Campaign V slice V6e, first of three. mesh_modern is the shader every world
static, every piece of scenery and every EnvCell surface draws through, and it
was one of the four production pairs the SPIR-V toolchain still refused.
The blocker was a varying. Since V2 the vertex stage looked a batch's table slot
up in the binding=9 handle table and forwarded the resulting 64-bit
GL_ARB_bindless_texture handle to the fragment stage as a `flat uvec2`. That
works on GL because a bindless handle is just a number a shader may carry
anywhere. It cannot work on Vulkan at all: the equivalent object is a descriptor
in set 2, and a descriptor is not a value a stage can hand to another stage. So
what travels between the stages is now the SLOT — a `flat uint` — and the
fragment stage does the lookup at the point of sampling.
That relocation needs one shared idea, because the two backends disagree about
what the lookup IS. `ACDREAM_SAMPLE_ARRAY(slot, uvw)` asks the dialect-neutral
question — "sample table slot N" — and expands to
`texture(sampler2DArray(gTextureTable[slot]), uvw)` under GL and to
`texture(uTextures[nonuniformEXT(slot)], uvw)` under Vulkan. It is deliberately
a SAMPLING macro rather than a sampler-returning one: `nonuniformEXT` belongs on
the indexing expression itself, and binding the result to a local
`sampler2DArray` first is exactly where an implementation is free to drop it.
That is the same shape V6d already used for the retained UI's 2-D reads, and it
now covers the array reads the world path needs.
`ACDREAM_TEXTURE_NONE` lands alongside it, unused here and used by the next
commit. GL can ask "does this slot hold a texture" of the payload, because an
unregistered slot holds the null handle; Vulkan cannot, because set 2 is opaque
and reading an unwritten element of a partially-bound array is undefined rather
than zero. The sentinel moves that answer into the index, where both dialects
test it identically.
On GL nothing about the sampled result changes — the same slot resolves to the
same handle to the same texel. The SSBO read simply happens one stage later,
and `flat` keeps it one scalar load per primitive rather than per fragment.
Also: RenderBootstrap has been loading mesh_modern without common.glsl since V2,
which cannot have linked — `ACDREAM_UBO_SET` sits inside a layout qualifier
there. The UI Studio path is the only caller. One argument, same pair, same way
WorldRenderComposition has always loaded it.
Gates: Release build clean; App tests 4,057 passed / 3 skipped (baseline);
offline pixel gate against 95f8c25f differing fraction 3.37e-05 (~19 px of
563,200), inside the documented 15–23 px same-commit noise band and ~30x under
the 0.001 threshold. mesh_modern is the shader that gate covers most heavily,
so this is the strongest automated evidence any V6e commit gets.
Manifest: 4/9 pairs compile (debug_line, mesh_modern, ui_text, vk_probe).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Campaign V slice V6d, commit 2 of 3. TextRenderer and DebugLineRenderer were the only two renderers speaking the RHI, and both refused any device that was not a GlGpuDevice. They now refuse nothing: this is the first production rendering acdream can do on Vulkan.
Three things had to go.
The loose uniforms. debug_line declared uView and uProjection separately and DebugLineRenderer set them straight against the compiled GL program, because the pinned push-constant block carries one combined matrix and IGpuPassEncoder has no verb for arbitrary named uniforms. That was never portable — Vulkan has no default uniform block at all — so the shader converged on uViewProjection and Flush multiplies on the CPU. System.Numerics is row-vector convention while GLSL reads the floats column-major, which transposes, so the CPU equivalent of the old per-vertex uProjection * uView is view * projection. The product now rounds once per frame rather than once per vertex; these lines only draw when collision wireframes are switched on, so the offline gate sees nothing of it. ui_text's uScreenSize became the block's two spare scalars, uParamA and uParamB, with the same two divisions and the same NDC mapping around them.
The sampling mode. uUseTexture selected between font coverage, RGBA modulate and flat colour, and no field of the 96-byte block means that. It did not need one: which of the two texture-table slots is assigned IS the mode. uTextureIndexB assigned means a single-channel coverage source, uTextureIndexA assigned means an RGBA colour source, neither assigned means the vertex colour alone. GpuTextureSlot.Unassigned is already a loud sentinel for exactly this kind of question, and both branches guard so it never reaches a sampler. That also retired the 1x1 white fill texture: DrawFill routed solid quads through the sprite bucket relying on white times colour, and the untextured branch produces the same value with no texture at all. Multiplying by 1.0 changes no bits, and the gate agrees.
The texture binding. The classic glActiveTexture/glBindTexture path survived V4a because DrawSprite takes an arbitrary texture from sixty-odd widget call sites. But TextureCache had already registered every one of those into the device's table — the classic path was consuming the raw GL name that registration also produced. The UI's currency is now UiTextureTableHandle, a one-based table index whose zero is the same "no texture" every widget already guards on; a raw slot index would have turned all of those guards into silent false negatives, since slot 0 is perfectly valid. One-based rather than the slot itself because GpuTextureSlot is internal to the pinned contract while UiRenderContext.DrawSprite, TextureCache.GetOrUploadRenderSurface and a dozen widget properties are public, and neither publishing a contract type nor converting the retained UI to internal belongs in this slice.
Two consequences worth stating. The two backends disagree about what a 2-D table entry is — GL reconstructs a sampler2D from the bindless handle, Vulkan reads layer 0 of its sampler2DArray descriptor array — and ACDREAM_SAMPLE_2D is the one place that lives. Keeping GL on sampler2D is what leaves the UI's textures exactly as they are, including the paperdoll/appraisal FBO colour texture, which is an externally-owned GL_TEXTURE_2D from the §7.1 transitional seam and cannot become an array before V4g. On the Vulkan side, sampled views are now always layered, which also removes a latent invalid usage V6c shipped: it registered a Type2D offscreen view into a descriptor array whose element type is sampler2DArray.
And one real fix. Sampling through the table means a bound sampler object overrides the texture's own parameters. Nearest-requested UI art used to get its point filtering from a glTexParameter applied before the bindless handle went resident, so registering it with the stock WorldRepeat sampler would have made every retail icon and dat-font glyph silently bilinear. Those now register with a nearest-and-repeat sampler.
Supporting moves: GlGpuDevice.CreatePipeline splices common.glsl the same way Shader does, since an RHI shader that reads the table needs the table declared; GlGpuPassEncoder binds the device's table with the pipeline, which is the GL analogue of Vulkan binding descriptor set 2 per draw, and has to be per-bind because every raw-GL world renderer puts its own privately-numbered table at that binding; and the encoder derives GL_MULTISAMPLE from the pass's SampleCount, which is where the retained UI's hand-rolled glDisable belonged all along. TextRenderGlStateScope is deleted — the encoder's ambient capture restored a strict superset of it — and its failure-safety test follows the guarantee to GlAmbientCapabilityState, which gains a fakeable seam and, with it, the multisample-dimension coverage #249 recorded as missing.
App tests 4,057 passed / 3 skipped, unchanged from commit 1. Offline pixel gate against 871c406b: differing fraction 2.31e-05, 13 pixels of 563,200 compared — below the documented 15-23 pixel same-commit noise band, on a change that redraws every pixel of the retained UI through a different sampling path. The capture was inspected: vitals, spell bar, radar, toolbar icons and slot digits, chat window and Send button all present and correctly placed. Both new .spv pairs compile; the manifest records ui_text and debug_line as Vulkan-ready, leaving six pairs blocked on the world-renderer slices.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The last of V6's three commits, and the one that makes the backend render.
Plan sections: 4.5 (pipelines and the persisted cache), 4.6 (shaders and the
committed .spv), 4.7 and 3.3 (clip space, the Y flip and winding), 4.9 and 4.10
(swapchain format and the scissor convention), 4.11 (the probe shader V5
deferred), 5.4 (Target: null means the swapchain image, literally).
WHAT RUNS. ACDREAM_RENDER_BACKEND=vulkan now renders a real scene through the
whole RHI on the RX 9070 XT: 60,000-plus frames per twelve-second run, 4x MSAA
resolving into a B8G8R8A8_UNORM swapchain, GPU timer scopes resolving, a
screenshot taken through IGpuDevice.CaptureBackbuffer, and a clean
CloseMainWindow exit with the allocator reporting three device-memory objects.
WHAT IT DRAWS, AND WHY IT IS NOT THE GAME. V6's milestone is "a full game frame
on Vulkan" and on this branch that cannot be the game's own frame. V4c and V4d
are parked by 5.5.5 so the world renderers are still raw GL; and the two
renderers that DO speak the RHI - TextRenderer and DebugLineRenderer, ported at
V4a - both throw for any device that is not a GlGpuDevice, because their loose
uniforms and their classic texture-unit sprite binding have no home in the
pinned contract yet. Converting them is a V4-class change with its own GL pixel
gate, outside this slice's file list.
So the backend is exercised through the contract by a scene of our own, and it
is not a toy. It uses a device-local mesh arena filled through the staging ring,
instance and batch data written straight into mapped ring memory, an offscreen
render target whose colour is registered into the global texture table and
sampled by a later pass, a BC1 texture with a CPU-built mip chain beside an
uncompressed one with a vkCmdBlitImage chain, one multi-draw-indirect covering
five quads with gl_DrawID selecting per-draw batch data, a second pipeline with
line-list topology bound mid-pass, dynamic cull/front-face/depth-write, push
constants, timer scopes, and an MSAA colour attachment resolving into the
swapchain image.
ORIENTATION, BY INSPECTION. Slice V5's screenshot was a uniform clear and its
orientation was right "by construction" - which a uniform clear cannot show. The
scene is therefore deliberately asymmetric in both axes: a quadrant card that is
red top-left, green top-right, blue bottom-left and white bottom-right, four
differently tinted markers at four different corners, and an open L of lines
whose short stub rises at its right end. The captured PNG reads correctly in
every one of those, including a miniature of the same card in the bottom-right
whose own quadrants are also the right way up. The negative viewport height, the
front-face inversion and the capture path agree.
THE SHADER TOOLCHAIN, AND WHAT IT FOUND. tools/compile-shaders.ps1 drives
tools/ShaderCompiler, a small out-of-solution .NET tool over Silk.NET.Shaderc -
the same shaderc glslc is built on, through the already-pinned Silk 2.23.0
family. glslc is preferred when a Vulkan SDK is present and reported when it is;
neither this machine nor CI has one, and requiring a 500 MB manual install
between a contributor and a working checkout is not a reasonable price for a
build step. The GLSL sources stay the single source of truth: the Vulkan dialect
arrives as a preamble injected after the #version line - ACDREAM_UBO_SET becomes
"set = 1,", the texture table becomes a set-2 descriptor array with a required
nonuniformEXT accessor, and the shared 96-byte push block is declared with each
loose uniform name defined onto its member. The only edits to a shader BODY are
mechanical and dialect-level: dropping default-block uniform declarations, which
Vulkan GLSL has no such thing as, and assigning explicit varying locations BY
NAME across a pair, because ordinal assignment would look identical today and
silently swap varyings the first time an author reordered a line.
Run over the eight production pairs, exactly one thing happened: none of them
compiled, and every failure is a specific source-level fact belonging to a
renderer-port slice that has not landed. debug_line needs uView/uProjection
converged into one uViewProjection - two matrices are 128 bytes and the shared
block is 96. mesh_modern and particle still pass a uvec2 bindless handle as a
varying, which is V4t's GpuTextureSlot retype. sky has ten loose uniforms and
wants a UBO. ui_text needs uScreenSize/uUseTexture/uTex. particle_mesh needs
uTextureIndex to become uTextureIndexA. terrain_modern needs V4d-1's matrix
convergence. mesh is the legacy pair with no RHI consumer at all. That inventory
is committed as shaders.manifest.json, with each source's SHA-256 and the
compiler's own message, and a test re-hashes it so an edited shader that never
got recompiled fails a build rather than shipping a stale binary.
vk_probe is the pair that does compile, and it is the shader 4.11 already asked
for: V5 recorded "build one real pipeline from the committed .spv" as its single
deliberate deviation because no toolchain existed. It is Vulkan-dialect only and
no GL renderer draws with it, so it forks nothing; it retires when the ported
world renderers become the backend's own proof.
DESCRIPTORS. Sets 0 and 1 are DYNAMIC buffer descriptors bound per flight slot,
so a per-draw range change costs a dynamic offset in vkCmdBindDescriptorSets
rather than a vkUpdateDescriptorSets in the hot path - which is what keeps 4.4's
zero-writes-per-frame property true for buffers as well as for textures. Ten
dynamic storage descriptors is above Vulkan's guaranteed minimum of four, so it
is a real requirement rather than a free choice, it fails loudly at layout
creation on a device that cannot serve it, and V9's lavapipe row must confirm
it. Unused bindings point at a shared dummy range so there is ONE set layout and
one pipeline layout; that is why binding a second pipeline mid-pass costs
nothing and disturbs neither the descriptors nor the push constants.
THE ONE MAPPING FUNCTION. VulkanViewportMapping holds the whole coordinate
reconciliation: negative viewport height, the front-face inversion that pairs
with it, and - separately - the scissor flip, which the viewport sign does NOT
perform. The V3 audit flagged that as a concrete V6 acceptance item and it is
the subtle one: vkCmdSetScissor is always top-left-origin, NdcScissorRect emits
GL bottom-left rectangles, and getting it wrong clips a doorway aperture from
the wrong edge in a scene that has one. Clip space needs nothing, as 4.7
concluded: the cameras already build [0,1]-convention projections.
CONTRACT GAP, RECORDED NOT PAPERED OVER. GpuPipelineDescription cannot name its
colour-attachment format, and Vulkan bakes that into a pipeline. Offscreen
targets therefore adopt the swapchain's B8G8R8A8_UNORM rather than a literal
RGBA order - invisible above the API, because an image is sampled through its
format's component mapping and the one CPU readback swizzles explicitly. The
honest fix is a colour-format field added in a reviewed contract commit, exactly
as GpuBlendMode.InverseAlpha and GpuVertexFormat.UByte4UInt were added when V4c
and V4d met the same wall. It is documented at
VulkanTextureFormatMapping.CanonicalColorAttachmentFormat.
The pipeline cache is persisted to the cache directory and validated by its
32-byte header against this device's vendor, device and cache UUID before use.
Drivers are required to ignore incompatible blobs, but "required to" is a poor
foundation for something that runs before anything else in the process, and the
check costs 32 bytes of comparison. Two consecutive launches report "cold" then
"reused".
Gates: Release build clean; App suite 4056 passed / 3 skipped (4037 at V6b plus
19 new); offline pixel gate PASS at a differing fraction of 5.15e-05 with a
same-commit control immediately after it at 2.84e-05 - 29 and 16 pixels of
563,200, the same class of ambient variation the campaign's 15-23 band records,
and roughly 19x under the 0.001 threshold on a commit that changes no GL code
path.
Validation layers could not be run: this machine has no Vulkan SDK, no
HKLM\SOFTWARE\Khronos\Vulkan\ExplicitLayers key, no VK_LAYER_PATH and no
VkLayer_khronos_validation.json anywhere on disk. Plan 7 already requires one
validation-clean run at V7; it needs the SDK installed first and is reported
rather than assumed here.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Runs the instrument section 5.5.2 asked for, on a V4c tree staged from
`git revert --no-commit 543bc79f` and never committed: GL_SAMPLES_PASSED around
the raw-GL terrain draw, the dispatcher's entity draws, and the retained-UI
flush, collected outside the frame that issued them, with the desktop witness as
the verdict. All probe code is stripped; what survives here is the two gate
scripts and section 5.5.3/5.5.4.
Building it found a fourth instrument fault. Reading a query result on the CPU
timeline - glGetQueryObject guarded by RESULT_AVAILABLE, one frame late -
deadlocks V4c at the first frame that draws the world: 4/4 runs, and five
dotnet-stack samples four seconds apart all show the render thread inside the
driver in that call. Not a probe defect - the same probe ran 4,420 clean frames
on the V4c parent, and instrumenting only the UI flush reproduces the wedge while
creating the query objects and never beginning one does not.
Routing the result into a persistently-mapped GL_QUERY_BUFFER instead - the
driver writes it on the GPU timeline, so no client wait is possible, and a
sentinel separates "reported zero" from "never reached" - does not wedge, and
gives the answer. On blank runs no query result is ever produced at any site for
the whole run, including the UI, in the same frames where the desktop grab plainly
shows the UI on screen. On the rendered run of the same binary, 1,068 frames, not
one missing result.
So the mission's fork resolves to "never completes", but not as a stall: frame
time holds at 5.5 ms for ~3,700 frames, the frame-flight fences keep retiring,
and present keeps working. Every channel that carries a result back from the GPU
is dead - pixel readback, CPU query read, GPU-timeline query write - and every
channel that carries none is fine. The transition is one sharp event at the first
world frame and never reverses, and that frame rasterizes correctly: 1,692,830
terrain and 317,561 entity samples, the same two numbers the parent reports for
its own first world frame.
Section 5.5.4 lays out the three options with their costs and recommends (C):
bring Vulkan up first and decide V4c afterwards, because running the identical
ported world path on the Vulkan backend on this GPU is both the cheapest test of
the driver-defect reading and work the campaign owes anyway. (B), accepting the
GL-side fork, is probably the right conclusion but should be adopted on a
measurement rather than an inference. No fix was attempted and V4c is not
re-landed.
Apparatus: run-repeat-connected-gate.ps1 and run-blank-world-ab-probe.ps1 now
assert on the desktop grab and record the client's own capture as a second
column, which is the re-arming section 5.5.2 required before re-land condition 2
can mean anything. Both verified end-to-end.
Gates: Release build clean; App tests 3,866 passed / 3 skipped; offline pixel
gate PASS at 3.37e-05 differing fraction (19 px of 563,200), inside the
documented 15-23 px band.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Investigating the V4c connected blank-world failure needed two things the tree
did not have: a way to tell whether a blank run is caused by the binary under
test, and a way to see GL state at the end of the world phase rather than only
at the frame clear. Both are apparatus only - no production behaviour changes,
and the new probe emits nothing unless ACDREAM_PROBE_GLSTATE=1.
run-blank-world-ab-probe.ps1 interleaves two client builds over the repeat
gate's exact connected route and reports the blank rate per arm. This exists
because the blank rate is not stable across blocks: the same V4c binary
measured 3/10 in one block and 7/10 in another an hour later, so a block of A
followed by a block of B confounds the change with whatever else moved on the
machine in between. Strict alternation shares that drift between both arms.
Run against V4c and its parent it reported 4/5 versus 0/5 (Fisher exact
p~0.024), which is what established the defect follows the binary.
run-blank-world-surface-probe.ps1 grabs the composited window off the desktop
with CopyFromScreen at the same moment the client writes its own screenshot.
No instrument inside the GL context can separate "the renderer drew nothing"
from "the read did not return what the renderer drew", because both live on
the same side of the readback; an independent witness can. It is what showed
the two disagree - see below.
EmitPostWorldGlStateIfChanged is a second sample of the existing [gl-state]
snapshot, taken at the end of the normal-world phase. The existing tripwire
samples just after the clear phase's RestoreFrameDefaults, so it can only
observe state that survives from one frame into the next, and the draw
framebuffer is restored by no frame-global path. A binding established during
the world phase and put back before the next clear was therefore invisible to
it. Sampling at both ends brackets the phase.
What the apparatus established, recorded here rather than in the campaign doc
because no fix landed and the doc's re-land conditions are unchanged:
* The world draw path is not what is missing from the frame. On a blank run
the desktop grab shows the atmosphere clear over the whole viewport and the
complete retained UI - chat, radar, toolbar, vitals - in their normal
places, with every 3-D surface absent. Terrain and sky are still raw GL and
V4c does not touch them, so whatever V4c disturbs is shared, not per-
renderer.
* The CPU issues the same work either way. With ACDREAM_PROBE_FLAP=1 the
render signature is identical between blank and rendered runs: same
RetailPViewInside branch, same resolved root, terrain drawn, 3,331 outdoor
statics and 6 live dynamics dispatched.
* Both GL-state samples read fbo=0, full 1280x720 viewport, scissor off and
err=0x0, byte-identical between blank and rendered runs.
* The client's own capture disagrees with the screen. glReadPixels returns
uniformly RGBA(0,0,0,0) on a frame the desktop grab shows as fog plus UI.
The default framebuffer is 4x multisampled (SampleBuffers=1, Samples=4 in
the capability report) and glReadPixels against a multisampled read
framebuffer is undefined per the GL spec, so the gate's blank-versus-
rendered verdict rests on undefined behaviour in both directions.
Baseline App tests 3,864 passed / 3 skipped, unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The V4c blank-world regression was intermittent - roughly one launch in three at the worst location, zero in seven at the parent - so a single connected capture passes the broken binary most of the time and gates nothing. The new gate runs N full connect-teleloc-render-screenshot cycles with graceful logout and a per-run verdict by screenshot content size, refuses to start if a client is already using the shared test account, and pins the teleloc because the failure rate is location-sensitive. Ten clean runs bound a one-in-three defect below roughly four percent.
Campaign doc 5.5 records the revert evidence and the binding re-land conditions: the GL ring write path moves to mapped unsynchronized writes, and V4c/V4d re-land only at 10/10 rendered plus a passing offline gate.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The offline capture window is minimised but still focusable, so a scroll or key press from whoever is at the keyboard can move the camera mid-capture. That yields two screenshots of the same scene from different camera positions and an enormous, entirely spurious pixel difference - which happened during slice V4b and was correctly discarded rather than interpreted. The gate now detects camera-affecting input in the client log and exits 2, so a perturbed run cannot be mistaken for a rendering regression in either direction.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Seven of the campaign slices (V2, V4a-V4g) are renderer ports whose entire
acceptance criterion is that no pixel changed, and the existing connected
lifecycle route needs both a live ACE server and the user watching. That would
have made the campaign advance only when someone is at the keyboard.
The client renders the world from the DATs without ACDREAM_LIVE, so the existing
UI automation probe can capture a settled frame with no session created and no
ACE state to disturb. The gate wraps that: capture at the parent commit, capture
at slice HEAD, compare through the existing compare-screenshots CLI at the
project's pinned tolerance 2 / 0.001.
Determinism was measured rather than assumed, and the first measurement failed:
two captures at the same commit differed in 0.29% of pixels. The differences
were confined to the top ~180 rows, which is correct behavior rather than a bug
- the sky animates and the Dereth clock advances with wall time, so two launches
cannot agree there. Below the horizon everything was stable. Masking the top 280
rows brings two independent same-commit pairs to 15 and 17 differing pixels out
of 563,200 compared, a ~33x margin under the threshold. Masking the animated
band keeps the rest a strict identity check; relaxing the tolerance instead
would have hidden real regressions everywhere else.
Covers terrain and blending, scenery, static meshes, water, fog, and the whole
retained UI. Does not cover sky (masked), EnvCell interiors, particles, or the
paperdoll viewports, since the offline scene is a fixed outdoor view - so V4e,
V4f and V4g keep a user visual gate on top of this one.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Carry one immutable prepared collision closure with each accepted near-tier generation and install graph plus flat views through the same retained publication receipt. Apply the same strict package-only rule to live entities, add exact sampled graph-authoritative comparison artifacts and lifecycle counters, and prove cancellation, demotion, rehydrate, revisit, teardown, reconnect, and the nine-stop route with 14,064 zero-mismatch samples.
Co-authored-by: OpenAI Codex <codex@openai.com>
Record route-lifetime streaming overruns and maximum operation costs, require canonical reveal/resource convergence at every connected checkpoint, and keep recenter semantics independent of the production wall-clock budget.
Co-authored-by: Erik Nilsson <erikn@users.noreply.github.com>
Starting dotnet-counters in the process-creation race can suspend the CLR before the graphical host creates its window. Wait for the guaranteed in-world boundary, then attach counters and contention tracing during the route's warm-up interval.
Co-Authored-By: Codex <noreply@openai.com>
Free-running sleeps could label a sample as the old scene exactly when automation began the next teleport. Give every route stop an explicit post-input liveness dwell and post-checkpoint hold, then capture process and frame facts only after its named canonical barrier.
Co-Authored-By: Codex <noreply@openai.com>
The first-player-position line is emitted only when initial streaming must recenter. Treating it as a required login event made a valid same-landblock login time out while route automation was already running. Gate on in-world and the route's authoritative materialization barriers instead.
Co-Authored-By: Codex <noreply@openai.com>
Correct whole-frame GPU timestamps so they bracket only the accepted render transaction and associate delayed query results with the owning CPU frame. Add route-wide frame-history summaries, fixed-camera screenshot comparison, process counters, contention traces, a pinned Arwic workload, and credential-safe launch disclosure.
The reference hardware/display contract now keeps local and RDP populations separate and defines the screenshot and re-baseline rules needed by later prepared-content gates.
Co-Authored-By: Codex <noreply@openai.com>
An adversarial performance review found our own instruments cannot
measure the project's own performance gates:
- FrameProfiler aggregated CPU/GPU/alloc/stage samples into ~5-second
windows and reset the ring buffers after each report, so route-wide
p50/p95/p99 distributions across a whole soak could not be
reconstructed after the fact. ACDREAM_FRAME_HISTORY=<path> now opts
into a separate per-frame history (one record per frame, ~72
bytes/record, accumulated in memory with zero frame-thread I/O) that
a shutdown-only Dispose() writes as CSV. The aggregated [frame-prof]
report format and its existing metrics are unchanged.
- The canonical checkpoint JSON tracked cache residency (entry/byte
counts) but never LOH size/fragmentation, process-wide allocated
bytes, or cache hit/miss/eviction traffic — a committed audit JSON
showed 65% LOH fragmentation that no tracked instrument recorded,
and "does a revisit portal hit or miss the caches" was unanswerable
from an artifact alone. WorldLifecycleResourceSnapshot now carries
loh_size_bytes/loh_fragmentation_bytes (GCMemoryInfo.GenerationInfo
index 3), process_total_allocated_bytes (GC.GetTotalAllocatedBytes),
and Interlocked hit/miss/eviction counters for the CPU mesh cache,
decoded-texture cache, and the four bounded DAT-object caches
(portal/cell/highRes/language, aggregated).
- run-connected-r6-soak.ps1 unconditionally forced
ACDREAM_UNCAPPED_RENDER=1 with no capped mode, while its sibling
lifecycle-gate script correctly gated it behind a switch. Added
-Uncapped (default capped, matching the sibling script's pattern),
fixed the stationary dwell (12s -> 26s, past the 25s
LiveEntityLivenessController deadline the adjacent comment already
cited), and now write an env-disclosure.json into the automation
artifact directory before every launch listing every ACDREAM_* var
the script sets plus -Uncapped, since the prior audit could only see
ACDREAM_DUMP_MOVE_TRUTH and nothing else was ever recorded anywhere.
Cache counters are wired via the existing composition path
(ObjectMeshManager already owns the CPU mesh cache and the mesh
extractor directly; content.Dats is threaded into
WorldLifecycleResourceSnapshotSource the same way every other
composition consumer receives it). The DAT-object cache lives behind
IDatReaderWriter, a third-party interface from the DatReaderWriter
package that cannot be extended; RuntimeDatCollection (the one
production implementation) exposes the aggregate stats directly and a
pattern match reads them, degrading to zero for any test double —
no new static registry was introduced (GpuMemoryTracker remains the
one precedented process-wide static).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(cherry picked from commit 1da2c33c875b41fa383dd79694ee2765f0e21896)
Pin every ACE teleloc to an explicit starting quaternion and distinguish stable-workload owner growth from legitimate authoritative population, visibility, and cache-retirement changes. Preserve the original process residency and frame-cost limits while making canonical deltas actionable.
Co-authored-by: Codex <codex@openai.com>
Make scripted lifecycle checkpoints acknowledged post-diagnostics render barriers, capture the exact frame outcome beside canonical resource ownership, and harden the nine-stop route with ordered same-location cache and lifetime gates without weakening process residency thresholds.
Co-authored-by: Codex <codex@openai.com>
Keep one-item endpoint and checkpoint results as arrays under PowerShell strict mode so the uncapped reconnect session is validated by the same gate as the multi-checkpoint capped route.
Send the active character id, drain until the authoritative server confirmation, then emit retail's zero-sequence connection disconnect with the negotiated receiver iteration. The connected gate now waits for ACE to remove the exact UDP session before reconnecting, eliminating fixed-delay races.
Drive turn, movement, jump, and combat through the production InputDispatcher so connected Release testing works without an interactive Windows desktop. Track held automation actions through normal completion and every shutdown path.
Add a seven-destination ACE route with post-liveness memory, allocation, update, fatal-log, outbound-movement, and graceful-close gates. Record dynamic ACE population changes as context instead of a false lifetime oracle, and document the accepted rebaseline.
Co-authored-by: OpenAI Codex <codex@openai.com>
Replace the incomplete package path with one DatCollection-backed compatibility seam for PhysicsScripts and Animations. Preserve CreateBlockingParticle's inherited payload and following cursor, route every production and audit consumer through the corrected loaders, and apply retail's post-UnPack StartTime ordering.
Add exact stored-order PhysicsScriptTable upper-threshold resolution, high-byte DID and embedded-ID validation, plus live effect profiles with Setup-to-PhysicsDesc precedence across top-level and attached entity lifetimes. Keep blocking execution deferred and narrow TS-11 accordingly.
Pin synthetic malformed/cursor/order fixtures, installed-DAT blocking and recall audits, high-index IDs, IEEE boundaries, profile teardown, and ordinary decoder parity; synchronize architecture, inventory, milestones, roadmap, research, and memory.
Co-Authored-By: Codex <noreply@openai.com>
Establish the executable-backed PhysicsDesc, sequence-gate, PhysicsScript, CreateBlocking, particle-anchor, projectile, and Hidden-state behavior before changing runtime code. Correct stale blocking/threshold claims and synchronize the project instructions with the current UI architecture and matching retail binary.
Add copyright-safe packet and DAT-container fixtures plus a failing installed-DAT conformance audit for projectile shapes, typed tables, recall motion, default scripts, and raw CreateBlocking inventory.
Co-Authored-By: Codex <noreply@openai.com>
Preserve the authored gray track, trained-Recklessness range, live bright charge meter, and independent desired-power thumb while keeping Speed and Power over gray side regions. Correct retail text justification value 2 to left alignment and retain direct RenderSurface decoding in the texture inspection tool used to verify the assets.
Co-Authored-By: Codex <codex@openai.com>
The indoor GREY flap at a top-floor connecting room. The render portal side-cull
reconstructed each doorway's "interior side" (PortalClipPlane.InsideSide) from the
cell's AABB CENTROID. For a THIN connector cell (0xF6820118, 5 render polys), the
bounding-box center falls on the WRONG side of the 0118->0116 doorway, so the eye
read as a back-portal and the forward room 0116 was culled -> the aperture showed
the fog clear color = grey.
Retail's PView::InitCell (0x005a4b70) and acdream's own PHYSICS path
(CellTransit.cs:190) both read the explicit dat PortalSide bit ((Flags&2)==0)
instead of guessing from geometry. Port the render path (GameWindow.BuildLoadedCell)
to the same bit.
Proven by a live retail cdb trace (retail draws 0116 from the 0118 root at the grey
pose; tools/cdb/issue186-connector-decider.cdb) + an offline dat diagnostic
(Issue186...PortalSide_CentroidVsDatBit_AtGreyEye): the dat bit matches the old
centroid on every portal of these cells EXCEPT the one #186 breaks, so the switch is
surgical. Full regression green (App 741 / Core 2631); the CornerFlood + Issue113
dat-loading helpers updated to the same bit confirm every real Holtburg/tower/hall
cell floods identically. Touches neither PortalSideEpsilon nor the deleted
EyeInsidePortalOpening rescue (the two DO-NOT-RETRY traps).
Live-gated: user-confirmed no grey at any camera angle; probe shows 216 root=0118
frames, 0 still grey (0118->0116 now TRV, vis=4).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 0 of the #182 verbatim rebuild. The classifier reproduces the design
baseline off acdream-crowd-resolve.jsonl (2883 move-intent resolves:
52.8% OK / 25.1% partial / 22.1% stuck / 107 airborne-stuck) — the A/B
'before' the rebuild measures against (retail target ~78% OK, 0 airborne-stuck).
The plan refines the design spec's §7: the airborne-stuck bleed is the
frames_stationary_fall counter (validate_transition increments; handle_all_collisions
zeros velocity at fsf>1), NOT the cached_velocity field (a separate reporting value).
Slices reorder accordingly; calc_friction (retail 0.25 vs acdream 0.0) is an
orthogonal L.3c divergence kept out of scope.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The #182 CSphere port (96ae2740) failed its visual gate and introduced an
airborne "stuck in the falling animation" regression. A player-attributed retail
cdb trace (tools/cdb/retail-crowd-jump3.cdb) proved retail's LOCAL client fully
runs player-vs-creature collision (76 land_on_sphere, 188 COLLIDED, 130 SLID,
~78% OK, glides across) -- NOT server-authoritative (an earlier unfiltered
land_on_sphere=0 read was a false lead the attributed trace refuted).
acdream's same-repro capture: 50.9% OK, 22.4% stuck, 115 airborne-stuck. Root
divergence: retail CPhysicsObj::UpdateObjectInternal (0x005156b0, pc:283688) sets
cached_velocity = (resolved - old)/dt -- velocity from ACTUAL movement, so a
blocked jump collapses to ~0 -> gravity -> the player falls/glides. acdream
integrates velocity + reflects on collision (PlayerMovementController ~:1008-1069),
so the jump velocity (~18) persists against the creature -> hang.
Fix = verbatim rebuild of the per-frame player-physics loop (UpdateObjectInternal
chain), velocity model first, transition internals kept. Full design +
retail function inventory + the capture apparatus + retail target numbers:
docs/superpowers/specs/2026-07-07-player-physics-update-verbatim-rebuild-design.md.
Implementation deferred to a fresh session (user decision). Also files #183
(floating distant scenery, observed during testing). #182 stays as the base.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Attached cdb to live retail (PDB MATCH), broke on PView::DrawCells (0x005a4840),
dumped cell_draw_num + cell_draw_list cell ids + the eye
(Render::FrameCurrent->viewer.viewpoint) while descending the Facility Hub spiral.
Retail's flood is dynamic and IDENTICAL in character to ours: from the spiral cells
it swings num 3->27 with gaze and collapses to 3 cells at many poses (cam=015d ->
{015d 015e 015f}). Our flood does the same (3->43). So retail does NOT keep the
staircase where we drop it -- the flood is exonerated as the cause.
Session trail (all in ISSUES #177): ruled out lighting, membership, camera coherence,
the collision sweep, the 0178/0182/0183 handoff cells, and edge-on eye-in-opening
(fix#1 shipped -> visual-gate-failed -> reverted, PortalVisibilityBuilder + AP-86 both
restored exactly). Freshest un-chased lead: the steps are STATIC objects (GfxObj
0x010000DE x6/cell) drawn via the viewcone cull, not cell shell.
Adds: Issue177StairDescentCameraFloodTests (real-camera+flood + composition +
flood-depth characterization pins) and a reusable retail-cdb capture toolchain
(tools/cdb/pview-verify.cdb, pview-spiral2.cdb with the correct top-level-qd detach --
qd in a CONDITIONAL bp action does NOT fire and strands cdb attached).
No production code change (fix#1 reverted). PARKED per user; M1.5 critical path next.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
cdb trace of LIVE retail (tools/cdb/issue176-floor-light.cdb, binary<->PDB MATCH) PROVED retail
applies ALL its dynamic lights — 4 intensity-100 magenta portal lights (d3dIdx 3-6, falloff 6) +
the viewer fill (d3dIdx 1, 2.25) — as D3D hardware lights to EVERY Facility Hub cell, every frame,
stable. So the faceted purple wedges on the floor are retail-FAITHFUL. acdream did a per-cell
SelectForObject sphere-overlap 8-cap for cells, so the portal set could differ/flip per cell.
- LightManager.SelectForCell (retail minimize_envcell_lighting 0x0054c170): ALL dynamic lights
applied unconditionally (shader range cutoff zeroes non-reaching = D3D hardware range), then
nearest static torches fill remaining slots. Wired into EnvCellRenderer.GetCellLightSet.
Objects keep SelectForObject (minimize_object_lighting). Pins:
SelectForCell_AppliesAllDynamicLights_EvenOutOfReach + _SameDynamicSet_ForCellsFarApart_NoFlap.
- Apparatus: [light-detail] gains owner/cell/dyn (pinned the culprit = 2 portal weenies
0x000F4247/48 in 0x8A020118/19, intensity=100 magenta); CellVertexNormals_SmoothOrFaceted_Dump
(corridor floor uses SMOOTH per-vertex dat normals, not flat); tools/cdb/issue176-floor-light.cdb.
#176 RESIDUAL is NOT this fix. It's a RUNTIME draw z-fight in the seam floor. Eliminated (evidence):
NOT lighting (per-light cap + this both no-change), NOT membership (render cell 0x8A020164 stable
100% of 188k frames / 526 angles, res=None), NOT dat geometry (coplanar sweep empty at z=-6 floor
incl. cell 0164). NEXT = RenderDoc pixel-history. Full handoff + DO-NOT-RETRY:
docs/research/2026-07-06-176-seam-floor-zfight-handoff.md. Suites green: Core 2599 + 2 skip.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pseudocode for the S2 port (unpack_movement case 0 / move_to_interpreted_state
/ apply_current_movement / apply_interpreted_movement / DoInterpretedMotion),
anchored on decomp lines + validated against a LIVE cdb trace of a retail
observer (per-UM DIM order confirmed: style -> forward -> sidestep-stop ->
turn-stop; empty UM = wholesale Ready stop).
Also settles the packer question: RawMotionState::Pack (0x0051ed10) is pure
static-default-difference — outbound L.2b port already verbatim; the
empty-vs-explicit walk variance between captures is driver-client state,
handled identically by the wholesale apply.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The paperdoll doll now matches retail: correct held pose, framing, and
facing. Three decomp-sourced fixes closed the visual gate.
Pose: cdb-confirmed m_didAnimation = 0x030003C0 (gmPaperDollUI), played
once + HELD (set_sequence_animation framerate=0, RedressCreature
0x004a3c22). Dumping the dat showed the 29-frame anim has only two
distinct keyframes — frame 0 (transitional, bent arm) and frames 1..28
(byte-identical: the settled stance, arms down + leg back) — so
ApplyPaperdollPose applies the LAST frame statically (no looping).
Camera: ported verbatim from UIElement_Viewport::SetCamera (decomp
0x004a5a39). position (0.12,-2.4,0.88); direction (0,0,0) => IDENTITY
view frame => look straight down +Y, ZERO yaw; FOV pi/4 (CreatureMode
ctor default 0x004543cf); ambient 0.3. The prior hand-tune aimed the
camera at mid-body, adding a ~2deg yaw that turned the doll's face away
— full-body framing comes from eye-height + FOV, not aiming.
Heading: retail Frame::set_heading(h) (0x00535e40) builds facing
(sin h, cos h); System.Numerics CreateFromAxisAngle(+Z, +h) rotates the
body's default +Y forward to (-sin h, cos h) — the X-lean was MIRRORED
(~22deg), the real cause of the turned-away face. Negate the angle to
land on retail's facing.
Wrap-up: stripped the temporary O/P pose-frame stepper + Slice2
diagnostics; divergence register AP-66 reworded, AP-67 (RTT doll render
vs in-cell CreatureMode::Render) + AP-68 (per-race UpdateForRace
unimpl) added; DollCameraTests pinned to the retail values + a zero-yaw
guard. tools/cdb/paperdoll-pose.cdb = the pose-DID capture script.
Build + full suite green (Core 1579 / Core.Net 343 / App 597 / UI 425).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Resolves the divergence-register conflict: kept the accurate per-VERTEX AP-35
(Fix A shipped per-vertex; main's row was the stale pre-Fix-A per-pixel text),
kept main's UI rows AP-37..AP-42, and renumbered this branch's torch-gate row
AP-37 -> AP-43 (AP-37 was taken by main's LayoutDesc row). AP count 41 -> 42.
Retargeted the AP-37 references in WbDrawDispatcher + the CHECKPOINT to AP-43.
Marked ISSUES #140 RESOLVED (b7d655b) with the corrected root cause.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Resolve the Fix D contradiction with decomp (workflow wf_f660eb88 + adversarial
verify) + 4 live cdb captures. The D3D-FF model was the WRONG oracle: retail has
TWO light systems — STATIC torches BAKE into wall vertices (calc_point_light,
triple-clamped: range gate + per-channel min(scale*color,color) + per-vertex
[0,1] from black), DYNAMIC lights go D3D hardware. The captured intensity=100 is
the purple PORTAL (magenta, dynamic), not a wall torch. Ground truth: 38 static
warm torches (orange (1,0.588,0.314)/cream, intensity=100, falloff 3-5) + 2 dynamic.
acdream over-brightness = two confirmed bugs: D-1 mesh_modern.vert folds
ambient+sun+torches into one UNCLAMPED accumulator (single frag clamp) -> warm
blowout; D-2 EnvCellRenderer never binds SSBO 4/5 so the cell shell reads a leaked
light set. Spec: D-1 in-shader clamp-split (clamp the torch sum on its own before
ambient/sun); D-2 bind the shell's own per-cell light set (mirror WbDrawDispatcher);
LightBake.cs is the C# conformance oracle. Adds the 4 reusable cdb capture scripts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
feat(D.2b): exact retail chat colors from a live cdb dump
Attached cdb to a live retail acclient (PDB-matched) and read the named RGBAColor
constants at acclient 0x81c4a8+ (colorWhite/colorBrightPurple/colorLightBlue/
colorGreen/colorLightRed/colorGrey), used by ChatInterface::BuildChatColorLookupTable
@0x4f31c0. Replaced the approximated RetailChatColor palette with the ground-truth
values: speech=white, tell=colorBrightPurple(1,.498,1), channel=colorLightBlue
(.247,.749,1), system/popup=colorGreen(.5,1,.498), combat=colorLightRed, emote=colorGrey.
Capture scripts saved under tools/cdb/.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@
The register's UN-2 row recorded a contradiction: the GetMaxSpeed XML doc
claimed the bare run rate was retail-correct (~5.9 m/s catch-up, calling
the xRunAnimSpeed multiply a misread), while the implementation multiplied
by RunAnimSpeed citing ACE. Settled against the binary, not the pseudo-C:
- BN pseudo-C (acclient_2013_pseudo_c.txt:305127) renders get_max_speed as
void with a bare `this->my_run_rate;` because it DROPS x87 instructions.
- Disassembling the PDB-matched v11.4186 binary at VA 0x00527cb0: all THREE
return paths end `fld <rate>; fmul dword ptr [0x007C8918]; ret`, and the
.rdata dword at 0x007C8918 is 4.0f. Sibling get_adjusted_max_speed
(0x00527d00) carries the same trailing fmul. Verifier committed at
tools/verify_un2_fmul.py (PE parse + byte decode, rerunnable).
- Retail paths: weenie null -> 1.0 x4; InqRunRate ok -> queried x4;
InqRunRate failed -> my_run_rate x4. ACE MotionInterp.cs:665-676 matches.
Changes:
- Doc-comment rewritten: the implementation is retail-correct; the catch-up
speed 2 x get_max_speed ~= 23.5 m/s at run 200 IS retail. The 1-Hz
remote-blip symptom the old comment attributed to this multiply is
therefore UNEXPLAINED by it (if it recurs: #41 family, not this).
- Weenie-null path aligned to retail's LITERAL 1.0 default (was MyRunRate).
- Tests re-pinned to the three retail paths (the old NoWeenie test pinned
the non-retail fallback).
- Register: UN-2 row deleted per the retire rule (6 -> 5 UN rows);
shortlist renumbered.
This is the 2nd confirmed instance of the BN x87-dropout artifact class
(memory: feedback_bn_decomp_field_names) deciding a register row.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Reconciles the wip/main-local-claudemd-condensation distillation (made
~2026-06-03 by a parallel session) onto post-merge main:
- Current state section: ONE status block (<=5 lines + pointers, by
rule) + canonical reading order + the two digest entry points + the
divergence register; replaces status sediment scattered through
Goal/Roadmap sections.
- Kept from the merged main (the condensation predated them): the
memory/digest rule in How to operate, the divergence-register
section + phase-checklist item, de-dated milestone rules.
- Dropped: shipped-phase ship-notes, stale next-phase candidate lists,
the superseded reference_render_pipeline_state pointer.
- Also salvaged from the wip branch: .gitignore entries (.obsidian/,
claude-memory junction) + pdb_extract.py __main__ guard. The wip's
TextureDump edit predates main's args support (discarded) and its
physics-probe edits were STRIP-marked leftovers (discarded).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Attaches to live retail, dumps CameraManager/SmartBox type offsets, captures viewer eye origin per frame (pub+sought) to measure boom jitter vs acdream. Used to pin the indoor flicker to the camera's published/swept eye (retail settled ~tens of um vs acdream ~1.3mm).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Attached cdb to the live 2013 retail client at the Holtburg doorway + read the decomp.
The indoor flap is a STRUCTURAL divergence, settled by measurement (not inference):
- Retail has ONE render path: DrawInside(viewer_cell) every frame. NO inside/outside
branch (RenderNormalMode's outside branch is dead code; is_player_outside only gates
sky/lighting). "Entering a building" is not a render event — only the camera sweep
resolving a different viewer_cell. Same path before/after threshold -> no seam.
- Retail's eye JITTERS ~36um at rest yet membership is stable -> robustness is
STRUCTURAL: many small per-building floods (~7/frame, ~2 cells each, via terrain BSP
-> DrawPortal -> ConstructView(CBldPortal)), not one giant knife-edge flood.
- Our 3 divergences: (D1) invented inside/outside branch (GameWindow.cs:7498,
clipRoot = viewerRoot ?? _outdoorNode :7396); (D2) synthetic _outdoorNode; (D3) one
unified flood.
DECISION (user-approved): Option A — rip out branch + outdoor node, root always at the
real viewer_cell, one DrawInside, per-building rendering. Phased, conformance-tested,
visual-gated.
REFUTED by measurement (do not retry): bounded-propagation/churn (maxPop=1, 0/63k
reciprocals empty); byte-stable eye (retail's jitters ~36um — rest-snap cd974b2 failed +
regressed, reverted 9b1857a).
Lands the canonical exhaustive handoff for a FRESH session
(docs/research/2026-06-08-full-retail-render-port-OPTION-A-handoff.md), the CLAUDE.md
READ-THIS-FIRST banner, and reusable cdb apparatus. No project code changed; working tree
at the known-good baseline.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Apparatus + handoff for the indoor flap. Confirmed (primary evidence): the flap is the
portal-flood clip being µm-sensitive at the threshold, driven by a ~1-8µm jitter in the
player RenderPosition (physics resting position not bit-stable; Lerp surfaces it). REFUTES
the 2026-06-07 see-through/EnvCell/outdoor-node diagnosis (ModelId GfxObj 0x01000A2B IS the
solid exterior) AND an enqueue-once attempt (retail propagates late slices via AddToCell;
the existing PropagatesNewSlicesToExit test caught it; reverted). Adds: Build determinism
test, A8CellAudit gfxobj dump, [pv-input] 6dp probe + [render-sig] outRoot/bshell fields.
No functional fix shipped. Next: higher-precision physics rest trace -> port retail
kill_velocity/contact rest-stability. Canonical: docs/research/2026-06-08-flap-rootcause-physics-rest-handoff.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Checkpoint of the unified retail-faithful indoor render. The two-week HANG/grey is fixed and the
interior seals (live-verified by the user). Commits the session render-rewrite foundation together
with the fixes that made it functional.
- HANG fix: PortalVisibilityBuilder.Build portal flood did not terminate (the faithful ProjectToClip
near-side clip drifts per round, defeating the CellView dedup; the BFS had no bound after U.2a removed
MaxReprocessPerCell). Fix = drift-tolerant snapped/canonical CellView.Add dedup (PortalView.cs) plus
restored MaxReprocessPerCell=16 bounded re-enqueue (PortalVisibilityBuilder.cs). Re-enqueue is kept
(load-bearing for late-slice propagation, Build_ViewGrowthAfterDoneCell_PropagatesNewSlicesToExit);
only its count is capped. CellViewDedupTests added.
- Seal (DrawCells Task 2): RetailPViewRenderer.DrawEnvCellShells draws EVERY visible cell via
IndoorDrawPlan.ShellPass (was gated on the ClipFrameAssembler slot filter, leaving slot-less cells grey).
- Look-in FPS: GameWindow exterior look-in candidates limited to the player landblock +-1 (was all ~81
loaded LBs iterated every outdoor frame). No behaviour change (far cells were >48m, already culled).
Remaining dominant issue = the FLAP at transitions: viewer-cell metastability (render roots at the
camera-eye cell, which oscillates outdoor-indoor as the 3rd-person boom drifts across the doorway,
confirmed in render-sig). SEPARATE fix, NOT the DrawCells port. Full handoff + flap fix plan + tracked
follow-ups (#78 terrain, look-in-from-inside, look-in FPS, L-spotlight):
docs/research/2026-06-07-indoor-render-session-handoff.md.
Baselines: build 0 err; App.Tests 210/210; Core.Tests 1331 pass / 4 fail (pre-existing) / 1 skip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diagnosis session: the indoor bluish void + grey/texture flicker is visibility metastability at cell boundaries, not a missing flood (R1's per-cell DrawInside is built; the cellar seals). Confirmed by named-retail decomp AND a live cdb capture of retail (viewer_cell rock-stable: clean monotonic transitions, zero oscillation across 4916 samples). Retail stays stable via boom stability + a 0.2mm viewer-cell dead-zone + clip-space portal clipping; acdream diverges on all three. Handoff documents the root cause, the cdb evidence, and the prioritized 3-part retail-faithful fix (boom stability -> dead-zone -> w-space clip) with decomp anchors + a planning/implementation kickoff prompt. Adds the reusable retail viewer-cell cdb capture script and the superseding CLAUDE.md banner.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P2 / M1.5 "blocked at the last step" cellar-lip wedge. This session built a faithful
deterministic reproduction and peeled the cause through six evidence-disproven framings
to one bounded question. NO fix landed — the last layers were each disproven by evidence,
and guessing at the load-bearing collision code is the saga's failure mode.
Apparatus:
- CellarLipWedgeTests.cs + Fixtures/cellar-lip/ (3 real cell dumps + wedge-records.jsonl =
29 captured ACDREAM_CAPTURE_RESOLVE wedge calls). Replays the exact calls + body-before
through the lip-cell engine: all 29 reproduce at 0% advance in <200 ms. Tests are
documents-the-bug / diagnostics (GREEN while the wedge exists).
- TEMP probes ([path5-wall]/[fw-enter]/[find-walkable] in BSPQuery; [neg-poly]/[stepsphereup]/
[stepdown-decide]/CheckOtherCells cn/sn/negHit in TransitionTypes), gated on
ACDREAM_PROBE_INDOOR_BSP, marked STRIP. TransitionTypes neg-poly shortcut has a reverted-fix
comment (slide attempt didn't clear the wedge).
- tools/cdb/retail-*-trace.cdb (retail cdb traces).
Findings (handoff: docs/research/2026-06-04-p2-cellar-lip-flatfloor-cp-handoff.md, see the
"NEXT-SESSION KICKOFF" at top):
- Flat-floor contact plane is retail-faithful (v1 trace, full-file correlation). NOT the bug.
- PosHitsSphere cull sign is retail-faithful (cdb -z verified; the Binary Ninja `test ah,N; jp`
parity-jump reads inverted — caught + reverted a wrong fix from that mis-read).
- Sphere radius correct (0.48 player / 0.30 camera probe).
- Retail connector cell 0xA9B40175 never blocks (CEnvCell::find_collisions trace: 0 Collided/Slid).
- PINNED: during the step-up's step-down, BSPQuery.FindWalkableInternal is never called for cell
0171, so the cottage floor (poly 0x0023, Z=94) is never tested as walkable -> no contact plane
-> step-up fails -> StepUpSlide=Collided -> wedge.
Next: trace FindEnvCollisions -> FindCollisions path dispatch for 0171 during StepDown=true (why
StepSphereDown/find_walkable is skipped), port retail, validate via CellarLipWedgeTests, regress
DoorBugTrajectoryReplayTests + visual gate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live retail cdb trace (tools/cdb/cellar-corner-escape.cdb) of the Holtburg
cottage cellar-top corner decodes the ground truth: retail escapes by
step_sphere_up->step_up (196x vs 38 near-misses), transitioning the contact
plane from the ramp (N.z=0.78) onto the flat cottage floor (N.z=1.0, 76
landings). acdream slides at the lip and never makes that ramp->floor
transition -> the intermittent cellar wedge.
So the remaining cellar bug is the #98-core step-up-onto-cottage-floor
(DoStepDown / step_sphere_down / find_walkable), which the shipped B1 (abbd761)
+ slide_sphere (0935a31) fixes got close to but didn't finish. Door still
blocks; generic step-up climbs; cellar went always-stuck -> works-mostly.
Next (handoff doc): instrument acdream's OWN corner path (does step_up fire at
the lip and fail to land on the cottage floor?) before porting the lip-climb --
no guessing (#98 saga rule).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The P1 "doorway membership lags retail" premise is FALSIFIED. acdream's swept
ResolveWithTransition already matches retail's true per-frame curr_cell: the
production gate ProductionPath_IndoorCrossings reads 9/9 on the indoor 0170<->0171
crossings with NO code change, once fed an aligned retail golden.
Root cause of the false 0/11: CPhysicsObj::SetPositionInternal calls change_cell
(acclient_2013_pseudo_c.txt:283456) BEFORE set_frame writes m_position (:283458),
so the original golden (find-cell-list-capture.cdb, read at the change_cell BP)
paired each frame's NEW cell with the PREVIOUS frame's position — a one-frame skew.
Verified 3 ways: the decomp ordering; golden_picked[i] == geom(golden_position[i+1])
for all 22 rows; acdream's static pick == golden_picked[i-1] for all rows. Both
retail and acdream pick with center-only point_in_cell on global_sphere[0] (no XY
lead; cache_global_sphere @ pc:274196). curr_cell commits via validate_transition
(@ pc:272608, curr_cell = check_cell) = the find_cell_list pick, structurally
identical to acdream's RunCheckOtherCellsAndAdvance -> FindCellSet -> SetCheckPos.
There was nothing to port; a swept advance would make membership LEAD by a frame.
- tools/cdb/find-cell-list-capture-aligned.cdb: re-capture reads the committed
position from the set_frame that follows change_cell (cell+position same instant).
- Fixtures/find-cell-list-threshold.log: replaced with the aligned capture.
- ThresholdPortalCrossingReplayTests / FindCellListConformanceTests: rewritten from
documents-the-bug to assert retail truth (per-segment / per-indoor-pick equality).
- handoff + notes + README + memory: banners correcting the disproven premise.
Still open (NOT indoor membership, which is DONE): outdoor->indoor 0031<->0170 entry
conformance (needs landcell + building stab in the gate cache); master-plan cleanups
(delete CheckBuildingTransit, unify find_env_collisions, demote ResolveCellId) refactor
working retail-faithful code -> need explicit user approval.
Conformance 60 pass / 1 skip / 0 fail; full Core 1309 pass / 5 fail (pre-existing
2 BSPStepUp + 3 door-collision = P2) / 1 skip.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Task 6 complete. Captured live retail membership at the 0031<->0170<->0171
doorway via cdb on CPhysicsObj::change_cell (symbol-driven; offsets verified by
discover-types.cdb; PDB MATCH). 22 transitions, clean monotonic sequence, NO
ping-pong (retail is correct-by-construction). Golden:
Conformance/Fixtures/find-cell-list-threshold.log.
ROOT-CAUSE FINDING (the central P1 work): retail transitions membership at the
PORTAL CROSSING (CEnvCell::find_transit_cells @ 0x52c820 pc:309968 — sphere crosses
the doorway polygon plane), while acdream's FindCellList re-picks by POINT-IN-CELL
containment at the foot. Retail commits room 0171 while the foot is STILL inside
vestibule 0170's BSP (in_0171=0); acdream lags. ALL 22 transitions diverge for this
one criterion mismatch — not a per-cell hysteresis or a building-entry-only split.
This is master-plan §0 'hysteresis gap' confirmed against the real client.
FindCellList_DoorwayThreshold_DivergesFromRetail_PendingP1 (documents-the-bug, GREEN)
+ ThresholdDivergenceDiagnosticTests (per-transition containment print) pin it; both
flip when P1 ports the directed portal crossing. Conformance 59 pass / 1 skip / 0 fail;
full Core 1308 pass / 5 fail (baseline) / 1 skip — no new failures.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P0 Task 5. RetailTrace parses the [fcl] golden format (seed/pos/picked,
RetailCellPick); 4 TDD tests green. find-cell-list-capture.cdb targets
CPhysicsObj::change_cell (commit-on-diff) to capture retail's accepted
membership sequence at the doorway; README is the operator runbook
(dt offset verification + decode_retail_hex float decode). The live run
is P0's one user-gated step (Task 6 mines existing traces first).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
