The piece the 0.2/0.3/2 pass explicitly deferred. Pins E's corroded tier
through the card's own pricing surfaces, off E's real resolver (not a
hand-typed 0.55):
- STEEL half (rigging.selftest): factoryExtras('corroded_post') -> hint 0.55
+ collateral key; _effRating on a corroded corner is rating x 0.55, so the
same load a carabiner holds on a sound post is over a carabiner on a
corroded one. The rust costs a tier.
- COLLATERAL half (gardenfly.selftest, real dressed world): a node-less
corroded placement -> collateralFor = 'the corroded post', summed by
collateralExposure -- through the resolver, not the baked GLB value direct
(the carport-to-null shortcut).
Cross-lane seam: the tier lives on lane/e, so both pins skip-with-disclosure
on bare lane/b (SKIPPED, visible in the report) and go live at integration.
Measured GREEN on a scratch merge of lane/a + lane/e (hint 0.55, collateral
). Browser 485/3/0, node sail 47/47 rigging 29/29.
Test-only + THREADS; adopted the landed 0.2/0.3/2 unchanged rather than
overwrite pushed work.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
site_04_pool_yard: E's ring (fourteen honest noes) east, the bed poolside, ONE
corroded post in the venturi channel where the NE corner wants steel. Authored
in the editor (palette shapes, iterative SCORE IT x3, canonical export).
Gauntlet: WINNABLE at $80 — 13 quads, 6 clean, 1 marginal; cheapest hold $40
(h1,t1b,p2,p3, 92.7 FULL, no rust); every 92%-cover line routes through c1 —
clean $65 rated on t1,p1,p2,c1, UNHOLDABLE on the two shapes that pull the eye
harder. audit.html over the shipped file reproduces the card to the digit.
Separation REFUSED with receipts (bare 83.7 FULL — mild-night canon, A's 0.4).
board.js (A's granted edit): pool_yard_southerly POOL entry with the audit
receipt. d.test: +5 pins (chosen-never-scripted, ring/rust facts, judged-XOR,
wreck apron, exposure $160). gardenfly.selftest (B's file, flagged): judged
walk 3 -> 4 in the same commit that ships the yard.
Selftest 515/0/0 (510 + 5, the count moved by exactly what was added).
Mutations: 1 full-suite (finding stripped -> DOUBLE red: B's walk 'UNJUDGED'
+ my XOR pin, 513/2), 10 probed (entry deleted/re-paired/receipt stripped/
scripted-slot, gnome unpriced/rust zeroed/fence priced, apron x3) — every one
red with its own words, controls green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GATE 0.3 — THE CHARTER, SAID OUT LOUD (D's soaker finding, A consulted on
wording and their every-card rule adopted). The audit always flew shade cloth
and never said so; on the soaker that silence read 'WINNABLE over an all-DEAD
garden — the night is broken'. Now: AUDIT.POROSITY is the one charter knob
(sweep.js + gardenfly.js both read it), scorecard.fabricCharter() names the
fabric and whether tonight's stones pass its weave (same hailBlockFor the sim
charges), and the SCORE IT header prints 'fabric · shade cloth (sweep
charter)' on EVERY card — plus, on leaky-storm nights, A's wording: the
fabric is the bet, and the audit does not place it.
GATE 2 — CLIENT CONSTRAINTS: A's shapes (seam contract, built against as
posted), this lane's teeth.
· rigging.js: validateNightConstraint (checked enum, both ends of the seam),
setConstraints (survives reset like _startBudget; [] is the only clear).
The house ban refuses the pick with the CLIENT'S WORDS as the ticker
reason; the cap refuses at SPEND time on every path (rig / cycle / set /
spares / fabric), exact at the boundary, refunds always free — and commit
is belt-and-braces: a caller who walked around the doors gets a THROW.
The cap covers the whole invoice, spares included — the client queries the
invoice, not the hardware line.
· sweep.js: applyNightConstraints, ONE function both drivers share — bans
shrink the candidate list (a card that recommends a forbidden line is the
card lying), the cap re-prices affordable/clean without touching a load.
· scoreSite carries constraints/budget/constrainedOut; the card prints the
client's terms next to funnel and fabric, and tells 'no geometry' from
'client forbids all of it'.
· spent getter now measures against _startBudget (the re-banked session was
measuring spend against the default bank).
Selftest 485/0/0 in the browser (474 + exactly the 11 added: 1 sail, 4
session, 2 charter, 4 audit-constraints). Wiring line for A filed in THREADS:
session.setConstraints(week.job?.constraints ?? []) at the night boundary.
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>
Both lanes added the identical param (same name, same 0.30 default, one-line
rig change) the same day. C's copy taken verbatim so integration merges clean;
my default-unmoved + membrane-heavier asserts in gardenfly.selftest.js stand
and cover both copies.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Measured post-ruling (garden_probe, funnel ON, 2026-07-20):
site_02 x earlybuster: bare 82.6 FULL (hail 10.7 + rain 6.8) · best flown 90.2 · gain +7.6
site_03 x southerly: bare 83.7 FULL (hail 10.7 + rain 5.6) · best flown 89.8 · gain +6.1
Bare WINS both nights by >30 points; 'bareMustLoseBelow' cannot be written
honestly at any threshold, so per the gate's own rule this is a FINDING, not
a fudged pin. Each yard carries _separation_finding (measured refusal with
numbers + the delete-on-pin instruction); scorecard exposes it and SCORE IT's
NONE PINNED branch becomes PIN REFUSED (measured). New assert: every shipped
yard is separation-JUDGED — pinned XOR refused-with-finding; silence and
stale-finding both red. Balance fight filed in THREADS.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Default stays the pinned 0.30 shade cloth (asserted: omitting porosity ===
explicit 0.30 to the digit, hp and every corner peak), so no pinned number
moves. porosity 0 flies genuinely heavier (per-corner strictly greater,
measured 1.23x average on the synthetic funnel yard). C's storm_06 gate-3.1
membrane-vs-cloth pin and B's membrane-balance tables ride this seam.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The first night the default fabric is the wrong answer, tuned by measurement
through the real chain (gardenfly, dressed anchors, funnel ON, post-ruling
tree): bare 17.5 DEAD, best cloth line 42.3 TATTERED (a loss at any price),
same $75 ring on membrane 96.5 FULL, $60 ring 68.8 FULL — one key, Δ54.2 HP.
Membrane on $20 carabiners inverts (19.1 < 24.5): the doubled load is real.
Measured over backyard_01 too: its geometry caps hail cover at ~31%, so the
night only works on site_02 — the pairing receipt for D. tools/fabric_bet/
probe.html is the measurement harness; flyGarden grows an additive porosity
opt (default 0.30 unchanged — B: contest at integration if wanted).
editor_storm.js owns THE storm selection and THE storm list; SCORE IT's
dropdown is a view of it (default = southerly, C's old default, kept on
purpose). The score panel renders scoreSite's progress ticks — flight x/y,
elapsed — and audit.html gets the same line. C's half: subscribe
editor.wind.js per the THREADS proposal.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
findCandidates/priceCandidate/judgeSweep extracted so the sync auditSweep and
the new auditSweepAsync are ONE copy of the math; scoreSite drives the async
path with onProgress ticks (sweep/fly/separation phases). Yield is a
MessageChannel task, not setTimeout — Chrome's intensive timer throttling
turned an occluded run into one flight per minute, measured. Three new
asserts, mutation-checked red-then-green in one sitting (impure chunk + a
swallowed tick = all three red with the intended diagnostics).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
C measured my proposed pin and it could not have failed. Converging rather than
landing a second pin that disagrees. Knobs are now C's:
site_02_corner_block / storm_02_wildnight / throat (-6,y,0) / t=60.0 / exact ===
Three wrong answers before this, all the same failure:
1. backyard_01 has venturi: [] — nothing to detect. (I found this one.)
2. site_02's BED is funnel-blind, Δ 0.0000 m/s. (Both of us, independently.)
3. t=30 on the wildnight is worth 0.44%. I MISSED THIS ONE: my throat
measurement read a healthy Δ +4.948 m/s and I trusted it, but that was on
storm_03b_earlybuster, not the wildnight the pin flies. A sensitivity
measured on one storm says nothing about another. The mutation check I was
proud of would have passed on the wrong storm.
Adopted C's vacuity guard and its 25% threshold — one number, not two. It is
the check that would have caught my proposal on its own: mutation asks 'did it
move', the guard asks 'is the funnel worth enough here that a wiring bug could
not hide in the rounding'.
Game side now imports main.js's createWindRouter instead of retyping its two
wiring lines (C's correction). My old version built BOTH sides with
windForSite, so it could only have caught a bad input, never a router that
routed wrong.
CONVERGENCE RECEIPT — my chain reads C's numbers to the decimal:
t=60 throat editor 47.38 game 47.38 exact === true funnel-off 31.60 = 33.3%
t=30 throat on 18.55 off 18.46 = 0.44% (C's finding, reproduced)
What my half adds over C's: C pins windForSite (the builder) against the game;
this pins buildScoringWorld — the path SCORE IT actually runs — against the
same router. Three chains, one number.
Kept the garden bed as a SECOND probe on C's explicit offer: what the sail
shades is what the audit is about, and equality there is a real claim even
though the funnel is not what decides it. Labelled as carrying no funnel
tripwire, because it cannot.
Mutation-checked: setting t back to 30 reds the guard and both tripwires with
the exact diagnostic (0.44%, 18.46 -> 18.55) while the EQUALITY asserts stay
green — which is precisely C's argument, reproduced on my own chain.
Selftest 375 passed / 0 failed / 0 skipped.
COLLATERAL ON THE CARD. 'Winnable at $80' is an incomplete sentence on a yard
where letting a corner go also bills $180 of carport, so the card now prices
it — through world.collateralFor(), the one resolver (A's structKey fix), never
a second copy of that lookup in a panel. Deduped per STRUCTURE the way main.js
bills ('two beams letting go is one carport gone'): site_02's four carport
anchors read $180 total, not $720. The bill also rides on the margin-flag line
itself, because 'knife-edge' and 'knife-edge AND $180' are different authoring
decisions. An unpriced label renders UNPRICED — not scored, in red, never
'free' — which is the gutter bug's shape and the lie the card exists to prevent.
Verified A's structKey fix rather than assuming it: an editor-shaped carport
(id s1 + collateralKey) prices to $180, a second one (s2) also $180, and the
negative control (id s1, no collateralKey) reads null = not scored. My first
test reported a false alarm because I renamed a SHIPPED spec's id without the
palette's collateralKey — A's editor writes that field explicitly, so the
editor path was never broken.
THE NIGHT-5 INVOICE LINE (pool item, offered S13, accepted). Night 5's garden
is beyond saving by design, and the job sheet was the last surface still
dangling 'up to $45 / paid on what's left of the bed at dawn' — a maximum
nobody can earn, phrased as a rigging problem the player could solve.
NO ECONOMY CHANGE, deliberately. week.js's own comment forbids quote and settle
diverging ('the job sheet promises a number the invoice doesn't pay'), and
zeroing the quote would break that in the opposite direction — settle really
does pay the bonus proportionally. So the NUMBER stays and its CONDITION tells
the truth, which is the job sheet's own idiom:
nights 1-4 'paid on what's left of the bed at dawn' (unchanged)
night 5 'scales with the bed at dawn — and tonight the ice takes it
whatever you rig'
and the ledger names it too, because an unexplained small number next to
'night lost' reads as the player's failure:
night 4 'garden bonus — 18% of the bed +$8'
night 5 'garden bonus — 18% of the bed, beyond saving from the start +$8'
Verified in the running game on all five nights: flag fires on 5 only, every
quoted and paid figure identical (garden $45, total $136, bonus $8).
Selftest 374 passed / 0 failed / 0 skipped.
Two chains built independently — GAME (loadSite -> createWorld -> dress ->
windForSite) and EDITOR (siteClone-shaped object -> buildScoringWorld ->
windForSite) — asserted EXACTLY equal at t=30, speed and vector, at every
probe on both shipped yards. Exactly, not nearly: a tolerance is where the
funnel-off bug hid, and it moved the S13 headline 91.5 -> 39.8.
The mutation checks caught my own probe TWICE before this went green, which is
the only reason it is worth anything:
1. I proposed backyard_01 in THREADS. It declares venturi: [] — no funnel to
lose, so a funnel-off regression could never turn it red.
2. So I moved to site_02 — and its BED is 6.08 m from a radius-5 throat.
Starving that yard of its venturi moved the bed probe by 0.000 m/s.
Both versions passed their equality assert and both were worthless.
Measured sensitivity on site_02/earlybuster at t=30 (funnel on/off, trees
present/hidden):
bed ( 0, 1) 12.452 dV +0.000 dT +0.000 <- blind, not a pin site
throat (-6, 0) 15.910 dV +4.948 dT +0.000 <- funnel tripwire
near tr1 ( 7,-1) 8.593 dV +0.000 dT -4.891 <- shelter tripwire
Mutation-checked: starving the editor clone of its venturi turns the @throat
equality red (plus both funnel asserts), while bed/tree stay green exactly as
the table predicts.
Also recorded as a pin, because it is load-bearing for anyone reading the
score card: the corner block's funnel does NOT reach its garden bed. It is
aimed at the rigging zone (the site JSON says so: q1 is 3.5 m inside the
radius), so on that yard the funnel is a hardware-load story, not a
garden-exposure one.
Selftest 374 passed / 0 failed / 0 skipped (362 baseline + 12).
The page keeps its rendering and loses its logic. A third front-end (A's
editor) was about to make sweep-fly-pick-judge exist twice, which is the exact
failure sweep.js was extracted to prevent, one level up.
Behaviour-preserving, and measured that way rather than reasoned: both shipped
sites were pinned BEFORE the refactor and reproduce to the digit after —
backyard_01/wildnight 16 cands, 3 clean, 2 marginal, bare 35.7 tattered, best
p1,p2,p3,p4 60.5, sep MEETS held 63.8; site_02/earlybuster 66 cands, 20 clean,
1 marginal, bare 82.6 full, best tr1,tr1b,q1,q4 91.3 full.
Caught in the act by that check: __audit.cands went undefined because scoreSite
returns the COUNT where the old local was the LIST. A refactor that only got
read would have shipped it.
Dims, tri counts and node names all pass on a bike leaning the wrong way and
a gutter that never fell. The Blender contact sheet is close but renders
through Blender's view transform — it showed this sprint's jacaranda as
near-white lavender when the material is #8C7FC0, so a colour call made there
is a call about Blender, not about the game.
Smallest thing that answers the question: the game's own vendored three.js,
daylight, a ground plane, and the 1.7 m ref capsule beside whatever you name.
/tools/assetcheck/look.html?a=swing_set_01,tree_jacaranda_01&cam=10,4.5,22
It caught two things this sprint the numbers could not: the lilac reads lilac
in the real renderer (the thumb was lying), and the swing set's wreck lands on
+Z with a 2.9 m reach against a 0.95 m standing footprint — which is now baked
as a placement extra and asserted.
Sized at draw time, every time: a tab that boots hidden reports 0x0 and you
get a correct scene rendered into nothing. That cost me the first screenshot
of this very page, so it is in the MANUAL gotcha list where it already was.
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.
scorecard.js: the audit run-logic that audit.html had grown moves into one
shared engine both front-ends import (build the scoring world, sweep, fly the
winners, pick the best line, judge the separation block). Takes a site OBJECT,
not a path — loadSite always returned an object, so the fetch was never the
audit's business. Builds its OWN dressed world on a re-pointable wind proxy so
the editor's calm stub can never reach a score; every m/s still comes from
windForSite() via sweep.js/gardenfly.js.
editor_score.js: front-end only, zero audit logic. Mounts at A's reserved
order 40, renders the card with A's contract classes, says the funnel state
first and loudly, and marks itself stale when the yard changes underneath it.
The scoring quantity is the FLOWN garden state now, not static cover%.
New gardenfly.js: windForSite (C's shared builder — venturi + shelters, one
copy in the repo) → real attach (throws on D's skipped-attach trap) → skyfx
exposure → garden.js, per candidate line; cover% demoted to a labelled
geometry diagnostic. audit.html flies every affordable line + bare bed and
judges the site's pinned separation block on the block's own storm; audit.mjs
(node) keeps fast winnability, gains p5 in its verified dump, and points at
the browser for garden truth.
On C's correction, three structural adoptions:
- windForSite everywhere (three harnesses independently mis-built site wind;
no fourth copy);
- LIVE anchors via the re-pointable wind proxy — the audit's own frozen-sway
remap was C's landmine 2 wearing my file's name (audit.html:69);
- the MARGIN rule (AUDIT.MARGIN = 0.15, one copy): every row prices twice —
$hw holds vs $cleanHw with >=15% headroom — and only clean lines are
winners; verdict code 'marginal-only' when the budget can only buy the
knife edge (D's 39.8-TATTERED wild night, C's dead 91.9 headline).
Also game-true flight: the phantom 12s calm settle is GONE from the sweep
(commit->attach->storm is one keypress; the settle skewed every storm sample
12 s off the authored curve — SailRig samples wind at its INTERNAL clock).
That skew is not academic: it is a.test's separation-flight harness too, and
gardenfly.selftest now pins BOTH chains (game-true 63.8 tattered vs skewed
68.4 'full' on the pinned p5 line — the disagreement is flagged to A in
THREADS, not overruled here). The pinned recipe also sweeps regardless of the
18-45 band (it is 45.9 m² — the band predates p5; flagged to A).
Selftest 356/0/0 on the scratch merge (b+a+c@45bdc2d). Mutation-checked:
MARGIN=0 reddens the margin test (node); dropping siteDef from gardenfly's
windForSite call reddens the venturi test (browser, exact test named).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
garden_probe.html drops its TEMPORARY COPY of HAIL_WEIGHT/RAIN_WEIGHT/
GARDEN_DRAIN and flies main.js's own createGarden (imported from garden.js),
so the probe's hp/state/damage-split are the model's, not a re-derivation.
balance.test.js's local GARDEN_DRAIN/W_HAIL/W_RAIN copy re-points at the
garden.js exports (A's flag on balance.test.js:38) — a future retune now
moves this suite the day it lands instead of drifting silently.
Selftest 351/0/0 on the scratch merge (b+a+c) — the constants are identical,
so no number moved; the SOURCE moved.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
D's live replay caught landmine 1: bench.js read def.wind.venturi off the
STORM def where a venturi never lives (it is SITE data, main.js:453), and the
probes never attempted it — every site_02 bench number was funnel-off, the
third harness to make B's Sprint-11 mistake. Verifying the fix exposed
landmine 2: the tools remapped anchors to a static sway:()=>pos, freezing the
gum tree whose sway is the dynamic load a tr1 rig eats (ablation on the $80
line's q4: 1.02 funnel-off/static, 1.03 funnel-on/static, 1.24 funnel-on/real
anchors). Backyard numbers matched D's UI to the decimal precisely because
p1..p4 are posts — both landmines are no-ops there.
Fix: windForSite(stormDef, siteDef, anchors) in weather.js is the one shared
wind builder (venturi + shelters, exactly main.js's wiring); all garden_bench
tools build wind through it, fly world.anchors themselves via a re-pointable
wind proxy, print the venturi in their headers, and drop the phantom 12 s
settle (commit->attach->storm is one keypress, and the settle skewed storm
sampling via SailRig's internal clock). c.test.js pins the helper with B's
strictly-raises assert against the real shipped funnel (mutation-checked).
Corrected numbers: $80 line wildnight 91.9 with q4 at 1.24-vs-1.2 rating - a
knife edge the real game falls off (D UI: q3+q4 break, 39.8 TATTERED; D's UI
is canonical there, and a corner over its rating is not a safe line either
way). Buster verdict stands (95.2 bench / 97.9 D live). Icenight 9 DEAD 2/4.
Bench still under-reads UI peaks ~5-15% on site_02 - residual in the pool,
stated in THREADS. Separation-target site_02 example withdrawn to A.
Selftest 341/0/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The gate-1 diagnosis the sprint chartered, inverted by measurement: the shadow
geometry is exact (probe2: raycast and rasteriser agree 1.0000/1.0000), the
canonical measuring lines just don't cover the bed against vertical hail, and
site_02's tr1+q1+q3+q4 at exactly $80 already separates 91.5 FULL vs 35.7 bare
(sep 55.8, 0/4 lost) with no model change. Full findings + separation-target
proposal in THREADS.
Re-verified after the authoring session died un-pushed: all headline numbers
reproduce; the bench drives RiggingSession.commit(rig) -> rig.attach(), not the
bypassed-attach seam D's landmine names; balance.test's canonical shade-cloth
peaks reproduce per-anchor to the last digit. One fix over the original staging:
probe3/probe4 labelled per-corner peaks in plan order, but rig.corners is ring
order — labels now come from corners[k].anchorId (values were always true; the
$80 line's two RATED corners read swapped). Selftest 335/0/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Gate 2.4 asked for lean/stagger/brace clips staged by wind band. The answer,
measured twice, is that the asset shouldn't exist.
1. The lean cannot be a clip. _rotOnly drops Hips.quaternion from every clip at
load (player.js:53, applied :239). Parsed the shipped pack to confirm rather
than trust the read: 17 clips, 195 channels each, 64 survive. A lean baked
into a clip's hips is discarded. And stripping it is right — the wind's
bearing swings (the change; site_02's venturi) and a canned clip has one
fixed lean axis, so only the sim can aim it. Same trick as the knockdown and
climbY. The lean is D's root pitch; thresholds proposed off Beaufort 7
("inconvenience felt when walking against the wind" = 13.9 m/s), which lands
storm_01 never leaning, storm_03 gusting through the line, storm_02 in gale.
2. Mixamo has no wind animations. Searched the real API, not assumed: shiver
returns zero, brace returns eleven ledge-hang clips, wind returns Catwalk
Turn. The factory header already recorded this in Sprint 3.
Rendered the two real candidates on a skinned rig, twice each — as authored and
as the game shows them. Pushing is a squat shoving a couch. Leaning is a man
waiting for a bus. Both rejected. Both would have passed every check we own:
they load, they're metre-scale, their nodes survive. Dims can't answer "is this
the right shape" — the bike rule, catching a whole feature this time.
Brace and stagger already ship (TakeCover, StumbleBack, Reaction). D should wire
the root lean first; I'll hand-author a WindPosture additive into the gap we can
both see, if one remains. Spine rotation survives _rotOnly, so it would land.
Evidence, harness and repro in tools/character/windclip_candidates/, plus the
four pipeline traps I hit in order so the next lane doesn't.
selftest 335/0/0, unchanged — no asset added, no shipped file touched.
Gate 1 evidence, posted early because C's model fix and A's export are both
downstream of it.
garden_probe.html flies the sim's own garden loop (skyfx gardenExposure +
gardenHailExposure, stepped the way main.js does) per rig line and prints the
garden HP the sim would report. Browser-only by necessity: skyfx needs
`document` (node throws), so garden prediction can never live in audit.mjs.
The finding, every corner held on rated shackles (the sail's best case):
night 1 gentle no hail bare 97.6 best 99.1 sail worth +1.5
night 2 southerly 0.7 bare 83.7 best 90.1 sail worth +6.4
night 3 buster 0.7 bare 82.6 best 90.1 sail worth +7.5
night 4 wild night 1.3 bare 35.7 best 60.5 sail worth +24.8
night 5 ice night 1.4 bare 0.0 best 27.7 sail worth +27.7
The separation IS hail size, exactly as A's GARDEN_DRAIN comment predicted:
cloth blocks 100% of 1.3+ stones, 74% of pea hail, and rain at 73 degrees walks
under and does nothing. No value of GARDEN_DRAIN can separate a good rig from
none — it scales both by the same factor.
Also found: my own audit's cover% is the SUN shadow, which C's skyfx comment
already called "a number about nothing" during a storm. On the wild night it
predicts the garden BACKWARDS (r = -0.42): the best garden line is p1,p2,p3,p4
at 25% cover / $65 / 60.5 HP, beating a 71% cover / $120 line at 52.8 HP. D's
"$20 rig banked 39/45" and the QA's +$97 are the same fact — I sorted the table
by the wrong quantity.
Flagged to A: the ice night's garden cannot be saved at any price (best 27.7 vs
WIN >= 50), and night 1's sail is worth 1.5 HP.
The probe carries a TEMPORARY marked copy of main.js's private garden constants;
asked A to export the model so audit and sim share one copy. Nothing ships
against the copy — the probe is evidence, not a feature.
audit.mjs's hand-typed backyard dump goes full-precision: the 2-dp rounding
moved t2 ~5 mm against t1's shelter cone and swung its ice-night peak
4.8 -> 9.3 kN — the node and browser front-ends disagreed by one whole
winnable line until the positions matched to the float. Reconciled by
measurement (node on full floats reproduces the browser exactly).
Re-audit verdict (THREADS, full tables): every night still passes, no retune
proposed — E's numbers stand. backyard: hints move exactly one number (night
2 honest $30 -> $40; fascia appears in zero bed-covering quads so its 0.35
is audit-invisible). site_02: 64/66 -> 23/66 affordable, every cheap carport
line dead, cheapest now $40 honest steel, cheapest >=70% line $45 through a
$30-rated carport post — the trap is priced, not fenced off.
Selftest 319/0/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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>
sail.js _checkFailure now fails a corner on load > hw.rating * anchor.ratingHint
(read live from the anchor — dress() mutates in place). Consequence asserted,
not the formula: on site_02's carport line (uniform rated shackles, funnel on,
12 s calm settle) a cb corner blows before any q corner — at DEFAULT tension
(t~29 s into the storm, honest posts hold) and at max. In-suite control pins
the attribution: the same rig with every hint forced to 1 holds 4/4, so the
break is the hint, not the load. Mutation-checked: unwiring the line fails both.
rigging.js weak-link (D's #7): summary.weakest reduces on rating * ratingHint
via _effRating, not bare hw.rating — uniform hardware now flags the genuinely
worst steel instead of the first click. Mutation-checked.
site_audit prices the anchor, not just the steel: tierFor(peak, hint) needs
h.rating * hint >= peak, both front-ends carry ratingHint through (browser off
world.anchors, node dump extended with the live-dumped hints), rows print the
hint. sweep.selftest asserts hints reprice without touching peaks and land
over-ceiling corners in unholdable. Mutation-checked.
Selftest 319/0/0 (315 + 4).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Second harness on B's post-ratingHint audit: measures what the STORM
delivers at every dressed anchor — peak speedAt, tPeak, dynamic pressure
(downdraft folded back in), dose, and Pa/hint (the wind-side failure-order
prediction under load > rating x ratingHint). No cloth, no rig, no tension:
independent by construction.
Browser front-end dresses the yard the way the game does (createWorld +
dress) and wires venturi + tree shelters exactly as main.js:446-447 does.
envelope.selftest.js wired into c.test.js, every assert written to fail if
a wiring line is deleted (sweep.selftest's pattern). Selftest 320/0/0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
site_audit's headless false-negative is fixed by the browser front-end
(A's loadSite/createWorld landed in main — that was all it waited on), and
fixing it surfaced a worse bug underneath: the sweep built its wind from the
STORM def alone and never called setVenturi, which main.js:424 does at every
site load. The venturi is SITE data, so every site_02 audit ever run — including
the numbers quoted at C in SPRINT10 — was of a corner block with no gap. The
SPRINT6 p1=7.4kN failure inverted: there the tool called a good site unriggable,
here it called a mean site cheap.
- sweep.js: setVenturi on both winds (the gap doesn't switch off for the settle);
both front-ends pass the site's funnel and print it, so a funnelled run is
legible as one.
- sweep.selftest.js (new, wired into b.test.js): the auditor gets an auditor.
Funnel must strictly raise every corner; a funnel-less site must be
byte-identical. Verified it fails when the setVenturi calls are deleted.
- rigging.js: setWorld(world) + session.setBudget(n)/setAnchors() — A's two asks.
Markers are rebuilt (the anchor set changes across sites) and disposed; stale
picks die with the old yard. Deletes A's "one private touch" into this module.
- dev_rigging.html: dead since SPRINT10 (called createWorld without a site, which
now throws) and nothing noticed, because a dev page has no selftest to go red.
Fixed, dressed, and given the site switch it should have had (N = next yard).
Verdict for C/A, in THREADS: site_02 PASSES with the funnel on — 64 of 66 lines
affordable, cheapest $20. The funnel kills nothing, so the gain does NOT need
dropping and E's tree stays put. It's a reach problem, not a gain problem: the
throat's radius is 5 m and the bed centre is 6.08 m away, so the gap screams over
the carport at +50% and reaches the garden at +0.0%.
selftest 300/0/0 (296 + 4 new).
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.
Gate 2's actual deliverable. The node tool cannot dress, so it can only audit
the built-in snapshot or a resolved export and REFUSES A's dress-source site
JSON. This is the honest audit: audit.html builds the site the way the game does
— createWorld(await loadSite(name)) then dress() — and reads world.anchors, the
single source of DRESSED positions (GLB fascia + branch anchors, 8°-raked posts,
all of it), then runs the SAME sweep. It is the only way to audit a site with a
carport (site_02) before it ships.
tools/site_audit/audit.html?site=backyard_01
tools/site_audit/audit.html?site=site_02_corner_block&storm=storm_03_southerly
The winnability math now lives in ONE place, sweep.js, imported by both
front-ends — a tool built to catch reimplemented-formula drift doesn't get to
keep two copies of its own sweep. audit.mjs is refactored onto it (output
byte-identical); the node refusal message now hands you the exact audit.html URL.
Validated in-browser against backyard_01: dressed 12/12 anchors (GLBs loaded),
PASS, best line p1,p2,p3,p4 $65+$15 spare — identical to the node snapshot
verdict and to balance.test, off A's real data with no snapshot. That
cross-agreement also re-proves A's extraction is byte-identical.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A's gate-1 merge made createWorld REQUIRE a site (throws without one), which
crashed this tool's post cross-check: it called createWorld(scene, {wind}) with
no site. Fixed by reading the shipped data/sites/backyard_01.json and passing it
(loadSite fetch()es, so it can't run in node — read the file directly). The
cross-check is now BETTER: it verifies the dressed-snapshot posts against the
actual shipped site data, raked in createWorld, rather than against code.
Built-in audit green again; dress-source refusal (exit 2) and resolved-export
(exit 0) paths unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A's committed site schema (SPRINT10 world.js loadSite/createWorld) is
dress-source, and my loadSite assumed resolved positions. Run head-on against
A's real data/sites/backyard_01.json the tool reported:
✗ FAIL — no quad in the band shades the bed at all. The site cannot be rigged.
That is a lie, and the dangerous kind: the site is fine; the TOOL couldn't read
it. It is the SPRINT6 unwinnable-site trap in reverse — a false negative that,
in a "every site runs the audit before it ships" workflow, condemns a good site.
Two things make the schema unreadable headless, both real:
· posts are pre-rake. JSON carries {id,x,z,h}; world.js:361 leans each post 8°
off centre. p4's spec (-3.2,-1.2) dresses to (-3.72,-1.40) — the exact 0.56 m
error this tool's own snapshot shipped in SPRINT9. Reading x/z raw re-makes it.
· house/tree anchors have no coordinates — only `node`, a GLB empty's name.
Their position exists only after dress() reads matrixWorld, and dress() needs
GLTFLoader + fetch, neither of which runs in node. Branch anchors are exactly
where the dangerous quads live (t2b at 10 kN), so dropping them silently is
the worst possible failure.
The correct source of dressed positions is createWorld(site).anchors, which is
browser-only. That path lands once A's data-driven createWorld is in main; until
then loadSite REFUSES dress-source JSON with the reason and the plan, and exits 2
(distinct from the winnability FAIL's exit 1). The built-in DRESSED snapshot and
a resolved { anchors:[{id,type,pos}] } export both still audit. Raised to A in
THREADS: audit in-browser off createWorld(site).anchors, or export resolved anchors.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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>
