Recovers an interrupted session and finishes its verification honestly.
RESTORED (the predecessor was killed mid-mutation-check):
- build_sail_post_corroded still had negative control M1 APPLIED — clean
galvanised steel, DROOP=0 — so both shipped GLBs were the untelled post.
Restored: steel_weathered/rust/rust_deep, DROOP_DEG = 38.
- asset_report.json held 1 of 37 assets and contact_sheet.png was a 196 KB
truncated write. Both were interrupted artifacts, not decisions; a clean
rebuild restores 37 assets and a 5.07 MB 1680x4200 sheet.
DETERMINISM: 37/37 GLBs byte-identical across two consecutive full runs;
report and contact sheet byte-identical too. Zero churn on the 35 pre-existing
GLBs.
ASSERT THAT COULD NOT FAIL, KILLED: the sag pin (honestY - corrodedY > 0.05)
reads 0.060 at ZERO droop, because the eye hangs off a weld line 160 mm down
the post — measured on the M-B build, anchor at (0, 3.920, 0). It measured the
weld offset and called it a sag. Replaced with the droop ANGLE derived from
the exported anchor and pinned to baked extras (padeye_droop_deg / arm_m /
weld_y_threejs / dir_threejs), all stated in three.js coords — the axis trap
closed the swing_set way, so note and mesh can only lie together.
ASSERT WHOSE REASON WAS BACKWARDS, FIXED: the 20 m pin claimed the corroded
shaft read darker. Measured off look.html's canvas at 20 m: honest lum 50,
corroded lum 94 — it renders twice as BRIGHT. Base colour is darker but
metalness drives the pixels (gal 0.90/0.35 goes near-black with no env map;
weathered 0.55/0.60 is diffuse). Test now pins both halves and names the
mechanism the player actually sees.
Three tells split into three tests so one failure cannot mask the others —
pre-split, the M-A mutation went red on rust area and never reached the
lightness or droop asserts at all.
Mutation-checked in the browser, all three witnessed red: M-A clean steel
(1/3 + 2/3 red, 3/3 green, factory GREEN throughout — the trap it cannot see,
confirmed); M-B flat eye (3/3 red, others green); M-C baked note lies while
mesh is honest (3/3 red, proving the pin reads the extra, not a literal).
Selftest 483/0/0 (474 baseline + 9 exactly). Tell verified by looking at
20 m at the game's own camera height, screenshots taken.
editor.js untouched — palette hunk filed verbatim in THREADS for A.
contracts.js ANCHOR_TYPE widening flagged for A.
Lane D flagged: the tier is placeable, gate 3.2 unblocked.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Preserved verbatim from the interrupted session. NOTE: build_sail_post_corroded
still carries the M1 mutation (clean galvanised steel, DROOP=0), so the two GLBs
in this commit are the UNTELLED post, not the shipped asset. Restored in the
next commit after the red-check is witnessed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
SPRINT14 gate 3.1. The editor is about to offer these GLBs to an author by
name, which turns every silent field in them into a trap for whoever places
one. Audited the placeable set, fixed the gaps in the factory (nothing
hand-edited), and added the two props D needs to author a yard worth playing.
THE AUDIT, and what it found:
- Four `*_anchor` nodes carried no rating_hint at all — door_anchor,
pickup_anchor, grip_anchor, window_light_anchor. adoptAnchor does
`rating_hint ?? 1`, so silence there does not mean "unrated", it means
RATED PERFECT: a shed door skin outranking a concreted post, conjured out
of a missing field. They now deny it explicitly (`tie_off: false`).
- The carport's price resolved only by coincidence: collateralFor() matches a
STRUCTURE whose site id equals the anchors' collateral string, and site_02
happens to name its structure "carport". The editor will generate
"carport_2" for the second one, and the trap would have gone free — the
gutter bug reborn in the sprint meant to bury it. The GLB now carries
`collateral_key`, like the house does for "gutter". Lane A: the runtime
half is yours (collateralFor could fall back to it).
- The trampoline is unpriced, and now says so rather than omitting it.
Three new rules make the audit permanent, and each is proven able to fail:
no silent anchors, no collateral string nobody prices, no baked anchor_type
outside the checked ANCHOR_TYPE enum. Rule 2 found the carport gap on its
first run — an unfaked negative control.
THE PROPS:
- swing_set_01 (+ wrecked) — the palette's honest middle. It had a ceiling
(gum fork 1.0) and a trap (carport beam 0.22) and nothing in between. Two
apex anchors at 0.45, typed `swing_frame` (NOT `post`: the type string is
what the player reads before committing, and "post" promises concrete),
and a crossbar that is the most anchor-looking object in the game and is
not one. Priced at $140 — proposal, reasoning baked beside it, A rules the
number: gnome 25 < gutter 90 < swing 140 < carport 180.
- tree_jacaranda_01 — a second species whose LADDER is the feature.
Gum: 1.00/0.88/0.76, climb freely, pay 24%.
Jacaranda: 0.95/0.52/0.40, climb at all, pay 58%.
That makes tree choice a real decision instead of "is there a tree".
Unpriced by ruling, same as the bike: no limb-failure event exists to see.
Both wreck/orientation claims are measured in browser coords, not reasoned —
`wreck_falls_toward: "+Z"` is baked AND asserted against the mesh.
Selftest 376/0/0. Two consecutive full factory runs: 44 files byte-identical.
The fascia anchors have said collateral:"gutter" since Sprint 6 and nobody
priced one — collateralFor('gutter') returns null, backyard_01's house is a
free failure. Proposal rides in the asset like the carport's did: 90, between
the gnome (25) and the carport (180), with the band reasoning next to the
constant. A rules the number.
house_yardside_wrecked_v1: same origin/footprint/facade (shared _house_facade
helper — intact house BIN chunk byte-identical to the committed GLB, only the
three new price extras differ), fascia torn through the anchored span, left
run hanging by its last bracket, right run kinked ON the grass, downpipe out
of plumb, offcuts flat on the lawn. No fascia_anchor_* empties survive — you
cannot re-tie to a ripped eave.
collateral_key is new and explicit: unlike the carport, the structure and the
thing you take have different names, so the key names which collateral string
the value prices.
All 33 GLBs byte-identical across two full factory runs; the house pair also
byte-identical between --only and full runs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The letterhead is the point: one business, two documents, same masthead, so
the job sheet and the invoice read as a pair of papers off the same outfit's
printer rather than two game screens. DESIGN.md has opened with "you run a
small landscaping outfit" since day one and it has never been on the glass.
tools/jobsheet/ — paste-ready for Lane A between CUT markers, purely additive
(.card/.row/.ledger untouched), rendered to four PNGs before handover, each
state addressable as ?v=. The ledger already WAS an invoice; it just never
said so. The line worth keeping is .void: a bonus that was on the job sheet
and didn't survive the night, struck through in the same column as the money
that did — a bonus you never watch fail is one you never learn to protect.
Two design calls left as A's with the reasoning beside them, same as the
carport's $180: the business name (proposal: HARD YARDS) and the ABN
(sequential digits on purpose — a plausible one could be a real business's).
bike_kid_01.glb — 1.15 x 0.33 x 0.74 m, 700 tris, lean baked in because a
bike doesn't stand up on its own. Two bugs caught by LOOKING at the render,
neither catchable by the existing suite: a comment claiming a step-through
above a diamond frame, and a fork with negative rake that drew head tube and
fork as one pole through the front wheel. Both passed the Blender verify —
a bounding box can't answer "is this the right shape".
The lean is an export-time axis fact (+Y in Blender is -Z in three.js), so
e.test.js now pins it in browser coords. Proved the assert fails: rebuilt at
lean=0 and Blender still said PASS while the selftest went red.
selftest 299/0/0. 32/32 GLBs byte-identical across three full factory runs.
The carport anchors have said collateral="carport" since Sprint 9, but the root
carried no collateral_value — so main.js, which reads world.gnome.collateralValue
for the gnome, had no number to reach for. A trap with no price is a warning
label. The root now carries it in the same shape and place, pinned above the
gnome's 25, with the proposed 180 and its reasoning as a named constant so Lane A
can argue with the number rather than the asset.
carport_01_wrecked is the payoff, on the intact one's origin for a mesh-for-mesh
swap: the pads stay (they were never the weak part), the frame racks over like a
parallelogram, the roof peels off downwind in one piece.
Two bugs found by measuring rather than reasoning, both now comments where the
numbers are: the beams were rotated when a beam spans the axis the frame racks
in, so it rides level — tilting it made the wreck taller than the carport; and
the roof sheets dipped below ground, which is what inflated the height. Asserts
now pin the wreck shorter than its twin and sitting on the grass.
291/0/0, Lane E 64 asserts, 40/40 output files byte-identical.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
carport_01 is site_02's dressing and its whole design. DESIGN.md wants the
corner block anchor-poor, and the interesting way to build that isn't to give
the yard nothing — an empty yard is just a smaller yard. It's to give it
something that looks like four free anchors and isn't. Light posts on small
pads, a beam sized for one sheet of roofing: tie 25 m2 to it in a southerly and
you take the carport, not the shackle. Ships with the worst ratings in the game
(beam 0.22 vs the fascia's 0.35) and collateral="carport", pinned in e.test.js
below the fascia with the reason attached, so nobody quietly fixes them upward
and turns the site back into a small yard.
The pyrrhic card is the third outcome on the same camera: dawn, bed green, gnome
standing, sail post down across the yard behind him. Warm light on purpose — a
pyrrhic result is a win, and the wreckage is the price, not the verdict. Copy
goes to Lane A in THREADS; cardpreview renders all three.
277/0/0, Lane E 60 asserts, 39/39 output files byte-identical.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
grass_atlas.png binned (six sprints, zero refs, Lane E's own recommendation).
Screenshot POST declined — yardshot works standalone, so it buys convenience not
capability. Saying no rather than carrying either an eighth sprint.
Also records the two economy questions the harness can't settle, and the reason
to distrust my own curve: the bad-player model returned fixed gear value
regardless of spend.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Each limb was one straight uniform tube — a peg poking out of a pole, and the
most visible art problem in docs/yard_day.jpg. They now leave the trunk thick,
sweep up on a bezier and taper to nothing, which is what a gum does. +96 tris on
gum_01, silhouette bounds unchanged.
Nothing physical moved. The tip is a bezier endpoint and the four rng draws per
branch are the same four in the same order, so every branch_anchor_* is
bit-for-bit where it was — verified before and after, all five identical to 6dp.
Lane A's winning line rigs off t2 and is untouched; no balance number can shift.
e.test.js now hard-codes those five world positions as a tripwire, so no future
art change of mine can quietly invalidate a sprint of someone else's physics. If
it goes red, the art moved an anchor — fix the art, not the numbers.
Card text and the dawn-tint treatment go to Lane A in THREADS. The tint is a CSS
snippet rather than a ninth texture: only two of my eight are consumed, and a
full-screen gradient needs no asset at all.
Selftest 262/0/0, 38/38 output files byte-identical.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
hide_render does not survive glTF export — verified in three r175, every node
arrives visible:true. So garden_bed has been drawing plants_full, tattered AND
dead superimposed since Sprint 1, and my THREADS note told Lane A the opposite.
The asserts missed it by checking the nodes existed rather than that only one
was on. The flag now rides in extras, which does export, and consumers apply it
in one line; e.test.js pins the flag instead of the node.
Night dressing: house_yardside gains window_glow, an unlit emissive pane marked
hidden_by_default, plus window_light_anchor. A warm window means someone is
inside, which is why the player is out in the dark keeping the sail on. Plus a
moon sprite whose halo, not its disc, is the part that survives cloud.
End cards are rendered from the game's own props as a diptych: identical camera
and yard, only the night's outcome differs. The left third is left empty for
Lane A's text.
Cards are opt-in behind --cards because EEVEE's shadow pass is not
byte-reproducible across processes in 5.1 — proven with a minimal repro, not
PYTHONHASHSEED, and present even with a hard sun. The dawn shadow is the art, so
the cards become committed art rather than build output. Everything the game
loads stays byte-identical: 38/38 verified.
Selftest 241/0/0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Hail (SPRINT5 §E-1) ships two ways because Lane C's rain is a BoxGeometry with a
flat material and skyfx loads no external texture at all: hail_stone_01 as
geometry that drops into their existing InstancedMesh pattern, and hail_pips.png
as a 4-cell atlas for impacts. A flat quad can't be round and an impact is round
— that's the only reason the atlas is a texture. Both come with a copy-paste
recipe, because a grep showed exactly one of my five shipped textures is
consumed: sail_weave, the one I wrote a recipe for. grass_atlas has sat
unreferenced for four sprints.
plant_shred (§E-2) is elongated after the first pass came out radial and read as
green potatoes — it's the long axis plus the midrib that says "leaf".
tools/yardshot/ finally lands the DESIGN.md pictures, carried since Sprint 2. It
does NOT touch server.py or main.js: it's a Lane E stdlib server that serves the
repo and takes POST /shot, so the browser posts a Blob and bytes go straight to
disk instead of base64 through a console. do_POST is ~25 lines and Lane A is
welcome to lift it and delete this.
Three traps found on the way, all documented: toBlob is async so the WebGL buffer
is already cleared (toDataURL is sync); a backgrounded tab lays the canvas out at
0x0 and toDataURL then returns "data:,"; rAF is paused there so step() must be
driven by hand.
Selftest 209/0/0, 36 output files byte-identical across two runs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pond (SPRINT4 §E-1, decision 10): pond_water.png and pond_normal.png, for a
patch Lane B builds from the cloth's own nodes — same ride-the-nodes rule as the
tears, since a rigid disc would sit still while the belly moves under it. Alpha
is a radial feather and RGB darkens toward the centre, both encoded radially so
the shading survives being scaled per pondMass (radius ~ sqrt(mass)). Water is
grey-green, not blue: rain caught in a sail is shallow, murky and mirrors an
overcast sky.
Both at 256², not 512²: smooth low-frequency content, and at 512 they were
256 KB + 320 KB against ~20 KB for every other texture here. The chop is seven
waves at golden-angle directions rather than three crossed sines, which
interfered into a lattice and read as basketweave; the normal map gets six
integer-frequency waves for the same reason, and still asserts its own tiling.
broom_01 (§E-2): poke_tip sits on the bristle end deliberately — a broomstick
jabbed at a loaded sail holes it.
Renames fence_panel_broken to fence_panel_snapped to match SPRINT4 §A-4/§E-3.
Lane A's dress() guards every load, so the mismatch would not have crashed —
the wreckage would silently never have appeared, which is worse.
All 21 assets pass, 33 output files byte-identical across two runs, 186/0/0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The aftermath screen needs something to point at (SPRINT3 §Lane E-2).
garden_gnome_01_broken snaps at the ankles with the base left where the player
last saw him, head rolled clear with the beard still on, hat off — recognisably
him rather than a shattered pile, because wreckage nobody can identify is just
gravel. fence_panel_broken loses a few palings and hangs one off a nail; most of
it stays standing, which is what makes the hole read as damage rather than as a
design choice. Both keep their intact twin's origin and ground plane so Lane A
swaps mesh-for-mesh with no offsets, asserted both ways.
Also asserts something claimed since Sprint 1 and never checked: that glTF
extras reach three's userData. They do — the gnome's collateral_value, the
canopy's sway_amp, branch_anchor rating_hint and the bin's mass_hint all arrive
as numbers. Had that silently dropped, Lane A's gnome would score $0 and every
anchor would rate identical, which reads as a gameplay decision rather than a
missing field.
Contact-sheet framing now keys the capsule off height, not max(dims): the broken
gnome is 0.39 m across but stands 0.11, so spread-based framing buried it the
same way it once buried the shackle.
Selftest 175/0/0, 30 output files byte-identical across two runs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
rake_pivot shipped as a childless empty, so rotating it moved nothing — and
rotating the whole GLB instead, which is the only alternative, tips the concrete
footing out of the ground along with the post. DESIGN.md makes raking away from
the load a player decision, so the handle has to actually work.
It now holds post + pad_eye + top_anchor, with footing left on the root: rake it
8° and the post leans while the concrete stays planted. Asserted both directions
in e.test.js — the head must travel over 0.3 m and the footing under 0.01.
Same class as the canopy bug, found the same way: by driving the handle in a
test instead of eyeballing the model. Exported dims unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Textures (SPRINT2 §Lane E-2). sail_weave.png is a seamless 512² knitted
shade-cloth weave with the stripe banding real cloth has; the script proves the
wrap by evaluating a second tile and requiring an exact match, because Lane B is
being told to set RepeatWrapping and a bad wrap would seam across the whole
sail. Luminance rides in a narrow band so it multiplies the base colour instead
of replacing it — high contrast reads as burlap, not HDPE. Dropped the per-pixel
noise: invisible at ±0.012 and incompressible, it cost 305 KB of the 323.
sail_tears.png is a strip of 4 escalating rips. Each is a lens, not a slit —
fabric under tension parts widest in the middle and tapers to a point, and a
constant-width gap reads as a drawn line. Threads bridge the gap, scaled to the
local width; without them a dark lens is a hole rather than a tear.
Dressing (§Lane E-3): wheelie_bin_01 (240 L, 12 kg, `lid` on its own node so it
flaps before the bin goes over), washing_line_01 (a Hills Hoist — the `head`
freewheels, giving a second wind tell at head height), garden_gnome_01
(collateral bait: a smashed gnome reads where a damage number doesn't).
Bin and tramp land in models/debris/ for Lane C to glob. All 19 assets pass, 28
output files byte-identical across two runs, selftest 129/0/0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
world.js sways a tree by rotating a `canopy` group whose origin sits at the
trunk top, so the blobs swing about the trunk. The trees shipped canopy_01..03
as siblings of `trunk` with each origin at its own blob centre — rotating one
spins a sphere in place, which is visually nothing. Lane A could not have swayed
these trees, and the canopy lean is the gust telegraph the player reads a beat
before it hits the sail, so the tell would have gone missing rather than looked
wrong.
Adds the `canopy` empty at the trunk top with the blobs parented under it, so
A's existing code works unchanged, plus sway_amp / sway_phase / sway_pivot_y
for the per-tree tuning SPRINT2 §Lane E asks for (gum_01 is big and leans less
at 0.85; gum_02 is whippy at 1.20 and shows a gust first).
sway_phase draws from its own RNG stream: taking it from the shared one would
consume a value and shift every blob draw after it, resilhouetting a tree the
other lanes have already tuned against. Exported dims are identical to Sprint 1.
e.test.js asserts the pivot by rotating the handle and measuring that a blob
actually travels — red before this change, green after.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Verifying the GLBs in three.js (not just Blender) showed four assets reporting
inflated bounds: tramp_01 came back 3.29 x 1.27 m against a true 2.96 x 0.78.
Box3.setFromObject expands each mesh's LOCAL box by the world matrix, so a node
carrying a rotation over-reports — the same trap Blender's obj.bound_box sets,
and what three uses for frustum culling. Joined nodes inherited parts[0]'s
rotation, which for an arc is half a segment step off-axis.
Applying rotation at join time makes every local box axis-aligned, so the
default Box3 path is now correct for consumers and culling is tight. World
geometry is unchanged; the exported dims are identical.
Adds tools/assetcheck/, the three.js harness that caught this. It's the only
check that can: a Blender round-trip exports Z-up->Y-up and imports Y-up->Z-up,
so a broken export_yup flips back and passes green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
16 generated assets plus the grass billboard atlas and the four debris models
copied verbatim from the 3D-STORE library (copies rule, §0). All meter-scale,
Y-up, and far under the 15k tri budget — garden_bed is heaviest at 2,580.
Regeneration is byte-deterministic: two consecutive runs produce identical
files, so re-running the factory causes no churn.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Character is a byte copy of 90sDJsim's man_casual_01 ped (already metre-scale,
head bone 1.75 m, bind pose proven in that game). Clips ride beside it in an
anim-only carrier built from the FBX libraries on the M1 Ultra.
The peds cannot round-trip through Blender: they encode metre scale as a node
scale of 0.01 on mixamorig*:Hips with child bones in centimetres, and Blender
bones carry no rest scale, so the glTF importer silently drops it — the rig is
exploded on import (LeftToe_End 96 m below the floor) before any merge happens.
So merging clips onto the ped (PLAN3D 5-D.1) is not viable; we ship the ped
untouched and retarget rotation-only clips onto it at runtime, the same shape
90sDJsim already uses for peds/idle.glb + peds/walk.glb.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
