostic-only wiring defect that falsified five audit rows). gapSegmentsOf lives with it; rail.ts re-exports. Bench asserts finite mesh bounds, frames empty plots safely, and ghosts cross-over's host mesh (a deliberate noMesh phantom). - APERTURES (T2/T5): PieceDef.holes → LoopSegment.holes opens a real bore in the ribbon (lip decks, side slivers, walls, under-deck guide ring). Drop-stacks get an authored measured compound: annulus deck, octagonal cup (r 1.0 world) with the entry arch open, octagonal shaft, receiver cup opened toward the true exit yaw, pan law kept. - WIDTH (T1): widening is an additive wall SHIFT (0.576 world/side), not the old x2.43 section scale that put wide walls outside the visible +-1.7 mesh; the bench passes def-authored port widths so an isolated funnel tapers honestly. - RIM (T6): CHANNEL_PROFILE tops at 0.62 over floor (measured rim 0.5 + 0.12 solver cap), wall face 0.62 lateral; junction composite walls follow the same law (diamond/spur/braid tightened ~0.4/side to the measured split-double faces). - TURNS (T3): arc stations bounded by outer-boundary chord sagitta (0.008 world) instead of a fixed angle; crease rounding stays as the documented kicker-prevention law. - JUNCTIONS (T2): the cross cell emits ONE authored 4-way composite (deck + solid corner posts, all four port arches open) from the mesh visit; the phantom is silent — two same-height route ribbons used to wall each other's route off. QUANTITATIVE CI (tests/collision_metrics_test.ts + collision_width_test): ring floor/lateral/rim budgets (3009 rings, worst floor err 0.041 at slant-d's intentional crest rounding), RENDER-derived directed distance (40k GLB channel verts within 0.2 of the collision surface; corridor excludes port-face bands — shared boundaries the neighbor's ring covers — and drop shells' cosmetic corners), bore-openness probes (13), drop compound probes (12), diamond compound probes (3), cross probes, and per-variant coverage accounting (rings 113 / render-dist 101 / probes 17 / paired-by-design 1; bench-build failures fail loudly). The CI caught three real defects during bring-up: approach ribbons overhanging drop bores (lip-trimmed), missing bore side slivers, and thin cross corner Ls (now solid posts). 174 tests green, deno task check clean. d32bc7844122c0aa1810d61f2f1524fe6c9ba911 Dami 2026-08-09T17:09:43-06:00 60c3c011a4f108bec9312e652ab76388af5b0c9f Dami 2026-08-09T17:09:22-06:00