docs(overhaul): confirm dungeon static-light starvation in live frame

This commit is contained in:
Erik 2026-09-05 14:02:10 +02:00
parent 59b0b6e72f
commit 491e8adefa
3 changed files with 75 additions and 0 deletions

View file

@ -788,6 +788,12 @@ Ordinary short Neftet correctness follow-up PASS/no crash on the same binary,
graceful exit0; not a long soak or recurrence guarantee. Evidence:
477-integrated-validation-r2.md. FPS/C2 deferred; G4FAIL/no merge.
**Lighting observation (packet§58):** integrated59b0b6e72 Release0W0E/128PASS.
Town High/off/High all show the activated defect: eleven cells take eight
dynamic lights and omit all static room lights; seven selected portals are
4068m away with9m range. No lighting repair yet. Bounded static/dynamic
separation design follows exact retail ranking-frame trace; no FPS work.
## 14. Risk register (condensed)
- **R1 second scene graph.** Graph records own only topology/membership facts;

View file

@ -64,3 +64,63 @@ confirmed, exit0, absent before next client. Next bounded read-only trace
targets the candidate authored assemblies in479-rynthid-location-evidence.md;
do not guess an origin, quaternion, animation or package fix. #479 remains
unresolved/FAIL acceptance, G4 FAIL. No main merge.
## #478 consumed-light R2 — omission reproduced, not fixed yet
Exact clean Release59b0b6e72f61d5484afded27dd82bea0a68dbfb2,
full build0W0E/10.71s and128 related tests passed, no skips. Seven diagnostic
blobs match reviewed41128d816; ordinary lighting code is unchanged.
Run `logs/selfgate-20260905-135801-478-town-consumed-lights-r2/`, PID2840,
unchanged Town High/off/High route/session/profile. ACE9000 PID13340 up,
both clients absent before launch, no build/test alongside the client.
Actual Khronos validation DLL observed. Route/graceful logout/exit0; both
clients absent afterward. No native validation/device-loss/unhandled matches.
All3 checkpoints complete/visible, failures0, publication/retirement/upload/
warmup0. All3 original1600x900 PNGs opened: corridor remains dark in both
profiles, High darker. **Visual FAIL remains; diagnostic observation PASS.**
The actual consumed input establishes an activated defect, not merely a
possible eight-light-limit explanation:
- All3 checkpoints have126 registered/selected global sources:48dynamic,
78static. This is below the current128 global cap, so registration order
survives; no global overflow sorting occurs.
- All11 consumed cell sets are identical: `[0,79,80,81,82,83,84,85]`.
Index0 is the viewer fill. Indices7985 are the first seven registered
portal lights, owners000F42B9/BA/BB/C3/C4/CC/CD. All eight slots are dynamic;
**zero static room lights are submitted to any of those eleven cells.**
- Those seven portal lights have9m range and sit40.3367.67m from the player
at(70,-60,.005). Their range cannot illuminate the player vicinity; the
viewer light remains. Distances are explanatory calculations from the
captured positions, NOT a measured vertex/pixel contribution.
- Authored nearby static owners40007113/40007114 in00070143 and4000712F in
corridor00070144 are present as global indices38/39/40, but none appears
in the consumed sets. Ambient is exactly0.2 in each channel throughout.
This demonstrates static-light starvation plus registration-order dynamic
selection. The named/paired evidence in478-480-readonly-diagnosis.md confirms
retail's separate products and pre-cap insertion order. A genuine repair must
preserve the static product separately from the dynamic hardware pool; merely
raising ambient or sorting the same eight shared slots does not repair both.
No new lighting algorithm is committed. Bounded repair design/ranking-origin
trace is in progress; do not infer exact matched-retail brightness yet.
| Under run artifacts/screenshots/ | Verdict | SHA256 |
| --- | --- | --- |
| 478-00-town-high.png | FAIL: dark corridor, omitted static lights | 0965F915B9FF1523BCBBADE19F188823972A0A6B453B1CA0275944A9A138060A |
| 478-01-town-retail.png | FAIL: omission persists without High | AB2DC90032995A24AB5187899D7FED28A2AA829B5CE54D81D0DFFB7ED5594764 |
| 478-02-town-high-repeat.png | FAIL: repeated omission/darkness | EBB07B172CD9E3A14B70064FE6A604402C4A813FF2CC487716AD177B2FA762EF |
Additional SHA256 seals:
- App.dll3485E287961E299D468ECA96DE12AFBE83F59038FAC5E1E7FDE29510D47BF0CF;
App.exeCA7612EE3E4A096FF5061870E548E27537CBD1ED81104BB98E4766BA13C2F27D.
- artifacts/478-integrated-59b0b6e72/478-integrated128.trx:
DA1B960CD42826262B3B065EAA7AB7D86FCBDCB236F40782DFF8CC09FB886412.
- run artifacts/world-lifecycle.checkpoints.jsonl:
05130A6A126C72D975FFFDFC4C09A18A826C2F0D859D02572DCA7BED6D67E5C5.
- run client.log894F46794B6D7B6258B774ACD79A0816D564D5196D3EFB903DF12B61863CB1C4;
client.err.log0FB4E5A49893B406771DB1339712C10DAF191413718969999D1B427145554A3D.
FPS/#481/C2 remain deferred; incidental checkpoint performance fields are not
analyzed. #479 still needs the specific displaced component. G4FAIL/no merge.

View file

@ -4924,3 +4924,12 @@ Wegener narrow production/lifetime PASS/no blockers. Lead verification and
seals in478-observation-lead-verification.md. No per-frame path/algorithm
change or new divergence. Integrate exact blobs, then fresh build/tests and
unchanged Town Network run. FPS deferral and G4FAIL remain.
Integrated59b0b6e72 full Release0W0E/128PASS; unchanged Town route PID2840
graceful exit0, native validation clean. Actual inputs confirm the defect:
126lights(48dynamic/78static), eleven consumed cells all use viewer plus the
same seven distant portal lights, zero statics. All3High/off/High checkpoints
agree; all3PNGs opened and darkness remainsFAIL. Observation evidence/seals:
478-479-graphical-observations.md R2. Next: bounded genuine static/dynamic
separation repair, after exact ranking-frame provenance is statically checked.
No FPS work or speculative ambient increase; both clients closed, G4FAIL.