Fr0zkaandClaude Opus 5
ce53552543
I predicted this would recover the D-section's 30% outright, on the grounds that the
floor relief is FMath::PerlinNoise2D and therefore a function of (X,Y) only, so every
voxel in a column recomputes the same pair of octaves. Codex checked the premise and
it does not hold:
production applies a 3D CAVE WARP before the floor is evaluated, so the warped X/Y
fed to the noise change with Z. A world-XY cache would have changed the field.
That is the third arithmetic prediction of mine that did not survive contact with the
code (after the double->float round-trips, and "the candidate scan is the last big
win"). Recording it because the pattern is consistent: I keep reasoning about cost
from a call site instead of from the data flow reaching it.
WHAT WORKS INSTEAD
Cache keyed on actual SDF-domain integer columns plus feature, owner, seed and
frequency -- correct in the warped domain, where world-XY is not. Fixed size, 32,768
entries, ~1.25 MiB per worker, no per-tile or per-region retention.
28,154,092 column lookups
25,842,374 hits / 2,311,718 misses (91.8% hit rate)
71,133,837 conservative relief skips (amplitude bound, only where SmoothMax is
provably the identity)
77,109,353 actual two-octave evaluations
my build, 3 runs: 1.384 / 1.405 / 1.415 s (ff8bd7a was 1.509 / 1.542 / 1.543)
mean 1.401 vs 1.531 = -8.5%
geometry SHA-256 bc0dc059..a8ae, IDENTICAL to ff8bd7a's output
Capability unchanged exactly: player-fit 20,835, reachable 10,901, connected,
40,429 verts / 75,416 tris, dihedral p50/p90 2.5529/33.7191, hash 4520DC6F,
determinism repeat passed.
Codex also declined a sign-only shortcut for the main SDF, correctly: marching cubes
interpolates on the field MAGNITUDE, not just its sign, so a sign-only answer would
move vertices. It kept it for the landing helper where only the sign is consumed.
⚠️ NOTE ON MY OWN "17% NOISE FLOOR": that figure came from a profiled-vs-unprofiled
comparison, not from run-to-run variance. Within one binary the spread here is ~2%
(1.384-1.415). An 8.5% shift that is consistent across every run in both directions
is real. I have been using the wrong number as a threshold and should stop.
Net position: the D-section still costs ~17% against pre-swept-floor (1.401 vs 1.15),
in exchange for +682 player-fit and +418 reachable cells, floors that undulate instead
of presenting a slab, and the retirement of the 1.5M-call validator and two dead ops.
Build 47 actions, 0 errors. Only morphology source changed; zero reparse points;
RealtimeMeshComponent untouched.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aheo2crrZmHuYtPpdY2Cb