Open the books

RAW FILE: timing.md, as written during the jam

Source game/devlog/timing.md in the jam repo. Last git commit of this file: 2026-11-14. About 5,341 words. Scrubbed only for private details and one line about another entrant. Mistakes left in on purpose.

Director's time sense vs the clock

Don, Thu 23:21: "I think you also have a misconception of time, so let's see how long it takes and what time it really is when you are done." The director writes an estimate before each ticket starts; the actual comes from date at the worker's start and end. Misses get a line on why.

ticketstarted (real)director's estimateended (real)actualmiss
Sets build, rounds 1-5 (laptop)Thu 22:001 hFri 00:112 h 11 minunder-estimated 2.2x: five package rounds chasing size, plus a clip-overwrite bug found and fixed
Seam audit plus batch 16 and 16b recovery moves (tower 1)Thu 23:051.5 hFri 00:221 h 17 minclose (over-estimated 1.2x), even with a second round
Proof tests P1-P6, Blender halves (tower 2)Thu 23:161.5 hFri 00:0852 minover-estimated 1.7x
Batch 17 brawl and cannon motion (tower 1)Thu 23:211.5 hThu 23:4120 minover-estimated 4.5x: generation is 3 min; the queue wait and ranking are most of it
T1 world-bake tooling (laptop, after sets)2 h
T3/T4 brawl rebuild: branches, heavy's moves, chair, rope, table, bandstand; the Fable 5.2 low-effort test (laptop)Fri 03:443 h (my laptop estimates run 2x low, batch estimates 2-3x high)Fri 05:582 h 14 minover-estimated 1.3x
T7 cannon (laptop)2 h
Fun and UX tickets 1-8 plus QA fixes (laptop, Sonnet)Fri 00:124 h (plan says 13 h; my estimates run 2-4x high on batches, 2x low on laptop packaging)Fri 01:471 h 35 minover-estimated 2.5x: packaging took 1 to 1.5 min (the cook was cached), the builds and 36 headless tests 2 min each, and the long part was about 45 game runs on the packaged build
Heavy's chair swing keyed in Blender plus baked chair break (tower 2, Opus)Fri 01:051.5 hFri 01:3328 minover-estimated 3.2x
Batch 20 constrained regeneration of all flub branches (tower 1, Sonnet)Fri 01:331 hFri 02:2552 minclose (1.15x), even through the outage
Cart vault and rope swing keyed in Blender (tower 2, Opus)Fri 02:0550 min (chair swing took 28)Fri 02:4540 minclose (over by 1.25x), through the outage
Set pass 2 placement plus MOSS sound swap (laptop, Sonnet)Fri 02:121 h 15 minFri 03:431 h 31 minunder-estimated 1.2x (the face rule and a coverage round)
Three cheap extra faces (tower 2, Sonnet)Fri 03:0545 minFri 03:3833 minover-estimated 1.4x
Peanut wagon plus keyed vault (tower 1 then 2, Opus)Fri 02:451 hFri 03:0520 minover-estimated 3x
Cannon prep: helmet, box stack, keyed real-speed flight and back-first landing (towers, Opus)Fri 03:441 hFri 04:481 h 4 minon the mark
T7 cannon build plus chase and cannon rebuild with the builder fixes (laptop, Opus)Fri 05:592 hFri 07:411 h 42 minover-estimated 1.2x
Brawl contact ticket 1: staging solver (laptop CPU, Sonnet)Fri 06:161 h (plan says 1.5 h)Fri 06:5438 minover-estimated 1.6x
Brawl contact ticket 3: batch 23 new moves (tower 1, Sonnet)Fri 06:5450 min (plan says 1.5 h)Fri 07:4147 minon the mark
Brawl contact ticket 2: round trip and SOMA converter (tower 2, Sonnet)Fri 07:4145 min (plan says 1 h)Fri 07:5918 minover-estimated 2.5x
Package profiles jam and ultra, size back under 860 MB (laptop, Sonnet)Fri 07:411 h 15 minFri 08:461 h 5 minover-estimated 1.2x
Batch 23d: four brawl moves re-pinned with the real converter (tower 1, Sonnet)Fri 07:5925 minFri 08:1213 minover-estimated 1.9x
Cannon fixes: ladder climb keyed, belly flop and stage fright branches (towers, Opus)Fri 08:0550 minFri 09:281 h 23 minunder-estimated 1.7x (two belly flop runs failed)
Brawl contact ticket 4: retarget to real bodies, plus extra faces and lip sync imports (laptop, Sonnet)Fri 08:461 hFri 09:1933 minover-estimated 1.8x
Brawl contact ticket 4 retarget and imports (laptop, Sonnet)Fri 08:461 hFri 09:1933 minover-estimated 1.8x
Brawl contact ticket 5: two-actor bake (laptop export rerun, then tower 2, Opus)Fri 09:191 h 30 min (plan says 2 h)Fri 10:441 h 24 minon the mark
Sink root cause, cannon wiring, walk-throughs and beats in (laptop, Sonnet)Fri 10:052 hFri 11:161 h 11 minover-estimated 1.7x
Brawl ticket 6: batch 24 branches plus two bridges (tower 1, Sonnet)Fri 10:4445 min (plan says 1 h)Fri 11:1228 minover-estimated 1.6x
Glass Box dev log first draft (writer, Sonnet)Fri 11:0550 minFri 11:149 minover-estimated 5.6x: the sources were already written down, and the time went on cutting 5,600 words to 4,600
Brawl ticket 7: build the baked brawl (laptop, Opus)Fri 11:162 h (plan says 2.5 h)Fri 11:2812 min, stopped at a blockerthe plan-first step paid off: one check on bone lengths stopped 2 h of building on a bake that misses by 20 cm
Brawl rework 4b/5b: real-body export and re-bake (laptop then tower 2, Opus)Fri 11:312 hFri 12:421 h 11 min, of which about 30 min was waiting for the fun builder's GPU lockover-estimated 1.7x: the one-clip probe showed the game offset is a constant per bone, so the export only adds it and the import only takes it off; the bake ran in 7 min
Fun tickets 10 and 15: hit feedback, barks, streak, subtitles (laptop, Sonnet)Fri 11:413 h (plan says 6 h)Fri 12:2746 min, of which about 8 min was waiting for the other builder's lockover-estimated 3.9x: both tickets sit on log lines and tables the earlier tickets built (KKRUMBLE, KKSFXLIVE), so the work was a few hundred lines and two builds
Barks, call/subtitle overlap, pause ticket 14 (tower 1 voice, laptop, Sonnet)Fri 12:291 h 30 minFri 13:2556 min, of which about 14 min was waiting for the brawl builder's lock (code written meanwhile)over-estimated 1.6x: the barks were two short queue jobs plus one retry round, the pause is one gate flag and one card, and the bot flags (-KKBotPauseAt, -KKBotPauseVia) made every check a log comparison
Brawl ticket 7b: build and package the baked brawl (laptop, Opus)Fri 12:432 h (shares the laptop with ticket 14)
SPOILED rail shooter prototype: track, weapons, five archetypes, The Wurst, bot, package (laptop, Opus)Fri 13:156 h 30 min (PLAN.md)Fri 15:472 h 32 minover-estimated about 1.5x: compiles took seconds (no engine rebuild) and the bot runs found the real bugs fast; the six tuning rounds were the slow part
Rig bake-off: Glaze and jerky stalker down four paths (FLUX>TRELLIS>rig, Claude-built Blender, Claude SVG, FLUX-referenced build) (towers 1 and 2, Opus)Fri 13:353 hFri 14:571 h 22 min, of which 58 min was the Path A job waiting in tower 1's queue (Paths B, C and D ran meanwhile)over-estimated 2.2x: Blender on the Threadripper built, rigged, tested and rendered each creature in seconds, so the only long pole was the queue
SPOILED: The Wurst proof, Blender on tower 2 plus Kimodo on tower 1 (build, mouth, regen, spider re-form, export, EEVEE renders; Cycles queued)Fri 13:08none written before the start (worker ticket)Fri 14:451 h 37 min to EEVEE clips, FBX and notes; the Cycles renders (spoiled-wurst-render-1 and 2) still wait in the tower 1 queueno estimate to compare; four look rounds and three mid-task scope additions (rename, AAA master, spider re-form) were absorbed
SPOILED audio: script, voices, MOSS effects, music, mix-direction reel (tower 1 queue plus laptop, Sonnet)Fri 13:08not estimatedFri 14:46about 1 h 40 min wall, of which about 35 min was Maya1 on the GPU, 20 min MOSS, 3 min music, and 5 min waiting for the queue; the sfx retry (spoiled-sfx-2) was still queued behind other jobstwo bugs found by checking numbers, not by ear: Pedalboard's Limiter adds make-up gain (voices came out at -12 LUFS), and a len() on a stereo array left the first mix demo silent
SPOILED concept art: all subjects, three looks, v2 fix pass (tower 1 klein queue, laptop sheets, Sonnet)Fri 13:08not estimated (about 1 h)Fri 15:031 h 55 min, about 75 min of it waiting in the tower 1 queue behind voice, bake-off and SFX jobs; GPU time about 6 min (about 260 images at 3 to 6 s)scope grew four times mid-run (Wurst rename and regen frames, new enemies and reveals, behavior shots and realistic painted look, spider form, fix pass)
SPOILED lead engineer round 2: feel, the swarm Wurst with latching spiders, track v2 (laptop, Opus)Fri 15:475 h to the r2 packageFri 17:52 (r2 package and its bot runs; track v2 not started)2 h 05 minearly by about 3 h: the code compiled clean on the first full build and most bot failures were data fixes in beats.json
SPOILED audio round 2: PA intercom and engine A/B, monster Sir Loin, Pru barks, fast Wurst, jingle remake, Score A fight loops, SFX prompt-versus-code test, Hybrid mix (tower 1 queue 2a to 2c, laptop DSP, Sonnet)Fri 15:05not estimatedFri 17:00about 1 h 55 min wall: 2a ran 15:19 to 16:04 (45 min of Maya1 and Chatterbox, then I stopped its 1 h CPU Whisper pass and redid it in 7 min on 12 threads), 2b 16:03 to 16:31, 2c 16:43 to 16:53; the rest was CPU DSP and reelsscoring on 4 default CPU threads was the long pole; found only by watching the log rate. Nothing was heard by the builder, picks rest on Whisper, Audiobox and CLAP
SPOILED concepts batch 2, art director pass on Don's review: new creature anatomies, Wurst swarm, new enemies, rail-POV sets, weapons, key art; four klein rounds plus refine upscale (tower 1 queue, Opus)Fri 15:05not estimated (about 2 h)Fri 17:092 h 4 min, about 75 min of it waiting in the tower 1 queue behind audio jobs; GPU about 25 min (about 360 images at 4 s, 51 refines at 15 s)one mid-task brief change (Don 15:05 notes: jerky out, new enemies, no hole-mouth) folded into round 2 before it ran; round 2 lost most frames to a porthole vignette from one prompt clause
SPOILED concepts round 5 after the reviewer gate: key art (one rail, cart under the wave) and bolt gun and torch as single two-handed tools (tower 1 queue, Opus)Fri 17:10not estimatedFri 17:2515 min; three short queue jobs (c2-5, -6, -7) ran with no wait; the text-to-image key art lost the wave, so two instructed Klein edits of our own round 3 wave fixed rails and cart insteadthe edit route beat regenerating for a one-detail fix
SPOILED: The Wurst passes 2 and 3 (director's look notes, then Don's review: legged giant dropped, rebuilt as a swarm of franks; four EEVEE shots on tower 2)Fri 14:47none written (mid-task director notes)Fri 17:362 h 49 min: 20 min on the pass 2 look, 25 min of pass 2 renders lost to the pivot, 40 min to build and test swarm_build.py, about 85 min of EEVEE renders (4 s a frame), including 40 min re-rendering two shots for Don's mouth noteno estimate to compare; two scope changes arrived mid-render
SPOILED weapons in Blender: nitrogen sprayer, blowtorch, bolt gun built from code, baked, FBX, 3/4 and first-person renders, one improvement pass (tower 2 CPU Cycles, Opus)Fri 17:26not estimatedFri 18:1145 min; builds take under 1 s, bake about 50 s and both renders about 90 s a weapon on 24 CPU threads, so iteration was cheap; most time went to look passesthe bolt gun still reads toy-like after the pass; black rubber shows grey under the big soft key lights
SPOILED weapons pass 2 on the director's review: bolt gun redesigned as a 1979 captive-bolt stunner, gloved first-person arms per weapon, rime frost and vapour, hung tanks, torch flame (tower 2 CPU Cycles, Opus)Fri 18:12not estimatedFri 18:4129 min; three full runs of about 8 min each on tower 2, the rest was code and five quick first-person framing teststhe bolt gun's grips sit 15 to 20 cm under its bore, so its hands stayed out of frame until the gun moved 62 cm forward and rolled toward the camera

| SPOILED weapons pass 3 after Don rated the three 4 stars: braided bolt-gun hose with brass crimps, LN2 rime geometry and drips, thicker torch with knurled grip and red wheel, wear and Larkmoor Foods tags, rebuilt rubber gauntlets (tower 2 CPU Cycles, Sonnet) | Fri 19:05 | not estimated | Fri 19:37 | 32 min; builds under 3 s, three parallel bake and render runs of about 5 to 8 min each, two look iterations; most time went to code | braid weave needed 9 strands per direction (16 mushed into noise at 2048 texels); flat zero-thickness text kept stencil tris down | | SPOILED: The Wurst swarm look pass (raw pink franks, hood, dark dressed factory, tiled partition; Cycles finals on tower 1) | Fri 17:36 | none written | Fri 19:48 for the look build, EEVEE review clips and the rise in Cycles; forms, leaper and reveal still rendering (spoiled-wurst-swarm-2) | Cycles on the A5000 is about 7 s a frame at 48 samples with volume haze; the hung encode cost 85 min of queue | no estimate to compare | | SPOILED round 3 lead engineer, laptop: lighting and post, Poly Haven and own-kit set dressing, food materials, weapon pass 2 with gloved arms, jam package, packaged bot runs, 82 s clip and 8 stills (Opus) | Fri 17:56 (lock taken; work began earlier in the session) | not estimated | Fri 19:59 | 2 h 3 min on the laptop since the lock; three packages of 35 to 67 s each, and two benchmark capture runs of about 15 min each (PNG writes run the sim at about 0.6x real time) | the mite and yellow-ball fix took three wrong guesses before tag-colour tests found the real causes (debris size, then subsurface Opacity running backwards) | | SPOILED creatures 1: frank spider rigged in Blender from code (octodog cut, 45 bones, 9 clips at 30 fps, 2048 bake, FBX, VAT for crowds, EEVEE review), tower 2 CPU (Opus) | Fri 20:01 | not estimated | Fri 20:25 | 24 min; a full build, bake, VAT, export and 543 review frames takes about 2 min on tower 2, so seven look passes were cheap | the first leg cross-section flipped 180 degrees a third of the way down, which put the pale cut face on top of every leg; seen only in the render, not in any number | | SPOILED factory environment kit: 97 modular pieces from code on a 100 cm grid (walls, coving, floors with drains, breakable grate, columns, drop ceiling, meat rail with switch and trolleys, smoke tree, catwalk, stair, ladders, freight lift parts, pipes, crawl duct, conveyors, fryer bank and hood, vats, lights, signs, office and canteen), 2048 PBR sets from Poly Haven scans, FBX LOD0 and LOD1, kit sheets, two EEVEE vignettes (tower 2, Opus) | Fri 19:58 | not estimated | Fri 20:43 | 45 min; a full kit build and export takes 8 s, thumbnails 3 min, a 96-sample vignette about 3 min | the first fryer vignette was a flat orange wash with a 1.6 m storage cart filling the lens; lower haze, narrow sodium pools and a moved camera fixed the look, not more geometry | | SPOILED creatures 2: frank spider rebuilt as 16 whole Wurst franks (Don's 20:28 note; the look pass and casing halves were dropped mid-way), Sir Loin (14-segment beef centipede on the rail, 104 bones, 9 clips, 4096 bake, parts), the Hanger (kit beef side rigged, 9 clips), tower 2 (Opus) | Fri 20:25 | not estimated | Fri 21:10 | 45 min; builds take 1 to 3 min, but EEVEE review frames slowed from 0.2 s to 3 to 5 s a frame while two other agents' Blender runs and the editor shared the W6600, so renders went to every second frame | three design changes arrived inside the hour; the look pass for the octodog spider was half built when the frank-swarm rebuild replaced it | | SPOILED env kit pass 2 from the director's review: beef sides from code (3 variants), cold-store unit coolers, icicles, cable tray and frost, fryer hall middle ground (tote dollies, rack carts, hose reel, basket hoist, outfeed line), detailed fryer bank, reworked trolley, small elbow and curve hangers, clean puddles, probe-baked stainless, both vignettes re-rendered (tower 2, Opus) | Fri 20:45 | not estimated | Fri 21:21 | 36 min; kit rebuild 40 s (the beef metaballs are most of it), a 96-sample vignette about 3 min | the beef took five looks: metaball lumps read as a snowman, a fat-rim mask measured against the back surface came out all fat, and fat albedo near 0.6 linear rendered white; still the weakest piece | | SPOILED creatures 2: yeast mite (6 dough legs, 32 bones, bud, pockets, 10 clips, VAT) and dairy blob (8 bones plus squash, stretch, reach and lunge morph targets, rind and curd material, 9 clips, 26-piece pre-fractured shatter set), built from code on the creatures pipeline, Blender 5.2 on tower 2 CPU (Sonnet) | Fri 20:26 | not estimated | Fri 21:30 | 64 min; each full build, bake and EEVEE review run took 3 to 11 min (three to four Blender jobs shared the tower), eight mite and six blob runs; the mite's surface took five looks | the first mite read as a pale clay cauliflower; two things fixed it: pockets cut in as real geometry with dark wet interiors and gluten strings, and killing low-frequency bump (8-bit normal maps terrace gentle slopes; a dither term and fewer broad lumps cured it); the blob's hat and clipboard only read once they stood out of the maw, and its cream drips still read as plastic tubes | | SPOILED creature look pass from the director's review: the Hanger (kit beef side with hinged rib-plate jaws, stretch strands, hook through the hock, two cracked bone legs, 18 bones), Sir Loin (14 fused beef cuts, fat caps, hanging rib ends, kit trolley hooks, brisket head, irregular legs), frank spider (23 franks, slim legs with cocktail tips, 7-frank body), Compound 7 C7 channel on all three, review stage built from the kit, Cycles heroes (tower 2, Opus) | Fri 21:10 | not estimated | Fri 22:51 | 1 h 41 min; Hanger build and renders 8 to 9 min, Sir Loin 36 min (4096 bake plus 400 EEVEE frames), spider 24 min | four passes on the Hanger and the Sir Loin hero rendered magenta three times: Cycles cannot read baked images still in memory (load the saved files fresh), and the Blender 5 use_nodes flag silently skipped the first fix; the Hanger's cut face is the kit's own flat graphic, which caps how real it can look | | SPOILED round 4 sound effects: 23 items, 96 MOSS-SoundEffect takes ranked by CLAP and Audiobox, 44 files to winners4 (audio worker, Sonnet) | Fri 20:02 | 2 h to landing | Fri 23:13 | 3 h 11 min; the job sat queued 20:04 to 22:39 behind spoiled-wurst-swarm-2 (Cycles, est 150 min, ran about 2 h 45 min), then ran 22:39 to 23:12 (96 takes, about 20 s each) | the wait for the GPU lock, not the work; the high priority flag does not preempt a running job | | SPOILED r4 content imports: 112 sounds, the 108-piece factory kit with LOD1 and 25 texture sets, five rigged creatures with 46 clips, morphs, shatter set, Sir Loin parts, two VAT crowds, masters M_Kit, M_KitMeat, M_Creature with Compound 7, r4_assets.json (laptop, content fork, Opus) | Fri 23:17 | none written (fork of the lead) | Fri 23:35 | 18 min; Interchange imports the whole kit in about 40 s, so five short headless runs and three bugs were the time | no estimate to compare | | SPOILED arms 2: rigged red rubber gauntlets over rolled navy sleeves (39 UE-name bones, 18,272 tris, 2048 PBR bake, posed per weapon on grip cylinders with a planar finger solver, trigger finger IK, one-frame pose FBX per weapon; tower 2 CPU Cycles, Sonnet) | Fri 23:06 | not estimated | Fri 23:42 | 36 min; build and bake 50 s, a pose plus three 128-sample renders about 2.5 min a weapon | first bake read pink and washed out only because the preview lights were 8x too hot; fingerprint voronoi rings looked like polka dots and were dropped; libraries.load crashed Blender until the view layer was updated, and removing the loaded dummy materials crashed it too | | SPOILED The Wurst real-size franks: SM_Frank 4 variants x 3 LODs (456/126/50 tris), 1024 PBR plus C7, wriggle and twitch VATs, hooded brute body (9,000 tris, tile textures, 5,000 placements), 4 Cycles review renders (tower 2 CPU, Sonnet) | Sat 00:11 | not estimated | Sat 00:40 | about 29 min; asset builds take 2 to 20 s each, renders 5 s to 90 s at 128 samples | the brute read as an armchair until the arms hung forward and the slick shrank; first frank normals were too strong (read as jerky) so strength went 9 to 3.8; a sin() power of a tiny negative put NaN into the VAT until clipped | | SPOILED beef sides (3) and dairy blob base mesh by the image-to-3D route: FLUX.2 klein 4B text-to-image, TRELLIS.2, a Blender clean-up (chain cut, neck saw, cut face paired with a fat side, 18.7k tris, 2048 bake, grade), corridor test on tower 2 (Sonnet worker) | Fri 21:22 | not estimated | Sat 00:40 | 3 h 18 min, of which about 2 h 15 min was waiting in the queue behind other jobs; the klein batch takes 1 min, TRELLIS.2 about 90 s a mesh on the A5000, a clean-up and bake 1 to 2 min on the laptop CPU, the corridor frame 5 min on tower 2 | the first klein round hung the carcass by its neck (rump at the bottom); one TRELLIS image gives a convincing front and an invented back (a second rib rack), so the final sides pair a cut-face mesh with a fat-side mesh; a voxel remesh to seal gaps deleted the thin shell | | SPOILED Hanger rebuild 3 on the image-to-3D beef sides: slot split (Lean, Fat, Bone) for the kit sides A/B/C with LOD1 into the kit root, jaw plates cut from the kit side, sawn rib teeth, two clawed bone legs, 9 clips, 1280x720 hero renders and a clip sheet (tower 2 Blender, Sonnet) | Sat 00:41 | not estimated | Sat 01:24 | 43 min; builds take about 10 s, the full review render about 9 min | the first stance read as stilts on a floating slab (legs thin, body 0.9 m up), so the legs got thicker with a knee kink and the body sat lower; the old stage's wall textures had silently gone missing (pink walls) until crlook's path fix; a heredoc ate a backslash in the first script | | SPOILED Lenny MetaHuman from KK Heavy_v1: navy work clothes, our hard hat, 62 lip sync takes (face and body-additive), 3 Kimodo clips retargeted, a code-project package on tower 2 without compiling (tower 2, Opus with two Sonnet workers) | Sat 00:34 | not estimated | Sat 01:57 | 1 h 23 min; the lip sync solve is 22 s a take (23 min for 62), Kimodo about 6 s a take, the package 80 to 110 s | the plugin pip install crawled at 15 kB/s until turned off for my project; the costume masters lacked the skeletal mesh flag (default material in game); rewriting pelvis keys lowered him both ways, so the floor fix became a 5.1 cm Body offset; the moustache binding crashed UE twice and was left for a rebuild; tower 2 fps is meaningless while the KKCast editor shares its W6600 | | SPOILED Hanger pass 2: meat legs with fat cap and cut-bone end, translucent fascia webs, fat-cap smear cut in half on the kit beef textures, body 10 percent lower (tower 2 Blender, Sonnet) | Sat 01:27 | not estimated | Sat 02:00 | 33 min; the full review render took 28 min because blended translucent membranes made the Eevee frames 5 to 10 times slower | the first leg pass folded into red blobs because the leg was longer than the hock-to-floor reach; shortening the bones fixed it | | SPOILED Lenny pass 2: moustache rebuild (one try), aviators, name patch, seed packet, gloves, flashlight with grip clips, seed crack and packet offer keyed in Blender, len_37/38 lip sync, take picks, hair re-check in a package (tower 2, Opus with two Sonnet workers) | Sat 02:00 | not estimated | Sat 04:38 | 2 h 38 min; the props worker alone ran about 2 h 20 min over three rounds; each UE headless step is 15 to 45 s, a package 60 to 110 s | the moustache rebuild needs the texture graph, which a commandlet cannot run, so it stopped after one try; tower 1 GPU was held by a queue chain all night, so the two new clips were keyed in Blender instead of Kimodo; the seed crack still enters the face and packet after one fix round; the grip needed the arm swung out up to 22 deg to clear the coat | | SPOILED Lenny pass 2b: flashlight arm under 12 deg (re-seated grip), seed-crack fallback (shoulder, neck and jaw curve, no hand raise), moustache one-command rerun script with a running-editor guard, stage2 handed to the integrator (tower 2, Opus with a Sonnet worker) | Sat 04:39 | not estimated | Sat 05:08 | 29 min | the hand-at-pocket seed crack could not keep the glove off the face (pocket 12 cm under the mouth), so the hand raise was dropped; paused at 05:08 by the director until 08:00 | | SPOILED area layouts, layout artist A: 13 areas from the kit plus Poly Haven (dock, cold store, cutting floor in this chat; six by forks), signage pack placed, look rig fixtures, Unreal import of all 13 plus B's grinder hall on tower 2 (nolights, 103 sign MIs), apply_look previews and the first in-engine 4K look captures (tower 2, Opus with four Opus forks) | Sat 00:33 | not estimated | Sat 05:32 | 5 h; a Blender area build is 20 to 40 s, an EEVEE preview pass 5 to 12 min, a 13-area UE import under 3 min, a 4K MRQ capture about 20 min a room on the W6600 | MRQ frames were pure black until the map URL forced GameModeBase (SPGameMode's director blacks out the view); add_level_to_world makes the sublevel current, so preview fog, PPV and cameras landed in L_Packing and the lighting levels; the sign master never compiled (sampler type vs default texture); a killed ssh left a remote lock waiter that held the GPU lock for 40 min | | SPOILED round 4 lead (laptop): HUD without film references, fun gate fixes, real-size Wurst (25k to 40k franks), layout streaming plus look rigs, Niagara flame and frost, escape counter, ride tickets (accel, ring keys, Forced turns, Big Screen, the turnaround), boss perf, gate package, 90 s capture and 8 stills | Fri 20:05 | not estimated | Sat 05:37 | about 9 h 30 min; editor builds 15 to 110 s, packages 1 to 3 min, a full bot run 15 min, the 6-segment capture 6 min | the boss was render-thread bound (25k franks in the Lumen scene); TSR's 200 percent history made 4K output an 8K buffer; DirectoriesToAlwaysCook skips .umap so no layout shipped until the -map list; this ffmpeg's NVENC needs a newer driver; a blocking helper call cost 27 min | | SPOILED Lenny lip sync on the final voice files (33 lines, game_final after the American English pass), body-additive copies, stage3 drop and delta to the integrator (tower 2, Opus) | Sat 09:07 | not estimated | Sat 09:23 | 16 min; the solve is about 25 s a line | none | | SPOILED Lenny lip sync lag fix: four late final lines ramp-shifted to a 0 to 50 ms jaw lead, stage3 refreshed and marked final (tower 2, Opus) | Sat 09:25 | not estimated | Sat 09:37 | 12 min | a plain shift cut the first syllable (speech starts 30 to 40 ms in), so the four lines were re-solved and ramp-shifted | | SPOILED creatures by the image-to-3D route: Sir Loin (segment and head), Yeast Mite, Dairy Blob as the Culture; klein stills, TRELLIS.2, Blender 5.2 and 5.3 rebuilds that keep every bone and clip name, review renders, FBX checks, director fixes (insect head, dough legs, shatter chunks 07 and 13) (Sonnet worker) | Sat 02:02 | not estimated | Sat 09:50 | about 7 h 50 min, of which roughly 4 h was waiting in the tower 1 queue behind other jobs; klein 1 min a batch, TRELLIS.2 about 90 s a mesh, a rebuild 1 to 2 min on the laptop CPU, a Cycles hero or clip sheet 5 to 15 min | TRELLIS decimation stalls at about half the faces on fused meshes (fixed by a voxel remesh first); the thin-shell head back looked like a red and white smear until it was remeshed; strand vertex indices were stale so the mite legs had no weights until an index update; shatter chunks with n-gons imported empty in UE |