erik/acdream
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 2

docs(render): record V7 - the instrument was measuring itself, and AD-46

Browse source 
Plan section 5.5.19, the V7 slice row, two rows in the section 5.1 uncovered
table, and one divergence-register row. No product code changes.

WHAT V7 TURNED OUT TO BE. Three of V6m's four numbers were taken through an
instrument that was not holding the world still. The route pinned the Dereth
clock by pressing AcdreamCycleTimeOfDay, whose mechanism is the transient
/time override that SyncFromServer clears -- so ACE un-pinned it seconds into
every run this campaign has taken. Two captures 45 s apart at ONE stop on ONE
backend differ in 22.3% of the frame because the sun keeps moving. Pinning it
(commit 2) took the interior stop from 12.16% to 0.78% on its own.

THE THREE LEADS, ANSWERED.

Lead 3 was WRONG and the section says so. V6m recorded the interior stop as a
route defect on the theory that the indoor spring-arm camera settles to different
distances in two runs. It does not. The interior was lit differently because the
sun had moved. With the sun held still the stop drops by a factor of fifteen and
its entire remaining difference map is the player character -- the EnvCell's
walls, floor, doorway and per-cell ambient are black. EnvCellRenderer's Vulkan arm
has its numeric pair, V6m's defect-list item 2 is discharged, and no route change
was needed or made.

Lead 2 is closed by pinning the cloud phase rather than masking the band, so the
gate keeps the sky under strict comparison.

Lead 1 is half fix, half finding. The residual was predominantly the anisotropy
gap (commit 1). What survives is one population -- dense alpha-blended distant
scenery -- and isolating it needed a better instrument than the connected route.

THE INSTRUMENT V7 RECOMMENDS FORWARD. An offline GL-versus-Vulkan pair, which is
just the two existing capture scripts run with ACDREAM_WORLD_TIME and
ACDREAM_SKY_PHASE_SECONDS set in the invoking shell. No session, no server, no
entities, no camera settle, no wandering NPCs, and unattended:

    GL vs GL   same commit (control)                       1,966 px   2.13e-03
    VK vs VK   same commit (control)                       1,039 px   1.13e-03
    GL vs VK   whole frame                                28,807 px   3.13e-02
    GL vs VK   everything below the tree band (rows 280+)     497 px   8.82e-04

Terrain, blending, roads, the water edge, fog, statics, scenery below the horizon
and the entire retained UI are at parity, inside the campaign's 0.001 threshold.

AD-46, FILED WITH ITS REFUTATIONS RATHER THAN ITS THEORY. The treeline band is an
anisotropic tap-pattern divergence between AMD's GL and Vulkan drivers. Three
competing explanations were tested and refuted, and section 4.7's predicted class
is one of them:

  - not a sub-pixel offset -- an integer shift search finds (0, 0);
  - not sharpness or LOD scale -- high-frequency energy matches within 5%;
  - NOT DEPTH PRECISION. Forcing the Vulkan viewport's window-depth range to
    [0.5, 1.0], which reproduces GL's compressed mapping exactly, moved the
    whole-frame number by 3% (28,807 -> 27,852). The experiment was reverted. The
    one pre-approved divergence class is not what this is, and the row says so
    rather than borrowing its approval.
  - It IS anisotropy, and there is no knob left: 41,509 differing pixels in the
    band at anisotropy 1, 22,266 at 16, which is GL's value and retail's.

THE VERDICT TABLE. Full route, both backends, tolerance 2, MSAA off, day group 0,
world time 0.5, sky phase 0 (artifacts/v7-diff-c2):

    holtburg_town           26,330 px   2.86e-02   EXCEPTION -- phase + AD-46
    facility_hub_interior    7,176 px   7.79e-03   EXCEPTION -- phase
    aerlinthe_island        62,892 px   6.82e-02   EXCEPTION -- AD-46 + dark floor

No stop passes and none of the three exceptions is a renderer defect; each is
named individually in the section, because "phase" is not an excuse unless it is
specific. Aerlinthe's is partly the instrument rather than either renderer: the
scene's mean luminance is 28/255 and half its differing pixels are exactly delta
3, one step over a tolerance that is absolute rather than relative. Changing that
tolerance is not V7's call.

CARRIED FORWARD, recorded in the section 5.1 table and the V7 row: a passing
connected stop needs authored per-stop masks that the gate script does not have
(it still has only the global -MaskTopPixels, deliberately defaulted to 0); the
portal depth mask has now gone three slices without drawing a pixel in an
automated run, and HouseExitWalkReplayTests names the cheapest target for it
(the Holtburg corner building, cell 0xA9B40170); whether AD-46 is visible to a
human is a user-stop question nobody has asked yet; and the R6 soak and RenderDoc
capture on Vulkan were not run.

Gates for this commit: docs only, so the code gates of commits 1 and 2 stand.
Complete Release suite 9,195 passed / 5 skipped with zero failures, and the GL
connected repeat gate at 3/3 RENDERED on both the desktop witness and the client
capture, both taken at this tree.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-07-28 19:44:33 +02:00
parent 1f25a60999
commit 45f58ac72b
2 changed files with 217 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
docs/architecture/retail-divergence-register.md
View file

@ -62,10 +62,11 @@ accepted-divergence entries (#96, #49, #50).
---
## 2. Adaptation (AD) — 38 rows (AD-11 retired 2026-07-23 — exact low-bit ItemUses predicate; AD-31 retired 2026-07-15 — the DAT-authored portal-space viewport replaces the black transit cover)
## 2. Adaptation (AD) — 39 rows (AD-11 retired 2026-07-23 — exact low-bit ItemUses predicate; AD-31 retired 2026-07-15 — the DAT-authored portal-space viewport replaces the black transit cover)
| # | Divergence | Where (file:line) | Why it is safe / justified | Risk if assumption breaks | Retail oracle |
|---|---|---|---|---|---|
| AD-46 | **Vulkan-arm only, dormant until the V10 cutover.** Dense alpha-blended distant scenery (the treeline) resolves differently on the Vulkan backend than on GL, by about 15% of the pixels in the band, even though both arms now request identical sampler state — trilinear, clamp-and-repeat, and the device's maximum anisotropy. What remains is the anisotropic TAP PATTERN, which both the GL and Vulkan specifications leave implementation-defined, and AMD's two drivers do not agree. | `src/AcDream.App/Rendering/Wb/WorldTextureArray.cs` (`RhiWorldTextureArray.WorldArrayAnisotropy`); measured in plan §5.5.19 | Not assumed — narrowed by measurement, on an offline capture with no session, no entities and both clocks pinned. Anisotropy 1 → 41,509 differing pixels in the tree band; anisotropy 16 (GL's value, and retail's `m_D3DCaps.MaxAnisotropy`) → 22,266, and the rest of the frame falls to 497 px of 563,200, i.e. 8.8e-04, inside the campaign's 0.001 threshold. The residual is not a sub-pixel shift (an integer shift search finds none), not a sharpness change (high-frequency energy matches within 5%), and not depth precision (forcing Vulkan's window-depth range to GL's compressed [0.5, 1] moved it by 3%). Monotone improvement toward GL's own anisotropy with no knob left is what makes it a driver property rather than a bug. | If it is NOT the tap pattern, some other Vulkan sampler or alpha-path difference is hiding behind this row, and it will present after cutover as distant foliage that shimmers or reads denser than retail's. The class is confined to alpha-blended dense overlap: opaque terrain, roofs, walls, water, statics, the character and the whole retained UI are inside threshold. | `RenderDeviceD3D::SetDefaultD3DStates @ 0x005a3800`, whose `SetSamplerState(stage, 0xA /* D3DSAMP_MAXANISOTROPY */, m_D3DCaps.MaxAnisotropy)` at `0x005a4230` is the value both arms now request |
| AD-38 | Outgoing teleport viewports retire when retail's quantized animation level exceeds the last captured visible level 1022 (index 96), suppressing levels 1023/1024 up to 20.2 ms before retail's literal `elapsed >= 1.0` state edge. Incoming fades retain the exact timer. | `src/AcDream.Core/World/TeleportAnimSequencer.cs` (`OutgoingViewportReachedTerminalProjection`) | An uncapped 2000 FPS pass can publish the finite tunnel at levels 1023/1024 even though the paired 2013 retail capture switches viewports after 1022. The table-level cutover preserves the captured visible viewport ordering without throttling the application. | Exit sound, viewport replacement, and logout tunnel entry can occur at most two easing-table quanta (about 20.2 ms) earlier than retail's logical timer. | `UIGlobals::GetAnimLevel @ 0x004EE540`; `gmSmartBoxUI::UseTime @ 0x004D6E30`; paired retail/acdream captures documented in `docs/research/2026-07-15-retail-portal-space-pseudocode.md` |
| AD-1 | Lost-cell machinery replaced by recoverable outdoor demote (**#107** safety net) + outdoor-restore `max(terrainZ, z)` under-terrain lift; retail goes `GotoLostCell` | `src/AcDream.Core/Physics/PhysicsEngine.cs:553` (+ :808) | acdream has no lost-cell state machine; outdoor landcell is the recoverable equivalent; the #107 auto-entry hold should make the demote branch unreachable | Gap in the hold → player committed to outdoor terrain inside/under a building (fake-grounded spawn, fall-through); a legit below-heightmap server restore is silently lifted — upward warp vs server | `GotoLostCell` pc:283418; `SetPositionInternal` 0x00515bd0, pc:283892-283945 |
| AD-2 | Async readiness gates replace retail's synchronous destination cell load. **#229 refinement (2026-07-20):** login and F751 portal-space exit now share `WorldRevealReadinessBarrier`, so neither path can expose the normal viewport until the same render-publication, composite-texture, and collision domains converge. A hydratable indoor claim requires its owning Near-tier static/EnvCell mesh set, destination composites, and exact EnvCell physics (`IsSpawnCellReady`); an outdoor claim requires those render domains plus terrain/collision residency for the required Near ring. Hard-recenter generations and tier-aware completion application prevent stale overlapping loads/unloads or Far/Near jobs from opening or erasing the gate; mesh upload remains separate from balanced landblock ownership. Claims beyond NumCells still take the loud unhydratable-placement path. `RuntimeWorldTransitState` owns the shared reveal generation, accepted readiness, transit correlation, and exact generation/cell-scoped host-acknowledgement suffix. `WorldRevealCoordinator` is a graphical adapter holding only App resource receipts; normalized Runtime checkpoints observe ownership without defining another readiness path. **Slice E3 refinement (2026-07-24):** the same generation now publishes an immediate `WorldGenerationQuiescence` edge: old-world drawing/spatial queries, simulation/effect clocks, reconciliation, targeting, and 3-D audio stop while retained physical teardown advances through metered cursors and destination network/UI/streaming/readiness remain live. **Slice E4 refinement (2026-07-24):** accepted render/physics/static publication may span update frames through retained exact cursors, but reveal still consumes only the completed spatial/render-ready generation; building and EnvCell snapshots remain invisible until complete and the final spatial identity swap stays observer-atomic. **Slice E5 refinement (2026-07-24):** the reveal generation owns one exact destination reservation across every typed budget dimension. Stale completion cannot consume or clear its replacement, and hydratable incomplete content is never force-revealed; portal transit retains the DAT tunnel and centered retail wait cue until readiness converges. The hold→materialize→regain-control lifecycle remains owned by `TeleportAnimSequencer`. | `src/AcDream.Runtime/World/RuntimeWorldTransitState.cs`; `src/AcDream.App/Streaming/WorldRevealCoordinator.cs`; `src/AcDream.App/Streaming/WorldGenerationQuiescence.cs`; `src/AcDream.App/Streaming/WorldRevealReadinessBarrier.cs`; `src/AcDream.App/Streaming/StreamingOriginRecenterCoordinator.cs`; `src/AcDream.App/Streaming/LandblockPresentationPipeline.cs`; `src/AcDream.App/Streaming/StreamingController.cs`; `src/AcDream.App/Rendering/PortalTunnelPresentation.cs`; `src/AcDream.App/UI/PortalWaitNoticeController.cs`; `src/AcDream.App/Streaming/GpuWorldState.cs` (`IsRenderReady`); `src/AcDream.App/Rendering/Wb/LandblockSpawnAdapter.cs`; `src/AcDream.Core/Physics/PhysicsEngine.cs` (`IsSpawnCellReady`, `IsNeighborhoodTerrainResident`) | This is the asynchronous equivalent of retail leaving `SmartBox::position_update_complete` false while `CellManager::blocking_for_cells` is set: neither initial login nor portal arrival may reveal or continue simulating an old/partial collision world, a terrain-only Far shell, or a published-but-not-drawable GPU landblock. Indoor does not require a terrain heightmap, only the owning render landblock and exact EnvCell. | Gate opens early → grey/untextured first login or portal reveal, free-fall, wrong-cell rooting, missing scenery, or a still-active old generation; predicate never satisfies (streamer/DAT/upload failure) → login remains behind the world render gate, while portal transit remains in the authored tunnel and presents the centered wait cue after five seconds. | `SmartBox::UseTime` 0x00455410; `gmSmartBoxUI::UseTime` 0x004D6E30; `gmSmartBoxUI::EndTeleportAnimation` 0x004D65A0 |