docs: retract G1.5 — both titles crack; partial coverage is not shippable

User verified seams on Ape Escape AND Tomba 2 with the exact-table plus
ambiguity-gate build. G1.5 claimed the gate fixed the cracking; it did not, and
G1.4's original conclusion stands.

Both wrong turns came from concluding on a single observation rather than a
controlled A/B: first from native captures that structurally could not show the
defect, then from one in-game frame where the corrected ~5% happened not to
border a visible silhouette.

The gate did remove vertices inheriting a neighbour's fraction, but ~5% coverage
still mixes corrected and uncorrected triangles inside a mesh and those seams
are visible. The crack is a property of the boundary, not of the error
magnitude, so no coverage ratio short of near-total is shippable.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Matthew Stanley
2026-08-05 16:02:18 -07:00
parent 9217ae203a
commit cd87fe403d
+28
View File
@@ -648,3 +648,31 @@ something that does not hurt.
**Still to verify first-hand** with `screenshot_hires`, before any of this is
called done: an A/B of the same frame with the knob off vs on, at
supersampling >= 2.
### G1.6 — G1.5 RETRACTED: both titles crack. Ambiguity gating is not enough.
User verification on the exact-table + ambiguity-gate build, **Ape Escape AND
Tomba 2**: thin seams across meshes in both. G1.5 claimed the gate had fixed the
cracking; it had not. G1.4's original conclusion stands.
**What went wrong in the analysis, twice:** each conclusion was drawn from a
SINGLE observation instead of a controlled A/B — first from native screenshots
that could not show the defect at all (G1.5), then from one favourable in-game
frame that happened not to expose it. A scene where the corrected ~5% of
triangles do not border a visible silhouette looks clean; that is not evidence
the seams are gone.
**Settled position.** Ambiguity gating removed one failure mode (vertices
inheriting a neighbour's fraction) but coverage of ~5% still mixes corrected and
uncorrected triangles within a mesh, and those seams ARE visible. Partial
coverage is not shippable at any ratio short of near-total, because the crack is
a property of the boundary, not of the magnitude of the error.
`geometry_correction` therefore stays OFF and must not be offered as usable
until precision propagates with the data (G1.2/G1.3). The launcher rows, the
`[video]` plumbing, the renderer-facade seam, `geom_correction` and
`screenshot_hires` all remain valid groundwork for that work.
**Method rule for the next attempt:** no conclusion about this feature from a
single frame. Same frame, same scene, off vs on, captured with
`screenshot_hires`, on at least two titles.