docs: record failed owner gate and bounded #477 validation recipe
This commit is contained in:
Erik
parent
de427d2c02
commit
8fd5e434b6
7 changed files with 422 additions and 8 deletions
|
|
@ -39,6 +39,61 @@ confirmed closed by the owner, 11 need a focused live gate, and 43 are safe to
|
|||
remain closed. See
|
||||
[`docs/research/2026-08-28-owner-closed-issue-validity-audit.md`](research/2026-08-28-owner-closed-issue-validity-audit.md).
|
||||
|
||||
## #481 — Low initial FPS after entering the world
|
||||
|
||||
**Status:** OPEN, owner-reported during the `de427d2c0` gate on2026-09-05.
|
||||
Owner felt FPS was initially low on world entry. No numeric FPS/frame-time
|
||||
capture was taken. Measure initial versus settled CPU/GPU frame percentiles,
|
||||
streaming/retirement debt and allocation on the same High/4xMSAA profile;
|
||||
do not label this fixed by #475 or reduce visibility/quality to conceal it.
|
||||
Evidence: `research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
## #480 — Wielded hand equipment is invisible
|
||||
|
||||
**Status:** OPEN, owner-reported during the `de427d2c0` gate on2026-09-05.
|
||||
Owner reports nothing wielded in either hand is shown. Attachment messages
|
||||
exist in the session log but do not prove render membership or draw admission.
|
||||
Trace exact root/child identity, canonical CELLARRAY propagation and the
|
||||
production draw consumer. No parent-cell fallback or duplicate membership owner.
|
||||
Evidence: `research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
## #479 — Hanging Rynthid structure partly shifted after recall
|
||||
|
||||
**Status:** OPEN, owner-reported during the `de427d2c0` gate on2026-09-05.
|
||||
The owner reports the hanging Rynthid over the crater is partly shifted after
|
||||
Rynthid recall. Exact object/part identity and cause are not yet established;
|
||||
do not assume wrong collision placement, streaming origin or Setup transform
|
||||
without reading the real asset and production transform path. No screenshot
|
||||
was supplied in this round. Evidence: `research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
## #478 — Dungeon lighting too dark, especially Town Network
|
||||
|
||||
**Status:** OPEN, owner-reported during the `de427d2c0` gate on2026-09-05.
|
||||
Town Network was barely visible; the run includes indoor00070143. The current
|
||||
High-profile indoor authored-light/material path needs comparison with retail.
|
||||
#469's outdoor direction fix does not establish indoor correctness. Do not
|
||||
blindly raise global ambient or change the owner's brightness settings.
|
||||
Evidence: `research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
## #477 — AMD watchdog / Vulkan device-loss crash while jumping in Neftet
|
||||
|
||||
**Status:** IN-PROGRESS, crash independently confirmed2026-09-05 on clean
|
||||
`de427d2c0`. Owner saw an AMD crash notice while jumping around Neftet.
|
||||
stderr records unhandled ErrorDeviceLost from the abandoned-frame timeline
|
||||
submit; shutdown also fails waiting for frame serial37509. Windows .NET1026,
|
||||
ApplicationError1000 for exact PID26920 and LiveKernelEvent141 corroborate it.
|
||||
The stdout graceful-logout lines are cleanup, NOT successful process exit.
|
||||
RX9070XT, reported Vulkan1.4.349/driver2.0.395, final1600x900, High/4xMSAA.
|
||||
|
||||
First raw acquire result is not retained: the swapchain adapter collapses
|
||||
fatal failure and ordinary unavailability to false, then the empty signal
|
||||
submission reports device loss. This is a diagnosis-obscuring boundary, not
|
||||
proof of the initiating GPU fault. No premature-release or semaphore-reuse
|
||||
cause established. Keep first-failure evidence distinct from secondary teardown.
|
||||
No driver replacement, TDR-delay changes, skipped waits or forced recovery.
|
||||
Exact stack, hashes, local Windows report and limits:
|
||||
`research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
## #476 — First post-resize graphical gate PNG is horizontally corrupted
|
||||
|
||||
**Status:** OPEN, observed2026-09-05 during #474 regression; capture-lifetime
|
||||
|
|
@ -61,7 +116,19 @@ and real first-frame/resize validation follow #475, not inside its chunk.
|
|||
|
||||
## #475 — Debt-free collision retirement yields a frame per prefix
|
||||
|
||||
**Status:** IN-PROGRESS, lead bounded contract2026-09-05; no fix landed.
|
||||
**Status:** REPAIR LANDED + REVIEW-CLOSED at `de427d2c0` (2026-09-05).
|
||||
Lead independently verified Release 0 warnings/errors, Runtime 1,891/1,891,
|
||||
Headless 48/48 and App 297/297. Retail review's test-only findings were
|
||||
reconciled in one fix round; production/lifetime review 2 PASS. Exact
|
||||
eleven-file reviewed return `ab989d717` matches the integrated Git blobs.
|
||||
Manual owner re-test completed five reveal generations with zero reported
|
||||
reveal failures. Owner narrowly accepts recall timing ("seems ok"); exact
|
||||
nine-hop benchmarking remains owed. Overall gate FAILED on #477–#481,
|
||||
including a confirmed device-loss crash. Logout cleanup is not clean exit.
|
||||
Evidence: `research/2026-09-01-overhaul/475-lead-verification.md` and
|
||||
`research/2026-09-01-overhaul/475-owner-regate.md`.
|
||||
|
||||
Pre-repair diagnosis:
|
||||
The #474 regression reaches all nine destinations, but uninstrumented
|
||||
adjacent hops take86–89s. The sampled retirement FIFO stops at the first
|
||||
resident-parking poll even with no pending withdrawal/restore or global
|
||||
|
|
@ -78,8 +145,9 @@ Do not increase budgets, reorder FIFO, bypass ACKs or reduce the reveal window.
|
|||
|
||||
**Status:** REPAIR LANDED + REVIEW-CLOSED at220bda797; lead nine-hop
|
||||
convergence PASS2026-09-05, provisional until owner inspection. Overall
|
||||
portal re-gate remains FAIL on separate latency #475 and capture #476;
|
||||
G4 is not accepted.
|
||||
portal re-gate initially failed on separate latency #475 and capture #476.
|
||||
After #475 the owner accepts recall timing narrowly, but G4 remains FAIL on
|
||||
#477–#481; #476 and the exact timing repeat are still owed.
|
||||
|
||||
Owner's repaired-exterior re-gate reaches portal destination30310005 but
|
||||
never reveals. In the captured heap, center3031FFFF has completed its render
|
||||
|
|
|
|||
|
|
@ -713,7 +713,7 @@ Update immediately when a slice changes state. Chat is not the ledger.
|
|||
| S5-#470 | **LANDED + REVIEW-CLOSED 2026-09-05; LEAD GRAPHICAL A/B PROVISIONAL PASS.** Campaign implementation stack `15a796c3a` -> `7506e5f14` -> `b333edb4f`; route `19b44e5e3`; reviewed scratch `51f974da4` -> `2cad9c84` -> `98c004aa7`; packet §§27–30. | G4 UNPASSED | Retail/behavior pass 2/10 and production pass 4/10 closed the two evidence-only findings. Fresh campaign Release 0W/0E and focused 48/48; exact scratch hermetic 16,976/16,976 and canonical InstalledDat 386/10/1, manifests 30/30. Pinned gate `logs/selfgate-20260905-032132-s5-470-pinned-shadow-ab-r1`: retail/off -> High -> retail/off, 3/3 PNGs, exit 0, graceful, no client; visual PASS provisional. The stale recipe-8 pre-route launch is excluded and recorded in §30. |
|
||||
| S5-#469 | **LANDED + REVIEW-CLOSED 2026-09-05; LEAD GRAPHICAL TRANSITION PROVISIONAL PASS.** Contract `809887524`; reviewed scratch `c09b6cf0f`; campaign implementation `94ddde69a`; route `62efc72cb`; packet §§31–32. | G4 UNPASSED | Retail/deviation pass 1/10 and production pass 2/10 both PASS with no finding. The two atmospheric vertex receivers keep retail's authored unnormalized `uLights` direction across every shadow gate; celestial direction remains for opt-in shadow/volumetric projection; IA-24 corrected in the same commit. Exact scratch: Release 0W/0E, focused 126/126, Vulkan 2/2, hermetic 16,983/16,983, InstalledDat 386/10/1. Fresh campaign Release 0W/0E, focused 82/82, Vulkan 2/2. Gate `logs/selfgate-20260905-040449-s5-469-lighting-transition-r1`: five PNGs, active 2,500-caster/four-cascade High rows, no near-black relight, exit 0/graceful, no client; visual PASS provisional. |
|
||||
| S5-c5 | **CLOSED + LANDED 2026-09-05; LEAD GRAPHICAL SMOKE PROVISIONAL PASS.** Contract `b77989c32`; campaign `bf53e2ad6` -> `e625dc4e6` -> `1b7ee4e58`; reviewed scratch tip `158656f0d`; packet §§33–35. | G4 UNPASSED | Retail/deletion pass 1/10 PASS. Production pass 2/10 found one static-field hole in the owner guard; fix round 1 reproduced both static-owner mutations, and production pass 3/10 PASS. Exact scratch hermetic 16,921/16,921, canonical InstalledDat 368/9/1, manifests 30/30. Fresh campaign Release 0W/0E, App 146/146, Core 8/8. A stale recipe-8 preflight is excluded; a non-destructive recipe-10 bake produced 2,237,865 keys with zero failures. Corrected gate `logs/selfgate-20260905-052130-s5-c5-landed-v10`: five PNGs, exit 0/graceful, no fatal/deleted-prefix match, no client left; geometry matches S5-c4, visual PASS provisional. |
|
||||
| S5 | **#473 EXTERIOR REPAIR OWNER-ACCEPTED `bf23673f3`; portal-reveal repair #474 LANDED + REVIEW-CLOSED, nine-hop convergence verified; latency475 in bounded repair (2026-09-05).** c1–c5 plus #470/#469 landed; C1c test/doc correction REVIEW-CLOSED, pending landing. | G4 UNPASSED | Packet §§7 and12–48 bind. §42 repaired opaque detail alpha-to-coverage, with both reviews PASS, Release0W/0E, affected158/158, mutation12/12 and all52 affected prepared payloads matching fresh extraction. §42.6 sampled exterior PNGs were provisional; the owner now accepts the repaired exteriors. §43 C1c retail/production reviews are closed; nine-file test/doc return remains unlanded, lead scoped68/68 only, no canonical full-lane claim. §44 manual owner session PID16892 ran unchanged bf23673f3. §45 geometry witness contract remains not dispatched. §46/new diagnosis474 proves a destination publication/placement acknowledgement cycle at3031FFFF: engine committed, exact valid Place into30310014 at global FIFO head, spatial backend absent and publication waiting on that acknowledgement. #474 round0 is integrated at220bda797 (reviewed scratch774a2cd1c), contract31960039b plus correctionfc1a24b6d: real reset/recenter converges publication before cancellation, so no new cancellation API. Retail pass1 found only a stale test comment, corrected directly; production/lifetime pass2 PASS. Lead Release0W/0E, App135/135 and Runtime42/42; three discriminating mutations with controller single-cycle byte restoration independently repeated. 474-round0-verification.md carries exact commands/hashes. Both clients are now absent; retail exited externally, never closed/attached by lead. Last privileged ACE check: UDP9000 PID13340 up. Clean c76f43c2f nine-hop run reached all destinations/occurrences with zero reveal failures, exit0/graceful; three30310005 visits revealed. Overall FAIL: adjacent86–89s, final1838retirements, first post-resize PNG corrupt476; six later PNGs coherent, settled cue check owed. 474-graphical-regression-r1.md records timings/PNGs/hashes. 475-debt-free-retirement-contract.md removes only synthetic no-debt frame waits while preserving exact callbacks/ACKs/meter/FIFO; nine-file implementation frozen in exact-base s5-475-openai-impl. Lead independently passes Release0W0E/Runtime122/App130 and all four mutation categories with byte-identical restoration. Wider caller census: Runtime296/300 and Headless47/48 expose five obsolete-poll test failures; App167/167 passes. Retail/behavior review1 FAIL, test/evidence only: those five obsolete expectations plus noncanonical direct-retirement fixture. Lead confirmed exact-key backend removal; contract§6 binds three-file test-only fix round1, actual first-activation measurement and effect-discriminating canonical content. Full hermeticRuntime1887/1891 finds only the same four failures. 475-lead-verification.md records lead commands/hashes/mutations. No production defect found; production/lifetime review follows corrected return. Not integration-ready. Both clients absent after run; ACE9000 PID13340 rechecked. #476 source defect is confirmed in476-capture-lifetime-diagnosis.md: pre-submit screenshot readback can consume a newly allocated buffer before its first GPU copy. No476 code changed; its bounded repair follows475. Repair475/review/re-gate and476 capture repair precede C1c landing, real geometry/membership hash, full-lane closure, C2 and owner G4. Never merge main before G4. |
|
||||
| S5 | **#473 EXTERIORS OWNER-ACCEPTED; #474/#475 LANDED + REVIEW-CLOSED; recall timing narrowly OWNER-ACCEPTED. Latest owner gate FAIL on #477–#481 (2026-09-05).** #475 `de427d2c0`; c1–c5 plus #470/#469 landed. | G4 FAIL / UNPASSED | Packet §§7 and12–49 bind. #474 `220bda797` repaired the publication/placement ACK cycle. #475 exact eleven-file return `ab989d717` is blob-identical to integration `de427d2c0`, including architecture/AD-2; one test-only fix round, both review lenses closed. Lead independent Release0W0E, hermeticRuntime1891/1891, Headless48/48, App297/297; evidence `475-lead-verification.md`. Fresh campaign Release0W0E; owner-requested clean-binary run PID26920 completed five reveal generations but CRASHED with ErrorDeviceLost/Windows LiveKernelEvent141 while owner jumped in Neftet (#477). Network logout cleanup is NOT successful process exit. Other owner findings: dark Town Network/dungeons (#478), hanging Rynthid partly shifted (#479), invisible wielded items (#480), initial low FPS (#481). Recall timing seems OK per owner, not an exact benchmark. `475-owner-regate.md` records hashes/logs/events and limitations; no PNG this round. Owner AFK authorizes autonomous continuation. First priority crash diagnosis; four-agent maximum, OpenAI only, no client/build overlap. Both clients absent, ACE UDP9000 PID13340 up at11:09+02, recheck before action. No current root-cause or repair claim for #477–#481. Still owed: exact nine-hop timing repeat, separate confirmed #476 capture-before-submit repair, preserved C1c nine-file test/doc landing (both reviews closed; lead68/68), §45 real geometry/membership witness, full canonical lanes/C2 and owner G4. Never merge main before G4. |
|
||||
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,16 +1,23 @@
|
|||
# #475 — lead verification and review record
|
||||
|
||||
Status2026-09-05: **Test-only fix round1 independently verified; production/
|
||||
lifetime review2 PASS with no actionable findings. Graphical timing pending.**
|
||||
Status2026-09-05: **LANDED at `de427d2c0`; test-only fix round1 independently
|
||||
verified; production/lifetime review2 PASS with no actionable findings.
|
||||
Owner narrowly accepts recall timing; overall gate FAIL on #477–#481.
|
||||
Nine-hop measured timing remains pending.**
|
||||
Contract: `475-debt-free-retirement-contract.md`, including §6 at3db6546eb.
|
||||
No graphical475 run or owner acceptance is claimed.
|
||||
See `475-owner-regate.md` for exact launch/crash evidence and owner findings.
|
||||
No automated timing or G4 PASS is claimed.
|
||||
|
||||
## Candidate and provenance
|
||||
|
||||
Scratch:
|
||||
`C:\Users\erikn\source\repos\acdream\.claude\worktrees\s5-475-openai-impl`.
|
||||
Branch `codex/s5-475-openai-impl`, base
|
||||
`6a364ec1a6a2d9248e47367953f107dd9567b781`; return is uncommitted.
|
||||
`6a364ec1a6a2d9248e47367953f107dd9567b781`; final eleven-file return committed
|
||||
as `ab989d71765c01a5a0bd1cd301aeb819f354ba9b` and integrated at
|
||||
`de427d2c02dbea48f9e96f6fd4f96fbae41bb883`. Lead compared every changed
|
||||
Git blob: all eleven match exactly. The round0 evidence below is historical;
|
||||
the fix-round1 section is the accepted test return.
|
||||
Implementer report: `artifacts/475-round0/implementation-report.md`.
|
||||
The lead independently read the complete diff, new test partial and affected
|
||||
production callers; all nine raw file hashes matched the frozen report.
|
||||
|
|
|
|||
107
docs/research/2026-09-01-overhaul/475-owner-regate.md
Normal file
107
docs/research/2026-09-01-overhaul/475-owner-regate.md
Normal file
|
|
@ -0,0 +1,107 @@
|
|||
# #475 — owner-requested immediate Release re-test
|
||||
|
||||
2026-09-05. **OWNER GATE FAIL; recall timing narrowly accepted.** This was
|
||||
the owner's explicitly requested latest-build check before leaving, ahead
|
||||
of the unchanged automated nine-hop regression. No G4 or timing PASS.
|
||||
|
||||
## Exact candidate and launch
|
||||
|
||||
- Clean source: `de427d2c02dbea48f9e96f6fd4f96fbae41bb883` on
|
||||
`claude/campaign-w-retail-frame-walk` in the campaign worktree.
|
||||
- Reviewed implementation: `ab989d71765c01a5a0bd1cd301aeb819f354ba9b`.
|
||||
Lead independently compared all eleven changed Git blobs against the
|
||||
integration commit: eleven identical. Review/test evidence is in
|
||||
`475-lead-verification.md`; no post-integration campaign tests were run.
|
||||
- Campaign Release build: 0 warnings, 0 errors, exit 0, 10.09 s.
|
||||
`logs/owner-gate-20260905-110507-475/build.log`.
|
||||
- App.exe SHA256:
|
||||
`98107B313788F7416D35BC282BF252A78C839A7A71C329DA03AD019AB447494C`.
|
||||
- App.dll SHA256:
|
||||
`8E056C0A627D629FA19364238A0DEBE596D799E7C35C3238D1CB7E56D4F15A03`.
|
||||
- Existing recipe-10 package `C:\Users\erikn\Documents\Asheron's Call\acdream-v10-s5-c5.pak`:
|
||||
previously verified SHA256
|
||||
`4E0FA6C589ECE12E13099A56385077964961EC52F888BC9161C6F5D34829B59B`.
|
||||
Package unchanged, not rehashed during the urgent launch.
|
||||
- Session: `logs/s5-c5-session-v10.json` (private credentials not copied).
|
||||
Retained retail UI; High profile, near 4/far 25, 4x MSAA/A2C. Interactive
|
||||
window, no automation route or auto-close. Startup 1280x720, persisted
|
||||
1024x768, then owner-session swapchain 1600x900 (stdout line223).
|
||||
Exact-framebuffer automation override cleared.
|
||||
- Both AcDream and retail clients absent before build/launch. ACE listening
|
||||
on UDP 9000, PID 13340. Launched PID 26920 at 11:05:18 +02:00; native
|
||||
window handle 394230 was independently observed. No second client,
|
||||
build, test, debugger attach or agent close during owner interaction.
|
||||
|
||||
## Observations and limitations
|
||||
|
||||
`logs/owner-gate-20260905-110507-475/client.log` records five complete
|
||||
reveal generations, each ready/materialized/world-visible/complete and
|
||||
`failures=0`: login `A9B40019`, then portals `00070143`, `F4180104`,
|
||||
`3032001C`, `8763000E`. The owner drove these visits, not an automated route.
|
||||
The log records graceful logout requested and confirmed during shutdown, but
|
||||
stderr and Windows events establish an unhandled GPU-device-lost crash.
|
||||
The stdout logout lines are NOT evidence of a successful process exit.
|
||||
Privileged process/UDP check at 11:09:01 +02:00 found neither client and
|
||||
ACE still listening, PID 13340. The lead did not close the owner's session.
|
||||
No retained process exit-code evidence is claimed.
|
||||
|
||||
No PNG was requested/captured in this manual session. Visual findings below
|
||||
are the owner's direct report, not inferred from logs. Five completed reveals
|
||||
do not establish exact latency or the original `30310005` reproduction.
|
||||
|
||||
## Owner verdict and confirmed crash
|
||||
|
||||
The owner reported an AMD crash notice while jumping around Neftet, and:
|
||||
|
||||
| Item | Outcome | Tracking |
|
||||
|---|---|---|
|
||||
| Client crash / AMD notice in Neftet | FAIL; stderr and Windows events confirm device loss | #477 |
|
||||
| Dungeon lighting, especially Town Network | FAIL; barely visible per owner | #478 |
|
||||
| Hanging Rynthid over the crater partly shifted after recall | FAIL; object identity/transform cause not yet established | #479 |
|
||||
| Recall times | Owner says they seem OK; narrow timing acceptance, not a benchmark | #475 |
|
||||
| Nothing wielded in hands is visible | FAIL; attachment logs alone do not prove drawing | #480 |
|
||||
| Initially low FPS when entering the world | Owner-reported performance concern; not measured yet | #481 |
|
||||
|
||||
The lead's early stdout-only interpretation was incomplete and is superseded
|
||||
here. `client.err.log` records an unhandled `VulkanCallException`:
|
||||
`vkQueueSubmit2 (abandoned frame timeline signal) returned ErrorDeviceLost`.
|
||||
Call chain: VulkanGpuDevice.SignalTimelineWithoutWork:560 ->
|
||||
TryBeginFrame:354 -> BeginFrame:318 -> RenderFrameOrchestrator.Render:220.
|
||||
Shutdown separately reports `AbandonedIncomplete`, blocked at mesh adapter:
|
||||
`vkWaitSemaphores (frame flight, value 37509) returned ErrorDeviceLost`.
|
||||
The lead independently read the full failure tail. No successful exit claim.
|
||||
|
||||
Windows Application log, local time +02:00:
|
||||
|
||||
- 11:08:01, .NET Runtime event1026: same unhandled exception/stack.
|
||||
- 11:08:07, Application Error event1000: AcDream.App.exe, exact campaign
|
||||
Release path, PID0x6928 (=26920), exception0xe0434352.
|
||||
- 11:08:07 and11:08:35, Windows Error Reporting event1001:
|
||||
LiveKernelEvent141, report b56f5566-5fe5-40aa-8460-ce3df39f1f06.
|
||||
- 11:08:11, WER event1001 CLR20r3 for this AcDream process, report
|
||||
f9a26760-f40f-4402-81df-71f9c744b580.
|
||||
|
||||
Hardware/capability log: AMD Radeon RX9070XT, Vulkan1.4.349, reported driver
|
||||
2.0.395 (raw0x0080018B), present ImmediateKhr, 4xMSAA. These records confirm
|
||||
device loss and a contemporaneous watchdog report; they do not identify the
|
||||
offending GPU command or prove a driver-only/hardware-only fault. The final
|
||||
logged arrival is8763000E at(40.4574,129.281,8.004999), not an exact crash pose.
|
||||
|
||||
Stable local log SHA256:
|
||||
|
||||
- client.log: `140E61353184C748961BE8CA59FA161968FF39F76364544ABCE1F9AD9B7F783C`.
|
||||
- client.err.log: `45E11F52BC438AD96312ADF1E893242DFD40D6B78F62C78932FC6612EDB30AF3`.
|
||||
|
||||
Existing Windows report, metadata only inspected (100766bytes,11:08:11+02):
|
||||
`C:\ProgramData\Microsoft\Windows\WER\ReportArchive\AppCrash_AcDream.App.exe_12501baeaa2b6e467d44cb735ff181f6b7b1c3_50b687fb_3e54b98c-2ce2-400a-98f2-a647a61371e0\Report.wer`.
|
||||
The event names `C:\WINDOWS\LiveKernelReports\WATCHDOG\WATCHDOG-20260905-1108.dmp`,
|
||||
but its existence/readability was not confirmed. No dump copied, read or
|
||||
uploaded. Logs and available Windows evidence remain local.
|
||||
|
||||
The owner then went AFK and authorized autonomous continuation. First diagnose
|
||||
the crash, preserving the other findings as separate bounded tasks; no driver
|
||||
changes, timeout suppression, forced reveal or blind global brightening.
|
||||
Read-only source diagnosis may proceed in parallel (four agents total).
|
||||
The unchanged nine-hop regression remains owed but is not the next launch
|
||||
until the crash investigation establishes a safe, useful run. #476 capture
|
||||
repair and the remaining campaign closure remain owed. G4 FAIL, no main merge.
|
||||
|
|
@ -0,0 +1,64 @@
|
|||
# #477 — one bounded GPU validation reproduction
|
||||
|
||||
Prepared2026-09-05; **NOT RUN YET.** No production behavior change. This is a
|
||||
diagnostic recipe under the owner's AFK continuation authorization, not a
|
||||
repair contract, new review round, timing benchmark or G4 PASS.
|
||||
|
||||
## Evidence and question
|
||||
|
||||
`475-owner-regate.md` records the exact owner crash and all five findings.
|
||||
The lead independently confirmed the reviewer's read-only source diagnosis:
|
||||
VulkanSwapchain.TryAcquire reads the raw result, classifies it with
|
||||
VulkanSwapchainRecreationPolicy, then SwapchainBackbuffer.TryAcquire collapses
|
||||
fatal failure and ordinary unavailability to false. VulkanGpuDevice then
|
||||
submits an empty timeline signal, where device loss is finally reported.
|
||||
Present similarly collapses fatal failure into a recreation Boolean.
|
||||
This hides first-failure provenance; it does not prove why the GPU stopped.
|
||||
The inspected frame-slot/retirement path has no demonstrated lifetime defect.
|
||||
|
||||
Before changing production diagnostics, use the existing validation layer
|
||||
once to ask whether this scene/jump workload reports an actionable API or
|
||||
synchronization fault. No driver/TDR changes, unsafe recovery, GPU dump upload,
|
||||
memory attach, shader modification, timeout suppression or additional owner.
|
||||
|
||||
## Run recipe
|
||||
|
||||
Fresh Release build green, no AcDream/retail client, ACE UDP9000 up. Same
|
||||
recipe-10 package/session as the failed owner run. High profile, near4/far25,
|
||||
4xMSAA, A2Ctrue, anisotropic16, completions4; final framebuffer1600x900.
|
||||
Process-local environment only: ACDREAM_RETAIL_UI=1, ACDREAM_DEVTOOLS=1,
|
||||
VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation,
|
||||
VK_LAYER_ENABLES=VK_VALIDATION_FEATURE_ENABLE_SYNCHRONIZATION_VALIDATION_EXT.
|
||||
The installed SDK1.4.350.0 layer manifest explicitly supports this legacy
|
||||
environment flag (its preferred setting is validate_sync); no global setting
|
||||
or registry edits. Validation output must confirm loading; without evidence,
|
||||
do not call the run validation-clean.
|
||||
|
||||
Use `tools/overhaul-selfgate/run-selfgate.ps1 -Label 477-neftet-validation-r1
|
||||
-Route tools/overhaul-selfgate/route-477-neftet-validation.txt
|
||||
-Session logs/s5-c5-session-v10.json -TimeoutSeconds 420` in one background
|
||||
command. The existing runner's exact-framebuffer mode uses a real hidden
|
||||
Vulkan window. No concurrent builds/tests/client. Route closes only its own
|
||||
client; if timeout cannot close gracefully, leave it and report the state.
|
||||
|
||||
The route uses the last observed owner arrival8763000E at
|
||||
(40.4574,129.281,8.004999) with an explicit identity heading, twelve bounded
|
||||
jump attempts, short movement/rotation and settled checkpoints. The owner's
|
||||
exact crash pose, heading and sequence were not captured, so this is a
|
||||
candidate reproduction, not an exact replay. Record actual jump outcomes;
|
||||
script inputs alone do not prove airborne transitions.
|
||||
|
||||
Capture first post-resize and pre/post-workload PNGs. Known #476 remains
|
||||
unfixed: retain and inspect every image, reject corrupt images explicitly;
|
||||
never discard the first capture or count checkpoint readiness as visual PASS.
|
||||
Validation overhead means frame times cannot establish #481 improvement.
|
||||
|
||||
## Decision after this single run
|
||||
|
||||
Record exact source/build hashes, stdout/stderr/checkpoints, validation
|
||||
messages, PNG paths and exit outcome. An actionable fault must be reconciled
|
||||
against production source before a bounded repair contract. A clean/no-crash
|
||||
run does not close #477; preserve the first-failure propagation defect and
|
||||
choose the next narrow diagnostic based on evidence. Do not enter an
|
||||
unbounded sequence of GPU-reset reproductions. Other owner findings and the
|
||||
remaining campaign gates stay open.
|
||||
|
|
@ -4477,6 +4477,13 @@ remains unchanged; main has not been merged.
|
|||
|
||||
## 48. #474 nine-hop convergence, #475 latency correction — 2026-09-05
|
||||
|
||||
**Current:** #475 LANDED + REVIEW-CLOSED at `de427d2c0`; owner narrowly accepts
|
||||
recall timing. Overall gate FAIL on new #477–#481, including a confirmed
|
||||
GPU-device-lost crash. See§49. The unchanged nine-hop timing repeat remains
|
||||
owed. The implementation/review checkpoints
|
||||
below are chronological history, superseded by the closeout at this section's
|
||||
end. No G4 PASS.
|
||||
|
||||
The prior unrun status in§47 is superseded by
|
||||
`474-graphical-regression-r1.md`. Clean campaign c76f43c2f / repaired code
|
||||
220bda797 ran9/9 exact materialization/visibility checkpoints, including all
|
||||
|
|
@ -4573,3 +4580,67 @@ Fix round1 is restricted to three test files. The full hermetic Runtime lane
|
|||
is1887/1891 with only the same four caller failures, no skips. No production
|
||||
behavior finding; paired retail hashes match. Production review is next
|
||||
after the corrected return and independent verification.
|
||||
|
||||
### #475 review closeout, integration and immediate owner re-test
|
||||
|
||||
One test-only fix round reconciled retail review1's findings. Lead independent
|
||||
Release 0W0E, hermetic Runtime 1891/1891, Headless 48/48 and App 297/297 all
|
||||
pass, no skips. Production/lifetime review2 PASS, no actionable findings;
|
||||
no extra wholesale retail review was run for the test corrections.
|
||||
`475-lead-verification.md` records exact commands, source/assembly hashes,
|
||||
lead-repeated mutations and the two implementer effect mutations that the
|
||||
lead inspected rather than re-executed.
|
||||
|
||||
Lead committed the eleven-file corrected return as `ab989d717` and integrated
|
||||
it at `de427d2c0`. All eleven changed Git blobs match exactly across commits.
|
||||
The canonical Runtime permission function is the sole behavior change;
|
||||
RuntimePhysicsState changes comments only. Architecture/AD-2 landed in the
|
||||
same implementation commit. Neither screenshot behavior nor budgets changed.
|
||||
|
||||
The owner explicitly requested the latest build within two minutes, so the
|
||||
manual check preceded the automated route. Fresh campaign Release 0W0E;
|
||||
clean `de427d2c0` launched PID26920 at11:05:18+02, with no other client and
|
||||
ACE UDP9000 PID13340. Five reveal generations completed with failures0:
|
||||
A9B40019,00070143,F4180104,3032001C,8763000E. Logout cleanup was confirmed,
|
||||
but stderr and Windows events prove an unhandled device-loss crash; the
|
||||
stdout-only clean-exit interpretation is superseded. At11:09+02 both clients
|
||||
were absent. Lead did not close the owner session. `475-owner-regate.md`
|
||||
records exact binary hashes, failure logs and owner findings. No captured
|
||||
PNG, exact timing or original30310005 reproduction is claimed.
|
||||
|
||||
The owner narrowly accepts recall timing but fails the overall gate on the
|
||||
new findings below. No main merge before G4. No new review is needed merely
|
||||
to record these facts.
|
||||
|
||||
## 49. Owner gate FAIL — crash, lighting, placement, held items and entry FPS
|
||||
|
||||
Owner2026-09-05 report on clean `de427d2c0`: AMD crash notice while jumping
|
||||
in Neftet; Town Network dungeon too dark to see; hanging Rynthid above the
|
||||
crater partly shifted after recall; wielded hand items invisible; initially
|
||||
low FPS on world entry. Recall times "seems ok" is narrow owner acceptance,
|
||||
not a measured performance matrix. Issues #477–#481 record the five findings;
|
||||
`475-owner-regate.md` is the full source/binary/log/Windows-event record.
|
||||
|
||||
Lead confirmed `ErrorDeviceLost` in the abandoned-frame timeline submit and
|
||||
secondary shutdown wait37509; .NET1026/ApplicationError1000 forPID26920 and
|
||||
LiveKernelEvent141 corroborate it. The original raw acquire result is lost
|
||||
at the existing Boolean adapter, so the offending GPU command/root cause
|
||||
is not established. Graceful network logout is not process success.
|
||||
No PNG exists for this owner session; its visual FAIL is the owner's verdict.
|
||||
|
||||
Owner went AFK and explicitly delegated continuation. First priority is crash
|
||||
diagnosis, then bounded repairs for the other findings. Three existing OpenAI
|
||||
agents perform independent READ-ONLY source diagnosis (GPU first failure,
|
||||
indoor lighting, held-item membership); lead owns the ledger and placement
|
||||
investigation, four agents total. This is not a new implementation review.
|
||||
No builds/tests or client launch by agents; no new behavior changes yet.
|
||||
No driver/TDR changes, skipped synchronization, forced reveal, fallback cell
|
||||
membership or blind brightness tuning. Each actual repair needs its own
|
||||
bounded committed contract, verification and the established review lenses.
|
||||
|
||||
The exact nine-hop repeat, #476 capture repair, preserved C1c landing, §45
|
||||
geometry witness, full lanes/C2 and final owner G4 remain owed. None is
|
||||
silently dropped, but the next graphical run must serve the confirmed crash
|
||||
investigation. Both clients absent; ACE UDP9000 PID13340 remains up. Check
|
||||
again before builds/launches; one run per background command, close only our
|
||||
own gracefully. G4 FAIL; no merge main.
|
||||
|
|
|
|||
97
tools/overhaul-selfgate/route-477-neftet-validation.txt
Normal file
97
tools/overhaul-selfgate/route-477-neftet-validation.txt
Normal file
|
|
@ -0,0 +1,97 @@
|
|||
# #477 bounded validation reproduction; no production changes.
|
||||
# Last owner arrival is observed, but exact crash pose/jump sequence is unknown.
|
||||
# Same High/near4/far25/MSAA4/A2C profile; 1600x900.
|
||||
# Run once with Khronos core + synchronization validation, not a performance gate.
|
||||
wait world-ready 180000
|
||||
wait world-visible 90000
|
||||
resize 1600 900
|
||||
wait framebuffer 1600 900 30000
|
||||
renderpack select high
|
||||
wait render-pack high 90000
|
||||
checkpoint 477_00_login
|
||||
screenshot 477-00-first-resize 15000
|
||||
|
||||
command /teleloc 8763000E 40.4574 129.281 8.004999 1 0 0 0
|
||||
wait materialized 1 180000
|
||||
wait world-visible 90000
|
||||
sleep 3000
|
||||
checkpoint 477_01_neftet_before
|
||||
screenshot 477-01-neftet-before 15000
|
||||
|
||||
# Six short moves/jumps while rotating, followed by six stationary jumps.
|
||||
# Every down has an explicit up; no endless stress loop.
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementForward
|
||||
sleep 150
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
input up MovementForward
|
||||
sleep 2500
|
||||
input down MovementTurnRight
|
||||
sleep 1500
|
||||
input up MovementTurnRight
|
||||
checkpoint 477_02_six_jumps
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
input down MovementJump
|
||||
sleep 350
|
||||
input up MovementJump
|
||||
sleep 2500
|
||||
checkpoint 477_03_twelve_jumps
|
||||
sleep 5000
|
||||
screenshot 477-03-neftet-after 15000
|
||||
checkpoint 477_04_settled
|
||||
close-client
|
||||
Loading…
Add table
Add a link
Reference in a new issue