83ef8c9234
194 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
m3ultra
|
83ef8c9234 |
Fable (integrator): V7 RATIFIED + round-30 instructions (the save and the crate, wave 1) + README manual (measured controls)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
a80ad55f47 |
Lane F R29 wave 0: the pose gate gains its subject — v6's first clip is in the game
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> |
||
|
m3ultra
|
10025d4082 |
Lane D R29 (v6 Spike 1): look.glb bound and wired — peds glance, feet planted (wave 0)
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>
|
||
|
m3ultra
|
09dff6b788 |
Fable (integrator): round-29 instructions — the v6 spikes (wave 0 now, the rest gated on John's footage) + charter amended (lean doesn't exist)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
cdce67c8f8 |
Lane F R28 (v5.1): the sweep gated and tagged — the spike never reached the game
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>
|
||
|
m3ultra
|
488067004c |
Lane C R28: audioEmitter — the till rings from the counter; the panner was inert
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). |
||
|
m3ultra
|
640bdbb55d |
Lane E R28 (v5.1): Spike 1 — look.glb shipped; lean does not exist (ledger #1)
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> |
||
|
m3ultra
|
999c9d653f |
Lane B R28 (v5.1): the awning fabric ripple — KILLED with the measurement
Ledger #2. Built it, proved it worked, measured what it was worth, killed it. A 10x-over-driven billow — 1.4m of flap on a 0.14m canopy — cannot move 1% of the frame. It is not the gain, it is the geometry. Nothing ships: buildings.js and wind.js are byte-identical to HEAD. IT WAS BUILT, AND IT WORKED. Split the stall canopies out of the rigid awning InstancedMesh (own mesh, own subdivided box — a 4-corner box has nothing to ripple), gave only them a cloth billow riding R17's wind.js uniforms: sin(pi u)* sin(pi v) pin (zero at the four posts by construction), two crossed travelling waves, per-instance phase so forty stalls don't breathe in unison. Proven by rendered-pixel hashes, not by eye: cloth displaces (calm→gusty) 40f59eaf → d6d3e1f1 moves back to calm 40f59eaf identical uWindTime=99 at amp 0 40f59eaf identical ?classic=1, pre- vs post-build 453c0154 → 453c0154 BYTE-IDENTICAL rest of town rides wind? (hidden) 76c8b518 → 76c8b518 still control: cloth back frame moves again Cost at market_square, true A/B: +0 draws (98→98) / +2,160 tris. +0, not the +1 my own comment claimed — the square's chunks hold only stalls, so the rigid list empties and makeInstanced returns null; the cloth mesh replaces it. AND IT IS WORTH NOTHING (amp is the same multiplier as gain): 0.02m CALM, the default weather amp 0.05% of frame 0.14m STORM, the most a player ever sees 0.21% 0.35m 2.5x over-driven 0.33% 0.70m 5x, absurd on a 0.14m slab 0.49% 1.40m 10x, flapping a body-length 0.77% At the player's actual browsing pose, storm-force: 0.00%, max delta 1. Not one pixel. THE REASON IS PHYSICAL — the same physics that killed the verandah in R17. A canopy is a horizontal sheet at y2.2 seen by a 1.7m player: edge-on or from underneath, so vertical displacement runs almost parallel to the view ray and makes almost no pixels. You cannot see a flat horizontal sheet flex vertically while standing under it. No gain fixes it; 10x buys 0.77% and stops reading as cloth. WHAT WOULD WORK (for whoever reads this in round 35): the motion must be perpendicular to the view — VERTICAL cloth: hanging stall skirts, a flag, a banner, an awning valance. That is an ASSET (Lane E), not a shader. B's applyWind/applyFabricWind seam is ready the day such a mesh exists; the shader was never the missing piece. Ledger #2 leaves the backlog forever. TWO VACUOUS MEASUREMENTS CAUGHT ON THE WAY. (1) `!!material.onBeforeCompile` to prove the rigid verandah untouched — every THREE.Material inherits a prototype no-op, so it is ALWAYS truthy; the assert could never go red. Replaced with hasOwnProperty + a behavioural hide-and-hash control. (2) A gain sweep reporting an identical 0.21% at gains 0.4→4.0 because I never stored the shader ref, so uFabricGain never moved — four identical numbers across a 10x range is a sweep that proves nothing. The constancy, not the value, gave it away. Both would have passed as green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
afad68b114 |
Lane D R28 (v5.1): instrument LOD — KILLED with the measurement (ledger #4)
E: STOP — do not build the _lo variants. That work is unnecessary; the item is dead.
The oldest tri item on the books, killed under the wind-sway rule attached to the brief
("if the measured win is <20k tris in practice, KILL it with the measurement").
MEASURED, from the shipped depot GLBs: each of the five instruments is exactly 14,000
tris; band + amp on stage = 70,000. On paper a 3-4k _lo each "saves" ~52k, clearing the
20k bar easily.
But the rule says IN PRACTICE, and the brief named the reason — "interiors are
close-quarters". So I measured the rooms an LOD would have to fire in (live, all three
venue archetypes, interiorMode driven by hand since rAF is throttled):
pub 9.7 x 9.7m, diagonal 13.7m, 12522 tris, 38 draws
band_room 8.9 x 9.6m, diagonal 13.0m, 11699 tris, 35 draws
rsl 12 x 9.4m, diagonal 15.2m, 15115 tris, 38 draws (the biggest)
VERDICT: practical win = 0 tris. A player can NEVER be more than 15.2m from the band —
corner-to-corner in the largest venue, and the stage sits at one end so the real max is
shorter. An LOD is a distance trade; inside a <=15.2m room the band is never small — it
is the subject of the gig and of F's tour money shot. Any sane threshold (15-25m) never
fires; an aggressive 8m one fires in half a room at a distance where a 3-4k _lo is
visibly worse with no perceptual cover. The win isn't "small", it's STRUCTURALLY ZERO —
the distance an LOD needs does not exist in this game. Off the backlog forever.
No budget motivates it either: a venue interior is ~12-15k tris, ~85k with the band —
inside the 200k stress ceiling E's R22 tri diet restored, at 35-38 draws vs <=350.
SIDE-OBSERVATION for E (not a proposal, no breach motivates it): all five instruments
are EXACTLY 14,000 tris — a decimator target, not a measurement of the objects. A mic
stand does not need 14k. If instrument tris ever do drive a real breach, the lever is a
flat re-decimation by object complexity (mic stand -> ~1k), which beats any LOD because
it needs no distance. Filed so the next tri hunt starts from it.
Spike 1 (ledger #1) pending E's lean.glb — look.glb has landed, lean.glb not yet.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
d9100962f6 |
Fable (integrator): round-28 instructions — THE SPIKE AND THE SWEEP (spike 1 + parked polish, tag v5.1)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
b2d29ba112 |
Lane F R27 (v5.0): the covenant re-pinned, the frame that took five tries, QA GREEN
qa.sh --strict --matrix: 7 passed / 0 failed / 0 warn. selfcheck ALL GREEN 156,352/156,352 ·
MATRIX GREEN 10 towns x 7 gates · scaffold_check GREEN (F's re-pins took) · flags harness green,
carrying the kill-the-server gate and the beta crate gate.
F RE-PINNED ITS OWN FILES per A's handoff (A doesn't edit other lanes' files, correctly):
GOLDEN_HASH 0x3fa36874 -> 0x5f76e76 · GIG_GOLDEN 0xb1d48ea1 -> 0xec7a2d39 · melbourne · katoomba
· the captions. AND A CAUGHT A LIVE HAZARD IN F'S GATE: hash_ok carried `or ('3fa36874' in
r.stdout)` — a bare-string OR-clause, dead the moment the covenant moved and worse than dead,
because it would go green off a SENTENCE ABOUT the old covenant appearing in stdout. Deleted.
The check derives from GOLDEN_HASH or it doesn't exist.
JOHN'S RULING IS THE EPOCH'S BEST LINE: the covenant is anti-drift machinery, not a defect
preservative. 0x3fa36874 did its job for five epochs — it never drifted. It just was never right,
and no hash can catch what was wrong in the first commit. A golden proves NOTHING CHANGED; it
cannot prove anything is CORRECT. Every gate we own was green over a world of back walls, and the
only thing that ever saw it was a human standing in the street looking at a wall.
THE FRONTAGE FRAME TOOK FIVE ATTEMPTS, FOUR OF THEM F'S FAULT:
1-3. cluster pose (the town's heart isn't this shop) / lot-anchored facing along the edge /
lot-normal facing — three photographs of empty road. Also never winnable: the shops faced
backwards the whole time. F was posing a camera at a wall and tuning the geometry.
4. B's helper — and F THREW THE RESULT AWAY. poseForShopFront RETURNS a pose; it does not
apply one. F called it, error-checked it, and never touched the camera, so the frame was
the spawn paddock — and F's error-check PASSED, because the call succeeded perfectly. The
vacuous check one last time, in the last frame of the epoch.
5. Applied it -> MONSTER ROBOT PARTY. Sign, door, awning, a ped walking in off Musgrave Road.
The measurement that would have found the facade bug in five minutes — compare the two
conventions against each other — is the one nobody ran for nine rounds, F included, right up to
the vacuous 72/72 F nearly filed as "B is wrong".
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
c1e862cf7c |
Lane A R27 wave 5: THE FACADE FIX — it was every town, not just the real ones (the covenant amended)
selfcheck ALL GREEN 156,352/156,352; consistency green. 51 goldens re-pinned in this one commit per the amendment law. scaffold_check is RED and CANNOT be green from here — the covenant hashes are duplicated in LANE F's files; exact values handed to F in LANE_A_NOTES §R27w5 (I don't edit other lanes' files). THE RULING'S PREMISE WAS WRONG, AND THE WRONG HALF WAS LOAD-BEARING. Wave 5 says "synthetic follows B's canon so every golden has been green over a wrong product since R18". Half right. Synthetic does NOT follow B's canon -- it is broken identically, and so is the marched path, and so are all three checked-in fixtures. It is not "every real town since R18". It is EVERY TOWN, EVERY PATH, SINCE v1. The ruling told me to measure the marched path rather than assume it was clean "because that's how this one survived nine rounds" -- so I measured all three, and the same discipline caught the premise itself. MEASURED FIRST, IN B'S CANON (+Z = the facade), BEFORE TOUCHING ANYTHING: synthetic seed 20261990 0 toward / 681 away (dot = -1.0000, exactly anti-parallel) marched katoomba/fremantle/godverse, fixtures melbourne/katoomba/silverton 0 toward, all away real-roads katoomba 0/72 <- B's number, reproduced 637 of 681 synthetic lots had their facade pointing at NO STREET AT ALL (44 hit a different street by corner coincidence). Then I confirmed it with my own eyes rather than trust the arithmetic: booted ?classic=1, stood on the synthetic main street at lot 0 -> three blank grey boxes. Walked BEHIND them into the block -> NUMBAT PLAYTHINGS, sign/door/windows/awning, facing the dirt. The classic covenant town has been a row of back walls since v1. After the fix, same camera, same seed, same ?classic=1: CROWN TOYS & GAMES · MARBLES & KITES · NUMBAT PLAYTHINGS -- a strip of shopfronts facing the road. FOUR SITES, ONE SIGN, AND EVERY ONE OF THEM SAID SO IN ITS OWN COMMENT: plan.js:122 synthetic strip "facade faces back down the outward normal, to the street" <- it didn't plan.js:244 market row "the facade then faces -(outward) = +x" <- it didn't plan_osm:330 marched avenue "facade faces +z (north) to the avenue" <- due south plan_osm:443 real roads B's proven one-liner (B measured it, then reverted it -- A's file) All four were atan2(outward) where the intent was atan2(-outward). The code never did what its comment said, and CITY_SPEC:71 blessed the result by writing the SIM convention into the LOTS contract. Nobody was wrong; the spec was. After: synthetic 681/681, marched + fixtures 100%, real-roads katoomba 72/72. THE COVENANT MOVED -- JOHN RULED IT, I DIDN'T. Fixing synthetic necessarily moves 0x3fa36874, frozen "forever" since R16. That is not Lane A's call, so I stopped and asked with the evidence. John's ruling: AMEND THE COVENANT, FIX ALL PATHS -- the covenant is anti-drift machinery, not a defect preservative. 51 pins in this commit: classic/synthetic 0x3fa36874 -> 0x5f76e76 gig 0xb1d48ea1 -> 0xec7a2d39 fixtures melbourne 0x9c7e76b3 · katoomba 0xf3aafec8 · silverton 0x6d74c4 23 real towns base + gig (46) Verified byte-identical across two fresh processes; pack-xor 0xa5058322. CITY_SPEC records the amendment as a ONE-TIME, ruling-gated precedent -- from this hash forward the law reads as it always did. A SECOND BUG THE FIX EXPOSED (fixed here, counted). buildRealRoads seated blocks at KERB = 4 m from the CENTRELINE, but a `main` carriageway is 10 m wide -- its kerb is at 5 m. EVERY MAIN-STREET BUILDING HAS STOOD 1 m INSIDE THE ROAD SINCE R18. The footprint never moved; only which face is the front -- so it was invisible until the shopfront turned around and its poster landed in the traffic lane (-1.07 m clearance, 6 real failures). Blocks now seat behind the kerb per A's own published corridor law (roadWidth/2 + FOOTPATH). Pack drops 6.4% -> 6.6%, counted, not hidden. THE CROSS-CONVENTION GATE -- built to F's lesson, not to a spec line. F proved that measuring lots against CITY_SPEC:71's own convention is SELF-AGREEMENT: 72/72 "facing" over a town of blank walls. So this gate restates nothing. It replicates LANE B'S OWN maths from buildings.js (toWorld(lot, 0, y, d/2 + 0.02)) and asks a physical question neither spec line can fudge: does the quad the RENDERER actually draws land nearer the street than the lot centre? It fires if A's ry sign drifts OR if B's zf sign drifts. It would have caught this on day one of v1. CITY_SPEC amendments: the lots contract now states facade/door/frontage = local +Z; the SIM/rig convention is documented separately as the different thing it is, with the explicit rule -- never assert a lot's facing against a spec line, assert it against the geometry the renderer draws. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
94fce8ea4b |
Fable (integrator): R27 wave-5 — THE SPEC RULING (facade = +Z, B's canon canonical); A one-commit fix+pins+cross-convention gate; F re-gates and tags
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
8a1d2e91e8 |
Lane F R27: HOLD v5.0 — every real town's shops face away from their street
B's micro-task found it while fixing F's frame, and F's three photographs of empty road are the
evidence: it was never the pose. plan_osm.js:443 aims the lot's -Z at the road ("facade (-Z)
faces road"); buildings.js:404 draws the facade on the local +Z face. Exactly opposite. The
shopfront — sign, door, R26 crate mark — lands on the far side and the street gets a blank wall.
Since R18. v4.0 shipped on it. katoomba_real's street_noon, the bookmark the tours have shot
since R9, is 46 draws of bare road.
Root cause: CITY_SPEC contradicts itself — :71 (lots contract) says -Z, :335 (R15 posters) says
"+Z, B's canon" — and each generator implements one half. Synthetic follows B's canon and looks
right, which is why every golden passes and nobody saw it for nine rounds. The epoch's signature
species and its worst instance: not a gate failing on correct behaviour, but a gate suite GREEN
over a product that is wrong.
F VERIFIED IT AND F'S FIRST CHECK WAS VACUOUS — recorded because it nearly ended the round the
wrong way. F measured each lot's ry against CITY_SPEC:71's own convention and got 35/35 and 72/72
FACING, a clean contradiction of B. But plan_osm IMPLEMENTS :71, so testing :71's data against
:71's rule is self-agreement — it cannot fail. A self-consistent check over a self-contradicting
spec passes under either convention, which is exactly why nine rounds of checks missed this, and
why F nearly filed "B is wrong" with a green number in hand. The only test that sees it compares
the two conventions. F's own frames already had.
Not F's to fix and B was right not to either: A's file, it moves every real-town plan (36 goldens
re-pin under the amendment law), and a self-contradicting spec is Fable's to rule, not a lane's
to pick a side of. B proved the one-liner (ry = atan2(-nx,-nz) → katoomba 0/72 → 72/72, F's money
shot 4m from Musgrave Road) and reverted it — the same discipline F applied to A's sweep and E's
validator, arriving from the other direction.
Order: Fable rules the spec → A lands the line + re-pins the goldens in ONE commit → F re-gates,
re-shoots, tags. Everything else is done and green: the kill-the-server gate, the reader, the
sandbox, the interior money shot (interiors pose themselves; the crate is real either way).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
4c2702fc58 |
Lane B R27 w4: poseForShopFront — and the real towns face the wrong way
The micro-task, done and verified. In doing it I found why F photographed
empty road: it is not the pose.
1. THE DELIVERABLE — DBG.poseForShopFront(sel, dist = 8)
Poses at a NAMED shop's frontage: door centred, sign band + R26 crate mark in
shot, under the verandah. clusterPose() answers "where is the town's HEART",
which is the wrong question for a named shop — Red Hill's heart doesn't contain
Monster Robot, which is exactly how F got three frames of empty road.
Keyed through ONE shop selector: F's R22/R24 resolver lived inside enterShop(),
so I extracted it rather than grow a second key scheme beside it. A selector bug
resolves the wrong shop AND looks like it worked (F's own first cut proved it),
so the id-namespace law now has exactly one home.
poseForShopFront(31) → Wholefood Cafe (PLAN id) + ambiguous: "31 is
also the godverseShopId of Silky Oak Furnature"
poseForShopFront('g:31') → Silky Oak Furnature (the prefix IS the fence)
poseForShopFront('g:999999') → {error} — never throws, never guesses
Framing proven by projection, not by eye: doorCentreNdcX = 0.0000 exactly, door
+ sign + mark inside the frustum. Derived from buildings.js, not hardcoded: door
is centred on the lot front (off = 0, unlike poseForShop's off = 1.0); sign band
= min(h,FACADE_H)-0.35 ± 0.25 ⇒ y 2.40-2.90; mark = y 2.45-2.52; camY 2.0 sits
below AWNING_Y 2.95, so a ray rising to a target ≤2.9 can never cross it — the
sign is clear BY GEOMETRY at any distance.
2. CORRECTING MY OWN R27 INTEL — I gave F two wrong numbers. I said the mark
"rides high (y≈3.3)" and the vinyl blocks 4% of it. Both wrong: the mark is at
y 2.45-2.52, and I had scanned y 3.0-3.6 — empty awning air above the sign,
which is why it read ~0. Re-measured: the novelty_record disc (top y 2.50)
reaches INTO the mark band and clips ~15% of the strip from EVERY offset;
moving the camera only changes which 15%. "Shoot head-on" survives, but for the
opposite reason to the one I gave.
3. THE FINDING — every real town's shops face away from their street, since R18.
From Musgrave Road (Monster Robot's own frontEdge 1316) the shop is a blank grey
back wall: no sign, no door, no mark. Measured:
redhill_godverse 35 shop lots: 0 facing, 35 backwards (dot = -1.000)
katoomba_real 72 shop lots: 0 facing, 72 backwards (dot = -1.000)
synthetic correct — always has been
dot = -1.000 exactly on every lot in both towns: a convention, not a defect.
Katoomba's street_noon — the bookmark the tours have shot for nine rounds — is
46 draws / 19,680 tris of bare road.
Root cause: CITY_SPEC contradicts itself and each generator implements one half.
CITY_SPEC:71 (lots contract) "World facing = (-sin ry, -cos ry)" ⇒ -Z facade
CITY_SPEC:335 (posters, R15) "the +Z street facade (B's canon)" ⇒ +Z facade
plan_osm.js:443 obeys the first (its comment: "facade (-Z) faces road").
plan.js, buildings.js, venue.js and the R15 posters obey the second. Synthetic —
the town every golden tests — follows B's canon and looks right, which is exactly
why nine rounds of real-town work never caught it. The epoch's own signature
species: a rule meeting a case written after it.
Proven, then REVERTED: one line in A's plan_osm.js (ry = atan2(-nx,-nz)) flips
katoomba 0/72 → 72/72 facing, and turns poseForShopFront's camera from 26m out in
a paddock into 4m from Musgrave Road — the money shot F specified. Not committed:
it is A's file, it moves every real-town plan (goldens re-pin under amendment
law), and a self-contradicting spec is Fable's to rule on, not mine to pick a
side of.
This also re-reads my R22 diagnosis: I recorded "chunk-buildings 348 tris (0.2%)"
in katoomba's busiest block and credited the instancing work. 348 tris is ~29
boxes — that was not efficiency, it was the buildings being ABSENT from frame. I
read the evidence as a win for five rounds.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
83ed848f16 |
Fable (integrator): R27 wave-4 — ratify the exactly-one-refusal + control-delta gates; B micro-task (frontage pose); F's tag path
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
630042fbe1 |
Lane F R27: the kill-the-server gate GREEN — and two corrections to F's own gate
THE KILL-THE-SERVER GATE, permanent and green (charter law #1 made falsifiable): ARM 1 server-up · gone[3] → 117 survivors · items[] 120, identity AND ORDER intact · COLLATERAL 0 · `gone` carries the atlas id verbatim (R26 fence) ARM 2 killed mid-session · enter 6ms vs control 6ms (Δ+0) · worst frame 204ms vs control 202ms (Δ+2) · room complete from files · ZERO game errors · EXACTLY 1 refused connection ARM 3 server-down boot · pure tier 1, gone[0] · ZERO game errors · exactly 1 refusal SANDBOX · POST/PUT/PATCH/DELETE all 501 · a mint shop 404s The subject exists only because G built `make-fixture`: production `gone` is EMPTY (0 sales since 2026-07-01), so this gate would have passed vacuously forever on live data. CORRECTION 1 — "ZERO console errors on server death" is UNACHIEVABLE, and F measured why. A fetch to a dead host logs net::ERR_CONNECTION_REFUSED at the BROWSER level; no .catch() sees it. To log zero the reader would have to know the server is dead WITHOUT ASKING. One refusal is the irreducible cost of discovering death. F's replacement is strictly stronger: ZERO errors from game code, and EXACTLY ONE refusal — which PROVES the circuit breaker. Without the breaker it is one error per shop entry forever, and "zero" could never tell a working breaker from a broken one because both fail it. F is the beneficiary of this change, so it is filed loudly in §27 for Fable to ratify or overrule, not quietly relaxed. CORRECTION 2 — F's frame assertion was weak: `< 250ms` is a number with no control. ARM 2 first reported worst-frame 201ms and PASSED. But entering a shop rebuilds an interior, which costs the same with or without a server — so an absolute threshold measures Lane C's room build, not the network. The arm now walks the same shop with the breaker tripped and asserts the DELTA: 202ms control, 204ms killed, Δ+2ms. The death cost nothing the room didn't. F told A the same thing one round ago; a number without a control is not a measurement. THE TOUR — 3 of 4 frames, and the 4th is FILED NOT SHIPPED. The real crate (the money shot: its own 120 records, price stickers, the keeper), a mint crate (captioned MINT — plausible, not real), the classic covenant. 0 page errors. The FRONTAGE frame does not land after three attempts and F will not ship a frame that doesn't show what its caption claims: the cluster primitive answers "where is this town's heart", which is not the question a frontage frame asks — Red Hill's heart is 2 shops/60m and does not contain Monster Robot. Next: ask B, who owns the frontage and shipped its mark this round. Re-deriving B's anchor in a tour script is how you get three bad frames. Also: the interior beauty shot came back with the HUD across it — R21's lesson, fixed on the street frames only. Now every DOM overlay is hidden, not a guessed selector list. NOT TAGGED. qa.sh --strict --matrix and the frontage frame remain. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
f91d734b32 |
Fable (integrator): R27 wave-3 — ratify ?live opt-in + the collateral assertion; F's remaining close enumerated
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
70743db47e |
Fable (integrator): R27 wave-2 status — F's three load-bearing inputs (pick-then-omit, make-fixture for gone[], the lifecycle switch)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
bee7563a76 |
Lane E R27 (v5.0): provenance v5-final — and the roster gap the freeze caught (ledger #6)
THE BUG THE FREEZE FOUND: redhill_godverse was ON DISK but NOT IN THE ROSTER — for three rounds. The v5 provenance table needed real numbers, so I measured instead of carrying the v4 row forward. The count didn't match: 23 caches on disk, 22 rostered. The missing one was redhill_godverse — THE TOWN THE ENTIRE v5 EPOCH WALKS TO. Mine, squarely. index.json is Lane E's file (write_index()). G's redhill_godverse landed in R24 ( |
||
|
m3ultra
|
6960bdd279 |
Lane C R27 (v5.0): the tier-2 seam ruling + LANE_C_PUB v5-FROZEN (ledgers #2, #6)
SEAM RULING (§8): bless the direct wire — NO lifecycle hook. F already holds currentAdapter (interior_mode.js:67) and passes it to dig.open() PER DIG (:111), so the live read lands, F updates the pack, the next riffle picks it up; not-landed/slow/dead => the adapter is simply the tier-1 pack => the dig never blocks (charter risk #1). No C lifecycle change. BUT THE QUESTION UNDERNEATH IT IS C'S, AND THE OBVIOUS WIRING IS WRONG. pickRealOffers indexes pack.items[(rr()*len)|0], so FILTERING THE ARRAY RESHUFFLES THE CRATE. Measured, 5 crates: one record selling changed 21 OTHER records (6/2/2/6/5 collateral). Pick-then-omit (draw from the stable full list, drop gone after) = 0 collateral: a crate with 3 sold shows 13, not a different 16. Selling one record must remove one record, not re-stock the shop. THE TRAP: both forms hide the sold record — so F's planned assertion ("a real gone item never renders") GOES GREEN ON THE WRONG WIRING. It cannot see the 21. -> F: assert on COLLATERAL (the survivors are unchanged across a gone delta), not absence. Vacuous-gate law, third application, and the cheapest to miss because it passes. LANDED (dig.js): pack.gone (Set|array) omitted at PICK time. F's invariant published: update item FIELDS in place (live price) freely; NEVER add/remove/reorder items[] — that identity+order IS the seeded pick's stability (R25's positional-id lesson, one layer up). DETERMINISM BOUNDARY sits exactly here: server down => byte-identical tier 1 (gate-able); server up => the crate legitimately changes (tier 2 is non-deterministic BY DESIGN), so no gate may assert dig determinism with the server up, and pack.gone never enters a seeded stream or golden. SCOPED: the bin fan is decorative (a different seeded pick from the dig's offers) — a sold cover may linger there until rebuild. That's when I'd publish a post-build hook; the epoch's claim doesn't need it, because the crate you RIFFLE is true. VERIFIED: pure-function measurement for the collateral claim, then end-to-end through the real dig's public surface (pull -> Esc -> scroll -> pull): sold record gone, survivors 4/4, collateral 0, items[] untouched at 120, live in-place price shows on the next pull. (My first harness returned the same title 4x — I never un-pulled — so its number was an artifact; re-ran it properly rather than report a metric I'd just watched lie.) DOCS FREEZE (§6): LANE_C_PUB is v5.0-FROZEN, whole document, with the case law at the top: three amendments here exist because a rule of mine met a case written AFTER it, and both times the lane building to my rule took the red. Header now states it: when a line meets a case it didn't foresee, the line is the stale thing — amend it, don't bend the case. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
01f398960b |
Lane B R27 (on-call): the R26 handshake verified against G's real manifest
R27 owes B nothing ("B/D: notes current; nothing else owed"; "B/D on-call").
This is on-call work on F's #5: the tour is specified as "Monster Robot Party's
frontage (B's mark visible)" — a claim R26 could not test, because G's manifest
did not exist and B proved the treatment against a temp manifest it scaffolded
and deleted. G shipped the real one (
|
||
|
m3ultra
|
839575b28d |
Lane A round 27 (v5.0): the docs freeze — CITY_SPEC v5 + the risk list judged (ledger #6)
Docs only. No code, no golden moved — correct: tier 2 touches no plan, no cache, no atlas bytes. selfcheck ALL GREEN 156,212/156,212; scaffold + consistency green; classic 0x3fa36874 and gig 0xb1d48ea1 frozen. The whole stock epoch moved zero plan goldens, which was the determinism boundary's entire promise. CITY_SPEC v5 SECTION — written from the ARTIFACTS, not the briefs (this epoch's own lesson): - The tier ladder: 0 parody / 1 static real / 2 live GODVERSE, each degrading softly to the rung below. - `sourcing` is NOT `tier`, and C's ruling is the precise one — both ship in every index. `tier: 1` is the LADDER RUNG (a mint atlas is still a static file, so still rung 1); `sourcing: "real"|"mint"` is WHERE THE CONTENTS CAME FROM. Different axes. Collapsing them is exactly how a mint crate comes to read as real. Measured: 15 keyed atlases = 1 real + 14 mint. - The three id-space fences, verified rather than restated (below). - The manifest: existence DECLARED, never probed — the root fix for fail-soft-by-404 colliding with the zero-console-errors law. No entry, no fetch, zero 404s, no attribution exceptions. 15 entries / 15 dirs. - Ruling 2 as law (mixed caches are the design; no gate may assert it away — mine did). - Lane A's seam is exactly one field: godverseShopId, and `'godverseShopId' in shop` is the tier test. THE ID FENCES ARE VERIFIED, NOT ASSERTED. The plan-id/godverse-id trap is STILL LIVE in the data and I reproduced it: in redhill_godverse, bare `31` is "Wholefood Cafe" as a plan id and "Silky Oak Furnature" as a godverse id — the exact ambiguity that made F's enterShop(31) enter the wrong shop and pass a soft-fall assertion vacuously on the wrong subject. So it goes in the spec as a STANDING RULE, not a fixed bug: no code or gate may compare a bare number across the two spaces. (Within the godverse space, G's two halves — census <=2992, dealgod store ids — are numerically disjoint by construction, which is what makes the opaque key safe.) THE RISK LIST, JUDGED LINE BY LINE — AND I DID NOT PRE-RETIRE ANYONE'S GATE: 1. CARRIED. Tier-2 latency/stutter is this round's defining gate and I'm writing in wave 2 — F's reader and kill-the-server gate do not exist yet. NOTHING retires this line except that gate going green. Retiring it from a doc while the code is unwritten would be the precise species this epoch spent four holds catching: a rule meeting a case that doesn't exist yet. The offline law is a promise until #3 makes it falsifiable; that's the whole point of #3. 2. RETIRED — measured, and the remedy the risk NAMED is the remedy that shipped ("crate rotation rather than bigger atlases"). Max 2 atlases/shop vs C's hard cap of 2; the 14 mint shops sit at 16 items / 1 atlas / 1024²; only the one real POS shop earns 2048² (120 items, 2 atlases, 1.2 MB) under the 60+ rule. 3. RETIRED — DESIGNED OUT, not solved. The risk assumed a fuzzy-matching problem ("names drift, shops move"). There is no matcher: identity is an explicit opaque key G emits. The predicted failure cannot occur in the predicted form — and the proof is the headline shop: Monster Robot is in NEITHER the census NOR OSM, so no name-matching strategy could ever have found it. Mis-stocking is fenced structurally instead (duplicate ids error, uniqueness over defined ids, the orphaned-atlas gate). 4. SPLIT. Write-back RETIRED by John's ruling recorded as MECHANISM, not config: /reserve and /buy are ABSENT, not stubbed — races and reservation semantics cannot occur because the verbs don't exist, and "prove the absence" beats any sandbox flag a config typo could flip. (Option B re-opens this line deliberately at v5.x, behind a hold budget.) The `gone` read half is CARRIED to F's #4. 5. Unchanged, extended: static-atlas determinism proved in practice (G's index-only sku re-emit moved no .webp bytes); the fail-soft ladder proved on every rung; the id fences now have case law behind them. 6. NEW, CARRIED — the orphaned atlas, with a live near-miss on the books. 7. NEW, CARRIED — "real" is one shop wide: 14 of 15 keyed atlases are mint, because exactly one shop has a POS. Honest and marked, not a defect — but the epoch's headline claim rests on ONE shop, and any future claim of scale needs either siblings for monsterrobot or continued care that `sourcing` stays legible to humans and not just to gates. THE GATE THAT ARMED ITSELF (the law working unattended, worth recording). My R25 orphaned-atlas gate SKIPped on newtown_godverse all through R25 — "no committed atlas keys to this town", subject absent, so it said so instead of passing quietly. The moment G's 14 mint atlases landed it STARTED BINDING ON ITS OWN and passed: 156,211 -> 156,212, and that +1 IS the gate arming itself. Nobody edited it. That is what the vacuous-gate law buys — a gate that waits, in the open, for its subject to exist. G: redhill_godverse's lift drops keyed shop 2286 "Empire Revival"; you stocked its numeric neighbour 2287. Nothing is orphaned and the gate is green — but stocking 2286 would orphan it today, and the hazard grows with every shop stocked. Carried as risk #6. F: nothing of mine gates your tag. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
78848ed254 |
Lane F R26 wave 4: the re-run — 6/1/0 → 7/0/0, QA GREEN, v5.0-beta tagged
The fence went in and the hold ended. C ruled, and both lanes titled their commit the same
thing: "my rule was the stale thing." E's one line:
SLOT_ID_PREFIX = {"real": "sku_", "mint": "mint_"} # sourcing-scoped, not one space
qa.sh --strict --matrix: 7 passed / 0 failed / 0 warn. selfcheck ALL GREEN 156,212/156,212 ·
MATRIX GREEN 10 towns x 7 gates · flags harness (carrying F's beta gate: 15 crates / 105 pairs /
0 shared titles, sourcing real+mint, ZERO 404s, 0 console errors) · atlas-QA OK, 15 atlases
valid, 15 LICENCE LINES SHOWN — R24's finding #4 at full scale, where one silently printed none.
The ruling is the round's keeper. Fable: prefixes are namespace fences. E's comment makes G's
judgment law: "`sku_` names a POS copy on a shelf, `mint_` names a dealgod listing standing in
for one... a prefix that lies doesn't fail a gate, it SELLS THE WRONG RECORD out of a real shop
to satisfy a minted stand-in. Never treat them as one space (same standing note as plan-ids vs
godverse-ids)."
That closing clause is F's R24 collision — plan id 31 vs godverse id 31, which walked F's own
harness into a cafe — promoted from a finding to a standing law, and re-derived independently by
G in the layer that moves money. The bug F found by entering the wrong shop is the bug G refused
to ship by selling the wrong record.
The rule bent; the id spaces didn't. That is the right direction: a rule describes an intent, and
when a legitimate case fails it, the description is what's wrong — PROVIDED somebody measured
that the intent still holds. F measured it before the ruling (non-positional, seeded, stable,
0 reuse across 15 crates), which is what made the ruling a decision instead of a guess. Holding
the tag was never about the rule being wrong; it was about who gets to say so.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
bd6f576c00 |
Fable (integrator): round-27 instructions — v5.0 THE LIVE CRATE (wave 0: beta tag; then tier-2 read-live, the kill-the-server gate, the tour, the tag)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
f2989858aa |
Lane E R26 w4: the mint-id fence — my rule was the stale thing (wave 4, E's one line)
THE RED WAS MINE. My R25 slot-id check demanded `sku_` on every real-stock pack — faithfully implementing C's R25 line, which was written BEFORE MINT EXISTED. It then failed all 14 of G's mint crates, whose ids honour its every intent: non-positional, seeded, stable, zero reuse. C's ruling names it exactly: "the rule was the stale thing, not E's check." Fourth hold of the epoch, and the fourth to trace to a rule meeting a case written before the case existed. THE FIX IS A FENCE, AND IT IS STRONGER THAN THE RULE IT REPLACES — that was the thing to prove. The easy green was to drop the prefix check. Instead the form is sourcing-scoped per C §7.2a (R26 w4): sourcing="real" => `sku_` (a POS copy on a shelf) · sourcing="mint" => `mint_` (a dealgod listing standing in for one). A crate wearing the other's prefix FAILS. WHY THIS IS A MONEY BUG, NOT A NAMING BUG — G's judgment, now law. G refused to file dealgod listing ids under `sku_` to go green the easy way. They were right: tier-2's sold-means-gone looks up POS skus. R24's id collision entered the WRONG SHOP; the same collision in the money layer SELLS A REAL RECORD out of a real shop to satisfy a minted stand-in. The gate's error says that, in those words — the cost, not the rule. The message also distinguishes a fence-crossing from a merely positional id, because they are different faults and the reader shouldn't have to guess which one bit. PROVED 7/7 — that the fence BITES, not merely that the red stopped. G's real crate (sku_...) -> PASS · G's mint crate (mint_...) -> PASS · mint filed under `sku_` (the exact shortcut G refused) -> FAIL, naming the fence and the real cost · real wearing `mint_` (the reverse) -> FAIL · positional ids still FAIL in BOTH sourcings (the old rule's whole intent, retained) · duplicate ids -> FAIL. Loosening would have passed the first two and lost the last four; this passes the first two and keeps all four. BOTH OLDER SUITES BROUGHT FORWARD RATHER THAN LEFT TO ROT. The R26 suite's mint fixture wore sku_ ids, so the fence correctly rejected it — a fixture that fails for a NEW reason has stopped testing what it was written for. Moved to the mint_ space: 17/17. R25 licence suite: 8/8, tally still firing VIA THE TALLY. Third round running where the near-miss was a test going quietly stale around a changing contract; it is the failure mode I now look for first. Gates: atlas-QA OK — 15 per-shop atlas(es) valid, 15 licence line(s) shown, 0 warning(s) (1 real + 14 mint, every one showing its flag to a human) · validate_manifest exit 0 · qa.sh --strict --matrix 7 passed / 0 failed / 0 warn — QA GREEN. -> F: your 6/1/0 is now 7/0/0. The beta is yours to tag. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
53e83308a0 |
Lane C R26 w4: the mint-id fence — my rule was the stale thing (wave 4, C's one line)
F's fourth hold traced to MY §7.2a. I wrote "sku_-prefixed for real-stock packs" in R25 — before mint existed — so it names one space where there are two. E implemented it faithfully and its validator did exactly what I told it to: demand sku_ on 14 mint crates. The check wasn't wrong; my rule was. (Second time a contract line of mine met a case written after it and another lane took the red.) AMENDED §7.2a — the id form is SOURCING-SCOPED and the prefix IS the fence: sourcing "real" -> sku_<POS id> the shop's POS inventory (a physical copy) sourcing "mint" -> mint_<dealgod listing> a dealgod listing (a plausible stand-in) WHY G WAS RIGHT TO REFUSE THE EASY GREEN: filing dealgod listing ids under sku_ would have gone green in one line and recreated the R24 collision INSIDE THE ECONOMY LAYER — where tier-2's sold-means-gone looks up POS skus. R24's collision entered the wrong shop; the same collision in the money layer sells the wrong record — a real copy, out of a real shop, to satisfy a minted stand-in. A prefix that lies costs money, not a gate. Blessed as design: disjoint namespaces; no gate, lookup or economy path may treat them as one (same standing note as plan-ids vs godverse-ids). VERIFIED against every crate before handing E the line: 15 crates (1 real + 14 mint), 0 fails under the amended rule (prefix matches sourcing, ids unique within pack), and 0 ids appear in both spaces across 344 distinct ids. The fence already holds in the data; only my wording didn't. -> E: assert the form against `sourcing` (already read at line 178) — sku_ for real, mint_ for mint; a crate wearing the other's prefix FAILS. Stronger than the check it replaces, not looser. -> F: 6/1/0 should flip to 7/0/0. Doc-only: no code, no atlas bytes, no goldens. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
0b4bc6c1f9 |
Fable (integrator): R26 wave 4 — the mint-id ruling (prefixes are namespace fences); C one line, E one line, F tags
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
7ba7ab20a8 |
Lane F R26: the beta gate's exact tally — 6 passed / 1 failed / 0 warn
The full qa.sh --strict --matrix run confirms the hold and sizes it. Green: selfcheck ALL GREEN 156,212/156,212 · MATRIX GREEN 10 towns x 7 gates · flags harness (carrying F's beta gate: distinctness, sourcing, zero 404s, 0 console errors) · consistency · figure-scale. Red: the manifest validator alone — atlas-QA, 14 identical errors, all 14 mint crates, one sourcing-blind assertion, one line. The real crate passes. Recording the ratio because it IS the argument: the beta is finished. What stands between it and the tag is a rule meeting a case written before the case existed — C's to rule, E's to land, and not F's to quietly loosen on its own behalf. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
528829eeba |
Lane F R26 (v5.0-beta): the manifest consumed, the tolerance retired — tag HELD on the mint-id rule
EVERY CRATE IS DIFFERENT, and F measured it rather than taking G's 15/15 on trust — from the indexes of record, not by entering one lucky shop: 15 keyed crates, 105 pairs, 0 sharing a title; 0 crates overlap the generic v2 pack; 0 ids reused across crates (G's sku fix holds at scale — the R24 vacuity dead at the source, not papered over in F's gate). 4 crates in Red Hill + 11 in Newtown = the manifest's 15 exactly, and the manifest lists nothing not on disk. SOURCING, NOT TIER — C was right and the gate follows C. `tier` is the charter's ladder rung and a mint atlas is ALSO rung 1: still a file, still offline, still deterministic. What differs is where the stock came from. 15/15 indexes declare sourcing, every one agrees with the manifest, a mint crate can never report as real, and the gate PRINTS it per room — provenance a human never reads isn't doing anything (R24's licence lesson, one round on). THE 404 TOLERANCE IS DELETED, NOT TUNED. R24 probed every keyed shop and ate a 404 per atlas-less one; F's smoke could only cope by NAMING them, and a tolerance is a confession. F filed the real fix as a contract ask rather than inventing it; G built the manifest; F now consumes it — the shell asks for nothing unlisted. Zero 404s because nothing absent gets asked for, not because the gate stopped looking. AND DELETING IT IMMEDIATELY CAUGHT A LEFTOVER: F had wired the boot preload but left a second blind probe at the door, 404ing on every keyed-but-unfetchable shop. While the gate said "8 probes, expected", a ninth was invisible too. A named exception starts protecting new bugs the moment the real fix exists. F's beta gate GREEN: distinctness · sourcing (real + mint rooms) · soft fall silent · ZERO 404s · 0 console errors, no attribution exceptions. Selfcheck ALL GREEN 156,212/156,212. TAG HELD — atlas-QA FAIL, 14 errors, and it is not the product. E's SLOT_ID_PREFIX = "sku_" (validate_atlas.py:73) applies to every pack; E reads `sourcing` at :178 but the slot-id check is sourcing-blind. All 14 mint crates fail; the real crate passes. A mint crate HAS no POS sku — it was never in that shop's till — so the rule demands what cannot exist, the same shape as R24's sweep demanding a cafe carry a godverseShopId. F measured the rule's intent and it holds: mint_<dealgod listing id>, non-positional, sorted by it, seeded bake, 0 ids reused across all 15 crates. Only the literal prefix fails. And G refused to go green the easy way for the best reason — F's own R24 collision grown up: a POS sku and a dealgod listing id are different id spaces, and filing one under `sku_` recreates that bug INSIDE the economy, where tier-2's sold-means-gone looks up POS skus. R24's collision entered the wrong shop; the same collision in the money layer sells the wrong record. F did not fix E's one line: E's file, C's contract, C's call as its author (the principle Fable ratified when C overruled `tier`) — and F is the party that benefits, since that line is the only thing between F and the tag. Cost of waiting: one line. Cost of F being quietly wrong: a gate that green-lights id-space collisions in the layer that moves money. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
a9a5ef05d8 |
Fable (integrator): R26 waves 1-2 complete — F's beta gate asserts C's 'sourcing' (the brief's 'tier' superseded, correctly)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
33577a9675 |
Lane E R26 (v5.0-beta): the sourcing arm + the atlas manifest, both ways (ledgers #3, #6)
THE FIELD IS `sourcing`, NOT `tier` — and reading C's contract instead of the brief is the only reason
this gate isn't already broken. The R26 brief says "the validator grows the tier arm... `tier` required
on every atlas". C ruled it CANNOT be `tier` (§7.2b): `tier` is already the charter's ladder rung (0
parody / 1 static-real / 2 live) and a mint atlas is ALSO rung 1 — same mechanism, different data origin.
Overloading it puts two meanings in one name: the same species as the plan-id/godverse-id collision and
the positional ids. Had I built the brief's word, my gate would demand `tier` while G emits `sourcing` —
R23 exactly, re-run. The contract's author is the authority for the contract's shape.
THE HONESTY SEPARATION, BOTH DIRECTIONS (C §7.2b). `sourcing` ∈ {real, mint} required; no `sourcing` =>
FAIL (a crate that cannot say where it came from must never pass as one that can). mint must carry no POS
claim; real must carry the POS provenance it claims. The licence line prints for both — verified, not
assumed.
-> C: §7.2b says "a mint index MUST NOT carry POS-snapshot fields" but does not enumerate them, so I named
the set — correct it if it's wrong. I assert on `crate`, `provenance.source.pos_db`,
`provenance.source.pos_dump` — the fields that claim a SPECIFIC SHOP'S ACTUAL INVENTORY. I deliberately
excluded `snapshot`: §7.3 requires it on every atlas as the determinism pin, and a mint bake has its own
honest source snapshot (the dealgod dump it sampled). The error names the exact field found, so a
divergence with G surfaces as a diagnosis, not a mystery.
THE MANIFEST (§7.2d, ledger #6), consistency both ways. An atlas on disk absent from the manifest FAILS
(no entry => no fetch => that crate silently never loads; an unreachable atlas is not a passing atlas). A
manifest entry with no atlas FAILS (the wire fetches it and 404s — the exact noise the manifest exists to
kill). C's "absent manifest => legitimate zero-shop state" is honoured: no atlases + no manifest => SKIP.
-> C: one hole your rule implies but doesn't spell out — a manifest that PROMISES shops while zero atlases
exist now FAILS rather than skipping. "Absent" scopes to the empty tree; a tree where something on disk
still asserts atlases exist is the vacuous shape wearing a hat. Flagging it as my reading, not your words.
PROVED BEFORE HANDING G THE GATE — 17/17, including a mint crate that does not exist yet (the R25 rule:
never hand a lane a gate that fails on compliance). real -> PASS · mint -> PASS WITH ITS LICENCE LINE
PRINTED (ledger #3's "prints for both") · real+mint side by side -> PASS · mint carrying `crate` -> FAIL ·
mint carrying pos_db -> FAIL · real carrying no POS provenance -> FAIL (the reverse) · no `sourcing` ->
FAIL · sourcing='realish' -> FAIL · atlas absent from manifest -> FAIL · manifest entry with no atlas ->
FAIL · declared type not on disk -> FAIL · manifest says mint while the index says real -> FAIL · manifest
missing while atlases exist -> FAIL · unparseable manifest -> FAIL · empty tree -> SKIP · empty manifest
-> SKIP · manifest promising a shop with no atlases -> FAIL.
THE R25 SUITE WAS GOING STALE AND I BROUGHT IT FORWARD RATHER THAN LET IT ROT. Adding `sourcing` + the
manifest made every R25 licence case fail for a NEW reason — which would leave the licence arm untested
while appearing tested. Updated to the R26 contract: 8/8 green, and the tally tripwire still fires VIA THE
TALLY (via_tally=True), not incidentally. Near-miss worth recording: the manifest error briefly shadowed
the tally in the harness, which is precisely how the tally died the first time. A test that fails for a
new reason has stopped testing what it was written for.
The gate is RED until G lands (2 errors: no `sourcing`, no manifest). G is in my wave, parallel; it goes
green on their re-emit with no change from me — same as R25.
-> F: THERE IS STILL NO v5.0-alpha TAG, and C's + B's BETA commits (
|
||
|
m3ultra
|
2f4ab5912f |
Lane B R26 (v5.0-beta): the stocked shopfront reads from the footpath
Ledger #5 — B's polish pick. A stocked godverse shop now tells you it's diggable before you reach the door: a crate of seeded spines under the shop name, and a warm-lamp-over-crates room behind the glass. Cost: +0 draws / +0 tris. The mark is painted into the chunk's EXISTING sign-atlas cell; the window cue rides the EXISTING per-shop aTint attribute. Nothing new is allocated or submitted. The manifest is the authority, deliberately NOT the godverseShopId key: keyed means "has a stock identity", the manifest means "a crate is actually here". Marking keyed-but-unstocked shops would advertise a crate the shop hasn't got — the street would lie. The mark does not leak tier: "you can dig here" is equally true of real and mint; provenance belongs on C's price card and in E's gates. (Second job for ledger #6's manifest, free.) Measured — true A/B at one pose, manifest 200 vs 404: spine-marked sign cells 0 → 3 (18,768 exact-palette px) stocked window tint verts 0 → 36 street_noon 65 draws / 115,502 tris → UNCHANGED Stocked shopfront (Goulds Books, footpath): 85 draws / 144,730 tris — inside street law (<=300 / <=200k). Cues agree with ground truth: 3 stocked shops live → 3 marked cells, 9 stocked window quads. Covenants re-proved (the prefetch is new code on the boot path), tested with the manifest present on disk so a stray fetch would 200 and be unmissable: ?classic=1 0 keyed, 0 stock fetches, 0 errors (zero-fetch covenant) default synthetic 0 keyed, 0 stock fetches, 0 errors manifest 404 18 keyed, fail-soft, 0 page errors Three traps, self-caught: (1) a module-load fetch would have put a fetch delta on ?classic=1 — restructured to a lazy reader + plan-driven prefetch, keyed plans only; (2) the lazy trigger alone raced the first chunk build, so the first stocked shop you approach built unmarked — prefetch moved to createChunkManager, before any chunk builds; (3) the first "zero cost" was VACUOUS (manifest never fetched, so nothing was applied) and the first pixel detector reported 60 marked cells with no treatment at all — replaced with an exact-palette signature that reads 0 in control. Vacuous-gate law applies to a lane's own verification too. Corollary: force-cache serves a deleted file (transferSize 0, status 200 while disk 404s) — poison the cache or you are testing your own memory. B consumes stock_godverse/index.json read-only and emits nothing into it; the temp manifest used to prove this round was deleted — the directory is G's. Shape-tolerant + fail-soft, so G's real emission needs no coordination with B. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
4f0adf12d8 |
Lane C R26 (v5.0-beta): three contract lines + the dig price sticker (ledgers #1, #4, #6)
Wave 1 — contract lines published first; G and E build to them. #1 THE BRIEF SAID `tier`. IT CAN'T BE (§7.2b). G's atlases already emit tier: 1 = the CHARTER'S LADDER RUNG (0 parody / 1 static-real / 2 live), and `kind` is already the stock type. A mint atlas is ALSO rung 1 — same mechanism (a statically baked per-shop atlas); only the data's origin differs. Real-vs-mint is an orthogonal axis, not a rung — overloading `tier` puts two meanings in one name, the same species as the plan-id/godverse-id collision just standing-noted, and as R25's positional ids. RULING: top-level `sourcing: "real"|"mint"`, REQUIRED; `tier` stays the numeric rung. A mint index MUST NOT carry POS-snapshot fields (the separation is the honesty). Gates read `sourcing`; absent => FAIL. #4 `condition` ISN'T IN THE INDEX (§7.2c) — the ledger assumed it was. The index carries artist/title/price/price_band; G's SNAPSHOT has condition + sleeve_cond and drops them at emit, so pickRealOffers mapped cond:'' and the card rendered dangling separators around a hole. That WAS the "database row". Ruled optional index fields, emitted when the source has them; no invented grades on mint items (a plausible-but-fictional grade is what `sourcing` exists to prevent). #6 THE ATLAS MANIFEST (§7.2d): stock_godverse/index.json is the authority for which shops carry atlases — no entry => no fetch => zero 404s, so the zero-console-errors law holds with no attribution exception. Shape + both-ways consistency specified. No manifest + no atlases = a legitimate silent zero-shop state, not an error. THE PRICE STICKER (code): the pull panel is now the shop's price sticker — manila card, rotation + shadow, typewriter caps, italic title, a stamped condition grade, and the round yellow $ sticker the sleeve art already uses (the panel echoes dig.js's own canvas language instead of fighting it). Fields compose only when present: a missing field never leaves a stray separator. VERIFIED (fresh context, 5 card shapes, both star branches): real grail VG+/VG + ★GRAIL; real standard VG+ + STANDARD; real ungraded COLLECTOR (grade omitted cleanly); parody NM + 1973 PUNK; parody rare (seed 4) VG + 1973 POP + ★RARE. No dangling separators anywhere. The verification caught a defect in MY OWN card: a real grail read "GRAIL ★ GRAIL" — `s` carried the price_band AND `rare` derived from that same band (parody is immune: its `s` is a genre). Fixed at the mapping. And I didn't leave the parody-rare branch on assumption — scanned seeds until a rare pull appeared and asserted it says RARE, not GRAIL. An unexercised branch is an unproven one. Budgets: the card is DOM — zero draws, zero tris, no 3D change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
e7d58eaad5 |
Fable (integrator): R26 wave 0 COMPLETE — E landed its own commit, F is GO to tag v5.0-alpha
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
9b1df5cca0 |
Lane E R25 (v5.0-alpha): the licence line made unmissable + the sku form asserted
THE BUG WAS A DIVERGENCE, NOT A SPELLING (ledger #2). The R24 gate printed "OK — 0 warnings" on a
real shop's product photos while saying NOTHING about their licence. Two paths read one field with
different spellings:
the CHECK: lic = prov.get("licence") or prov.get("license") <- permissive: passed either
the PRINT: lic = prov.get("licence") <- strict, behind `if lic:`
While G's artifact was British they agreed. G's R24 keyed re-emit moved it to US `license`: the check
still passed, the print silently resolved to {}, and FLAG BEFORE ANY PUBLIC/COMMERCIAL RELEASE stopped
reaching a human. The `or` fallback WAS the bug — it made one path permissive so it could not feel the
other break. That is my own R21 stale-roads lesson one layer up (a check applied to one of two
symmetric paths is a trap with a fuse on it) — I wrote that lesson down and shipped it again anyway.
Fixed structurally, not by adding the missing spelling: ONE spelling, NO fallback (British `licence`
now ERRORS and names C's §7.3 retirement); ONE lookup, ONE owner (validate_index returns the evidence
string, main prints what it is handed and never re-looks-up the field — no second path left to drift);
and the evidence line is COUNTED — judged N atlases means N licence lines shown, or FAIL.
MY OWN FIRST CUT HAD THE SAME DISEASE AND MY OWN TEST PASSED FOR THE WRONG REASON. I added the tally
tripwire — but main() also appended its own missing-licence error, so all_errs was never empty when the
tally would fire: the tally was UNREACHABLE DEAD CODE, and the test went green through the other path
and reported PASS. Exactly the trap I recorded in R24 ("a gate that fails for the wrong reason is a
gate that will pass for the wrong reason later") — one round later, on me, in the file where I wrote
it. Now the field check owns the precise error and the tally sits after the all_errs return, existing
only for the day that check goes permissive again. Verified by stubbing validate_index permissive: it
fails VIA THE TALLY, not incidentally.
Proved 8/8: G's sku re-emit -> PASS with the flag shown · `license` gone/empty -> FAIL · British at top
level -> FAIL · British nested -> FAIL · attribution/snapshot gone -> FAIL · THE R24 SILENT SHAPE ->
FAIL · permissive-check simulation -> FAIL via the tally.
Aligned to C §7.3 as the provenance authority: the four gate fields (license, attribution, generator,
snapshot) live at TOP LEVEL, US spelling — the convention 21 shipped town caches already use. This
reverses R24's #2/#3, which read the artifact of the day. Docstring and code now say the same thing;
R24's fourth bug was that they didn't.
C §7.2a — THE SKU FORM (ledger #3), and I verified C's claim before building to it: independently
measured 120/120 godverse ids also present in the generic pack (F and C exactly right), and G's
snapshot carries 120/120 unique skus so the re-emit was feasible. Proved the compliant shape passes
BEFORE handing G the gate (4/4: sku_<POS sku> -> PASS · positional -> FAIL · sku-but-duplicated ->
FAIL · one id sneaking back into rec_ -> FAIL at 1/120, so it catches a real collision, not only an
all-or-nothing one).
-> C: two things beyond your literal ask, say if you disagree. (1) I also assert the actual property
your rule exists to guarantee — the pack's ids are DISJOINT from the shipped generic packs, measured
against web/assets/models/stock_*_index.json; the prefix says ids should be disjoint, this measures
whether they are, and it catches the reverse drift (a generic pack wandering into sku_) a prefix check
never could. Absent generic packs WARN rather than passing silently. (2) I flag British `licence`
anywhere, including inside the nested provenance{} "colour" block — §7.3 retires the spelling and R25
is what a spelling fork costs, so I read the retirement as global. Scope it back if you meant
otherwise.
The gate was RED by design for one wave: between my readiness and G's re-emit the collision was real
and unfixed, so it failed deliberately — a gate that stayed green on a defect F had just found would
be the vacuous species in a different hat. It went green when G landed
|
||
|
m3ultra
|
4a80ea8505 |
Fable (integrator): R26 amended — wave-0 status (F green, tag pending E) + the atlas manifest (F's finding #4, structural)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
29942002eb |
Lane F R24 (v5.0-alpha): THE CRATE WALKS — the per-shop wire + a gate that can't lie
You can boot redhill_godverse, walk down Musgrave Road into Monster Robot Party, riffle its own real crate and buy a real record. Proven falsifiably, not by eyeball: base assets/stock_godverse/3962749/ (its own, not the town-wide) TITLES 120/120 the shop's atlas · 0/120 the generic pack TEXTURES 2 from stock_godverse/3962749/*.webp · 0 generic — the pixels' own file BUY rec_0000 "Abstract Latin Lounge II (Part II)" @ $15 → cash 191→176, exact REBOOT 191/0 restored — gone-this-session, back next boot Ladder: ?noassets → ZERO network requests · keyed-but-atlas-less shop → 404 → parody, live · newtown_real → base null, 350 items: the wire is a NO-OP off the godverse path. flags_check GREEN (0 fails, 0 warns). #6a THE WIRE: godverseBaseFor(shop), exported and used by BOTH the shell's boot preload and the room's lookup — C's cache identity is the LITERAL (type, base) pair, so two sides building that string differently would drop a real shop to parody, a bug indistinguishable from "no stock" by eye. One function, one string, one key. Gated on stockReal, never the id alone. #6b THE GATE: smoke_real_crate + DBG.stockInfo() — permanent, and it cannot pass without touching its subject (SKIPs with a printed reason when the town/atlas is absent). FINDING — THE BRIEF'S OWN HONEST-GATE SPEC WAS VACUOUS. It specified id equality. Both packs number items POSITIONALLY (rec_0000…), so all 120 atlas ids are generic-pack ids too: an id-equality gate goes GREEN on the generic pack — the exact failure the vacuous-gate law was written to stop, reproduced inside its first application. Ids there are slot numbers, not identity. Titles (0 overlap) and the texture URL discriminate; F asserts on those. FINDING — THE ID SPACES COLLIDE, and it bit F's own selector inside the anti-vacuous gate. Plan ids (1..N) and godverse ids (31, 767, 3962749) share a numeric space: Silky Oak's godverse id is 31 and another shop's PLAN id is 31, so enterShop(31) silently entered "Wholefood Cafe" and the soft-fall assertion PASSED ON AN UNKEYED CAFE. Fixed both ends: 'g:<n>' selects a godverse id explicitly, a bare number stays the plan id and now REPORTS the collision, and the smoke asserts it landed in the shop it asked for before asserting anything about it. FINDING — F'S OWN GATE CRIED THE LOUDEST WOLF WE OWN. classic_regression reported "plan fingerprint != golden 0x3fa36874 — determinism/prime-law broken" while the covenant was provably intact. F wrote it in R16 as `hash_ok and returncode == 0` — two assertions, ONE verdict, reported with the fingerprint's message — so any unrelated selfcheck red screamed that the byte-identical-forever covenant had broken. Two assertions, two verdicts now. It isn't enough to fail for a reason; you have to fail for YOUR reason. FINDING — fail-soft-by-404 vs the 0-console-errors law. The boot preload must resolve a keyed shop's pack before the door opens (getStockPack is sync), so it probes blind: 9 of Red Hill's 10 keyed shops have no atlas → 404 → parody. That IS fail-soft working, and it is also 8 console errors no .catch() can suppress. Fix is not to loosen the law but to stop probing for what isn't there: an atlas index in G's namespace (or hasStock on the cache entry, A+G). Filed as a contract ask, not invented. Until then the smoke counts and NAMES the probes and fails on any other error — tolerance stated out loud. Verified the lift is clean: cache 10 keyed → plan 9 keyed, every one a real secondhand shop. No cafe keyed, zero duplicate ids — there is no mis-stocking. F did NOT touch A's identity sweep or E's validator. F was the party that would benefit, and changing a red gate to green in another lane's file to unblock F's own tag would retroactively devalue every green F has ever reported. Filed instead — and R25 has since landed A's reconciliation and the six pins, which is the right hands doing it. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
533fa32b59 |
Fable (integrator): round-26 instructions — wave 0 closes the alpha, then v5.0-beta EVERY CRATE DIFFERENT
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
a493c0a13c |
Lane A round 25 (v5.0-alpha): the sweep reconciled with Ruling 2 + the six pins (ledger #1)
All 8 tree reds cleared. selfcheck ALL GREEN 156,211/156,211; scaffold + consistency green; classic
0x3fa36874 and gig 0xb1d48ea1 frozen. F: the tag is unblocked from my side, nothing of mine is pending.
SIX OF THE EIGHT REDS WERE MY OWN GATE ASSERTING RULING 2 AWAY. I built the R24 identity arm on an
assumption that was true when I wrote it and false the moment G shipped: that a godverse cache is ALL
census shops. Ruling 2 makes it MIXED -- a keyed GODVERSE layer plus an inherited OSM texture layer that
G measured to be load-bearing, not cosmetic (census-only Red Hill: 306 m spacing warn + a poster inside a
kerb). So newtown_godverse (18 keyed + 54 texture) and redhill_godverse (10 + 27) were HEALTHY TOWNS MY
GATE CALLED BROKEN -- it fired 54/72 on a good town. Keyed asserts now run PER-LAYER over the keyed subset
only; unkeyed texture shops raise nothing. The validator's "N/M missing" warn dies with it. The one honest
requirement left at that layer: a `godverse+osm` cache must have a NON-EMPTY godverse layer (else it's an
osm cache wearing the wrong `source`) -- still a warn, for the R24 charter reasons that still hold.
THE UNIQUENESS CHECK WAS BROKEN IN BOTH DIRECTIONS. `new Set(all ids).size === shops.length` collapses
every `undefined` into ONE Set entry, so it (a) failed structurally on any mixed cache -- 54 texture shops
became one entry -- and (b) could HIDE A REAL DUPLICATE behind that same collapse. Now over DEFINED ids
only. And per the vacuous-gate law I made it prove it bites rather than asserting that it does:
validateTownCache rejects duplicates at the front door, so a dup can never reach the plan that way -- the
test feeds one STRAIGHT TO THE LIFT (generatePlanOSM does not validate) and asserts the predicate catches
it, with a healthy mixed cache passing the same predicate. A check nobody can watch fail is not a check.
NEW -- THE ORPHANED-ATLAS GATE (my own R24 finding, now enforced instead of filed). The lift legitimately
drops shops (overlap-resolve, counted since R19). Dropping a shop that HAS a committed atlas means real
stock keyed to a shop that isn't in the town -- exactly the failure I filed in R24, and one F's #7b could
only catch in a browser at the end of a round. Now a node gate: for each godverse town, every
stock_godverse/<id>/ on disk belonging to that cache MUST be seated, else FAIL. Subject-aware -- it SKIPs
by name when a town has no atlas, rather than passing quietly.
It already found one, benign today: redhill's lift drops keyed shop 2286 "Empire Revival" (10 keyed in
cache -> 9 seated). No atlas, so nothing is orphaned and the gate passes -- but it prints, because a drop
must never be silent (the R19/R20 ruling). G: stocking Empire Revival at the beta would orphan it. Monster
Robot Party is seated (lot 8), which is the one that matters this round.
THE SIX PINS, cross-checked two ways before landing -- byte-identical across two fresh processes, AND
matching the three values G stated independently in
|
||
|
m3ultra
|
0d8d59a22a |
Lane C R25: sku-derived slot ids — F's id-collision finding made structural (ledger #3)
One contract line, published first so G re-emits against it. §7.2a added. THE COLLISION, MEASURED: both packs number items positionally from zero, so all 120 godverse ids are also generic-pack ids — 120/120 overlap. An id-equality gate cannot discriminate which pack a rendered item came from; that is exactly why the brief's own #7a spec was vacuous. (The vacuous-gate law's first application broke the law — and my own R23 metric had the same disease.) THE RULE (§7.2a): items[].id MUST derive from the source's stable key, never from position, and MUST be unique across ALL packs. Real-stock packs: id = "sku_" + <POS sku> (G's snapshot already carries `sku` per item, e.g. 20260314070442 — it's just dropped at emit, so this is index-only). Generic packs keep rec_NNNN as their namespace; real-stock must not emit into it. TWO REASONS, and the second outlives the gate: 1. positional ids collide by construction (the 120/120 above) -> id-equality gates can't discriminate. Sku ids fix that permanently, now, at two atlases instead of fifty. 2. POSITION IS NOT IDENTITY: a positional id renumbers on every re-bake — sell one record out of crate 550 and rec_0050 silently becomes a different record. Anything holding an id across snapshots (a wallet purchase, a saved dig, a gate assertion, and above all tier-2's sold-means-gone) would point at the wrong item. The POS sku is stable because it is the shop's own identity for that copy (release_id identifies the pressing, not the copy). So this is also the identity tier-2 will need, bought now for free. SCOPE: index-only — no atlas bytes, no town cache, no goldens, no C code (id isn't read by the loader, §7.3; the dig/buy consumers and gates read it). -> G: re-emit with sku_ ids. -> E: assert the form (present, unique within pack, sku_-prefixed). Committed by explicit pathspec — F's six staged files sit untouched, as its session left them. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
2c1ae37160 |
Fable (integrator): round-25 instructions — the tag, earned (sweep vs Ruling 2, licence line, sku ids, F closes)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
5815dfbd58 |
Lane E R24 (v5.0-alpha): the crate's real street, the honest gate, and Ballarat retired
THE STREET (ledger #3). redhill_real: Red Hill, Brisbane — 36 shops (9 heroes + 27 texture, -29 subsampled), 884 roads, median NN 73 m. **Musgrave Road is in roads[] (41 ways, primary + residential)** — that was the point: Monster Robot Party's real shop now has its real street. One bounded Overpass fetch under the standing authorization; raw snapshots in _raw/, ODbL, C's mapping. This is G's donor for godverse_town.py -> redhill_godverse. TWO HAZARDS HANDED TO G/A, both measured against the real geometry (notes R24 §1): (a) The shop is NOT in OSM — no shop=music node, no name match anywhere in the bbox (checked the raw snapshot, not the emitted cache). It cannot arrive via the fetch; G must inject id 3962749 from the census or the town ships without the shop the release is about. (b) THE SEATING HAZARD: A's lift seats on the NEAREST edge, and from G's geocode three streets are nearer than the addressed one — Zig Zag Street 38.6 m (residential), Craig Street 46.1 m, Musgrave Road 51.6 m. The shop would seat on Zig Zag St and the headline would quietly become false. Measured fix (G's call): the point on the real Musgrave centreline nearest G's geocode is -27.455169, 153.006902 — 0 m to Musgrave, next candidate 43 m. Snaps the street, does NOT verify the house number; I have no address data for where #147 sits along the road. THE HONEST GATE (ledger #4). F's catch was right and there were FOUR bugs: the glob used the published-contract name the runtime can never load (0 matches -> returned 0: it passed identically whether atlases were absent, misnamed, or perfect); the code read provenance top-level while my own docstring said nested (G nested it correctly — the gate would have failed G's real artifact); licence/license; and a stale ceiling. Rebuilt on the "runtime is the authority" ruling: globs stock_godverse/<id>/stock_<type>_index.json, type in {record,book,toy}. Vacuous-gate law implemented literally — 0 matches + stray files FAIL (listing them); 0 matches + empty dir prints "atlas-QA SKIP: no atlases present (0 files matched glob ...)"; plus an id equality (shop.godverse_id == dir) so it proves it touched its subject. Proved on four cases: empty -> loud SKIP, misnamed (F's exact bug) -> FAIL, unparseable -> FAIL, G's real atlas -> PASS (120 items, 2 atlases, licence printed). My own test found a fifth hole — the glob still matched the old stock_shop_* name, so the misnamed case failed only incidentally; added a STOCK_TYPES name check so it fails for the RIGHT reason. A gate that fails for the wrong reason will pass for the wrong reason later. BALLARAT — RETIRED, counted, and a DIFFERENT illness from Toowoomba (notes R24 §3). Toowoomba was dead everywhere. Ballarat is alive exactly where the game isn't: 103 POIs, densest 400 m disc = 47 of them (Toowoomba's was 3) — but that disc is cafes/clothes/bakeries while the op-shops sit 854 m, 1042 m and 1709 m out. Heroes and density don't coincide, so no bbox holds both. Real builds at every span: 1.0 km -> 8 shops/184 m/hub 4-8 but ZERO op-shops (a thrift town with no thrift shop); 2.0 km -> 12/186 m/3-12; 3.0 km as shipped -> 20/255 m/4-20. Every row worse than any shipped town. THE BINDING CONSTRAINT WAS MINE, NOT OSM'S: with 2 heroes the 3x texture cap keeps 6 of 47 texture POIs and the subsample is spatially blind, so survivors scatter at any span — part of the 255 m A's warn caught was my own cap's fingerprint. Seeding texture near heroes fixes the class but moves every town's golden -> filed as a v5.x proposal, not done in a release round. I OVERSTATED MY R23 ADVICE TO A: the hub fraction is NOT a standalone discriminator — glebe (shipped, healthy) reads 13% and redhill 14%; a hub threshold would false-flag both. Median NN is the robust primary; hub is only comparable at equal shop counts (toowoomba 2/12 vs darwin 7/12 was fair because both had 12). The raw-vs-seated calibration point stands. METHODOLOGY, recorded because it nearly shipped: I picked Ballarat's span from an offline simulation that predicted 68 m / hub 5-8, and had already written those numbers into a config comment as fact. The real build returned 184 m / 4-8 — same counts, different sample. Never trust a re-implementation of the pipeline's own selection; run the pipeline. -> A (batched pins): ADD REAL_TOWN_GOLDENS['redhill_real']=0x6046700f and REAL_TOWN_GIG_GOLDENS['redhill_real']=0x673785ee (the headline town currently only WARNS as UNPINNED — ungated until you pin it). REMOVE ballarat_real 0xf0a9e8a1 / 0xf097c331 — dead config by your own R23 note. -> F: tools/qa/tour_shots.py still shoots town=toowoomba_real, which I deleted in R23 — the v4 tour breaks on a missing town. Your file, your call. A retirement isn't done when the cache is deleted; it's done when the consumers are swept. -> C: built to the runtime + G's real artifact per the ruling; LANE_C_PUB §7 line 271 still names the contradictory stock_shop_<id>_index.json. I reconcile when your corrected contract lands. Determinism receipt: the retirement forced a full 21-town rebuild from cached raw (no refetches) and every existing cache came back byte-identical — only ballarat, redhill and index.json moved. Pack: 21 real towns · 1196 shops · 299 heroes (22 rostered with G's newtown_godverse). Selfcheck ALL GREEN 149,696, synthetic golden 0x3fa36874 unchanged. qa.sh --strict --matrix: 7 passed / 0 failed, MATRIX GREEN 10x7. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
1eb0b45bd0 |
Lane A round 24 (v5.0-alpha): the identity field — godverseShopId, published first (ledger #2)
The optional shop key that resolves `stock_godverse/<godverseShopId>/`. Schema STAYS v2 (optional field ⇒ every shipped cache still valid, the 21 osm caches untouched by construction). Published to web/assets/towns/README.md — the doc G and E build to — plus CITY_SPEC. selfcheck ALL GREEN (+15 checks), synthetic 0x3fa36874 frozen, NO golden moved. THE BRIEF'S "THE POS SHOP ID IS THE ID — NO MAPPING TABLE" IS FALSE FOR THE ONE SHOP THIS ALPHA IS ABOUT. G's own docstring says it: Monster Robot Party is NOT in thriftgod's census, so its id is its DEALGOD STORE ID 3962749, while census shops use thriftgod ids (max 2992) -- two disjoint spaces, deliberately kept clear of each other. Had I coded "godverseShopId == shop.id" or derived it from the census, the headline shop could never be keyed. So it is an OPAQUE key: no derivation from `id`, no equality assert, no mapping table (the R12 gigKey law -- one key, zero mapping; G emits, everyone reads). THE FIELD HAD TO SURVIVE THE LIFT OR IT WAS DECORATION. plan_osm builds a FRESH shop object and copies nothing unknown, and the runtime reads plan.shops[], never the cache -- so a cache-only field could never reach F's per-shop base. It's now carried through, CONDITIONALLY: absent ⇒ the key isn't written ⇒ JSON.stringify byte-identical ⇒ every pinned golden holds until G re-emits (that re-emit IS the sanctioned move). Proven end-to-end: a keyed cache puts `stock_godverse/3962749/` on the plan with census ids (101..107) coexisting alongside it. ARMS SPLIT BY HAZARD, NOT TIDINESS: - malformed -> ERROR. It names a path segment; "3962749" vs 3962749 is the same folder but different JSON, and that coercion drift is the exact species that broke C's atlas contract (licence/license). - duplicate -> ERROR. Two shops keyed to one atlas IS mis-stocking -- charter risk #3 forbids it by name. - missing on a godverse cache -> WARN + fail-soft to tier 0. RECORDED DEPARTURE (Fable's to overrule): ledger #2 says validateTownCache "requires" the field; I made MISSING a warn. Three reasons. (1) The charter's determinism boundary is "world = seeded, frozen, gated; stock = data tiers" -- an error inverts it, failing to load a TOWN because its STOCK identity is absent; tier 0 must always boot. (2) Charter risk #3 names the remedy for identity gaps: "fail-soft to tier 0, never mis-stocked" -- missing IS the soft case, duplicate is the hard one; they are different failures and now get different arms. (3) Sequencing: newtown_godverse carries 0 ids today, it is G's to re-emit in wave 2 (their namespace, no census DB on m3), so a hard error would red the tree for C and E through all of wave 1 to say what the warn already says. The requirement keeps its teeth where R23 taught us to put them -- the VACUOUS-GATE LAW: selfcheck FAILS on partial keying (a half-keyed town silently mis-stocks the rest) and prints an explicit "⊘ SKIP ... 0/18 shops carry godverseShopId. Subject absent ... binds the moment one id lands" at zero. It names its subject, proves it touched it with a count, and cannot pass vacuously. FILED FOR G BEFORE THEY START (measured in wave 1, so it costs them minutes not a wave): redhill_godverse cannot contain Monster Robot from EITHER source. godverse_town.py takes roads from the OSM donor and shops from the census, and the shop is in neither -- not in the census (G's R23 finding), and NOT in E's redhill_real, which I checked the moment it landed: 36 shops, ZERO of type `record`, nothing matching /monster|robot/. As written the adapter yields a Red Hill of census charity shops while the atlas keys to a shop that isn't in the town -- F's #7b id-equality gate would fail at the end of the round. It must be injected explicitly; the opaque-key design exists precisely to serve that case. Second trap on the same path: redhill_godverse must still clear MIN_TOWN_SHOPS (6) from the census bbox or the cache is rejected outright (hard error). E's donor is otherwise good: valid, no warns, 73.3 m median spacing. HOUSEKEEPING: removed toowoomba's two stale pins -- E retired the cache in R23, and a golden for a town that cannot load can never fire (the vacuous-gate law, one layer down). E's attempt also proved my R23 spacing metric is GAMEABLE: the re-bbox fixed the median (395.7 -> 89.4 m, "passes") while hub density fell 3/12 -> 2/12, the pack's worst. It is necessary, not sufficient -- D's hub test is the complement, and E used both and refused to game it. Ballarat (255 m) is still flagged and still E's call; I pin what moves. PINS NOT TAKEN YET, DELIBERATELY: redhill_real is untracked/in-flight (E still editing build_towns.py), and redhill_godverse + the re-keyed newtown_godverse don't exist until wave 2. Pinning a moving file buys a false green or a spurious red. The round's law says A pins these BATCHED -- that batch lands when E commits and G's caches are in. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
ba901d4a75 |
Lane C R24: the contract fix — the runtime is the authority (ledger #1, the root cause)
F held the tag because my published atlas contract and my own loader were mutually
incompatible. That was mine. Closed, and published FIRST so G and E re-emit against it.
WHAT WAS WRONG: my R22 §7 invented `stock_shop_<godverseShopId>_index.json`; the loader
(stockpack.js:25) fetches `${base}stock_${type}_index.json`. An index named per the contract
can never be fetched by the code meant to load it. G built runtime-compatible (correct), E
built contract-compatible (also correct — the contract said so), and E's validator globbed
stock_shop_*_index.json -> zero files -> vacuous pass. Nobody was wrong; the contract was.
§7 REWRITTEN AGAINST THE CODE (LANE_C_PUB):
- Authority rule: stockpack.js IS the contract; doc vs loader disagreement = doc bug.
- Layout ratified as G shipped it: assets/stock_godverse/<godverseShopId>/stock_<type>_index.json
(`type` is load-bearing — SLOT_FOR picks the slot). stock_shop_* RETIRED.
-> E: the glob becomes stock_godverse/**/stock_*_index.json (the vacuous gate's actual cause).
- Runtime-read vs ignored stated exactly: reads atlases[] + items[{atlas,uv}] (uv top-left, the
loader flips V); ignores version/atlas_px/cell*/count/kind/tier/shop/crate/provenance.
- Provenance drift dies: runtime reads none of it, so the authority is E's validator + the 23
shipped town caches -> TOP-LEVEL `license` (US), attribution, generator, snapshot. British
`licence` retired; G lifts+renames those four, keeps its rich nested provenance{} as colour.
atlas_px stays REQUIRED (checked against real dims). cell* informational — C reads uv, never
cell; do NOT shrink the cell for VRAM (the dig renders covers close-up).
CODE (F's filed item): stockpack cache identity is now `${base}|${type}`;
getStockPack(type, base = DEFAULT_STOCK_BASE); new opts.stockBase on buildInterior = the seam F
wires. Omitted => town-wide packs, byte-identical to today. Pre-v5 callers untouched.
VERIFIED FRESH — and I fixed my own metric first: R23's "8/8 atlas-backed" only checked that a
map exists, but a parody canvas has a map too (same species as the vacuous gate). Re-asserted
falsifiably: every sleeve's texture source IS stock_godverse/3962749/stock_record_atlas_00/01.webp,
pack items == index items by id; negative control — the parody room's textures are (canvas, no src)
and touch the atlas zero times, so the metric discriminates. Collision FIXED (shop 120 + town 350
coexist). Fail-soft FIXED with a populated cache (missing base -> null -> parody). 98 draws with
real covers, pathOK, 0 carves, 0 console errors.
SCOPED NOT SLID: the LRU+dispose (§7.6) lands at v5.0-beta — one stocked shop can't accrete, and
safe eviction needs a room-exit hook (a live room holds the pack's shared material) that arrives
with F's per-shop preload. Written into the contract so it can't slide silently.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
0740b4697d |
Fable (integrator): round-24 instructions — the alpha completion (contract fix, identity field, redhill, honest gates) + two new laws
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
m3ultra
|
57973e0540 |
Lane F R23 (v5.0-alpha): the alpha gate — HELD. The crate is real; it has no street.
v5.0-alpha is NOT tagged. #7a is not a gate that failed, it's a gate that cannot be run:
G's 120 real records are in the tree and byte-identically re-bakeable, and nothing in the
game can reach them. R18's law — release rounds don't ship known breaches.
GREEN (everything except the crate): qa.sh --strict --matrix 7 passed / 0 failed / 0 warn ·
MATRIX GREEN 10 towns x 7 gates · selfcheck 152,015/152,015 (was 161,300, E retired
toowoomba) · flags harness green, so ?classic=1 still holds through the merge · tier 0
?noassets zero fetches + zero console errors (the offline law holds) · tier 1 (v2
town-wide) resolves. Tier 1 (v5, per-shop) is NEVER FETCHED.
FINDING 1 — the atlas contract contradicts itself. LANE_C_PUB:271 publishes
stock_shop_<id>_index.json; C's own loader (stockpack.js:25) fetches
${base}stock_${type}_index.json. Incompatible. G built runtime-compatible (which is why
C's base-override check passed); E built contract-compatible (which is why E's validator
can't see it). Nobody was wrong — the contract was. F recommends amending to G's shape:
it's the only one the runtime can load, and it costs G nothing.
FINDING 2 — the gate built to catch that was blind and green. validate_atlas.py globs a
filename that can't exist at runtime, finds nothing, and returns 0 ("No atlases yet =>
clean pass"), which validate_manifest.py:158 propagates. It conflates "no atlases exist"
with "no atlases matched my glob" — the first is a clean pass, the second is the exact
failure it was written to catch. Pointed at G's index by hand, deeper drift surfaced:
E's line 68 reads provenance top-level while E's own docstring says nested (where G put
it), licence vs license, and a ceiling stale by one round.
FINDING 3 — #7a would have passed vacuously. Walked into The Vintage Record (real shop,
real road) via DBG.enterShop: the crates filled with real covers — from E's 350-item
town-wide v2 pack, not the crate. Zero requests to stock_godverse/, ever. Eyeballed as
written, the gate goes green and proves nothing. Two vacuous gates in one round, same
species: both test the presence of a thing rather than the identity of it.
R24 — three gaps, none F-only: A+G land a godverse_id (verified live: every godverse shop
reports null; the schema has no such field) -> E builds redhill_real (one Overpass fetch,
E's file, outward-facing — not F's to invent) -> C keys the pack cache by type+base
(today a hit returns before base is read, so two stocked shops collide and fail-soft
breaks) -> F passes the per-shop base and re-gates. F deliberately did not build its half:
untestable with no street and no identity field. Filed, not guessed.
Also: A's spacing warn earned on day one — ballarat now warns at 255 m, so the thin tail
has a second member. The README v5 teaser is held with the tag; they're one claim.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
d2eb9df03f |
Lane C R23 (v5.0-alpha): the first real crate verified in the room (ledger #3)
Verified G's stock_godverse/3962749/ (120 real records, Monster Robot Party crate 550, 2 x 2048^2, cell 256) in a fresh context — deliberately BEFORE the merge, so a contract failure would block it. It didn't. PASS — the contract holds; G re-emits nothing: - Resolves through the existing stockAdapter path via a base override (no C code change for the single-shop alpha). 120 items / 2 atlases / 8-of-8 stock meshes atlas-backed (real covers: peperecordings, DEEPCHILD, "39", mff FREAKS...). - Ceilings hold: 118 draws with real covers (<=350 law), pathOK, 0 carves. - Fail-soft holds (clean cache): missing atlas -> null -> parody (20 meshes, 110 draws); no adapter (?noassets shape) -> parody, zero fetch. 0 console errors. - G's index compliant: items[] sorted by id (my determinism ask honoured), provenance, uv in [0,1], 2 atlases = exactly the cap, cell_w/h harmless (C reads uv, never cell). CORRECTED MY OWN CEILING (LANE_C_PUB §7.2): I wrote "<=8 MB decoded" from an assumed ~16-item "typical shop". A real record shop is 120 records => 2 x 2048^2 ~= 32 MB, 4x my number. The structural rules held (<=2 atlases; 2048^2 for 60+ items — G landed on the cap); only the MB figure was wrong, because it described a shop that doesn't exist. New: <=2 atlases @<=2048^2, LRU = 1 shop's atlas set resident => <=32 MB peak (bounded; vs E's 181 MB unbounded today). Do NOT shrink the cell below 256 to save VRAM — the dig renders covers close up. G's ask #1 CONFIRMED and worse than reported (LANE_C_PUB §8 — C's spec, C's fix): stockpack keys _packs/_resolved by type alone, so base is ignored on a cache hit (measured: the 2nd preload returned the identical object). (1) two stocked shops collide; (2) it BREAKS FAIL-SOFT — an unstocked shop of a cached type inherits that crate instead of parody. Seam: key by type+base, per-shop preload on entry, LRU=1 + dispose; C consumes whatever identity field lands, C does not invent it. Two gaps block F's #7a, neither C's: the crate has no street (redhill_real absent; monsterrobot not in any shops[] — G's ask #2 to E/Fable), and interior_mode.js:192 pins base to assets/models/ (needs §8 + the identity field). Committed from a temp git worktree on main — the shared checkout stayed on lane-G-godverse (Fable's review) and G's branch keeps its disjoint namespace. Docs-only. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
m3ultra
|
84e0a1b764 |
Lane E R23 (v5.0-alpha): the atlas validator + toowoomba retired (ledger #4)
1. validate_atlas (the G2a format gate). pipeline/validate_atlas.py, wired into validate_manifest so it
runs in qa gate 3. Judges Lane G's tier-1 atlases from the COMMITTED FILES ALONE -- no dealgod DB, no
network. No atlases yet => clean pass, so it's in place before G's lands. Enforces: provenance
{license, attribution, generator, snapshot} (charter law #3 + determinism); NO private-individual item
fields (law #3); 1 atlas/shop hard cap 2 (C §7.1); <=1024², 2048² only for a 60+ item shop (C §7.2);
declared atlas_px == the atlas's REAL pixels + every listed atlas exists; schema/UV/bands/dup-ids
(C §7.3). The licence string is PRINTED, not pattern-matched -- clearing photos for public release is
a human call, so the gate surfaces it. Dependency-free PNG/WebP header parser. Tested against a
10-case matrix (real 2048² WebP + 1024² PNG): good passes clean, all 9 violations caught.
2. toowoomba RETIRED (my call, counted). Tried the re-bbox FIRST: 3.0km CBD-wide -> 1.6km on the Ruthven
St heart. It fixed the named metric (median NN 395.7 -> 89.4m raw, 'passes') but made the real thing
WORSE: hub density 3/12 -> 2/12 within 160m, the pack's worst (D's alive-darwin is 7/12). Its shops
are PAIRS strung along a road, not a high street; densest 300m cluster is 3 shops (< MIN_TOWN_SHOPS),
so no tighter box exists. Shipping it would have gamed the number while leaving a town nobody shops in
(D: 0 finds / 1417 checks). Config entry kept commented with the rationale. Pack: 21 real + 1 godverse
= 22 rostered, 1180 shops, all 8 states/territories (QLD keeps westend). selfcheck ALL GREEN 152015.
-> Lane A, before you freeze the spacing warn (#5): (a) your validator sees the CACHE (raw lat/lon) but
D's numbers are SEATED (post-lift) -- darwin is 27m seated vs 68.3m RAW, so a threshold picked off D's
table would FALSE-WARN darwin; calibrate on the raw spread (daylesford 11.6 ... braddon 119.2,
ballarat 254.6 -- a ~200m line has daylight). (b) median NN alone is GAMEABLE: my re-bbox improved it
4.4x while the town got worse -- the hub fraction is the discriminator, and my raw hub metric
reproduces D's darwin exactly (7/12), so it's directly encodable (alive ~38-75%, dead <=20%).
(c) ballarat is a SECOND spread town D's two-town comparison missed (NN 254.6m, hub 4/20 = 20%) -- it
will trip your warn, correctly. Not retired: the ledger named toowoomba only.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
m3ultra
|
47f05ecaa8 |
Lane D R23 (v5.0-alpha): the GIG_RANGE chunk-neighbour fix — scoped to the gig path (ledger #6)
My carried v4.x finding, fixed post-release as scheduled. Runtime-only: no plan, no
golden, no flag, no fetch. selfcheck ALL GREEN 161300/161300, 0x3fa36874 frozen.
THE FINDING GREW WHILE IT WAITED: filed at 8.7% (measured at katoomba's isolated pub);
A's R21 cluster bias made it WORSE — venues now sit inside the retail cluster where peds
converge from every side, so 42.6% of in-gig-range peds (5094/11971) were chunk-blind.
THE OVERRIDE (measurement wins): the brief asked for "extend to chunk + 8 neighbours"
AND "byte-identical flags-off". I built the literal version and measured that those
cannot both hold: on the synthetic town (what ?classic=1 boots, gig layer absent) a full
3x3 scan sourced 127 of 855 finds (14.9%) from neighbour chunks — finds v2 never made.
The frozen v2 crowd would shop ~15% more. Covenant breach, in a clause of the same brief.
So the fix is scoped to the GIG path — which is what I actually filed:
- ordinary shops: own chunk only (v2 semantics, frozen — noticed from its own block)
- gig venues: the full 34m radius, honest across chunk edges ("follow the sound" is the
whole reason GIG_RANGE 34 > PATRON_RANGE 18)
- no gig on -> _gigVenues empty -> neighbour loop skipped entirely -> ?classic and every
flags-off boot take the IDENTICAL code path and cost to v2. Byte-identical BY
CONSTRUCTION, not by measurement luck.
Both ranges < CHUNK(64) so 3x3 is exact, not approximate; fixed dz/dx order keeps ties
deterministic as the old single-list scan was.
BEFORE -> AFTER: cross-chunk in-range peds seeing the gig 0% (structural: v2 scanned only
the ped's own chunk) -> 100% (katoomba 4/4, fitzroy 11/11). Same-chunk unchanged 100%.
Classic neighbour-sourced finds: 0 (127 under the rejected full scan). Determinism
identical. 0 NaN on katoomba/fitzroy/synthetic. Sim 0.056 -> 0.057 ms/frame (+2%, i.e.
0.006% of a 16.7ms budget); lookup 0.58us unchanged with no gig, 1.3-2.7us with a gig on
(fires per PATRON_STRIDE, not per frame). Fitzroy clean at 376 peds.
CARRIED, documented: PATRON_RANGE keeps v2's own chunk-blindness — fixing it costs the
classic covenant, and the covenant wins. A frozen v2 quirk now, not an oversight.
Known minor: a scheduled-but-closed gig still runs the loop for nothing (0.58->1.3us).
D->F: no blocker for v5.0-alpha; classic untouched by construction.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|