minimap.js drew every lot in the plan from the first frame (493 shops on the synthetic, one M
press, against 5,330 m / 4.8 game-days of walking to earn it). It now draws what has been walked
past AND FRAMES ITSELF TO IT — a shop lot goes 0.83x1.11 px -> 29.4x39.2 px on bowral_real, 35.3x.
ZERO DRAWS, measured as an A/B: two port-isolated no-store servers, HEAD vs treatment, the same 15
stations x 4 yaws compared as JSON before either sweep runs. Draw calls IDENTICAL on 60/60 samples
at noon and at night; with the gig layer off both sides (the venue queue's async rigs were the only
noise) NIGHT is identical in draws AND triangles. Map mode cannot cost a draw structurally: the
frame loop never calls composer.render() in the map branch.
- discovery.js (new): a 25 m PROXIMITY PROBE, not onChunkBuilt (R=2 => the outermost live chunk
corner is 181 m out), on a plan-derived spatial hash — 1.09-1.37 us a probe, every 6th street
frame = every 0.46 m at WALK. Plus a side test: you must be in FRONT of the shopfront. Control
with a red arm: walking the synthetic's two service laneways, radius-only reveals 93 shops, the
shipped probe 59 — 34 refused through the back wall; on the arcade it refuses 0 of 17.
- save.js: optional `known` block under the delta law, keyed by TOWN (re-derived: 23 caches x 3
seeds, shop/edge ids and lot geometry all identical; the synthetic alone keys @<seed>). Bounded
by the plan (max 493 shops / 2,593 edges) at 8,192, 64 towns FIFO, ints only. Whole corpus fully
learned = 137,547 B (2.6% of quota); 482 B after a 500 m outing. Eight reject arms demonstrated.
- classic BY CONSTRUCTION: createMinimap(plan, null) is the pre-R39 map and the file has no flag
test at all. ?classic=1 / ?fog=0 / ?game=0 map canvases are hash-identical to HEAD's.
- Lane A's address layer consumed (getTownCache + createAddresses): street names along walked
streets, deduped; the synthetic gets district labels from the same line. No town-type branch.
FILED, NOT FIXED: index.html:384 (Lane D patronage door points) and :450 (the R32 spawn) use
(-sin ry, -cos ry) — the BACK of the building. buildings.js:410/583-597, dbg.js:53 and minimap.js
all use local +Z. Measured against each lot's own frontEdge the +sin point is 0.00 m from the kerb
and the shipped one 15.25 m the wrong side, 493/493 and 72/72 shops. At the shipped spawn 0 of the
shops within 25 m are in front of you on all four towns tested; at the corrected one, all of them.
Two characters, but it moves the default boot's opening pose mid-round — Fable's call.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
THE TABLE (LANE_C_PUB §10, the deliverable): what a player can actually do in each of
the 12 shop types, counted in 48 rooms per type. `record` digs (4.6 bins/room) and is the
only room where anything can be sold. `stall` digs — in 27 of 48 rooms; the other 21 (44%)
have NO crate, a measured correction to the round's binding fact #5. The other ten types
offer looking. `?stock=real` is opt-in, so the default boot has zero `buyMesh` anywhere;
garments/snacks/magazines/treasures are unbuyable in any mode; `interactable:true` on
cases/fridges/returns/arcade is read by nothing.
THE MECHANISM: `rummageBin` — a castored galvanised dump tub, rim 0.76 m, heaped with
stickered bric-a-brac and a hand-lettered "EVERYTHING $2" card. 72 rooms: placed 72/72,
reachable 72/72 (0.77–1.15 m), 0 carves, pathOK 72/72, deterministic 72/72, and the rest
of the room byte-identical to pre-R39 72/72 (the spec is appended last). Armed end-to-end:
aim → dig → 16 offers → BUY → bin depletes. Default: 0 bins, dig never opens.
THE CONTENTS ARE NOT SHIPPED. `o.dig` defaults false, so the tub is furniture and the
economy is untouched; `opts.fittingOpts.rummageBin.dig` (or `?opshopdig=1`) arms it. Held
for John with the numbers: 44 record + 36 stall vs 147 op shops across 23 towns.
THE BUDGET: `C-progress.md:841`'s GLB-on 344 is R9-era and stale — today's tree measured
319 before I touched it, so the margin was 31, not six. The bin costs +5 GLB-off / +10
GLB-on at opshop/hall; every other type byte-identical. Margin now 21. Also: `slot.pool`
stops the scatter fill minting a material per item, so 12 goods cost less than 9 did.
FILED, NOT FIXED: Lane E's bookshelf GLB ships transmission:1, which triggers three.js's
transmission pre-pass and issues every draw in the room twice — opshop/hall 191→104,
book/hall 101→51. Roughly half the interior budget. E's asset, so E/F rule.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- The magpie's white, restated exactly. The wing-bar occlusion claim is the strong one and it now
carries its footprint proof: the bar's (x 0.174-0.316, z 0.005-0.055) lies ENTIRELY inside the
wing's (x 0-0.34, z -0.085-0.065) with the bar 3 mm above it, so from below the wing hides it.
The nape box tops out at 0.067 against a head surface at 0.0728 (I had written 0.065, and had
the sliver on the wrong side). And the flank argument is now the honest one: nape and rump are
14 mm and 20 mm slabs on a 166 mm bird = 1.3 px and 2 px of white at 64 px.
- THE CLASSIC ARGUMENT FOR THE ARCADE IS INVERTED relative to R38 and I nearly carried the wrong
one over. R38's ground palette is classic-safe BY ABSENCE (character.js returns frozen literals
for a null town key; classic has none). The arcade district exists ONLY on the synthetic, and the
synthetic IS the classic boot — so the kit lands squarely on the covenanted town and needs an
EXPLICIT flag test, ground.js's VERGE shape, not the by-absence one.
- arcade_measure.mjs now also counts chunks: arcade lots span exactly 2 (0,0 and 0,1) on all five
seeds, so a per-chunk arcade class is +1 draw in 2 chunks and +0 in the other 23.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- LANE_E_NOTES §39: RECOMMENDATION TAKE THE GLB, on three findings in the order they decide it
(80.5% of drawn frames are PERCHED and the perch is the pose E's asset already is; E reads pied
at 64/48/32 px and under a 25% smear while B reads as a dark aeroplane; and B's white is
geometrically buried — the wing bar sits 3 mm ABOVE a horizontal wing plane, so from underneath,
which is where the player always is, the black wing occludes its own white flash).
- Three asks filed for Lane B, none of them edits: magpie.js's two stale 154-tri comments (:9, :123
against a measured 182); the wing bar's side (y 0.048 -> ~0.042 is +0 tris and fixes ?classic and
?noassets whatever Fable rules); and two GROUND_FALLBACK entries so the arcade is not grey
without assets.
- The arcade measured, with the awning-post finding: two post rows 1.00 m apart down a 5.00 m lane.
- AUDIT §Round 39 §0 makes the sizing law a STANDING INSTRUCTION with two new clauses: draws are
counted per CLASS (town-wide InstancedMesh +1 vs per-chunk +1 x 25 live chunks), and for a SKIN
the budget is the SLOT, not bytes (36 facade-atlas slots, 22 used, 14 free — past 36 skins.js
silently REUSES one).
- pipeline/skin_sheet.py: wrap-preview + measured seam error, the check R38's stretched grounds
needed before the harvest instead of after.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The contract, published for B and C in LANE_A_NOTES §39 (first commit, per the half-day rule).
createAddresses(plan, cache|null) → streetOf(edgeId) · localityOf(shopId) · cohort(pred)
· streets() · stats()
New pure module web/js/citygen/address.js: ZERO imports, no THREE, no fetch, no DOM, no module
state, no plan mutation. Two suppliers, ONE consumer contract — a real town fills `label` from the
cache's named ways, the synthetic fills it from district.kind + block. Consumers must never branch
on town type; that constraint is the design.
MEASURED (Fable's binding facts re-derived independently, all three land exactly):
· 19,132 road ways, 17,835 named (93.2%)
· median plan-edge-sample → nearest named way 0.03–0.90 m (worst katoomba)
· 1,192 / 1,219 corpus shops resolve to a street name (97.8%); 29,865/30,986 edges named (96.4%)
· all 27 unresolved shops front a way OSM genuinely leaves unnamed (adelaide 21 arcade) — null is
the honest answer and the module returns it rather than borrowing a neighbour's name
TWO DEPARTURES FROM THE SYNTHESIS, both with numbers:
· the shift is recovered EXACTLY (node↔waypoint lattice intersection), not voted for. The proposed
centroid-align + nearest-waypoint vote returns the WRONG shift on 5 of 23 towns (braddon,
fremantle, hobart, northbridge, westend) — island culling drags the plan centroid off the
cache's. A wrong shift renames every street in the town. Exact on 23/23, nothing to tune.
· the resolver samples FIVE points along an edge, not the midpoint — a midpoint cannot tell a
street from the street that crosses it.
THE ONE-LINE UPGRADE: TAKEN. plan_osm.js writes norm.shift = {shx,shz} into the normalization log,
reaching the caller only via the existing opts.report sink — never onto the plan. It is a
cross-check, not a dependency: selfcheck compares plan_osm's published shift against address.js's
independently recovered one and fires on a 0.01 m disagreement.
GATES (+240 checks), each proven to fire on a broken world:
· 0 wrong street names on 884 shop-bearing edges, against an INDEPENDENT way-membership resolver
(857 agree, 0 disagree). Corpus-wide 4/30,986 disagree — all 2D-stacked ways, none carrying a
shop — PINNED at 4, not asserted > 0.
· CONTROL: with roads[].name stripped (exactly the pre-change state) the layer resolves 0 streets.
· CONTROL: no cache ⇒ supplier 'district', visible in stats(), never a silent wrong name.
· createAddresses does not mutate the plan — asserted byte-for-byte, per town.
· tolerance 8 m, and the finding that sets it: correctness SATURATES AT 6 m (857 agreeing frontage
names at 6, 8, 12, 16 and 24 m). Every metre past 6 buys only a name borrowed from an unnamed
way; it can never buy a correction.
Additive: getTownCache(key) — index.html registers the cache and drops its reference, so roads[]
was unreachable downstream. Read-only, no new state, no shell change.
GOLDENS: NOTHING MOVED, ZERO RE-PINS. selfcheck 157,407 → 157,647 ALL GREEN, fingerprint 0x5f76e76.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
THE RECONCILIATION. The brief's naive sum (292 + washing 1 + magpie 1 = 294) is wrong, in our
favour. The pinned worst-DRAW frame is a NIGHT frame and both of wave 1's draw-spending layers are
structurally zero at night: the washing comes in (count 0), the magpie roosts (count 0). Wave 1's
one draw is spent in daylight, where the walked peak sits ten draws lower.
Two pinned trees, assets symlinked to the live tree on BOTH sides, 802 stations at 2 m down the
main spine at yaw 0 (the worst bearing), citizen sim and tram LIVE, 1280x720:
NIGHT worst draws 292 -> 292 (+0) tris 120,145 -> 122,879
NOON worst draws 282 -> 283 (+1) tris 99,872 -> 103,118
?classic=1 269 / 90,580 unmoved
fitzroy_real 172 / 130,727 unmoved bendigo_real 204 / 131,492 unmoved
(even with ?magpie=1 forced)
The control reproduced R37's pin TO THE BYTE, twice, on a fresh checkout. Every revert is exact:
?yards=0 at night returns all three control numbers; ?yards=0 at NOON returns 282/99,872, which
isolates the noon +1 to the washing and nothing else. MARGIN IS STILL EIGHT DRAWS.
LANE B's "control 346", answered rather than dismissed. As a delta it is the wall-clock ped-spawn
artifact R37 named, and B was right to hide the sim. But 346 > 300 is a claim about the world, and
neither stepwalk (walks the chunks in, photographs a half-dressed street) nor dwell (fills the
street, arrives clean at 27 chunks) can adjudicate it. budget_walk.py gains --method stepsettle:
the audit's fixed-yaw stations, each DRAINED then held N real rAF frames — R+1 dispose residue AND
a fully-acquired crowd in the same frame. Like for like on the same 101 stations, the settle is
worth +3 draws at noon and -9 at night, and the SIGN FLIP is the finding: at night the settled
street retires near rigs to impostors (tris 120,145 -> 100,939) — the crowd goes home, and a
sampler that never waits does not see it happen. The highest number anywhere in the reconciliation
(4 methods x 2 trees x 2 segments x 5 towns) is 292, the pin. NOTHING READ OVER 300.
A CORRECTION TO MY OWN R37 PIN: the corpus TRI worst is 130,727 on fitzroy_real, not 124,793 — and
it reads 130,727 on the CONTROL tree, so it is my R37 station spacing (12-bearing 8 m sweep) versus
this round's 2 m stations, not a wave-1 cost. Station spacing, not settle policy, is the biggest
lever on a walked worst frame. A pin is only as good as its spacing, and mine had not said so.
FOUR GATES, FOUR CONTROLS DEMONSTRATED (smoke_ambient / smoke_character / smoke_magpie /
smoke_yards, wired into flags_check.main()). A gate that cannot fail is not a gate — this round
earned the sentence twice, R37 killing two vacuous gates and B catching its own season mechanism
producing a constant that would have gated green under ?magpie=1.
1.1 audio — 6 sources anchored 63/8/81/32/39/57 (B's numbers exactly); dog AUDIBLE at 5.9 m with
1 play scheduled (for a one-shot, `audible` is only published once plays>0, so one word covers
fetch/decode/gain/schedule/counter); HOURS enforced (sprinkler audible 06:30 -> idle 12:30, same
pose); DAY-OF-WEEK enforced (mower idle Mon -> audible Sat, same pose); PAN is a bearing (yaw
0/90/180/270 -> +0.00 +0.85 -0.00 -0.85); ZERO DRAWS byte-identical to ?ambient=0. THE CONTROL:
a second server whose root is web/ symlink-for-symlink except one patched manifest with
sfx['dog-bark-fence'] renamed -> no-entry, 0 plays, never audible, other five unchanged, 0
console errors. Printed on every run: audible OUTPUT is unverifiable headlessly. Nobody in this
loop has ears.
1.2 character — classic is the frozen v1 literals, all tints null, 5 ground meshes; all 23 shipped
keys matched PER TOWN including the two carrying state:''; the control fires (unknown key ->
matched:false); 21 distinct palettes across 23 towns (one shared palette FAILS by name); and the
two orphans are ON THE GPU — ground-gravel.jpg bound on Castlemaine, ground-reddust.jpg on
Darwin, read off the live material.
1.3 magpie — the latch arm verbatim with its positive control FIRST; day-roll clears it;
localStorage gained NO new key (baseline taken after a game.save() — the first cut blamed the
magpie for v7's own write); the season is 23 distinct positions across 24 towns run UNFORCED,
against the first cut's 24-of-24-at-214, newtown_real in season on day 1, 129/1000 seeds against
a 11.5% design ratio; territory 6 -> 157 synthetic -> bendigo; +1 draw / +182 tris at ONE pose
reused across both boots.
1.4 yards — 181 lots, 181 hoists -> 394 items, 256 active, 0 at night; cost decomposition at one
pose with the sim quiet: box props +0 draws (in fact -2, the 36 `yard` lots used to fall through
to buildShopfront) with +1,936 tris, washing exactly +1 draw / +512 tris; and on real towns it
SKIPS LOUDLY — a ⊘ SCOPE line printing 0 house/yard lots / 0 hoists / mesh never created, beside
the synthetic 181 that makes the zero attributable.
Caught in my own work: the pan arm's first draft used two bearings and read +0.00/-0.00 because
that anchor sat on the forward axis — it would have called a working panner inconclusive. Four
bearings now. The gate's first draft failed the standard the gate exists to enforce.
SUITE: tools/qa.sh --strict --matrix GREEN on the true final tree (74d6257) — 7 passed, 0 failed,
0 warn, 0 skipped; flags_check GREEN 0/0 with all four new gates inside it; MATRIX GREEN 10 towns
x 7 gates, 0 fetches under ?noassets on all ten.
Held: nothing. Recorded not fixed: magpie.js says 154 triangles at :9/:123 and 182 at :42 (measured
182); E's three published GLBs are manifest-and-depot but not under web/assets/models (staged in
pipeline/_normalized, nothing loads them, the validator's contract covers it — not a defect).
NO TAG — wave 1 is content; the epoch tags when it is whole.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A. DEPOT — closed, under John's R22 standing Lane-E authorization (the authority that
shipped sit.glb in R28 and the nine R36 sweep assets). Pushed via the passwordless
tailnet ingress: _published.json 53 -> 56, drift check clean. Then verified the way
this lane verifies — FETCHED BACK OFF THE PUBLIC DEPOT and hashed, not "the upload
said 200": magpie/paling_fence/water_tank all HTTP 200, byte counts exact, sha1
identical to local, 3/3. _r38_results.PENDING_PUBLISH.json renamed, build_manifest
re-run (GLBs 33 -> 36), validate_manifest 0 errors / 0 warnings.
B. THE MAGPIE TINT — done, not offered. The bird shipped as an all-black crow (R38 4d
traced the loss to trellis2_mlx's PBR bake). "Hand-tint the atlas" is not a paint job:
the bake is a smart-UV unwrap of ~100 islands and none of them announces itself as a
nape. So pipeline/tint_atlas.py defines the tint in the MESH's frame and projects it
through the UVs — rasterize every UV triangle to texel->(position,normal), evaluate
soft anatomical windows on the positions (white nape behind a black crown, shoulder
bar on the flanks only so the dorsal midline stays black, rump + tail base stopping
short of the black terminal band), paint keeping the bake's own luminance as shading,
then dilate 5 texels into the UV padding so no black fringe survives bilinear/mip.
3-D masks give one consistent soft edge across every island at once; a 2-D blur could
not (it would bleed island-to-island through the black background).
GEOMETRY UNTOUCHED byte for byte: still 894 tris / 1 mtl / 1 img / 512^2 WebP / no
Draco / 0.155x0.358x0.404 m. Painted luma median 0.84 (p90 0.95) vs the bake's 0.13.
VERIFIED IN THE GAME'S OWN RENDERER: the PUBLISHED file loaded off digalot.fyi through
the vendored GLTFLoader + three.js — 1 mesh, 894 tris, 512x512 texture decoded (which
also proves the PIL-written WebP decodes in a browser).
VERDICT: it reads. Pied and unmistakable at 64 and 48 px, still black-and-white at 32.
Honest caveat: from directly below-and-in-front it stays dark — correct, a real
magpie's bib and belly ARE black. Render: docs/shots/laneE/r38_magpie_tint.png.
Standing after B's 4514cb9: an UPGRADE PATH, not a dependency — B's 182-tri procedural
bird is 4.9x lighter and the right call. Nothing probes for the GLB and nothing should.
C. CORRECTIONS FOLDED INTO THE RECORD so the next asset round does not re-learn them:
skins are 9.7 s not 5.7 s (1024^2, 1.78x the pixels, per-pixel identical) - the reuse
list's "one line each" holds for 5 of 9 (extractGLB keeps only the FIRST material) -
normalize.py's collapse decimator floors at the shell count on TRELLIS output while
bake_lowpoly.py clears it every time - the 489 s outlier at 3.8x median means
throughput plans on a tail, not a mean - furniture.js:30's stale comment is fixed by B.
NEW, and the one worth keeping: an asset's tri budget is tris x THE INSTANCE COUNT THE
PLACEMENT RULE IMPLIES. paling_fence is a good 1,121-tri panel and a rejected asset —
905 instances = 1.01M tris, ~47x over. Ask the consuming lane for the instance count
BEFORE generating anything that tiles, edges or repeats per-lot.
New tools, all on-device and $0: tint_atlas.py, render_views.py (view sheet + --eevee +
roster-consistent --thumb), view_sheet.py (the distance strip no 256px thumbnail can make).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The town is inhabited: 23 towns finally differ, 181 back yards have a Hills Hoist with washing on it,
a magpie owns one street for six weeks of the year, and six point sources put a life off-screen.
MEASURED, walked, not bookmarked — two port-isolated no-store servers (:8472 = `git archive HEAD`,
:8471 = treatment, assets symlinked so E's mid-round wave is identical on both sides), 102 8-metre
spine steps at yaw 0, each frame fully streamed, through the shell's own composer:
worst draws 224 -> 225 (+1)
worst tris 89,502 -> 93,298 (+3,796 of ~75k spare)
sum draws over 102 frames 16,676 -> 16,663 (-13)
The citizen sim and the tram are hidden identically on both sides, and that IS the measurement: with
them live the same step read control 346 vs treatment 232, because peds spawn on wall-clock time.
`?yards=0` reproduces HEAD exactly (224 / 89,502 / 16,676). Isolated at a fixed pose: washing +1 draw
/ +512 tris, magpie +1 draw / +182 tris, every yard box prop +0 draws.
§1.2 THE CHARACTER VECTOR — +0 draws, +0 tris, classic byte-identical BY CONSTRUCTION.
`townCharacter(null)` returns the frozen v1 literals and a null tint keeps the SAME material cache key,
so classic gets the same object, not a copy of it: default-boot ground hashes identical to HEAD
mesh-for-mesh, `?classic=1` hashes unmoved from R37. There is no `if (classic)` in the ground path.
The sky feeds the BASE skyName, never `setSky` (a one-way latch that would have killed R32's dawn dome
on all 23 towns). The vacuous arm is per-town: `newtown_godverse` / `redhill_godverse` carry `state: ''`,
so a per-corpus arm would have dropped exactly those two silently. 21 distinct palettes across 23 towns.
`gravel` and `reddust` were not unwired but UNWIREABLE — their `use` values named no slot ground.js had;
making the slot addressable fixed it at the root, and both are selected now. E's three new grounds
landed mid-round and the upgrade was a name change in one table, which is what procedural-first is for.
§1.4 BACKYARD AUSTRALIA — inside boxMats, +0 draws. Houses inset off their boundaries so there is a
yard; 36 `yard` lots that had rendered as blank shopfront sheds since v1 are now fenced blocks with a
hoist, a shed and often a tank. Collision byte-identical (the lot rect is where the fence stands).
THE WASHING SHIPS, and I wrote the inverted weight rather than dropping it. The shipped top-weighted
mode moves a hem 0.10 cm calm / 0.67 cm gusty; `pegged` gives 4.5 cm / 31 cm — ~52x, and at the hem
instead of the pegs. The `canopy` string is byte-for-byte unchanged, so the gum-tree covenant argument
is untouched. It is ONE TOWN-WIDE InstancedMesh (1 draw at any N, cap 256, proximity-first off the R37
chunk seam) rather than per-chunk, which would have been +1 x 31 live chunks = 4x the whole margin.
Zero on every real town (no hoists ⇒ no mesh) and zero at night.
§1.3 THE MAGPIE — +1 draw on screen, exactly 0 otherwise; 182 tris of 900. Territory scales with the
road graph: 6 on the 6.78 km synthetic town, 157 on bendigo's 188.6 km / 2,593 edges, nearest 120 m
from spawn. The latch sets, the day roll clears it, localStorage gains no key and the save never
mentions it. `swoopPos()` is exported and pure. The hat stays cut.
MY OWN SEASON MECHANISM WAS VACUOUS AND THE SWEEP CAUGHT IT: keyed on citySeed alone it produced a
CONSTANT, because the town key is deliberately off the plan — 24 of 24 towns were 214 days from
swooping season, and it would have gated green because `?magpie=1` forces it. Re-keyed on
(citySeed, townKey): `newtown_real` is in season on arrival on the default seed, marrickville is 8 days
out, and 12.9% of 1,000 seeds put the synthetic default in season against a design ratio of 11.5%.
§1.1 THE AUDIO LAYER — six new mixer layers, zero draws, zero tris, seeded anchors, distance falloff,
AND A StereoPannerNode PER LAYER, so direction is real (measured spread -0.71..+0.85) instead of a
limitation written into the notes. Day-of-week derived locally. All six anchored on the default boot;
3 of 6 report `no-anchor` on real towns rather than being faked onto shopfronts. The falsifiability
control — one manifest key renamed away, served from a third port — makes that source report
`no-entry`, 0 plays, never `audible`, with the other five unchanged and zero console errors.
TWO DEFECTS MY OWN GREEN NUMBERS COULD NOT SEE, both caught by a screenshot: the washing shipped at 256
instances / +1 draw / cap respected and was INVISIBLE behind its own back fence (per-instance peg height,
and a fence taller than the hoist), and the magpie perched on THIN AIR over the road. R37's lesson
relearned at full price. Fence 1.60 / hoist 2.15 measured against the object; territories now snap to
furniture.js's own streetlight rule.
LANE E'S PROPS, ADJUDICATED: hoist + aerial permanently primitive (E's spike killed them, and primitive
is also the +0-draw mechanism). `paling_fence` at 1,121 tris x 905 runs = 1.01M triangles — no. The
`magpie` GLB is HELD FOR THE TINT: it reads as a crow, and a magpie seen for 1.2 s in peripheral vision
IS its white bar. My 182-tri bird already carries it, at a fifth of the triangles and no fetch.
furniture.js's "novelty_record is 26.5k tris" was stale by six epochs — corrected to 8,000.
Zero fetch delta: `?noassets=1` fetches the same 23 assets, same list, on HEAD and on R38.
0 console errors on 10 boots (default, classic, yards=0, washing=0&magpie=0&ambient=0, noassets,
magpie=1, darwin, hobart, bendigo, the patched-manifest control) beyond the known PointerLock ones.
Shots: docs/shots/laneB_r38/. Detail + every number: LANE_B_NOTES §38.
A. PIPELINE FIX: gen_props.py trellis_mac -> trellis2_mlx, contract read from its
manifest.json. The two operators are NOT drop-in on output — trellis2_mlx writes
BOTH candidate.glb (raw) and candidate_pbr.glb (baked), so the old
sorted(rglob('*.glb'))[0] would have silently shipped untextured meshes. Added
_pick_glb() (result.json first) + the bg_remove_local cutout stage, fail-soft.
MEASURED: median 128.0s vs trellis_mac's 245.4s p50 = 1.92x. 5/5 first attempt,
0 retries against the 4-5% budget. One 489s tail outlier, no error — named.
B. AUDIO (1.1) — six PACK entries, +0.86MB, ZERO draws/tris, all originalby
construction: dog-bark-fence, mower-distant, radio-through-window, sprinkler,
roller-door, magpie-carol. Seeded via crc32(name); beds fold seamlessly
(wrap-jump <=0.18x internal step). Verified by spectrogram + band-energy, and
one defect found by LOOKING: the sprinkler's spray masked its own ticks and read
as a hose — rebalanced and re-rendered. Distance is spent in the SPECTRUM, not
the level, because encode()'s loudnorm undoes a quiet render.
C. GROUNDS (1.2) — gen_skins.py had NO ground template; added one with its own
outdoor style prefix (the house STYLE says "warm fluorescent", an interior cue).
4 new 512^2 square tiles. Review's 5.7s is 9.7s here — same per-pixel, mine are
1.78x the pixels. Found: gravel/reddust are selected by nothing partly because
their `use` values name no slot ground.js:177 has.
D. MESHES — 3 shipped (magpie 894 tris, paling_fence 1121, water_tank 500), all
GLB-law clean. 2 FAILED and were DELETED: hills_hoist + tv_aerial. Thin wire
returns as thousands of disconnected fragments, so collapse floors at 250k/89k
AND a remesh coarse enough for 600 tris erases the pole. Lane B's boxMats
primitives cost +0 draws and are strictly better. The magpie's geometry passes
but TRELLIS crushed its white nape/wing to black — traced to the operator's bake
(the cutout and our decimation are both exonerated); flagged, not hidden.
E. REUSE — all 9 idle assets re-measured; the review's "one line each" holds for
only 5. furniture.js's extractGLB keeps only the FIRST material, so park_bench/
coffee_table/box_crate render in material #1 and bench_wood (0 materials) falls
to grey. Nothing was regenerated that already existed.
The 3 GLBs are staged, NOT published: a depot push is outward-facing and had no
in-session authorization. Held at _r38_results.PENDING_PUBLISH.json; publish +
rename + rebuild lands them. Gate green: validate_manifest 0 errors / 0 warnings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
qa.sh --strict --matrix on the true final tree (Lane A e4d7762 + Lane B 47db478/79177bc + Fable
7bd96e5/4a7c0f1 + F 2396885/6afd1ed/f522021), port-isolated on :8975.
SUMMARY 7 passed · 0 failed · 0 warn · 0 skipped — QA GREEN
selfcheck ALL GREEN 157,407/157,407 (was 156,352 — Lane A's +1,055 poster-clearance
assertions, three arms, all green on the ruled street)
manifest 0 errors / 0 warnings; atlas-QA 15 atlases, 15 licence lines shown; depot live
flags 0 fails / 0 WARNS
classic fp 0x5f76e76 · dig INERT at the bin · gig layer wholly absent ·
162 draws / 16,642 tris · fetch surface CLEAN against v2.1 (125 requests:
68 code, 52 same-origin v2.1 assets, 4 depot, 1 NAMED EXCEPTION printed) ·
allow-list control fired (2 injected -> 2 named violations)
default four flags on · 14 gigs / 14 posters / venue geo / tram · gig golden
0xec7a2d39 · 131 draws / 23,382 tris @ venue block · dig REACHABLE naked
kerb all six arms green, control at EXACTLY 14 under ?verge=0
tram all arms + BOTH controls green
R11-R36 audio, gigs, continuity, glance, the R30-35 game gates, djdance, real
crate, tier-2 kill-the-server: all re-green on the ruled street
fig-scale 29 figs across 6 shops in [1.4,2.0] m incl. the R36 DJ-body leg
matrix GREEN, 10 towns x 7 gates; every town byte-equal on re-boot; 0 fetches under
?noassets across all ten; tram verdicts unchanged (fenced on bendigo, fremantle,
katoomba_real — the same three the rebuilt smoke_tram uses for its absence arm)
One caveat kept in the notes rather than buried, so this round's central finding is not undone by
its own suite: the matrix's budget column (69-161 draws) is DBG.shot('venue_night') CLEAN-LOAD
BOOKMARK numbers, and 37.1 is the proof that a bookmark under-reports a walked-and-turned frame by
~120 draws. That column is a regression signal, not the budget of record. Making the matrix
walk-and-turn every town is a real improvement and a real cost — FILED TO THE v8 DEBT LEDGER, not
smuggled into a floor round.
NO TAG. Wave 0 levels the floor; it does not ship a version.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two sessions disagreed by 90 draws and neither was wrong. Worst draw per BEARING, 12 bearings x
102 stations down the same spine, one tree, one box: 0deg 284 · 30deg 248 · 60deg 208 · 90deg 177 ·
120deg 196 · 150deg 214 · 180deg 211 · 210deg 207 · 240deg 178 · 270deg 182 · 300deg 223 ·
330deg 267. 110 draws of spread from the camera direction alone, on identical positions -- and the
peak sits at yaw 0, the DEFAULT ARGUMENT of DBG.teleport(x, z, ry = 0). The city-patterns audit
took the default and landed on the worst bearing (280, reproduced at 282-292); the animal spike
supplied a yaw and landed mid-range (191, reproduced at 205-215). Correct numbers, different
quantities. Lane B's independent R37 walk (276 draws / 99,880 tris) and mine (282 / 99,878) agree
on TRIANGLES TO 2 -- the same frame through two harnesses.
THE PIN, default boot, synthetic golden seed, tree 7bd96e5, walked the spine:
292 draws of 300 (NIGHT, 31 live chunks -- the R+1 dispose window at maximum)
120,145 tris on synthetic; 124,793 on fitzroy_real, the corpus worst
?classic=1: 269 draws / 90,580 tris
margin: EIGHT draws (the audit said "about twenty") and ~75k tris
The worst-draw frame and the worst-tri frame are NOT the same frame. Real towns are draw-cheap and
tri-dear; the synthetic town is the opposite. The audit's "99k-124k" tri range is confirmed and
pinned at its top.
Also measured: viewport is IRRELEVANT (1280x720 and 2560x1440 identical to the byte -- culling is
frustum-shaped, not pixel-shaped) · the R+1 dispose residue is worth +19 draws, not the 60-80
folklore (the rest is place and bearing) · CitizenSim's per-frame rig-acquisition budget means any
one-render-per-station sampler photographs a half-dressed street · ?r=3 measures 307 draws and
BREACHES the 300 gate through an ungated override (filed, not fixed) · ?shadows=1 costs +38 draws
and +38% tris · Lane B's kerb change independently confirmed at +0 draws / -8 tris at the worst
frame, by a harness B did not write.
THE SHARED INSTANCE CAP: 256, one InstancedMesh, town-wide. Measured +1 draw at every N from 32 to
2048, and N x tris/instance ALWAYS submitted (the instance bounding sphere spans the town). Draws
are not the constraint; triangles are. 256 x a 2-4 tri cell = 512-1,024 tris = 0.5% of budget;
0.01 ms/frame to rewrite every matrix. Binding companions: <=24 tris/instance at the cap, and spend
it PROXIMITY-FIRST -- anchor supply runs from synthetic's 246 poles over 6.8 km to bendigo_real's
7,325 over 188 km, so 256 sprayed town-wide there is one instance per 730 m.
Honest limit stated where the number is quoted: headless SwiftShader cannot change a draw or
triangle count (both are counted in JS, and the viewport control proves it), but frame TIME remains
unmeasured on any real GPU and no v8 proposal may claim a millisecond figure.
Re-pinned in LANE_F_NOTES 37; the four other docs that assert the number are flagged with what each
line should now say rather than edited in another author's file. README flags table gains ?verge=0
(Lane B's control, F's file) and a row naming the ?r=3 breach.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The measurement tables (+0 draws at every walked step on two towns, tris DOWN not up, the
raycast surface table, the ?verge=0 and classic hash pairs, the §0.4 hook proofs), plus
three things worth keeping:
- the defect my own numbers could not see (every intersection painted shut; all metrics
green, only the picture wrong) and why ?verge=0 + a screenshot caught it;
- a full-frame pixel A/B is not an A/B while a live sim is in the frame — hiding
everything but `ground` is what made the classic covenant measurable;
- the market plaza reads worse under the wider footpath, and that is the STANDING item
awaiting John's amendment, not new damage from §0.1.
The audit's claim against my lane, re-measured and CONFIRMED exactly: on the default
boot (synthetic 20261990, gigs default-ON) there are 14 posters, all 14 stand inside
e.width/2 — the bitumen ground.js paints — and 0 stand inside the ruled kerb
roadWidth/2. One refinement it did not make: 12 are spine posters at poleOffset=6m,
8m inside the paint (genuinely mid-road); 2 are frontage posters clipping the corridor
edge by 0.06m (FACADE_PROUD past their venue's facade plane).
WHY the gate was green: my placer seats posters via poleOffset, which is DERIVED FROM
roadWidth, and my selfcheck asserted clearance against roadWidth/2. Gate and subject
were the same function agreeing with itself — the R27 cross-convention trap recurring
inside the very file whose header explains it. It never read a number that came from
Lane B, so no change to where the road is painted could make it fail.
The correction (selfcheck.js, districtInvariants), three arms:
1. no poster on a carriageway — measured through vergeBand(e)[0], the published law
itself, the same call B's ground now consumes, so gate and renderer move together
2. every SPINE poster stands INSIDE some edge's vergeBand — the outer bound the old
check never had ("not on the road" also passed for a poster in a paddock).
Frontage exempt BY MEASUREMENT: 18/108 gig plans have a frontage poster past every
corridor edge (worst 3.98m, launceston_real) via plan_osm's lot setback.
Spine violations over the same corpus: 0/108.
3. THE CONTROL — count posters standing on bitumen under the PRE-RULING ground
(?verge=0, road quad = full e.width) and assert it is NOT ZERO on every gig plan
(min 3 / max 212 / never 0 across 510 plans), pinned at EXACTLY 14 on the default
boot. A control allowed to drift toward zero is how a gate goes vacuous.
The replica is pinned to B's FLAG, not to a line number that will drift.
FALSIFIABILITY PROVEN. Simulated ?verge=0 honestly — vergeBand switched to the
pre-ruling reading, poleOffset pinned to the true band so the PLACER stays
byte-identical (it is a render flag: paint changes, plan does not; fingerprint and
every gig golden green in both runs, so the gate's reading is the only variable):
corrected (R37) gate → FAIL, 1020 failures (2 arms x 510 plans)
pre-R37 (HEAD) gate → ALL GREEN 156,352/156,352 — structurally blind
GOLDENS: NOTHING MOVED, ZERO RE-PINS. 0x5f76e76, GIG_GOLDEN 0xec7a2d39, 25 base town
goldens and 24 real-town gig goldens all untouched — correct and predicted, since the
only generator-adjacent file changed is the selfcheck, which validates output and never
produces it. 156,352 -> 157,407 checks (+1,055), 10.3s.
Corridor law CONFIRMED for Lane B, semantics unchanged, contract published in full
(LANE_A_NOTES §37) with four hazards flagged loudly while B consumes it in parallel:
roadWidth and edge.width are both FULL widths but one is carriageway and one corridor
(that naming IS why this bug existed); lane's band is NOT the literal [2,2] the round
instructions state (2 synthetic edges are, 56 real lane edges are [4,4] — branch on
outer<=inner); arcade is all footpath with coincident kerbs to skip; the inward
extension never inverts (0 of 31,039 edges). Two charter numbers corrected: the "9m
band / 18m of every main" figures are synthetic-only — on every real town a main is
24m => 7m band, and the corpus's dominant edge is side@12 at 22,415/31,039 (72%) with
a 3m footpath.
FILED, NOT FIXED — v8 debt ledger, four entries (LANE_A_NOTES §37): D1 chunkIndex
(two impls, THREE divergences, header comment promises a swap that would break three
ways — owner B) · D2 isOpen (three impls + an ad-hoc fourth site; A's canonical one has
ZERO importers; the live shell copy treats no-hours as CLOSED where A's law says OPEN —
latent until v8 adds an hours-less lot — owner F) · D3 two shop registries (B's missing
all three venue kinds => buildings.js:344 renders op-shop signage on every gig venue
since R13 — owner B) · D4 NEW, mine: the poster gate's inner bound is NEAREST-EDGE and
an arcade (kerb 0) camouflages a real carriageway — 8 real-town plans hide a spine
poster up to 1.78m inside a main, 0 on synthetic. Cost measured decision-grade:
synthetic 0/4 plans move, 11/26 town gig goldens re-pin, poster counts unchanged.
Deferred to R38 on purpose — it moves plan output for a reason that is NOT the kerb
ruling, and the "did the kerb ruling move a golden?" answer must stay a clean no.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
§36.4 held -> ruled -> RESOLVED: Fable ruled 'classic keeps its people', D fixed
it in rigs.js alone (2fd4c1c: base-17 restored, the five DJ bodies append behind
djs=dance riding the shell's existing !CLASSIC signal; classic's 129-citizen
census byte-identical to a b339402^ shadow tree). F's full-suite re-gate on the
true final tree (run 3): SUMMARY 7 passed / 0 failed / 0 warn / 0 skipped, QA
GREEN, MATRIX GREEN 10 towns x 7 gates, flags harness 0 fails / 0 WARNS — the
smoke_djdance warn flipped to '?classic=1: zero DJ-body fetches'; default keeps
all 22 bodies, the organic gig-crowd proof, and the DJ figure-scale statures
exactly. Honest record: re-gate runs 1-2 died to a box-level fork famine
(~10,600 leaked m3panel pmset|tail orphans, fixed at source) — starvation
artifacts, not gates; noted in §36.5.
The five-holds law end-to-end in one round: held by F, ruled by Fable, fixed by
the owning lane, re-proven by the gate that held it. Nothing improvised.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
b339402 appended the 5 trellis'd DJ bodies to PED_NAMES.normal unconditionally, so
?classic=1 fetched them and its seeded crowd pool widened 17->22 (F's held §36.4 warn).
Ruled option (b): the five gate OUT of classic, mirroring the sit/look/dance !CLASSIC
clip gates. classic is the v2 town INCLUDING ITS PEOPLE; the DJs walk every other boot.
MECHANISM (rigs.js only — zero shell edits): PED_NAMES.normal restored to the covenanted
base 17 (exact pre-July-20 list); the five move to PED_NAMES.djs; loadPedFleet gains
djs = dance (default tracks the existing dance flag, which the shell already passes as
!CLASSIC). Roster = djs ? [...normal, ...djs] : normal — the five APPEND, so no base
slot renumbers: default boot's 22-list is byte-identical to the pre-ruling list and
classic's 17-list is byte-identical to pre-b339402. pickRig maps through pool.length,
so 17 restores the exact pre-July-20 crowd mapping, not merely the width.
MEASURED (fresh contexts, port-isolated no-store servers, seed 20261990):
- ?classic=1: ZERO of the 5 DJ GLBs fetched (21 ped-GLB fetches = 17+2 bodies +
walk+idle) · pool 17 normal / 19 all · 129-citizen census all base-17 u comical,
zero DJ · 0 console errors.
- default: all 5 fetched · pool 22/24 · 37 DJ-bodied street peds · F's organic gig
proof preserved (pinned seed's venue crowd includes dj_phrtt_01 + woman_dj_02) ·
0 console errors.
- IDENTITY vs pre-July-20: classic booted on a shadow tree with b339402^ rigs.js —
full 129-citizen (id, pedRoll, pedIndex, pedName) census BYTE-IDENTICAL to the
gated tree's classic boot. Same crowd, ped for ped, as before July 20.
Known side effect (noted §36): citizens_test.html passes no opts -> its dev harness
shows 17+2; not a covenanted surface, {djs:true} restores it if wanted.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
README: the v7.0 section (the loop dig->guide->want->travel->list->paper, the
delta-law machinery, the classic covenant) + the morning-paper manual row.
V7_THE_GAME.md: the five charter risks get their retired-or-carried verdicts,
v4-style, written by F under the single-lane-close ruling: economy RETIRED-
structural (the named bot verifier carried to the live tier) · schema drift
RETIRED (three optional widenings, zero migrations) · re-emit identity CARRIED
(never fired; owned-item stability check unbuilt — the next re-emit round owns
it) · scope gravity RETIRED (the stage gates held) · v6 collision RETIRED (the
2 a.m. commits were the one contact, swept R36w1 + gated R36w2).
LANE_F_NOTES §36 + F-progress: the close record — both playtests, E's three
flags closed by measurement, the DJ-roster hold, the full gate numbers
(7 passed / 0 failed / 0 warn / 0 skipped · flags 0 fails / 1 warn (deliberate)
· selfcheck 156,352/156,352 · figure-scale 29 figs incl. the 5 DJ bodies ·
MATRIX GREEN 10 towns x 7 gates · R32-35 smokes re-green on the final tree).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
046668e + b339402 provenance'd: 4 dance clips are ROKOKO free mocap (263 pack,
Mixamo-skeleton export — the brief's Mixamo guess corrected by measurement; new
licence-table zone) and there are FIVE trellis DJ bodies, not three. All 9
sha1-identical to their 90sDJsim originals; chain recorded to source FBXs.
Structural: clips mesh-free 68n/198t/52-joint skin+IBM, 0 NaN, bind 65/66 (only
mixamorigReference unbound); bodies 3998-4000 tris, 1024 webp, no Draco, ped law
PASS. FLAG -> F w2: bodies' bind-pose ~0.99 units vs roster ~175 (cm) — the
figure-scale gate's expected number; rigs.js:163 absorbs it at spawn. No file
touched, nothing fixed in place.
Depot: published per R22 standing authorization via tailnet ingress,
_published.json 44->53, all 9 fetched back from the public depot byte-identical.
validate_manifest 0/0, publish --verify 53/53 live.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
qa.sh --strict --matrix: 7 passed · 0 failed · 0 warn · 0 skipped, MATRIX 10 towns green. New
STRICT smoke_first_five: the fresh game hunts a roster artist and the splash names hunt + cash
· the beats walked the REAL paths (② at the crate, ③ on the buy, the send-off on the ⚖) · a
returning save gets no tutorial · a second fresh browser hunts the SAME artist (seeded, not
random) · ?game=0 shows nothing · 0 console errors. THE CROSS-ROUND CATCH: the seeded want
moved a fresh-boot invariant (wants 1, not 0) and the R33 gate went red on the FULL run while
green solo — amended to a baseline snapshot, lesson filed in §35 (a feature that moves a boot
invariant must sweep every gate that assumed it; full runs exist for exactly this). Money shot:
docs/shots/v7_beta/first_five_splash.png. Remaining for v7.0: the tour + docs freeze + THE TAG
— which waits for John's hands (the alpha precedent).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
qa.sh --strict --matrix: 7 passed · 0 failed · 0 warn · 0 skipped, MATRIX 10 towns green. New
STRICT smoke_godbay: consigned through the real crate DOM · ONE-HAMMER determinism (a frozen
pre-sleep export slept twice lands byte-identical — the seeded gavel, measured not trusted) ·
honest bounded resolution (net $19 ≤ 2×$24×0.88) · the wake toast carried the morning paper ·
the ride to melbourne hammered a listing on arrival · ?game=0 has no crate to list from · 0
console errors. Money shots in docs/shots/v7_beta/ (the dawn paper + the ⚖ crate). The full
v7 loop now closes end to end: dig the gem → keeper cash now or the gavel tomorrow → the
wantlist points at another town → the ride costs a day and hammers your listings on arrival.
Remaining for the v7.0 tag: THE FIRST FIVE MINUTES (onboarding), the tour, docs freeze — and
John's hands on the whole thing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
qa.sh --strict --matrix on the final tree: 7 passed · 0 failed · 0 warn · 0 skipped, MATRIX 10
towns green. New STRICT smoke_wantlist_travel: ☆→★ through the real card (the save carried
{THE VERANDAHS — 'Servo at Midnight'}) · a keeper muttered the hunted artist within 2 shops
('saw one in Castlemaine just last week') · buying the wanted sleeve cleared the hunt · the
synthetic→melbourne ride cost EXACTLY one day with $179 and the find riding along · the
same-town reboot was free (the control discriminates) · the ride home charged again (day 3) ·
?game=0 built zero want DOM and muttered nothing · 0 console errors. Matrix note: town_matrix
shares one browser across towns so travel drifts the day as it walks — safe because
determinism double-boots the SAME town (free by the same-town rule); reasoned before landing.
Money shots: docs/shots/v7_beta/rumor_and_wantlist.png (the keeper's mutter over the toy
shop's 70s wallpaper + THE WANTLIST open in the crate) + want_card.png. README manual row.
v7.0-beta systems are now ALL BUILT — what remains of v7: Esc-in-person (John's hands), then
the v7.0 close (GodBay offline auctions, THE FIRST FIVE MINUTES, the tour, the tag).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
qa.sh --strict --matrix on the final tree: 7 passed · 0 failed · 0 warn · 0 skipped, MATRIX 10
towns green. New STRICT gate smoke_bible: band law over EVERY bin off the digOffers surface
(80 offers/5 bins), the buy-everything ledger runs on THE GEM'S BIN so the finder-margin leg is
non-vacuous (bin1: 2 gems, grail asking $19 vs keeper $30; ledger closes to the cent — Σask
$333 → Σoffer $135, gems alone +$21, every non-gem strictly negative), scarcity measured
(emptied bin 0/16 on re-open AND across a full reload; 16 pulls in the save) with the ?game=0
negative control re-opening 16/16 (the detector discriminates). sell_routing amended from the
alpha inequality (credited < paid) to the exact formula (credited == min(basis−1, basis//2),
basis = bandLow-or-pricePaid) — the fence moved from the receipt to the band, which is what
charter law #2 always said; it passed on a banded standard item (+$4 on basis $8). _at_bin
grows an idx param (back-compatible). Money shots in docs/shots/v7_beta/ (the guide card + the
crate at the counter). LANE_F_NOTES §32 (incl. the JING5 fresh-checkout trap: gitignored
web/assets/models synced from m3ultra — also un-warned the validator's 15 cross-pack checks).
README manual: the guide/gem/scarcity rows. Remaining beta charter queued in the round doc:
wantlist + rumors · travel costs a day · Esc-in-person · then GodBay.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
R28's gate had no subject and SKIPPED with its reason printed (look.glb was published, verified,
sha1-proven off the depot — and nothing bound it). D bound it in wave 0. The gate is now
permanent, and flags_check is GREEN:
✓ default boot: look.glb BOUND — 6.5s, 64 tracks
✓ the glance REACHES THE GAME — peak 46 of 396 peds glancing · 0 console errors
✓ ?classic=1: look.glb NEVER FETCHED — the zero-fetch-delta covenant holds
✓ ?classic=1: glance inert at the source — 0 of 269 peds, AND the rng stream still draws
That last line is the design, and it is D's cleverness: THE DETERMINISM IS IN THE DRAW, NOT THE
OUTCOME. D takes from a dedicated glanceRng stream always — so the seeded sequence is identical
whether or not a clip exists — and only flips the flag when fleet.lookClip is real. Under
?classic the roll still happens, the flag never sets, nothing is fetched, the covenant stays
byte-identical. A behaviour added without touching a hash.
The gate asserts the clip's ABSENCE under classic, not merely "0 glancing" — because "0
glancing" alone passes on a fleet that failed to load at all. Present-on-default + absent-under-
classic is the pair a broken fleet cannot satisfy.
F'S OWN GATE CRASHED THE SUITE FIRST: the subject check used pg.goto() on a .glb, which starts a
download, aborts the navigation, and took flags_check down before one assertion ran. A gate that
DIES looking for its subject is worse than one that skips — it reports nothing at all and takes
the suite with it. Now it reads the disk. The vacuous-gate law has a sibling: a gate must survive
its own subject being absent.
No tag — wave 0 rides Spike 2's close, gated on John's footage.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The spike reaches the game. E's look.glb (mesh-free, 65 joints, 6.5s 'Look') is bound
through the fleet gate; street peds now glance around during window-shop stops. `lean`
is absent and never existed (E's R28 measurement). Goldens untouched — clips sit outside
every plan hash; the synthetic anchor stays R27's amended 0x5f76e76.
WIRED: loadPedFleet(base, {sit, look}) -> fleet.lookClip, the exact twin of the R16 sit
gate; the shell passes look:!CLASSIC so ?classic fetches NOTHING (a raw fetch would have
breached the zero-fetch-delta covenant — E flagged it, the gate is the answer).
makeActor gains lookA, played only on setLooking(true) — crossfaded (0.3s), not snapped
like sit, since idle->idle-variant reads naturally and there's no replant for a fade to
fight. sim.js gains a dedicated `glance` rng stream, rolled ALWAYS (deterministic) but
only flipped when a clip exists => classic inert at the source. Mutually exclusive with
bench-sit (sit rolls first at 35%, glance takes 40% of the rest). RigPool threads
lookClip; release snaps the glance off so no rig is pooled mid-look.
THE FINDING — my own prediction was wrong, and the measurement caught it. I wrote that a
STANDING clip needs no foot replant ("no hip descent to recover — that trap was sit's
alone"). Measured over the full loop: idle keeps the lowest bone within 4cm of the ground
(-0.04..0.00); look.glb ranges -0.009..+0.205 — the ped HOVERS up to 20.5cm, drifting
through the loop. Same trap, different coat: _rotOnly drops the Hips POSITION track for
rig-height independence, so the clip's authored weight-shift dip can't move the hips and
comes out as the FEET RISING. Sit's one-time replant can't fix it — the float varies per
frame.
FIX: plantFeet() — the per-frame twin of sit's replant. Reset to bind plant -> measure
posed feet -> drop inner by float/scale. ABSOLUTE (reset->measure->correct), never
incremental, so it cannot drift. Reproduces the authored motion exactly: the hips dip
relative to the ground instead of the feet leaving it. Scale-aware like sit's. Foot bones
cached once; cost paid only by glancing peds, on frames their mixer ran (both mixer sites
— the drummer's post-mix lean taught that coupling rule).
PROOFS: foot float 0.205m -> 0.000m (pinned); head yaw sweep span 1.735 (-0.85..+0.89) so
the glance still animates; ?classic — no look fetch, no Look action on any rig, glance
never flips (0); identity signature unchanged, no glance term; sit AND glance never
simultaneous (0); 0 NaN; goldens 0x5f76e76 untouched. A glancing ped is NOT tagged
seated/posed — standing pose at standing stature, so F's no-giants gate needs no arm.
D->F: the pose gate HAS ITS SUBJECT — un-SKIP it. Fresh ?localdepot=1 boot shows a ped
glancing on the footpath, head sweeping, feet planted.
Spike 2 readiness: plantFeet() is the general machinery the brief asked for ("data, not
new code") — riffle and wall-lean should land as clip + roll; the lean's wall-contact
placement is the only genuinely new part.
FLAGGED (A's file, A rests): selfcheck.js:301 still says "the synthetic golden above stays
0x3fa36874" while line 267 pins the amended 0x5f76e76. A stale comment, not a rule — but
the same species as the epoch's five holds, so: flagged, not touched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
qa.sh --strict --matrix 7 passed / 0 failed / 0 warn. selfcheck ALL GREEN 156,352/156,352 —
UNCHANGED, confirming the brief's prediction that nothing this round moves a golden. MATRIX
GREEN 10 towns x 7 gates.
F'S POSE GATE HAS NO SUBJECT — SKIPPED with the reason printed, per the vacuous-gate law. The
brief asked for "lean/look poses through the scale gate" and "leaning peds in old frames". There
are no new poses in the game:
· `lean` DOES NOT EXIST. E measured instead of trusting the list — the stash's 34 clips have no
lean, and find -iname "*lean*" across ultra returns nothing. Brief and kickoff both called it
already-cached. The premise was half wrong, and E found it the only way these are ever found.
· `look.glb` is shipped, verified, published, sha1-checked by fetching the bytes BACK off the
public depot — AND NOTHING BINDS IT. loadPedFleet() loads walk, idle, sit. No look. E's
handoff said it plainly ("-> D: gate it behind an opt the way rigs.js gates sit"); D's entire
R28 commit is LANE_D_NOTES.md, 48 lines, docs only. The run order said "when E's clips land:
D (#1 wiring)" — the log says D committed BEFORE E shipped.
Nobody was careless; the wave order was. Every check E ran was real and passed, and none of them
could see that no consumer exists. A GATE ON THE ARTIFACT CANNOT SEE THE WIRE — fifth appearance
of the R17 pattern this epoch and the cheapest yet to have missed, because the artifact was
perfect.
v5.1 SHIPS ANYWAY, AND DOES NOT CLAIM THE SPIKE. An unbound clip is inert: nothing is broken,
the game is unchanged and correct. This is not R18's "known breach", it is an unfinished item,
and holding a polish tag over an inert file would be precious. What ships is the sweep:
C's audioEmitter (its own v3.0 note had specified it wrong on both halves — "the panner was
inert"), and TWO backlog items KILLED WITH THE MEASUREMENT — B's awning ripple and D's
instrument LOD, the oldest tri item on the books. They don't leave the backlog "deferred"; they
leave it gone, with the number that killed them.
Filed for R29 (D, one opt): rigs.js gains `look` the way it gates `sit`; F passes it from the
shell. Then the pose gate has a subject and F runs it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
R28 ledger #3. My own idea, parked since R7. Brief stated acceptance only, so I scoped it.
THE TRAP MY OWN v3.0 NOTE SET: it said "userData.audioEmitter = musicKey so Lane B can
position/spatialise the bed". Both halves wrong. (1) Shape: the engine holds no reference to
the room group and would traverse a scene it cannot reach — emitters belong on room.audio as
data, riding the EXISTING playInterior() call (no lifecycle, no traverse, no dispose hook).
(2) Method: "spatialise" invites a PannerNode. NOTHING in this tree sets ctx.listener (zero
hits), so a panner resolves against a listener frozen at (0,0,0) — which, because rooms are
built room-local at the origin, is the room's own CENTRE, forever. Measured, OfflineAudioContext,
real pub stage bed:
panner, listener never set (the engine as it SHIPS): RMS 0.106905 door / 0.106905 stage
-> ratio 1.0000, bit-identical
panner + listener tracking the player: RMS 0.055665 / 0.500000 -> 8.98x
It plays, throws nothing, warns nothing, and never changes by one sample as you cross the room
— while failing at the only thing it exists to do. "Routed through a panner?" yes. "Audible in
a venue?" yes. "In a shop?" yes. All green, all worthless.
-> ASSERT THE DELTA (gain at door != gain at stage), NEVER THE PRESENCE. Vacuous-gate law, 4th
application — and this one had no author to blame but me.
The house idiom was already right: tram/spill/gig-spill are all hand-rolled distance-gains vs
PROCITY.camera.position. No listener, no panner.
SHIPPED (C's files only):
- room.audio.emitters {stage?, counter?} — room-local, 3dp, 18/18 same-seed deep-equal. `r`
(reach) is the ROOM DIAGONAL, not a constant: measured diagonals 9.81-15.63m, and scaling to
the room holds door->stage proximity in a 0.42-0.59 band from a pokey band_room to a hall pub.
`floor` is the room-filling residue, never 0 — a PA fills a pub; what changes as you walk in
is the BALANCE, not the presence. Swing 2.65-4.52 dB (John R20: subtle changes are great).
- THE TILL RINGS FROM THE COUNTER — audible positioning in a normal shop with NO ENGINE CHANGE,
via B's already-public playSfx(key,{gain}): C does its own distance math, hands B a scalar.
Measured through the real BUY button, real record shop: 0.9489 @0.60m vs 0.6436 @4.19m =
3.37 dB; negative control (no emitters) fires at exactly 1.0. sfx-till.ogg has been on disk
since 15 Jul and had NEVER ONCE PLAYED. Routing, not content — as the brief said.
- ?mute=1 now actually silences the crate riffle (below).
WHY THE VENUE HALF IS AN ASK: a one-shot resolves its gain once, at fire time, from a position C
already knows. A bed must be re-gained every frame inside the engine's update() loop — B's file.
Structural, not political. B has a ~12-line paste-ready diff in the house idiom; audio.js
untouched. The letter of "consulted not modified" could have been met by reaching through B's
exposed state.layers diag to rewire its graph from outside — that backdoor is strictly worse
than the edit it avoids. Refused.
TWO BUGS FOUND:
- F DROPS EMITTERS ON THE ONE NIGHT THEY MATTER. interior_mode.js:289 rebuilds the spec on a gig
night ({musicKey: ra.gigKey, toneKey: WALLA_KEY}) and every unnamed field dies. Measured on a
real gig-night pub: quietNight_emittersSurvive TRUE, gigNight_emittersSurvive FALSE. Silent,
fail-soft, no error. -> F: don't just spread `ra`; pass emitters DELIBERATELY, gig-night only —
on a quiet night the radio + room-tone genuinely ARE room-filling.
- ?mute=1 NEVER SILENCED THE RIFFLE (R5->R28). dig.js opened a SECOND, private AudioContext
straight to destination, breaking both stated house laws (ONE AudioContext; mute forces
silence). It survived three epochs because NO GATE ASSERTS MUTE => SILENCE: the engine returns
its silent surface and looks honest from its own side, while the sound came from somewhere it
never knew existed. A gate that asks the engine whether it is muted can only ever answer yes.
Fixed via B's PUBLIC .muted getter (consulted, not modified), URL fallback when standalone.
Measured with a positive control so the probe provably CAN see sound: live 1 ctx / 1 source;
muted 0 / 0. Blips stay procedural on purpose — zero-fetch, deterministic, still work under
?noassets=1 where a sampled riffle cannot. Only mute was broken; only mute was fixed.
-> F: assert SILENCE, not muted-state.
HARNESS HONESTY: the first till rig reported the same clean 3.37 dB — from a CONTAMINATED rig
(createDig appends a panel per call, close() doesn't remove it, so querySelector clicked trial
1's stale button every time). Tell: panelUp false on all three, and the NEGATIVE CONTROL
returning 0.6436 where it must return 1.0. The gains matched the formula exactly; I refused them
anyway and re-ran isolated (panelCount 1, panelUp true, control 1.0). Never report a number from
a rig you have watched lie — even when you like the number.
QA GREEN 6/0/0/0. Docs: LANE_C_AUDIO.md -> v3.1 (contract + both asks + every measurement).
look.glb — DONE, published, verified. Looking_Around.fbx (675 KB) from the CACHED tailnet stash (not
re-downloaded — the asset-pipeline skill's law) -> fbx_to_clip.py -> web/models/peds/look.glb: mesh-free,
65 mixamorig joints, 195 tracks, inverseBindMatrices present, 136 KB. Structurally identical to the
sit.glb precedent on every measurable field (meshes/nodes/skins/joints/bind), so D's _canon binds it
exactly like walk/idle/sit. Depot: published to digalot.fyi/3god under John's R22 standing authorization
for Lane E assets (_published.json 43 -> 44), verified by fetching the bytes BACK from the public depot:
https://digalot.fyi/3god/a/look.glb -> HTTP 200, 139,640 bytes, sha1 identical to local.
-> D: it's ready. Gate it behind an opt the way rigs.js gates sit (loadPedFleet(base, { sit })) — a raw
fetch would breach the ?classic=1 zero-fetch-delta covenant, same as sit.glb.
`lean` DOES NOT EXIST — the spike's premise is half wrong. The brief and kickoff both say "two of the 34
already-cached clips (lean, look-around)" and "Spike 1 needs nothing from anyone". look-around is cached;
lean is not, anywhere. Measured, not assumed: the stash's 34 clips contain no lean; `find ~/Documents
-iname "*lean*"` on ultra returns nothing (not mixamo-fetch, not character_kit, not 3D=models, no FBX by
that name at any depth); and the fetch tool cannot help unattended — `node fetch.cjs search "lean"` ->
"FATAL nothing on debug port 9222 — quit Brave and relaunch with --remote-debugging-port=9222". It needs
John's logged-in Brave session (his browser, his Mixamo account); I won't relaunch it.
I did NOT substitute a lookalike. `Bored` is the nearest stash clip and it is a standing weight-shift idle,
not a wall-lean — D's half explicitly needs shopfront adjacency ("lean against a wall wants the same care"),
so a standing idle would quietly change what Spike 1 proves. Half a spike, honestly, beats a whole one that
lies. -> John: one small unblocker — relaunch Brave with --remote-debugging-port=9222 and open a Mixamo
tab; the fetch+convert is minutes once the token exists.
Instrument LOD — D killed it and D's measurement is better than mine. I had built the five _lo variants in
scratch (14k -> 3500/3500/4000/6000/3000) when D's commit landed saying "E: STOP — do not build the _lo
variants". Nothing shipped; scratch discarded. D measured the rooms an LOD must fire in: max diagonal 15.2m
in the largest venue, so the player can never be far enough — practical win 0 tris. My own reading reached
the same verdict by another route, recorded as corroboration: C's §4 already says "the interior gate is
DRAWS; tris are the STREET budget (200k)", and an LOD swap does not change draws at all (same object, same
material, fewer tris) — the item's entire win landed on a metric no interior gate reads. Two independent
measurements, one verdict. Salvage if it ever returns: at ~3k the guitar_amp SHREDS (white gaps punched
through its silhouette — confirmed against the shipped 14k thumb as a decimation artifact, not a source
defect, per the R22 food_cart discipline); it needs 6k to read clean. The other four were clean at 3-4k.
-> A: your own file's comment now contradicts its code. Chasing why the fingerprint read 0x5f76e76 rather
than the 0x3fa36874 I'd seen every round, I confirmed it is LEGITIMATE — your R27 w5 facade fix re-pinned
51 goldens under the amended covenant. But selfcheck.js:267 pins hash: 0x5f76e76 while selfcheck.js:301
still says "the synthetic golden above stays 0x3fa36874". Code and comment disagree by 34 lines — the exact
species as R24's validator bug (my docstring said nested, my code read top-level); the next reader trusts
whichever they read first. Your file, your call; flagging per the round's "flag loudly if anything
disagrees".
Gates: selfcheck ALL GREEN 156,352/156,352 · qa.sh --strict 6 passed / 0 failed. look.glb moves NO golden —
proved by moving it aside and re-running: the fingerprint is identical either way, because an unreferenced
clip is never fetched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
