Fr0zkaandClaude Opus 5.5 a176d0743a wip(assembler): round L14 -- far-LOD cost counted; no exact-output lever above 3% (opt-in counters) ...

Plain W12 far work: LOD6 2.59, LOD8 2.05, LOD4 1.57, LOD5 1.41, LOD7 1.16 worker-s. Ground/cap
refinement ~3.95 worker-s and LOD6 connector repair 1.40 worker-s (82% uncached formula calls)
build visible geometry; no bit-identical reduction proved. Adds opt-in far-shell column counters
(reviewed: no cost when off, interleaved 1.429 vs 1.419 s). All hashes, the crossing hash and route
parity are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CSFYgsdW8ejZSyiMf4qRSX

2026-10-20 02:33:48 +02:00

..

Linux1

wip(assembler): round L1 -- Linux baseline on abyss: Linux world identical to Windows

2026-10-19 18:18:34 +02:00

Linux2

wip(assembler): round L2 -- the cross-platform hash check can fail; FP contraction changes the world

2026-10-19 18:18:34 +02:00

Linux3

wip(assembler): round L3 -- Core forces FP contraction off and refuses to generate if it can't

2026-10-19 18:51:07 +02:00

Linux4

wip(assembler): round L4 -- lazy per-thread caches: 17.3 MiB -> 43 KiB per thread, view 7% faster

2026-10-19 19:16:14 +02:00

Linux5

wip(assembler): rounds L5+L6 -- Linux far-terrain captures and screen-space gate (world unchanged)

2026-10-19 21:10:08 +02:00

Linux6

wip(assembler): rounds L5+L6 -- Linux far-terrain captures and screen-space gate (world unchanged)

2026-10-19 21:10:08 +02:00

Linux7

wip(assembler): round L7 -- shared column cache across workers: whole view 16-19% faster, world unchanged

2026-10-19 21:46:15 +02:00

Linux8

docs(assembler): round L8 report -- where the time goes after the shared cache

2026-10-19 22:10:05 +02:00

Linux9

wip(assembler): round L9 -- LOD1 skips proven-uniform blocks: 95% fewer samples, view ~2% faster

2026-10-19 23:14:16 +02:00

Linux10

docs(assembler): round L10 report -- spawn waits on the full shape; 24 m crossing hits bad_alloc

2026-10-19 23:44:52 +02:00

Linux11

wip(assembler): round L11 -- fix 24 m+ crossing crash; near ring ready 2x sooner at spawn

2026-10-20 00:28:58 +02:00

Linux12

wip(assembler): round L12 -- region pages keep LOD1 block proofs: 24 m crossing 5x faster

2026-10-20 01:07:47 +02:00

Linux13

wip(assembler): round L13 -- one density field for every route (multiplayer determinism fix)

2026-10-20 01:54:54 +02:00

Linux14

wip(assembler): round L14 -- far-LOD cost counted; no exact-output lever above 3% (opt-in counters)

2026-10-20 02:33:48 +02:00

Round1

wip(assembler): M3 round 2 -- islands get real mass: deep crust, separate tapered roots, arched spans

2026-10-10 12:24:35 +02:00

Round2

wip(assembler): M3 round 2 -- islands get real mass: deep crust, separate tapered roots, arched spans

2026-10-10 12:24:35 +02:00

Round3

wip(assembler): M3 round 3 -- separated islands + Unreal island route (owner build 18:46)

2026-10-10 20:53:03 +02:00

Round4

wip(assembler): M3 round 4 -- plans built inside Unreal from the owner's strates; no import needed

2026-10-10 22:48:46 +02:00

Round5

wip(assembler): round S1 -- one Core voxel-size source; per-tile feature lists (islands)

2026-10-18 18:48:38 +02:00

Round6

wip(assembler): round S2 (Sol) -- shared per-tile feature queries for every Core family; island placement 30x faster

2026-10-18 19:50:45 +02:00

Round7

wip(assembler): round S3 -- far ground from the native heightfield; LOD7/8 tiles 36-66% cheaper

2026-10-18 21:29:52 +02:00

Round8

wip(assembler): round S4 -- continuous far ceiling at the strate's upper boundary

2026-10-19 00:41:13 +02:00

Round9

wip(assembler): round S5 -- streaming gates + trace; coarse-first order kept opt-in (no win)

2026-10-19 04:42:41 +02:00

Round10

wip(assembler): round S6 -- in-game cold view counted; workers mostly idle, apply cap not the floor

2026-10-19 06:46:54 +02:00

Round11

wip(assembler): round S7 -- workers kept busy: in-game cold full view 17.5 s -> 7.8 s

2026-10-19 08:10:10 +02:00

Round12

wip(assembler): round S8 -- region rebuilds no longer stall admission; result backlog capped

2026-10-19 09:43:03 +02:00

Round13

wip(assembler): round S9 -- collision look-ahead: ground ring ready on arrival up to 40 m/s and 50 m/s falls

2026-10-19 10:56:54 +02:00

Round14

wip(assembler): round S10 -- LOD8 horizon sampled at MC spacing: LOD8 worker time 18.8 s -> 4.3 s

2026-10-19 12:23:18 +02:00

Round15

wip(assembler): round S11 -- bounded spawn apply budget: in-game full set 7.1 s -> 6.0 s

2026-10-19 13:03:44 +02:00

Round16

wip(assembler): round S12 -- short ready queue keeps workers fed between ticks: full set 6.0 s -> 4.9 s

2026-10-19 14:29:38 +02:00

Round17

wip(assembler): round S13 -- tile cost counted per LOD (diagnostics only, no output change)

2026-10-19 15:40:25 +02:00

Round18

wip(assembler): round S14 -- 4-way column cache + memoized MC gradients (output unchanged)

2026-10-19 17:58:54 +02:00

HANDOFF-ABYSS.md

docs(assembler): handoff for the abyss (Linux) research machine

2026-10-19 16:11:06 +02:00