Nishi Family › Compare › Browser Performance -- 4K-class fidelity on the device the visitor actually has
Nishi Compare · capabilities, evidence and delivery
Browser Performance -- 4K-class fidelity on the device the visitor actually has
Compare capabilities and delivery evidence. Source discovery, qualification, deployment and accepted outcomes are separate; an open contract names work still to verify.
The subject is not whether it renders. It is what a high-fidelity experience COSTS to deliver and to run: bytes on the wire, time to first frame, frame pacing after the phone gets hot, and whether the renderer can trade resolution for RECONSTRUCTION rather than for blur. Judged against the in-browser delivery field. Nishi figures measured 2026-08-28 on /world/beach and the sovereign compressor.
Overview
The purpose, declared position and evidence coverage of this domain. Source presence and completed acceptance are different measures.
Where we are. The delivery path is measured for the first time and one rung is already shipped. /world/beach is 477701 bytes on disk served as 252724 gzip, and 275468 of those bytes -- 56 percent of the document -- are one base64 string literal carrying a 206600 byte wasm module. Base64 costs a MEASURED 29.6 percent on the wire even after compression, WebAssembly.instantiateStreaming is ABSENT-PROVEN across 23320 runtime files at corpus_complete=1, and Cache-Control no-cache means the byte-identical module is re-sent on each of the six worlds while crossing between worlds is the advertised feature of every one. Against that, lazy matching shipped today: the sovereign DEFLATE went from 252105 to 237977 bytes on this page and 107878 to 103082 on the module, both inflating byte-identical with CRC32 and sha256 verified, contentdiff GREEN at lost_from_live=0, promoted with prev banked. The frame side is in better shape than the delivery side and was NOT the gap: the adaptive controller derives every line from a device-measured budget with a one-over-q-squared cost model and non-touching hysteresis bands, and a p50 p95 p99 jank harness already exists.
Where we need to go. A visitor on a phone, a laptop and a 4K panel each receives the highest fidelity their device can actually sustain, and the cost of that is a published number rather than a claim. Concretely: the module is fetched once and cached across every world instead of re-sent per crossing, compilation overlaps download instead of following it, the renderer trades resolution for RECONSTRUCTION rather than for blur by reusing the motion-vector primitive the estate already built and verified, and the device budget is re-measured continuously so a hot phone is not judged against the frame budget it had when it was cold.
The unit. 1 u = one measured session-leg. Calibration: the lazy-matching rung -- read the compressor, isolate the levers against a reference, edit two copies, build, content-rule, behaviour-probe, round-trip prove and promote -- was one leg and is the anchor for every estimate below.
Where we are: 2 of 12 rungs. PF3 dynamic Huffman SHIPPED 2026-08-29: lazy+fixed 254216 vs lazy+dynamic 217388 on the same 509018-byte page, a 14.49 percent gain against an 8 percent floor, 1.3 percent from zlib -9, proven by nx_gunzip round-trip, zlib-strict inflate, an adversarial stress set and GNU gzip 1.13 on another machine -- and its first build was REFUSED BY ITS OWN ORACLE (a one-Kraft-unit over-subscription zlib named as an invalid literal/lengths set) and fixed the same hour, which is the round-trip tooth doing exactly what it exists to do. Lazy matching is shipped and live. Eleven contracts are open, eight of them named for the first time by this board. The largest single win and the only LIVE DEFECT are both open, and so is the pairing that would have made the whole board honest from the start -- PF11, because a frame-time number published without an image number on the same clock is how a day of measured improvements reads to the operator as pixels.
What it costs today: 252724 bytes per world entry. Measured on the served response 2026-08-28. With no-cache and six worlds sharing one byte-identical module, an isekai visitor pays that on every crossing.
What is already proven: 29.6 percent. The wire cost of base64 that survives compression, measured on the real module bytes rather than argued from the 4-over-3 expansion.
Research bar. zlib 1.3 DEFLATE level 9 is measured on compressed size of the same shipped wasm module bytes through raw deflate. Theirs: 0.429 of input. Ours: 0.499 of input after lazy matching shipped today and 0.522 before it. The residual gap is dynamic Huffman which was isolated at 11.3 percent using zlib's own Z_FIXED strategy so the figure is the Huffman table alone and not the matcher.
Research bar. Google Core Web Vitals good threshold is measured on largest contentful paint judged good at the 75th percentile of real visits. Theirs: 2500 ms. Ours: UNPINNED -- we have no field LCP for /world/beach. nx_uiq_perf is a STATIC predictor and returned PASS on this page while blind to 275468 bytes of inline payload, so no honest LCP figure exists on our side yet and none is asserted. Closing this needs a real browser harness rather than a static ruler.
Research bar. Unreal Pixel Streaming reference is measured on delivering high-fidelity 3D to a thin client by encoding frames server-side. Theirs: full fidelity bounded by encode and network latency rather than by client GPU. Ours: UNPINNED -- not a path we have taken. Recorded as the honest escape hatch for genuinely weak devices and as the reason the native render path must stay the default rather than as a target.
Beach adapter diagnosis, 2026-09-09: MEASURED, NOT FIXED. Same served sand candidate, 1254x627 q1 and 480 stationary free-roam frames after 4/4 visitor load: default Chrome selected Intel xe-lpg, p50 25.005ms p95 45.790ms; dedicated high-performance Chrome selected NVIDIA Blackwell, p50 4.165ms p95 4.205ms. Last GPU samples 24.969216ms and 0.851968ms are individual samples, not averages. Dynamic simulation time differs between captures. No persistent user setting or product shader changed; this is hardware-selection evidence, not a shipped performance fix.
Beach full-run stutter - 2026-09-09: OPEN - operator reports super laggy. Review beach-hqshift-f13272bb-walk-1 recorded3600 frame intervals on NVIDIA Blackwell: p50=4.175ms p95=16.695ms p99=52.010ms worst=162.495ms;191 intervals above16.667ms and38 above50ms. These supersede any impression of smooth delivery from the final256-frame HUD window. Full-run frame tails and stalls must be published alongside rolling metrics; fast median and no GPU error do not imply smooth gameplay. Investigating CPU/simulation/render correlation before assigning cause. The user's active URL/adapter is not yet established.
Beach GPU telemetry overhead comparison - 2026-09-09: STUTTER UNRESOLVED. Same served corrected Beach page bc745117, NVIDIA Blackwell,1254x627,3600 W-held frame intervals per run. Native timing: median7.180ms,p95=26.795ms,p99=39.020ms,max73.930ms,8 intervals above50ms. Timer stopped via existing native resource cleanup in disposable browser: median4.175ms,p95=16.655ms,p99=33.325ms,max116.760ms,10 intervals above50ms. Disabling GPU readback did not remove the stalls. Single sequential pair; not randomized, equivalent simulation duration and scene evolution not established, host load and observer overhead uncontrolled. No causal cost estimate or performance acceptance. Production timing remains enabled.
WebGPU performance programme - research cutoff 2026-09-09: TARGETS, NOT ACHIEVED. Operator requires September2026 state-of-the-art and beyond on Beach. Judge complete frame-time distributions and delivered visual quality at declared native resolution, actual adapter, browser/driver, power mode and scene complexity. Separate cold launch, warmed walking, streaming, character arrival, interactions and recovery. A rolling fast median, renderer submission or generated/interpolated frame count cannot establish delivery performance.
WG1 - Desktop frame pacing: Target ladder: 60 then120 then higher refresh. Proposed Beach qualification profiles: native1920x1080 on verified RTX5080Laptop, p99 at or below16.667ms for60Hz then8.333ms for120Hz, with unchanged declared scene fidelity and physics. Report every deadline miss and all stalls above50ms over a declared10-minute route; startup is reported separately, never silently discarded. Native1440p and higher-refresh profiles remain separately measured stretch targets. These are acceptance targets, not predicted hardware guarantees or current results.
WG2 - VR pacing and latency: Target: headset cadence with explicit evidence. Qualify90Hz at11.111ms or120Hz at8.333ms where the headset/runtime supports it, at declared per-eye render resolution and fidelity. Record missed/reprojected frames separately from newly rendered frames; require actual headset walking and interaction. Input-to-photon latency needs appropriate presentation/hardware evidence; rAF and GPU pass timings are not substitutes. VR remains unverified.
WG3 - Work placement and streaming: Target: no surprise work on interaction. Audit asynchronous pipeline preparation, stable resource reuse, bind/draw batching, render bundles where repeated work warrants them, dirty-region terrain uploads, worker-based decode/conversion and bounded upload admission. Preserve deterministic simulation and explicit interpolation. Every optimization needs equal-quality before/after evidence and fault tests; do not turn off world physics, NPC variety or detail to claim a win. Shared-memory verifier remains a required gate.
WG4 - Explain stalls before tuning: Target: attributable frame gaps. Correlate native simulation/render clocks, GPU queue/pass/readback times, browser Long Animation Frames, GC and layout, shader compilation, upload volume and compositor/scheduling evidence where exposed. Record unsupported/unattributable intervals. LoAF detects frames over50ms and omits some worker/extension attribution, so it complements per-frame records rather than proving high-refresh smoothness alone. Browser API profiling is a diagnostic source; Nishi owns records and comparisons.
WG5 - Current API and feature discipline: Target: portable measured gains. Primary guidance checked2026-09-09: Chrome144 improved writeBuffer/writeTexture; Chrome151-152 documents feature-detected subgroup size control for compute, which is not evidence that Beach fragment rendering becomes faster. Use asynchronous pipeline creation where applicable; benchmark supported subgroups, immediates and transient attachments only for matching workload and correctness. Never require experimental unsafe flags or infer availability from browser version alone. Matched competitive benchmarks required before any exceeds claim.
Research bar. Chrome151-152 August2026 release is measured on Subgroup size control and API updates. Theirs: Feature-gated capability. Ours: WG5; no Beach FPS gain measured.
Research bar. Chrome144 upload changes is measured on writeBuffer and writeTexture implementation improvements. Theirs: Browser implementation optimization. Ours: WG3; measure our upload path.
Research bar. Long Animation Frames is measured on Main-thread animation-frame and script attribution. Theirs: Frames over50ms; presentation not included. Ours: WG4; complement per-frame telemetry.
Research bar. WebGPU specification is measured on Asynchronous pipeline creation, queue/mapping and render-bundle contracts. Theirs: Portable API semantics. Ours: WG3 WG5; performance requires runtime evidence.
Resident character startup hitch - 2026-09-09: REPRODUCED - correction pending. Browser Long Animation Frames identified a606.025ms script callback after Response.arrayBuffer resolved in loadNPC, inside a607.380ms startup long frame. Source location is offset643011 relative to the main inline script (HTML script starts at5380), verified against the served bc745117 document. Resident NXA parsing/checks and geometry preparation execute on the main thread until their next await. Move applicable work through the estate's existing native conversion/worker owner and verify equivalent buffers before rendering; do not omit the resident or weaken validation to hide the hitch.
Startup profile attribution - 2026-09-09: REPRODUCED - worker migration incomplete. Real served bc745117 review with NVIDIA Blackwell reproduced a512.630ms resident loading callback within a541.460ms long frame. CDP sampling through visitor readiness attributed335.621ms self time to loadNPC across startup. Its222 source-position ticks include57 in nearest-body dress binding,97 in hair preparation and40 at synchronous texture getImageData; sample counts are not exact per-stage durations. Existing native geometry conversion alone does not cover these costs. Preserve rig, garment and hair outputs when moving preparation out of the rendering thread. Full walking stutter remains open.
Native garment preparation - 2026-09-09: QUALIFIED OPERATION - Beach integration pending. Native binding passed 19/19 gate checks on the resident asset. All 4,032 resulting joint-and-weight components matched the unchanged served loader reference for 504 dress vertices. The same native code emitted to WebAssembly produced all 504 mapping indices identically in two real Chrome worker runs. Memory was 35,717,120 bytes per instance; isolated call times 64.7 ms then 16.5 ms are diagnostic samples, not frame-rate gains. Beach has not yet switched to this operation. Hair preparation and texture readback remain startup costs.
Beach worker integration - 2026-09-09: RUNNING IN REVIEW - severe stutter remains. Canonical emitted review /releases/beach-garment-808ee6a1.html now prepares 504 dress bindings in the native worker. Real free-roam run completed with four visiting draws and no reported page errors; worker cleanup verified. Full 3,600-frame sample: p50 8.30 ms, p95 50.99 ms, p99 74.625 ms, worst 118.28 ms, 198 intervals above 50 ms on NVIDIA Blackwell. This fails the smoothness target. No controlled performance gain is claimed. Visual defects remain visible.
Beach presentation stalls - 2026-09-09: MEASURED - no accepted fix. Real Chrome trace on the current Beach review found 304 DXGISwapChainImageBacking::Present calls, nine above 50 ms and a worst 90.494 ms on the GPU thread. The 300-frame walking sample had p50 8.295 ms, p95 44.855 ms, p99 68.145 ms and worst 91.385 ms. Rendering time alone cannot establish delivered smoothness. Diagnostic DirectComposition disabling worsened the median to 33.325 ms despite reducing spikes; normal composition is retained. Single sequential pair, trace overhead and host power/load uncalibrated. Precise per-frame clock joins remain open.
Clock-aligned Beach trace - 2026-09-09: MEASURED - presentation association. Follow-up 300-frame run linked all 17 intervals above 50 ms to overlapping presentation calls; 16 overlap a call above 50 ms. Worst frame interval 103.610 ms overlaps a 102.979 ms Present call. Start/end clock brackets intersect within 0.020 ms; this observed bracket is not certified accuracy or photon timing. Temporal association does not identify the driver, scheduling or hardware cause. Instrument overhead remains uncalibrated. No frame-rate improvement claimed.
Native GPU submission batching - 2026-09-09: DELIVERED FOR REVIEW - speedup unproven. Nishi's canonical emitter now resolves and copies GPU timestamps in the render submission, removing the second submission while preserving measurements. Real GPU checks cover occupied slots, busy handling, reuse and named failure cleanup. Public review /releases/beach-timing-4618858f.html completed 3600 windowed walking frames, four visitor draws, no reported page errors and 3909 completed timing samples with zero timing failures. p95 8.335 ms, p99 33.135 ms, worst49.985 ms. The old run used a different viewpoint/route, so no controlled speedup is claimed. Art, warping and sustained smoothness remain unfinished.
Native ownership and optimization references: OPERATOR DIRECTION - not achieved superiority. Emscripten and other tools are references and benchmarks, not requested dependencies. Grow Nishi-owned source/data layouts, compiler, shader and command emission, telemetry and delivery from the first byte upward. Browser WebGPU and GPU drivers remain external execution dependencies today; do not claim their replacement. Evaluate SoA, quantization, fusion, bundles, indirect draws and asynchronous compilation against measured bottlenecks with fidelity/error gates. Require adapter evidence; powerPreference alone cannot force the discrete GPU on Windows. Wasm heap views avoid extra staging allocations, not GPU upload copies. Compare equal routes and explicit browser modes before superiority claims.
Native route and tick evidence - 2026-09-09: MEASURED - mixed performance result. Review receipts now read native camera, orientation, simulation tick and world-origin integers without moving the camera. Equal frame counts covered different world durations, so a new tick-duration condition reports actual elapsed ticks and overshoot. Fresh old/new runs start at matching coordinates and reach the ocean after600/601ticks. Old p95/p99=25.365/44.165ms; new=26.185/39.570ms. One metric worsened while another improved; sustained speedup remains unproven. Start worldtime differs39ticks and hostload is uncontrolled. Early frame-cap refusal was tested against real Beach.
Shared-Wasm reservation evidence - 2026-09-09: QUALIFIED COMPONENT - integration pending. Native layout records now derive initialized and zero-initialized reservations from the WAT compiler's existing allocator, including the initialization state cell. NAS gate passed28/28 on mixed-data, page-boundary, empty-module and refusal fixtures. This closes a missing evidence representation, not the shared-memory safety proof. Compiler/package integration, source and binary identity binding, independent interval verification and real shared-worker tests remain required. Existing pair rejection is retained; Beach remains on its verified review artifact.
First-byte WebGPU audit scope: REFERENCE TRANSLATION - not blanket flags. Audit Nishi toolchain output and SIMD correctness, reflected memory layout and transfer bytes, initialization/async pipelines, command counts/binding reuse, then GPU work and presentation. Current verified environment is Chrome152 on Windows with RTX5080 Laptop GPU plus browser traces, native telemetry and NAS gates; no verified Nsight/RenderDoc capture yet. Emscripten compiler flags are not this build's controls. SoA does not guarantee coalescing; binding-offset alignment differs from member layout; Wasm views do not eliminate GPU copies; adapter hints do not force hardware. Preserve diagnostics and validation, measure their cost, and require numerical/visual checks before aggressive optimization or quantization. Loading every possible variant upfront is not assumed optimal; measure startup and runtime tradeoffs.
Sovereign research lab - 2026-09-09: REQUIRED - experimental programme pending. Turn research into reproducible Nishi experiments from compiler bytes through displayed frames. Record hypothesis, dated primary source and pinned revision, factor levels, interactions, baseline, immutable asset/compiler/backend identities, seed, route, actual adapter/driver, power/load and observer cost. Use randomized repeated runs and factorial screening before focused tuning; retain failures and negative results. Separate exploration from held-out confirmation to avoid selecting noise. Rank Pareto tradeoffs across fidelity, numerical error, physics correctness, latency tails, memory, energy and build/deploy cost; never declare a single universal winner from average FPS. Bit-level candidates include packing, quantization, alignment, SIMD, specialization and cache layouts; every candidate needs boundary tests and equal-quality Beach evidence.
Browser and device research matrix: REQUIRED - Chrome-only evidence insufficient. Qualify Chrome and Edge separately, Firefox, and Safari on real supported Apple hardware, plus mobile and headset profiles when available. Record browser engine/version, OS, driver, actual adapter, supported features and windowed/headless mode. A Playwright WebKit run is not Safari qualification; software rendering is not RTX evidence. Missing hardware and unsupported capabilities remain visible untested states. Keep common scene, input sequence, simulation duration and fidelity fixed; retain per-platform regressions and fallback behaviour. Existing Beach measurements only establish the explicitly recorded Chrome configurations.
Nishi browser and native GPU backend: DIRECTION - implementation and gains unproven. Preserve a backend-neutral Nishi scene/simulation/resource contract and evidence format. Continue interoperable WebGPU delivery while qualifying a future Nishi browser native GPU path through supported OS/driver APIs. Track capability negotiation, resource ownership and synchronization, shader correctness, crash/device-loss recovery, security isolation and presentation latency. Removing WebGPU does not remove GPU drivers or automatically improve speed. Admit native capabilities individually through equivalent Beach scenes and replay/visual/numerical checks; keep export/interoperability and stable APIs. No firmware writes or hardware safety bypasses.
RedCube reference review - 2026-09-09: REFERENCE ONLY - not benchmarked or adopted. Repository README advertises glTF2 rendering through WebGL2 and WebGPU plus headless parsing, with material/texture extensions useful for Nishi ingestion and fidelity comparisons. README discloses global renderer state and one-instance-per-page limitation; Nishi /compare requires independent simultaneous viewports, lifetimes and recovery. Current Playwright config lists only Chromium, so it does not prove cross-browser qualification. Package reports5.1.0; MIT license requires retained notices for copied substantial code. Research read only; no dependency added, code copied or superiority established. Pin source revision and asset licenses before executable comparisons. Sources https://github.com/Reon90/redcube and https://raw.githubusercontent.com/Reon90/redcube/master/playwright.config.ts .
Web3D Survey benchmark reference - 2026-09-09: REVIEWED - no local benchmark run. https://web3dsurvey.com/benchmark links https://github.com/bhouston/webgpu-bench . Upstream documents read/write bandwidth and arithmetic kernels, randomized round-robin sampling, feature skips and GPU/wall-clock checks. Its best-run headline and exclusion of samples slower than a best-relative threshold target peak throughput; those exclusions must not define Beach sustained performance. Nishi research must retain all samples, exclusion reasons and timing source, publish peak and sustained distributions separately, and distinguish suspected throttling from measured thermal evidence. Connect hardware microbenchmarks to equal-quality Beach workloads before assigning a bottleneck or claiming gains. External result pages show browser/device metadata, not proof of Nishi compatibility. No benchmark run or external upload authorized by merely reviewing this reference; default research evidence stays sovereign with user-controlled sharing.
WebGPU capabilities report reference - 2026-09-09: REVIEWED - runtime collection unverified. https://web3dsurvey.com/report/webgpu describes browser features and limits with links to survey comparisons. Retrieved page remains at Collecting capabilities; no local device report was obtained. Nishi needs versioned capability snapshots before benchmark selection: actual adapter identity where exposed, adapter/device features and limits, requested/enabled feature distinction, format and presentation support, timestamp availability and device-loss state. Preserve unavailable or privacy-reduced fields explicitly. Capability presence is not speed or correctness evidence; link snapshot identity to microbenchmark and Beach run receipts. Keep collection local by default with opt-in sharing and retention controls; external collector not executed as part of this review.
Firefox operator benchmark identity ambiguity - 2026-09-09: VERIFIED REPORT - WebGPU adapter unknown. Sovereign HTTPS retrieved https://web3dsurvey.com/benchmark/907df1f4-1337-4cd5-964f-027d2e00aedc (fetch job1788979692). Report records Firefox153.0 Windows10; WebGPU vendor/device/architecture/description all blank. Intel ANGLE identity belongs to the separate WebGL device section. Headline65GB/s and1.25TFLOP/s do not prove WebGPU ran on Intel. Do not attribute WebGL metadata to the compute adapter or infer adapter from throughput. Retain API-specific identity provenance and unknown fields; request actual Firefox diagnostics before a scoped reversible GPU preference change, then verify adapter and equal-workload result. Prior NVIDIA Chrome qualification does not establish this Firefox session's selection.
Waterfox default adapter confirmed - 2026-09-09: OPERATOR DIAGNOSTICS - correction pending. User supplied about:support identifies Waterfox6.7.2 build20260908212929, advertising Firefox153. WebGPU requestAdapter({}) reports Intel(R) Graphics, IntegratedGpu, Dx12, driver32.0.101.8991; RTX5080Laptop is detected but GPU2 Active=No. This confirms default adapter selection, not the adapter returned for a high-performance request or every prior benchmark. The forced fallback probe reports Microsoft CPU rendering and does not mean normal operation uses software rendering. Record actual application identity separately from user-agent inference. Three displays reported at240/60/60Hz and device pixel ratio1.5 need representation in performance profiles; no causal stutter claim from topology alone. Windows per-app high-performance preference for waterfox.exe is the supported next reversible action; restart and actual adapter verification required. No settings changed by this session. Public record intentionally excludes unrelated profile/add-on/device details. Source https://support.microsoft.com/en-us/windows/hardware/display-graphics/optimizations-for-windowed-games-in-windows-11 .
Compiler-attached memory evidence - 2026-09-09: QUALIFIED - pair proof pending. Native compiler --layout-record now attaches the allocator reservation record to WAT in the same source compilation. Real garment-binding compilation with and without record produces identical15247-byte Wasm SHA c788f9964d0452e51eaf068f365fbec503c2b25c5ff2dff761e2ee377d8a319d, also matching the reviewed worker. Shared source compilation emits896B record describing26globals,365pages and8B initialization reservation at23855920. Native comment boundary gate passed10/10 after correcting a one-byte test-fixture prefix miscount; failed run retained. Comment metadata is not authenticated binary provenance. Independent parser, source/toolchain/Wasm binding, BSS interval checks and shared-worker race tests remain open; existing shared pair refusal remains.
Beach graphics preference control - 2026-09-09: PUBLIC REVIEW - cross-browser qualification pending. Reusable Nishi game-page emitter now provides Graphics / Performance with browser default, high-performance and power-saving requests, active WebGPU identity and Windows per-app guidance. Apply explicitly reloads; the preference travels in the URL. Public review /releases/beach-graphics-2543b601.html SHA2543b601ea0381592d254b982d2ffd2f61c83e43ec69a09edcfb5a71f9d3f644,786383B. Real Chrome default and high-performance runs both reported Intel, despite correct requested option; UI correctly did not claim a switch. Unknown identity stays unknown. No website permission to change OS preferences is invented; future installed Nishi controls need verified authority and rollback. Control UI, guidance, reload and120 free-roam frames exercised without reported page/GPU errors. General world rollout and Waterfox/Safari/Firefox interaction qualification remain pending.
Waterfox Windows preference applied - 2026-09-09: SETTING VERIFIED - runtime switch pending. At explicit operator request, the current user's Windows graphics preference for verified C:/Program Files/Waterfox/waterfox.exe was set to GpuPreference=2; and read back identically. Previous value was absent; rollback receipt retained locally at nishi-ops/waterfox-gpu-preference-20260909.json. User reports full restart. No post-restart adapter diagnostics received yet; NVIDIA selection and frame-rate improvement remain unverified. Direct Windows registry boundary used because available sovereign MCP tools do not expose local Windows graphics settings. No firmware/driver modification or process termination.
Independent memory record verification - 2026-09-09: 24/24 COMPONENT GATE - shared proof incomplete. Independent native reader validates record shape/version, count, alignment, exact reservation order, memory extent, full state cell and supported flags without invoking the allocator. Mixed initialized/BSS fixtures and malformed variants pass24/24, including overlapping BSS/state, truncation, overflow and false lengths. Gate ELF52376B SHA8f01c6efe0c1fd0ba77d212b53ed9332d76c368ec304055e8457fc8786367f9f. Consistent forged metadata can still pass structural checks: source/toolchain/Wasm identity binding and binary reservation cross-checks remain mandatory before pair acceptance. No shared-worker execution enabled.
Real compiler records joined to Wasm - 2026-09-09: 12/12 REAL-ARTIFACT CHECKS - execution gate remains closed. Native reader extracts the final NXWLAY01 compiler record from WAT. New join compares plain/shared reservations, declared memory pages, initialized payload destinations/extents and initialization state, alongside existing application-code, interface and section checks. Actual garment-binding plain/shared pair passed12/12; swapped roles, truncated binary and corrupted record were rejected. Supplying real WAT without a record returned RED as intended. A shared variant using the existing67108864-byte garment memory budget also passed12/12. This proves agreement of supplied records and executables, not trusted source/BSS provenance or shared-worker race safety. Production pair acceptance remains unchanged.
Shared garment workers runtime qualification - 2026-09-09: BOUNDED CORRECTNESS PASS - production pair gate remains closed. Real public shared module7a4e60cb, plain modulec788f996 and11728120-byte resident asset were SHA-256 verified in isolated headless Chrome. Two workers over16 synchronized rounds produced identical4032-byte bindings; initialized segments, padding and output guards passed. Late instantiation and READY/IN_PROGRESS initializer calls preserved a mutated live byte. Allocated35717120 bytes within configured67108864 ceiling. Initial test incorrectly expected padding to be copied; retained failed receipt and corrected test to check26 actual declared segments with untouched padding. No runtime/compiler fix was needed. No FPS improvement, exhaustive race proof, BSS/source provenance or Waterfox qualification claimed.
Waterfox Beach identity unavailable after restart - 2026-09-09: USER-OBSERVED UI - selected adapter unknown. Windows high-performance preference readback remains GpuPreference=2;. Operator first reported NVIDIA, then supplied exact Beach panel text: WebGPU is active; this browser does not disclose the adapter identity. Therefore this Beach session's adapter remains unknown; earlier NVIDIA observation does not establish this page's device. UI correctly distinguishes requested preference from undisclosed actual selection. No measured speedup recorded.
Waterfox NVIDIA diagnostic clarification - 2026-09-09: OPERATOR CONFIRMED - webpage identity withheld. Operator clarified that Waterfox about diagnostics show NVIDIA after restart while Beach reports undisclosed adapter identity. Retain NVIDIA as operator-observed browser diagnostic evidence and Beach identity as unavailable via the page API. These observations do not conflict. Exact diagnostic field and automated capture are not retained, and sustained FPS improvement remains unmeasured.
Delivery workflow acceleration 2026-09-09: Measured small-call median 10457.262 ms to 187.170 ms. Three interleaved reads per sovereign client returned identical file-size data. Old client SHA256 a371fe6262978df77184637ee6c228649662074141abae1573ae1cb4a0238eac; newer client92cd943e34528086da9142e96d2e5d64f7de937c1da527f105f54cac3e820d42. About55.9x improvement for this operation, not compiler throughput or Beach FPS. This task now uses the newer client, including verified file arguments and a real source write. Raw local receipt C:/Users/elder/nishi-ops/browser-review/sovereign-client-ab-20260909.json. SOTA remains unproven.
Delivery acceleration next acceptance: Open. Measure edit-to-production-reviewed-URL critical path with per-stage queue, transfer, compile, test, emit, publish and browser-review duration; retain unique job-scoped logs and artifact/closure identities; explain cache misses and reuse unchanged actions; batch independent reads; resume by existing terminal job handles; publish live readiness and failures on existing Compare/release surfaces. Calibrate performance targets from comparable repeated workloads rather than invented fixed ceilings. Every speed variant must retain byte/behavior correctness, privacy and rollback evidence. Nishi implementation stays sovereign.
Research references checked 2026-09-09: Reference designs, not dependencies. OpenTelemetry CI/CD span conventions describe pipeline/task execution: https://opentelemetry.io/docs/specs/semconv/cicd/cicd-spans/ ; Bazel remote caching keys reusable actions from their inputs/commands/environment: https://bazel.build/remote/caching . Study and compare native Nishi capabilities; no framework adoption or superiority claim follows from these references.
Beach continuous-water integration 2026-09-09: Candidate, not delivered. Native UR_WATER_Q8 uses word41 in the existing176-byte frame block. Shader surface mode skips voxel water and intersects the simulation-level continuous plane against the terrain field; voxel mode retains cell water. Native gate run1788985776 passed11/11 including4096 generated-field samples, packed simulation-level equality and next-frame level change. Candidate shader backend SHA25627f33cf205343917d4c10983bc35d4e92761e2a507d9f76cda3ed09ad957fe01,1249472 bytes; repeated build1788985777 reused the byte-identical verified cache artifact. No public renderer integration accepted: matched Wasm emission, GPU compilation, visual/interaction A/B, waves/optics and vegetation separation remain open. Visitor world anchors, camera replay and support/jump/edge bugs remain open.
Build evidence isolation defect 2026-09-09: Open. Shared /tmp/buildrun.out was overwritten by a different target between shader build completion and review. The unrelated log was rejected as shader evidence. Subsequent shader cache run log was read fully and matched nx_wgsl. Persist full logs by job and artifact so concurrent work cannot replace review evidence. Do not treat this one observation as evidence that immutable per-job full build logs already exist.
Latest recorded release
2026-10-01 · PF11
THE LADDER TO SOTA IS DECLARED AND ATTESTED FOR OCTOBER 2026 (cross-board ladder seat, T0 job 1790857487): four sotabar rows (SB-WEBVITALS-2609, SB-FRAMEPACING-2609, SB-UPSCALE-2609, SB-VR90-2608), four sotatarget rows (best-in-class T-BIC-FIELD, T-BIC-PACING, T-BIC-RECON; frontier T-FR-VR90) and thirteen rungrole rows (substrate 5, arm 3, contender 5); performance.refs CREATED with fourteen rows (the board had no refs file): ten mirrors fetched through nx_research_fetch on 2026-10-01 and read from the mirrored bytes, four shared pinned mirrors re-hashed today. nx_ladder performance reads LADDERED (13 of 13 placed, no unclimbed target); nx_barfresh reads FRESH with 4 of 4 ATTESTED by attestfile from this seat's reads (nx_frontier_scan attest answered ATTEST-UNATTESTABLE for all four refs and its fan-out over 68 fetches read newer_sources=0 stale_bars=0); nx_compare_rank performance is not refused.
Release entries describe recorded changes; they do not establish that every capability passed evaluation.
9 of 23 capability source symbols found|3 of them authored stronger-claims|14 open|coverage 391/1000|adoption 6 full / 3 partial
Evidence profile — what the gaps on this board actually are
Measured by nx_swcompare_evidence, read back by nx_evprofile_lib. Every figure is a count with its denominator — there is deliberately no score, no grade and no percentage anywhere in this band, because a stored scalar is a field a seat can edit and a counted partition is not.
evidence|grounded 9/9|unsupported 0|gates green 1/1|proven able to fail 1/1|never bitten 0|green at 0/0 0|open gaps 13|of them unnamed 0|of them proof withheld 0|flips ready 0
proven able to fail counts the gates that have a RECORDED RED — nx_gate_bite mutated the gate subject, rebuilt it, watched the gate go red, and that record is inside the shared TTL. never bitten is its complement over the same denominator: those gates ran and were green, and nothing has ever shown them able to detect anything, so their green is a statement about this run and not about the gate. green at 0/0 is a separate and much weaker observation — the gate printed GREEN on a zero denominator, so its own tooth counter says it examined nothing. A gate can be green, non-zero, and still never bitten; that is the common case and it is now visible instead of implied.
partition: grounded + unsupported = 9 vs present 9 · named + unnamed + withheld = 13 vs open 13 · both reconcile
liar-kill conj=GPQN · all four conjuncts held
graded document: BUILDROOT tree, 24494 bytes · gates map: BUILDROOT · stamped 37d 6h ago · STALE against the shared TTL of 604800s — this is evidence about a build nobody can vouch for now · source ../knowledge/status/evstamp_performance.verdict
| Gap class | What it is, and the work it names |
|---|---|
| STALE | This profile is older than the shared TTL, or its clock ran backwards. It is reported as unusable rather than as fresh, because being wrong in the direction of refusing is the safe direction. |
This band reports the referee counts. The per-axis worklist rows — which axis is unsupported, which watch is ready to flip, which gap is unnamed — are printed by
nx_swcompare_evidence performance itself and are not carried on the stamp, so this page names the classes and the producer names the rows. That split is stated rather than hidden: a count without a worklist is not actionable, and this band is honest about which half of that it is.Production map
Follow the dependencies, declared acceptance criteria and recorded priorities. Inspect source binding before treating a rank as executable work.
Ranking source binding: STALE_OR_INVALID_BINDING. Recorded priorities require current acceptance evidence and resource checks before execution.
Recorded ranking withheld
Retained ranking is not bound to the current plan and matrix: STALE_OR_INVALID_BINDING. Historical data remains retained. Refresh the ranking after resolving the input differences. Latest execution outcome is unverified.
Declared roadmap — contract, acceptance, executor, effort
| Rung | Closes with | Definition of done (pre-declared) | Executor | Est. |
|---|---|---|---|---|
| External cacheable wasm with streaming compilation (PF1) | gpe_wasm_external | the emitter writes the module to a content-hashed file served immutable and the page loads it with instantiateStreaming and an arrayBuffer fallback AND nx_wasm_vm_verify and nx_wasmpage_census follow the external reference instead of anchoring on the base64 magic AND a second world entry transfers the page without the module | seat | 2 u |
| Continuous device re-measurement under thermal throttle (PF2) | wc_budget_remeasure | the frame budget is re-derived on a rolling window and on visibilitychange and resize instead of latching at frame 45 AND a gate proves a budget captured at one interval is replaced after the interval changes | seat | 1 u |
| Dynamic Huffman blocks in the sovereign DEFLATE (PF3) | gz_dh_lens | gz_deflate emits BTYPE=10 with a length-limited canonical code and the code-length alphabet RLE and chooses the smaller of fixed and dynamic per block AND every output round-trips byte-identical AND the measured gain on the shipped module is at least 8 percent. SCOPE MEASURED 2026-08-28 by a check-before-build pass rather than estimated, and it is SMALLER than a from-scratch encoder. ALREADY EXISTS, on the DECODE side: nx_deflate.nx implements all three BTYPE handlers end-to-end and carries _deflate_cl_order which IS the HCLEN permutation, plus the RFC length and distance base and extra tables and _deflate_build_dynamic_trees which already reads the code-length alphabet and the 16 17 18 RLE. nx_huffman.nx carries a canonical table builder. MUST BE WRITTEN, and this is the whole of it: frequencies to code lengths, which is the Huffman tree build and exists NOWHERE in the corpus, plus 15-bit length limiting, the ENCODE direction of the code-length RLE, symbol buffering so the emitter can make two passes instead of emitting inline, and the fixed-versus-dynamic per-block cost comparison. ⚠THE EXISTING PIECES ARE REFERENCE, NOT COMPOSABLE: nx_deflate.nx and nx_huffman.nx use nx_int and nx_size and structs and import nx_runtime and nx_tier and nx_bitstream, while nx_gzip_lib.nx is plain i64 and *u8 with no runtime -- importing across that boundary would drag a type system into the compressor that serves every response. Copy the CL order VALUES, not the code. ⛔PRECONDITION, AND IT IS NOT OPTIONAL: DECIDE THE EMITTER CONSOLIDATION FIRST. nx_deflate_enc.nx is a SECOND self-declared fixed-Huffman DEFLATE compressor. Writing dynamic Huffman into one copy while another remains deepens exactly the duplicate-source split that cost lazy matching a double write and a double proof on 2026-08-28. ⚠COUNT HONESTLY BOUNDED: a sibling investigation reported FIVE fixed-Huffman emitters and its verification agents died before refuting it; a literal grep for the phrase over the whole runtime at corpus_complete=1 finds TWO self-declared. Those are different membership tests, so the true number is between 2 and 5 and is NOT pinned -- pin it before consolidating | seat | 3 u |
| Temporal reconstruction from the existing motion-vector primitive (PF4) after PF2 | gpe_temporal_reproject | the renderer keeps a history buffer and reprojects it with the same world-point reprojection the optical-flow lane already verified AND a reconstructed frame at a quarter of the ray count scores closer to native than the current spatial upscale on a declared image metric | seat | 4 u |
| Quarter-resolution volumetrics with a temporal pattern (PF5) after PF4 | wc_cloud_quarterres | the cloud pass renders to its own quarter-resolution target with a temporal update pattern and composites AND frame time falls with no verdict change from the panel judge | seat | 2 u |
| Mip-mapped voxel cone stepping (PF6) | wc_voxel_mip | a mip pyramid is built over the voxel volume at world commit and vmarch selects a level from cone footprint AND the DDA step count per camera ray falls while the rendered frame stays within the declared difference bound | seat | 3 u |
| Foveated and variable-rate raymarching (PF7) after PF4 | wc_foveate | ray density follows an importance field with a reconstruction pass AND the 90Hz VR verdict is met on a device that fails it today | seat | 3 u |
| A per-frame media budget for in-browser calls (PF8) | pf_rtc_budget | encode and transport cost per frame is published by the same instrument that publishes frame_ms so a call and a world are judged against one clock | seat | 2 u |
| Device-tiered mesh and texture delivery (PF9) | pf_mesh_tier | one supercompressed asset is transcoded to the device's native GPU format and the tier chosen is reported AND a tier the device cannot decode is refused rather than shipped | seat | 3 u |
| A real browser field harness for this board (PF10) | pf_field_harness | a real browser run produces LCP and INP and a frame-time distribution for a published world page so the Core Web Vitals bar stops reading UNPINNED AND the harness works on the browser the operator actually mandated. ENGINE CONSTRAINT, named 2026-08-28 because it invalidates the obvious implementation: the operator has directed Waterfox and directed the workstreams off Edge. Waterfox is GECKO and CDP does not exist there, while this estate's existing capture wire IS CDP -- the emitted page carries its own measurement of that wire, that requestAnimationFrame NEVER fires under it, rAF count 0 over 120 to 400 ms across five loads, which is why a one-shot frame door had to be emitted inside the closure at all. So a CDP harness would be built on a wire the mandated browser does not have. The cross-engine answer is WebDriver BiDi, the W3C successor Gecko implements, or in-page instrumentation which is engine-agnostic by construction and is the half enginelab's overlay already covers. Whichever is chosen must be NAMED in the row rather than assumed | seat | 2 u |
| Frame cost and the quality it bought emitted as ONE record on ONE clock (PF11) after enginelab:EL7 | pf_quality_cost_record | the record defined by enginelab EL7 reaches this board so a performance claim can be read on both axes AND no rung here may be reported on one axis alone once it exists. DELIBERATELY NOT A SECOND RULER: EL7's binding second clause already requires the overlay to publish frame time and image quality on ONE clock, so this rung CONSUMES that record and defines only the reporting rule for this board. Two guards for one invariant is the duplicate-ruler defect and enginelab owns the instrument | seat with enginelab | 2 u |
| Quality becomes an INPUT to the quality controller (PF12) after PF11 | wc_quality_term | the controller decision reads an IMAGE-QUALITY term expressed in the RENDER units defined by PF11 rather than frame time alone AND a gate proves a change that improves frame time while degrading the image is REFUSED where today it is rewarded. SCOPE CORRECTED 2026-08-28 and this rung is a DESIGN gap not a wiring gap: the first draft said sourced from nx_quality_balance and nx_quality_metric. nx_quality_balance measures MODEL-WEIGHT fidelity in perplexity-permil and bits-per-weight and its consumer is the quantisation builder -- a different quantity that shares the word quality, with no conversion to render units anywhere in the corpus. nx_quality_metric is SSIM and PSNR which are FULL-REFERENCE and need a ground-truth frame the live renderer does not have. So neither drops in. PF11 is what DEFINES the missing unit, which is why this depends on it. THIS RUNG MUST NOT CLAIM TO ENFORCE THE EXISTING GOVERNED QUALITY FLOOR: that floor is a perplexity bound and cannot be evaluated here, and claiming to enforce a floor you cannot evaluate is worse than not enforcing one | seat | 4 u |
Milestones
| Milestone | Rungs | Cumulative |
|---|---|---|
| M0 · Nothing is reported on one axis -- the cheapest win taken and the pairing that makes every later claim honest | PF2,PF11 | 3 u |
| M1 · The module is fetched once and compiled while it downloads and quality can refuse a speed win | PF1,PF3,PF10,PF12 | 9 u |
| M2 · Resolution is traded for reconstruction rather than for blur | PF4,PF5,PF6 | 18 u |
| M3 · Every deliverable in the browser is judged against one clock | PF7,PF8,PF9 | 26 u |
| Rung | Closes with | Definition of done (pre-declared) | Executor | Est. |
|---|---|---|---|---|
| Simulation off the render thread over the isolation the page already pays for (PF13) | pf_worker_sim | the world tick runs in a Worker over a SharedArrayBuffer view of the engine arena and the frame loop only reads it -- the performance.matrix row that ABSENT-PROVED zero Workers across 404 world pages on 2026-08-28 now has a rung -- so the GPU thread is never delayed by economy or NPC routines; done when the served beach reads crossOriginIsolated true, constructs the worker from an emitted blob, p95 frame time at the same resolution falls with the tick moved, and the frame digest stays bit-identical to the single-thread frame (the GE27 tooth reused, never a second ruler). BLOCKED BY NAME 2026-09-03 on lang LN33: the sovereign compiler emits no wasm threads (atomic.rmw ABSENT-PROVEN over 23506 sources), so the isolation the router pays for on every response is spent by nothing until that rung lands | seat | 2 u |
Ladder to SOTA
Two dated targets, and every rung's role toward them. Best in class is the proven, deployed leader today; the frontier is the research edge as of its month. Each target is dated by a sotabar row and judged against the current month by the bar-age ruler (FRESH, ATTESTED with the fallback month named, or refused). Each rung is substrate (what the target's own methods consume), an arm (a sovereign arm that triangulates without competing for the number), a contender (its done-rule is a number against the target) or superseded (kept as capability, retired as a claim), and carries the rungs it stands on.
best-in-class (T-BIC-FIELD), dated by SB-WEBVITALS-2609 = 2026-09 ATTESTED. A world page judged in the field at the 75th percentile: Largest Contentful Paint within 2.5 s, Interaction to Next Paint at or under 200 ms, Cumulative Layout Shift at or under 0.1, measured by a real browser harness and not by a static predictor ref webvitals_releases_202610
best-in-class (T-BIC-PACING), dated by SB-FRAMEPACING-2609 = 2026-09 ATTESTED. Frame delivery reported the way the reference instrument reports it: displayed against presented, dropped frames, animation error and every stall attributed to its cause, as percentiles over a whole declared run on one clock ref presentmon_releases_202610
best-in-class (T-BIC-RECON), dated by SB-UPSCALE-2609 = 2026-09 ATTESTED. Resolution traded for reconstruction rather than blur: a temporal reconstruction pass that scores closer to native than a spatial upscale at a fraction of the rays, with any generated frame counted apart from rendered ones ref dlss_releases_202610
frontier (T-FR-VR90), dated by SB-VR90-2608 = 2026-08 ATTESTED. Stereo at ninety hertz inside an 11.1 ms frame budget at native panel resolution with a foveated periphery, sustained from a browser-delivered renderer with no reprojected frame counted as rendered: the budget gameengine GE82 spends ref meta_refresh_rate_202610
| Rung | Role | Toward | Stands on | Why |
|---|---|---|---|---|
| PF1 | substrate | T-BIC-FIELD | - | the module fetched once and compiled while it downloads: the bytes and the compile the first paint waits on |
| PF2 | substrate | T-BIC-PACING | - | a budget re-measured continuously; every pacing judgement on this board stands on it |
| PF3 | substrate | T-BIC-FIELD | - | transfer bytes; shipped 2026-08-29 |
| PF4 | contender | T-BIC-RECON | PF2 | its done-rule is a number: a reconstructed frame at a quarter of the ray count scores closer to native than the spatial upscale |
| PF5 | contender | T-BIC-RECON | PF4 | its done-rule is frame time falling with no verdict change from the panel judge, on the temporal pattern PF4 lands |
| PF6 | substrate | T-FR-VR90 | - | fewer march steps per ray is what the headset budget is paid from |
| PF7 | contender | T-FR-VR90 | PF4 | its done-rule is the 90 Hz VR verdict met on a device that fails it today |
| PF8 | arm | T-BIC-PACING | - | media cost on the same clock as frame_ms: a second witness of the frame budget, not a competitor for it |
| PF9 | substrate | T-BIC-FIELD | - | device-tiered delivery: the bytes a weak device must not be sent |
| PF10 | contender | T-BIC-FIELD | - | its done-rule is LCP and INP from a real browser run, the numbers the threshold is read against |
| PF11 | arm | T-BIC-PACING | enginelab:EL7 DONE | cost and quality on one record: the referee instrument every other rung reports through |
| PF12 | arm | T-BIC-RECON | PF11 | quality as a controller input: the guard that refuses a speed win which degrades the image |
| PF13 | contender | T-BIC-PACING | - | its done-rule is p95 frame time falling with the tick off the render thread and the frame digest bit-identical |
Ladder verdict. LADDERED. Targets, rung roles and bars are counted in the tables below, each partition re-derived beside its total. Bars read FRESH against the current month 2026-10.
Ladder verdict, counted (17 rows, more than the 12-row disclosure bar)
| Class | Count | Note |
|---|---|---|
| targets | 4 | sotatarget rows on the plan |
| best in class | 3 | targets whose class is best-in-class, by code |
| frontier | 1 | targets whose class is frontier, by code |
| undated | 0 | targets naming no bar that exists |
| bad class | 0 | targets whose class is neither word |
| duplicate | 0 | target ids an earlier row already carries |
| rungs | 13 | rung rows on the plan |
| placed | 13 | rungs carrying a rungrole row |
| unplaced | 0 | rungs with no role toward any target |
| bad role | 0 | rungrole rows naming a role the ladder does not know |
| bad target | 0 | rungrole rows naming a target no sotatarget row declares |
| rungs, summed | 13 | RECONCILED: the parts sum to the total 13 |
| substrate | 5 | what the target's own methods consume |
| arm | 3 | a sovereign arm that triangulates without competing for the number |
| contender | 5 | its done-rule is a number against the target |
| superseded | 0 | kept as capability, retired as a claim |
| placed, summed | 13 | RECONCILED: the parts sum to the total 13 |
| Class | Count | Note |
|---|---|---|
| bars | 4 | sotabar rows on the plan |
| fresh | 0 | the bar month is the current month |
| attested | 4 | an older bar with a this-month barscan reporting no newer entry |
| stale | 0 | a this-month scan saw a newer month the bar ignores |
| unattested | 0 | nobody looked this month |
| malformed | 0 | a sotabar row missing a field |
| duplicate | 0 | bar ids an earlier row already carries |
| bars, summed | 4 | RECONCILED: the parts sum to the total 4 |
| current month | 2026-10 | the month every bar is judged against |
Treatment verdict. NOT DECLARED. No rung on this board declares its three-step treatment (a treat row naming the MCP row, the API route, the primitive it composes, the workflow that composes it and the agent beat that runs it); the ranker labels every rung TREATMENT-UNDECLARED until it does.
Inspect a rung and its prerequisites
Declared nodes 13. Rank input binding: STALE_OR_INVALID_BINDING. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.
Plan SHA-256 6cd46202b8b441aabceab0fe46829455f3b00395d7e67c3cb8ad1a5d6e9b045b. Target rows 4; role rows 13. Existing risks and release worklog retain their own scope; no node completion is inferred.
Runtime observation coverage: 0 of 13 rungs have current, identity-bound execution observations. Observation may report RED. Accepted completion remains unknown; source presence, qualification and deployment do not establish it.
Use Enter or Space on a rung to inspect its contract. Prerequisite links locate another rung in this list; open its summary to inspect it. Estimates are authored effort, not forecasts.
PF1 — External cacheable wasm with streaming compilation
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gpe_wasm_externalAcceptance: the emitter writes the module to a content-hashed file served immutable and the page loads it with instantiateStreaming and an arrayBuffer fallback AND nx_wasm_vm_verify and nx_wasmpage_census follow the external reference instead of anchoring on the base64 magic AND a second world entry transfers the page without the module
Authored effort: 2. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF2 — Continuous device re-measurement under thermal throttle
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
wc_budget_remeasureAcceptance: the frame budget is re-derived on a rolling window and on visibilitychange and resize instead of latching at frame 45 AND a gate proves a budget captured at one interval is replaced after the interval changes
Authored effort: 1. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF3 — Dynamic Huffman blocks in the sovereign DEFLATE
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gz_dh_lensAcceptance: gz_deflate emits BTYPE=10 with a length-limited canonical code and the code-length alphabet RLE and chooses the smaller of fixed and dynamic per block AND every output round-trips byte-identical AND the measured gain on the shipped module is at least 8 percent. SCOPE MEASURED 2026-08-28 by a check-before-build pass rather than estimated, and it is SMALLER than a from-scratch encoder. ALREADY EXISTS, on the DECODE side: nx_deflate.nx implements all three BTYPE handlers end-to-end and carries _deflate_cl_order which IS the HCLEN permutation, plus the RFC length and distance base and extra tables and _deflate_build_dynamic_trees which already reads the code-length alphabet and the 16 17 18 RLE. nx_huffman.nx carries a canonical table builder. MUST BE WRITTEN, and this is the whole of it: frequencies to code lengths, which is the Huffman tree build and exists NOWHERE in the corpus, plus 15-bit length limiting, the ENCODE direction of the code-length RLE, symbol buffering so the emitter can make two passes instead of emitting inline, and the fixed-versus-dynamic per-block cost comparison. ⚠THE EXISTING PIECES ARE REFERENCE, NOT COMPOSABLE: nx_deflate.nx and nx_huffman.nx use nx_int and nx_size and structs and import nx_runtime and nx_tier and nx_bitstream, while nx_gzip_lib.nx is plain i64 and *u8 with no runtime -- importing across that boundary would drag a type system into the compressor that serves every response. Copy the CL order VALUES, not the code. ⛔PRECONDITION, AND IT IS NOT OPTIONAL: DECIDE THE EMITTER CONSOLIDATION FIRST. nx_deflate_enc.nx is a SECOND self-declared fixed-Huffman DEFLATE compressor. Writing dynamic Huffman into one copy while another remains deepens exactly the duplicate-source split that cost lazy matching a double write and a double proof on 2026-08-28. ⚠COUNT HONESTLY BOUNDED: a sibling investigation reported FIVE fixed-Huffman emitters and its verification agents died before refuting it; a literal grep for the phrase over the whole runtime at corpus_complete=1 finds TWO self-declared. Those are different membership tests, so the true number is between 2 and 5 and is NOT pinned -- pin it before consolidating
Authored effort: 3. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF4 — Temporal reconstruction from the existing motion-vector primitive
Prerequisites: PF2 (acceptance unverified)
Contract:
gpe_temporal_reprojectAcceptance: the renderer keeps a history buffer and reprojects it with the same world-point reprojection the optical-flow lane already verified AND a reconstructed frame at a quarter of the ray count scores closer to native than the current spatial upscale on a declared image metric
Authored effort: 4. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF5 — Quarter-resolution volumetrics with a temporal pattern
Prerequisites: PF4 (acceptance unverified)
Contract:
wc_cloud_quarterresAcceptance: the cloud pass renders to its own quarter-resolution target with a temporal update pattern and composites AND frame time falls with no verdict change from the panel judge
Authored effort: 2. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF6 — Mip-mapped voxel cone stepping
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
wc_voxel_mipAcceptance: a mip pyramid is built over the voxel volume at world commit and vmarch selects a level from cone footprint AND the DDA step count per camera ray falls while the rendered frame stays within the declared difference bound
Authored effort: 3. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF7 — Foveated and variable-rate raymarching
Prerequisites: PF4 (acceptance unverified)
Contract:
wc_foveateAcceptance: ray density follows an importance field with a reconstruction pass AND the 90Hz VR verdict is met on a device that fails it today
Authored effort: 3. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF8 — A per-frame media budget for in-browser calls
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
pf_rtc_budgetAcceptance: encode and transport cost per frame is published by the same instrument that publishes frame_ms so a call and a world are judged against one clock
Authored effort: 2. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF9 — Device-tiered mesh and texture delivery
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
pf_mesh_tierAcceptance: one supercompressed asset is transcoded to the device's native GPU format and the tier chosen is reported AND a tier the device cannot decode is refused rather than shipped
Authored effort: 3. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF10 — A real browser field harness for this board
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
pf_field_harnessAcceptance: a real browser run produces LCP and INP and a frame-time distribution for a published world page so the Core Web Vitals bar stops reading UNPINNED AND the harness works on the browser the operator actually mandated. ENGINE CONSTRAINT, named 2026-08-28 because it invalidates the obvious implementation: the operator has directed Waterfox and directed the workstreams off Edge. Waterfox is GECKO and CDP does not exist there, while this estate's existing capture wire IS CDP -- the emitted page carries its own measurement of that wire, that requestAnimationFrame NEVER fires under it, rAF count 0 over 120 to 400 ms across five loads, which is why a one-shot frame door had to be emitted inside the closure at all. So a CDP harness would be built on a wire the mandated browser does not have. The cross-engine answer is WebDriver BiDi, the W3C successor Gecko implements, or in-page instrumentation which is engine-agnostic by construction and is the half enginelab's overlay already covers. Whichever is chosen must be NAMED in the row rather than assumed
Authored effort: 2. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF11 — Frame cost and the quality it bought emitted as ONE record on ONE clock
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
pf_quality_cost_recordAcceptance: the record defined by enginelab EL7 reaches this board so a performance claim can be read on both axes AND no rung here may be reported on one axis alone once it exists. DELIBERATELY NOT A SECOND RULER: EL7's binding second clause already requires the overlay to publish frame time and image quality on ONE clock, so this rung CONSUMES that record and defines only the reporting rule for this board. Two guards for one invariant is the duplicate-ruler defect and enginelab owns the instrument
Authored effort: 2. Executor kind: seat with enginelab. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF12 — Quality becomes an INPUT to the quality controller
Prerequisites: PF11 (acceptance unverified)
Contract:
wc_quality_termAcceptance: the controller decision reads an IMAGE-QUALITY term expressed in the RENDER units defined by PF11 rather than frame time alone AND a gate proves a change that improves frame time while degrading the image is REFUSED where today it is rewarded. SCOPE CORRECTED 2026-08-28 and this rung is a DESIGN gap not a wiring gap: the first draft said sourced from nx_quality_balance and nx_quality_metric. nx_quality_balance measures MODEL-WEIGHT fidelity in perplexity-permil and bits-per-weight and its consumer is the quantisation builder -- a different quantity that shares the word quality, with no conversion to render units anywhere in the corpus. nx_quality_metric is SSIM and PSNR which are FULL-REFERENCE and need a ground-truth frame the live renderer does not have. So neither drops in. PF11 is what DEFINES the missing unit, which is why this depends on it. THIS RUNG MUST NOT CLAIM TO ENFORCE THE EXISTING GOVERNED QUALITY FLOOR: that floor is a perplexity bound and cannot be evaluated here, and claiming to enforce a floor you cannot evaluate is worse than not enforcing one
Authored effort: 4. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.PF13 — Simulation off the render thread over the isolation the page already pays for
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
pf_worker_simAcceptance: the world tick runs in a Worker over a SharedArrayBuffer view of the engine arena and the frame loop only reads it -- the performance.matrix row that ABSENT-PROVED zero Workers across 404 world pages on 2026-08-28 now has a rung -- so the GPU thread is never delayed by economy or NPC routines; done when the served beach reads crossOriginIsolated true, constructs the worker from an emitted blob, p95 frame time at the same resolution falls with the tick moved, and the frame digest stays bit-identical to the single-thread frame (the GE27 tooth reused, never a second ruler). BLOCKED BY NAME 2026-09-03 on lang LN33: the sovereign compiler emits no wasm threads (atomic.rmw ABSENT-PROVEN over 23506 sources), so the isolation the router pays for on every response is spent by nothing until that rung lands
Authored effort: 2. Executor kind: seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
Learning and practice paths
No structured learning path is declared for this plan. Existing research, roadmap and worklog remain available above.
Rung dossiers — every record that bears on a rung, in one place
Joined by symbol, never typed. Each dossier is assembled on every publish from the rows this board already carries: the plan's rung row, the matrix row that carries its contract symbol (its status re-measured by the one symbol ruler), the gate the organ declares and the verdict the evidence referee recorded on its last run, the reference marks in the rung text, the design rows that name the rung and the work log filed against it. The capability matrix links each cell to its dossier and each dossier links back to its cell and forward to the rungs it stands on; a dependency naming no rung on this plan is written as UNRESOLVED rather than linked. Absences are named: NO-MATRIX-ROW, NO-GATE-DECLARED, UNRECORDED.
Gate verdicts: UNSTAMPED — no per-gate verdict artifact is published for this domain yet (
nx_swcompare_evidence performance writes it beside its stamp); every gate below reads UNRECORDED until it runs.Gate declarations:
knowledge/organ_gate.conf read from this working directory (1 rows) and from the estate root (196 rows); rows in both counted once: 0; declarations read: 197. Every declaration in either copy counts, so a stub beside the publisher can never hide the root.PF1 External cacheable wasm with streaming compilation · cell WATCHING · gate NO-GATE-DECLARED
| Contract | gpe_wasm_external · executor seat · 2 u · stands on nothing |
|---|---|
| Definition of done | the emitter writes the module to a content-hashed file served immutable and the page loads it with instantiateStreaming and an arrayBuffer fallback AND nx_wasm_vm_verify and nx_wasmpage_census follow the external reference instead of anchoring on the base64 magic AND a second world entry transfers the page without the module |
| Matrix cell | External cacheable wasm with streaming compilation · _ABSENT_:gpe_wasm_external · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:gpe_wasm_external; declare _ABSENT_:gpe_wasm_external TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-09 measure 2026-09-09 review.cjs now validates gpu_timing_mode=native or disabled-for-diagnostic and records native timer.stop intervention before/after state. Timer-off run confirmed stopped=true,inflight0,cancelled3, no cleanup errors; both dedicated browsers closed without reported capture failures. Receipts nishi-ops/browser-review/beach-timing-native-20260909-1 and beach-timing-disabled-20260909-1. These are explicit diagnostic counterfactuals, not shipped feature removal. GPU readback alone is insufficient explanation; broader scheduling/compositor/GPU attribution and representative user-browser reproduction remain open. 2026-09-09 measure 2026-09-09 beach-longframes-20260909-1:3600 W-held intervals p99=24.995ms,max37.505ms on this run. Bounded Long Animation Frame observer retained9 records,dropped0, callback accounting0.545ms (not total observer cost). Eight ended before walking began. Remaining132.175ms record starts27197.495ms and straddles sampling endpoint27201.195ms; its128.140ms unattributed FrameRequestCallback starts27201.450ms AFTER that endpoint. Do not classify it as a reproduced in-walk application stall or dismiss it as proven browser overhead. Evidence capture/serialization is a confound. Contexts closed; startup hitch verified, sustained user lag still open. Primary API scope reference https://developer.chrome.com/docs/web-platform/long-animation-frames . 2026-09-09 measure 2026-09-09 beach-startup-profile-20260909-1 captured real served document, startup.cpuprofile, bounded LoAF records and600 free-roam frames; no reported page failures; browser closed. Profile begins before navigation and stops at requested visitor readiness; sampling overhead uncalibrated, not an FPS acceptance run. Local receipt C:/Users/elder/nishi-ops/browser-review/beach-startup-profile-20260909-1/receipt.json retains artifact hashes. Browser adapter now accepts explicit profile_startup and uses caller evidence/deadline budgets. LoAF scope corrected: retain runtime-provided presentation fields where exposed; these are not an input-to-photon measurement. 2026-09-09 land 2026-09-09 Performance board delivery: September WebGPU targets and retained startup evidence published through scoped sovereign generator; public nishifamily.com/compare/performance returned200 and new startup attribution card was visually reviewed. Local review tooling now captures optional startup CPU profiles; real served Beach run and cleanup verified. This is a telemetry/record delivery, not a Beach frame-rate fix. Worker migration, sustained-walk stutter, competitive benchmarks and structured API parity for these narrative targets remain unfinished. LoAF research-bar correction: current browser records may expose presentation fields; retain actual supported fields, without treating them as measured input-to-photon latency. 2026-09-09 land 2026-09-09 Native garment-binding preparation and isolated browser-worker qualification added; corrected review module https://nishifamily.com/releases/nxa-garment-bind-c788f996.wasm is 15,247 bytes, SHA c788f9964d0452e51eaf068f365fbec503c2b25c5ff2dff761e2ee377d8a319d. Exact public module and resident asset verified in Chrome; native reference, original loader output and worker results agree. Earlier e62fd282 review module failed because the general compiler predates the staged string-data fix; retained as failed evidence. No compiler promotion, shared-memory verifier bypass or Beach FPS improvement claimed. Canonical loader integration and startup/walking A/B remain required; full goal stays open. 2026-09-09 land 2026-09-09 Scoped Beach review delivery at https://nishifamily.com/releases/beach-garment-808ee6a1.html?gpu=1&hud=1&w=beach&review=0 ; canonical native emitter, byte identity, actual free-roam and start/end images verified. Native garment preparation runs off-thread, preserving existing weight composition. First ab7bbcae review failed on blob-worker relative URL; owner corrected, failed evidence retained. Review tool now captures named character failure promptly, verified against failed page. Follow-up startup capture retained all 87 long-frame records: 244.315 ms loader continuation and 106.295 ms texture-related continuation remain, plus unattributed stalls. Worker integration is delivered for review; overall stutter, art quality, shared-WASM proof and general deployment remain unfinished. 2026-09-09 measure 2026-09-09 Presentation evidence retained in master worklog and local hashed receipts beach-gpu-scheduling-20260909-1 and beach-composition-diagnostic-20260909-1. Both captured current 808ee6a1 document, 300 W-held frames, no page failures and closed browser cleanup. Trace SHA 415522f8095629f001c93beb295fb5d42729af5ac710262ffa870b393a9b1c30. Optional browser backend inventory and isolated composition diagnostic now available in review adapter. No persistent hardware settings changed; no frame-rate fix, equal elapsed route, exact user-browser identity or competitive superiority claimed. Next work: join delivered intervals to GPU/compositor scheduling evidence and retain unsupported telemetry explicitly. 2026-09-09 land 2026-09-09 Diagnostic trace clock anchors added to the existing review adapter and verified in real Chrome against public Beach 808ee6a1, 300 W-held frames, failures=[] and cleanup=closed. Retained trace SHA e25e9bd7f8b39553d784d58b46d3459741c4c1c44fc6da8bb8510f475747dc22 and receipt SHA 3d115ba8d2b30030e3b603c8f5f63901cc65be0515ddb5e041653ea2483e013e. Existing presentation board was published and visually checked on nishifamily.com. This delivers diagnostic evidence, not a smoothness fix. Severe stutter, presentation root cause, startup work, NPC warping, photorealism and structured API parity remain unfinished. Local analysis: beach-presentation-clock-20260909-1/presentation-clock-analysis.json. 2026-09-09 land 2026-09-09 Native timing batching review served and walked at https://nishifamily.com/releases/beach-timing-4618858f.html?gpu=1&hud=1&w=beach&review=0 SHA4618858f003f0469da2cb674dc149182acd6c8be0d1762066a6597c87286ff15,782424B. Canonical build and real GPU qualification passed; actual public document identity, four visitor submissions, timing completion and start/end images reviewed. Windowed/headless mode now explicit in review receipts. Changed viewpoints invalidate a causal performance comparison; zero stalls above50ms in this sample does not meet sustained SOTA acceptance. No Emscripten dependency, general compiler promotion or shared-memory verifier bypass. Next: reproducible routes, submission/upload attribution and measured quality-preserving improvements. 2026-09-09 land 2026-09-09 Review adapter native-state observations and simulation-duration stopping verified on both public Beach releases. Positive runs captured with no page errors and closed browsers; one-frame negative control correctly reports simulation-window-incomplete. Endpoint screenshots reviewed: both ocean route, sky/ocean fidelity still unfinished. Evidence retained in beach-tick-route-pair-20260909.json and master contract. This is stronger comparison instrumentation, not a completed performance fix or forced camera. Old/new documents contain identical worldWasm and shader bytes. Native slot batching remains in4618858f; severe tails, wider-route stability, NPC warp, startup and shared-WASM proof remain open. 2026-09-09 land 2026-09-09 Native reservation-record component and gate staged and executed through sovereign APIs. nx_wasm_layout_gate_review returned28/28, ELF48952B SHA66136b4417e10fa1305468309f8f436ef264efe8d32bec068a3fc46919a0c7c3. Full build and gate output reviewed. Qualification uses representative compiler IR; real Beach packaging/pair acceptance remain unfinished. No compiler promotion, shared verifier bypass, Emscripten addition or frame-rate gain claimed. Reusable native memory ownership proof is required before further shared off-thread preparation. 2026-09-09 measure 2026-09-09 Operator research direction retained: multivariate experiments beyond published recipes, browsers beyond Chrome, and eventual Nishi-browser native GPU capabilities. RedCube README, package, license and test configuration inspected. These are research requirements and source findings, not implemented experiment automation, cross-browser passes or a Beach performance fix. Beach remains the delivery pilot; API parity and public visual review of these new board records remain pending. 2026-09-09 land 2026-09-09 Compiler measurement component qualified through sovereign build/run/hash APIs. nx_compile_wat staged759488B SHA36018e9d6240d6868ced3fd6c940576ea5b50337c2028f61e77c419026c78d1e; gate47240B SHA f628e20be2be3bfd56e4c895b576de993418147d051091eb0e5fce300d3d274f,10/10 run1788979868. Full logs reviewed; real module byte identity checked. Measurement and browser research records prepared for existing /compare/performance surface. No compiler promotion or Beach visual/FPS improvement claimed. Adapter preference change and public visual review pending. 2026-09-09 land 2026-09-09 Graphics control delivered on scoped Beach review and visually inspected at nishifamily.com. Native emitter build1788980345, SHA10acd927177c6e7afe3c04e548283ae5fd8a18915d24b3fb55f2d009e586c292,707976B; publish1788980437 verified via actual document hash and interaction. Receipt beach-gpu-choice-20260909.json retains requested/actual adapter distinction and closed-browser cleanup. Same Chrome chose Intel for both preference modes; no speedup claim. Shared verifier component24/24 run1788980250; complete proof pending. Existing Beach art, sustained stutter, NPC warping and broad production rollout remain unfinished. 2026-09-09 land 2026-09-09 Native real-artifact layout join built102040B SHA9ebe6ef93ef874f09c34d9da81fe264e0b17835a509075c5497ae49e1ab81e63; full build output reviewed. Gate runs1788980827 and1788980960 passed12/12 on actual compiled garment artifacts; missing-record control1788980887 returned RED. Budgeted shared Wasm SHA7a4e60cbdd202d599f3435a29c924b079cb3de82411a1f1f692b38d9941493c1 remains staged on NAS for runtime qualification. No Beach rendering or FPS change claimed. Next required evidence: shared workers, initialized-data persistence, output equality and provenance binding. Existing graphics-choice review remains the evaluation URL. 2026-09-09 measure 2026-09-09 Shared worker runtime receipt shared-garment-runtime-20260909-v2.json passed with no errors and browser closed. Output SHAe77e70ef0d6ed5be9eb3a3e5aaae3505097fe86c5df9113719155e70ce9d2e71. First failed receipt retained as test-instrument defect evidence. Shared artifact is public test-only; existing Beach worker configuration and general pair acceptance unchanged. Next: source/BSS provenance and state ownership before production shared execution; sustained GPU-specific Beach frame pacing remains open. |
PF2 Continuous device re-measurement under thermal throttle · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_budget_remeasure · executor seat · 1 u · stands on nothing |
|---|---|
| Definition of done | the frame budget is re-derived on a rolling window and on visibilitychange and resize instead of latching at frame 45 AND a gate proves a budget captured at one interval is replaced after the interval changes |
| Matrix cell | Continuous device re-measurement under thermal throttle · _ABSENT_:wc_budget_remeasure · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:wc_budget_remeasure; declare _ABSENT_:wc_budget_remeasure TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-09 measure 2026-09-09: /releases/beach-sand-2e93c4ce.html served SHA256 2e93c4ce615672cd78ceb0251ea3e32f7f784ba7a17f1126237861663f05cec6 in both captures. Evidence C:/Users/elder/nishi-ops/browser-review/beach-freeroam-adapter-20260909-1 and beach-freeroam-adapter-20260909-nvidia-1; both receipts captured, no page errors, browsers closed. Chrome docs confirm Windows powerPreference limitation: https://developer.chrome.com/docs/web-platform/webgpu/troubleshooting-tips . Native emitter active-adapter telemetry built SHA9091b6de5e912d314381e6b351a93b6b165ab08534fdbc0b7c2799f95d03e6f4; canonical emit passed, not yet production reviewed. GPU task 1788970740 stays open. |
PF3 Dynamic Huffman blocks in the sovereign DEFLATE · cell PRESENT · gate NO-GATE-DECLARED
| Contract | gz_dh_lens · executor seat · 3 u · stands on nothing |
|---|---|
| Definition of done | gz_deflate emits BTYPE=10 with a length-limited canonical code and the code-length alphabet RLE and chooses the smaller of fixed and dynamic per block AND every output round-trips byte-identical AND the measured gain on the shipped module is at least 8 percent. SCOPE MEASURED 2026-08-28 by a check-before-build pass rather than estimated, and it is SMALLER than a from-scratch encoder. ALREADY EXISTS, on the DECODE side: nx_deflate.nx implements all three BTYPE handlers end-to-end and carries _deflate_cl_order which IS the HCLEN permutation, plus the RFC length and distance base and extra tables and _deflate_build_dynamic_trees which already reads the code-length alphabet and the 16 17 18 RLE. nx_huffman.nx carries a canonical table builder. MUST BE WRITTEN, and this is the whole of it: frequencies to code lengths, which is the Huffman tree build and exists NOWHERE in the corpus, plus 15-bit length limiting, the ENCODE direction of the code-length RLE, symbol buffering so the emitter can make two passes instead of emitting inline, and the fixed-versus-dynamic per-block cost comparison. ⚠THE EXISTING PIECES ARE REFERENCE, NOT COMPOSABLE: nx_deflate.nx and nx_huffman.nx use nx_int and nx_size and structs and import nx_runtime and nx_tier and nx_bitstream, while nx_gzip_lib.nx is plain i64 and *u8 with no runtime -- importing across that boundary would drag a type system into the compressor that serves every response. Copy the CL order VALUES, not the code. ⛔PRECONDITION, AND IT IS NOT OPTIONAL: DECIDE THE EMITTER CONSOLIDATION FIRST. nx_deflate_enc.nx is a SECOND self-declared fixed-Huffman DEFLATE compressor. Writing dynamic Huffman into one copy while another remains deepens exactly the duplicate-source split that cost lazy matching a double write and a double proof on 2026-08-28. ⚠COUNT HONESTLY BOUNDED: a sibling investigation reported FIVE fixed-Huffman emitters and its verification agents died before refuting it; a literal grep for the phrase over the whole runtime at corpus_complete=1 finds TWO self-declared. Those are different membership tests, so the true number is between 2 and 5 and is NOT pinned -- pin it before consolidating |
| Matrix cell | Dynamic Huffman -- the measured remainder of the compressor gap · runtime/_hdl_build/nx_gzip_lib.nx · PRESENT decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gzip_lib; declare nx_gzip_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF4 Temporal reconstruction from the existing motion-vector primitive · cell WATCHING · gate NO-GATE-DECLARED
| Contract | gpe_temporal_reproject · executor seat · 4 u · stands on PF2 |
|---|---|
| Definition of done | the renderer keeps a history buffer and reprojects it with the same world-point reprojection the optical-flow lane already verified AND a reconstructed frame at a quarter of the ray count scores closer to native than the current spatial upscale on a declared image metric |
| Matrix cell | Temporal reconstruction -- trade resolution for reconstruction, not for blur · _ABSENT_:gpe_temporal_reproject · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:gpe_temporal_reproject; declare _ABSENT_:gpe_temporal_reproject TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF5 Quarter-resolution volumetrics with a temporal pattern · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_cloud_quarterres · executor seat · 2 u · stands on PF4 |
|---|---|
| Definition of done | the cloud pass renders to its own quarter-resolution target with a temporal update pattern and composites AND frame time falls with no verdict change from the panel judge |
| Matrix cell | Quarter-resolution volumetrics with temporal reprojection · _ABSENT_:wc_cloud_quarterres · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:wc_cloud_quarterres; declare _ABSENT_:wc_cloud_quarterres TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF6 Mip-mapped voxel cone stepping · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_voxel_mip · executor seat · 3 u · stands on nothing |
|---|---|
| Definition of done | a mip pyramid is built over the voxel volume at world commit and vmarch selects a level from cone footprint AND the DDA step count per camera ray falls while the rendered frame stays within the declared difference bound |
| Matrix cell | Mip-mapped voxel cone stepping · _ABSENT_:wc_voxel_mip · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:wc_voxel_mip; declare _ABSENT_:wc_voxel_mip TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF7 Foveated and variable-rate raymarching · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_foveate · executor seat · 3 u · stands on PF4 |
|---|---|
| Definition of done | ray density follows an importance field with a reconstruction pass AND the 90Hz VR verdict is met on a device that fails it today |
| Matrix cell | Foveated and variable-rate raymarching · _ABSENT_:wc_foveate · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:wc_foveate; declare _ABSENT_:wc_foveate TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 measure THE VR BUDGET IS A CITED NUMBER ON THIS BOARD NOW: Meta's Set Display Refresh Rates page (its own lastUpdated field 2026-08-27, mirrored whole at 716752 bytes and read) states the frame budget as 13.9 ms at 72 Hz, 11.1 ms at 90 Hz and 8.3 ms at 120 Hz and that an app which cannot hold its chosen rate fails store review; the WG2 answer row's 11.111 ms and 8.333 ms are the same budgets by arithmetic. WITNESS COUNT STATED, NOT CLAIMED: this bar has ONE vendor page with a number, the Khronos loop definition (no number) and two research listings; a second headset vendor's published budget and a standards-body or peer-reviewed figure are owed before it is a triangulated bar. SB-UPSCALE is three vendor feeds and is labelled as vendor claims on the bar itself. |
PF8 A per-frame media budget for in-browser calls · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pf_rtc_budget · executor seat · 2 u · stands on nothing |
|---|---|
| Definition of done | encode and transport cost per frame is published by the same instrument that publishes frame_ms so a call and a world are judged against one clock |
| Matrix cell | Real-time media path for in-browser calls · _ABSENT_:pf_rtc_budget · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:pf_rtc_budget; declare _ABSENT_:pf_rtc_budget TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF9 Device-tiered mesh and texture delivery · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pf_mesh_tier · executor seat · 3 u · stands on nothing |
|---|---|
| Definition of done | one supercompressed asset is transcoded to the device's native GPU format and the tier chosen is reported AND a tier the device cannot decode is refused rather than shipped |
| Matrix cell | Explode-view and heavy-mesh delivery on a phone · _ABSENT_:pf_mesh_tier · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:pf_mesh_tier; declare _ABSENT_:pf_mesh_tier TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
PF10 A real browser field harness for this board · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pf_field_harness · executor seat · 2 u · stands on nothing |
|---|---|
| Definition of done | a real browser run produces LCP and INP and a frame-time distribution for a published world page so the Core Web Vitals bar stops reading UNPINNED AND the harness works on the browser the operator actually mandated. ENGINE CONSTRAINT, named 2026-08-28 because it invalidates the obvious implementation: the operator has directed Waterfox and directed the workstreams off Edge. Waterfox is GECKO and CDP does not exist there, while this estate's existing capture wire IS CDP -- the emitted page carries its own measurement of that wire, that requestAnimationFrame NEVER fires under it, rAF count 0 over 120 to 400 ms across five loads, which is why a one-shot frame door had to be emitted inside the closure at all. So a CDP harness would be built on a wire the mandated browser does not have. The cross-engine answer is WebDriver BiDi, the W3C successor Gecko implements, or in-page instrumentation which is engine-agnostic by construction and is the half enginelab's overlay already covers. Whichever is chosen must be NAMED in the row rather than assumed |
| Matrix cell | A real browser field harness for this board -- LCP, INP and a frame-time distribution from a live browser run of a published world page, on the browser the operator mandated (PF10) · runtime/nx_field_harness.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_field_harness; declare nx_field_harness TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 queue ABSTENTIONS AND OWED READS 2026-10-01: web.dev's vitals article carries dateModified 2024-10-31 and its thresholds are unchanged, so the month on SB-WEBVITALS is the web-vitals library feed's (v6.2.2, 2026-09-14); the W3C specifications behind the thresholds were not re-read this month (www.w3.org answered 403 to the estate fetcher on the gpu board's two attempts) and the Long Animation Frames draft is the gameengine mirror of 2026-09-04; export.arxiv.org answered 429 to this seat, so the foveated listing is the mirror of 2026-09-27. OUR side of the Core Web Vitals bar is still UNPINNED: no field LCP or INP exists for /world/beach until PF10 lands, and the frame-pacing bar's axes (displayed against presented, dropped, animation error, stalls attributed) are exactly what the 2026-09-09 beach stutter rows lack. |
PF11 Frame cost and the quality it bought emitted as ONE record on ONE clock · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pf_quality_cost_record · executor seat with enginelab · 2 u · stands on enginelab:EL7 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | the record defined by enginelab EL7 reaches this board so a performance claim can be read on both axes AND no rung here may be reported on one axis alone once it exists. DELIBERATELY NOT A SECOND RULER: EL7's binding second clause already requires the overlay to publish frame time and image quality on ONE clock, so this rung CONSUMES that record and defines only the reporting rule for this board. Two guards for one invariant is the duplicate-ruler defect and enginelab owns the instrument |
| Matrix cell | One record joins what a frame COST to what it LOOKED LIKE · _ABSENT_:pf_quality_cost_record · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:pf_quality_cost_record; declare _ABSENT_:pf_quality_cost_record TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 land THE LADDER TO SOTA IS DECLARED AND ATTESTED FOR OCTOBER 2026 (cross-board ladder seat, T0 job 1790857487): four sotabar rows (SB-WEBVITALS-2609, SB-FRAMEPACING-2609, SB-UPSCALE-2609, SB-VR90-2608), four sotatarget rows (best-in-class T-BIC-FIELD, T-BIC-PACING, T-BIC-RECON; frontier T-FR-VR90) and thirteen rungrole rows (substrate 5, arm 3, contender 5); performance.refs CREATED with fourteen rows (the board had no refs file): ten mirrors fetched through nx_research_fetch on 2026-10-01 and read from the mirrored bytes, four shared pinned mirrors re-hashed today. nx_ladder performance reads LADDERED (13 of 13 placed, no unclimbed target); nx_barfresh reads FRESH with 4 of 4 ATTESTED by attestfile from this seat's reads (nx_frontier_scan attest answered ATTEST-UNATTESTABLE for all four refs and its fan-out over 68 fetches read newer_sources=0 stale_bars=0); nx_compare_rank performance is not refused. 2026-10-01 measure TWO GRAMMAR DEFECTS IN THIS PLAN WERE HIDING THE BOARD FROM THE RANKER, both repaired in place under a size compare-and-swap (65380 bytes before and after each edit): (1) PF11's deps field read enginelab EL7 and nx_compare_rank answered RANK REFUSED: malformed, missing, duplicate or cyclic authored graph, GRAPH OFFENDER rung=PF11 rule=bad-dep-id; spelled enginelab:EL7 (the cross-board token nx_xdep_lib reads) the graph reads malformed=0 and XDEP edges=1 done=1 unresolved=0; (2) the four ms rows listed their rungs separated by spaces, so with the graph repaired the ranker read ALL DECLARED VERSIONS MET -- ranking by value alone on a board with 1 of 13 rungs done; with commas it reads TARGET VERSION 0.1 (milestone M0). PF13 is still in no milestone row: named, not changed. |
PF12 Quality becomes an INPUT to the quality controller · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_quality_term · executor seat · 4 u · stands on PF11 |
|---|---|
| Definition of done | the controller decision reads an IMAGE-QUALITY term expressed in the RENDER units defined by PF11 rather than frame time alone AND a gate proves a change that improves frame time while degrading the image is REFUSED where today it is rewarded. SCOPE CORRECTED 2026-08-28 and this rung is a DESIGN gap not a wiring gap: the first draft said sourced from nx_quality_balance and nx_quality_metric. nx_quality_balance measures MODEL-WEIGHT fidelity in perplexity-permil and bits-per-weight and its consumer is the quantisation builder -- a different quantity that shares the word quality, with no conversion to render units anywhere in the corpus. nx_quality_metric is SSIM and PSNR which are FULL-REFERENCE and need a ground-truth frame the live renderer does not have. So neither drops in. PF11 is what DEFINES the missing unit, which is why this depends on it. THIS RUNG MUST NOT CLAIM TO ENFORCE THE EXISTING GOVERNED QUALITY FLOOR: that floor is a perplexity bound and cannot be evaluated here, and claiming to enforce a floor you cannot evaluate is worse than not enforcing one |
| Matrix cell | Quality is an INPUT to the quality controller, not only its victim · _ABSENT_:wc_quality_term · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:wc_quality_term; declare _ABSENT_:wc_quality_term TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-23 queue INTAKE 1790172869 routed by nx_intake_route (nishi-review-via-codex, epoch 1790172936) from operator-via-codex: Multimodal reality measurement composer for the Nishi estate. Build one owned, reusable, versioned evaluation workflow that accepts a qualified asset or scene plus a declared style and domain (photore -- sponsor intake-1790172869 plus2 |
PF13 Simulation off the render thread over the isolation the page already pays for · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pf_worker_sim · executor seat · 2 u · stands on nothing |
|---|---|
| Definition of done | the world tick runs in a Worker over a SharedArrayBuffer view of the engine arena and the frame loop only reads it -- the performance.matrix row that ABSENT-PROVED zero Workers across 404 world pages on 2026-08-28 now has a rung -- so the GPU thread is never delayed by economy or NPC routines; done when the served beach reads crossOriginIsolated true, constructs the worker from an emitted blob, p95 frame time at the same resolution falls with the tick moved, and the frame digest stays bit-identical to the single-thread frame (the GE27 tooth reused, never a second ruler). BLOCKED BY NAME 2026-09-03 on lang LN33: the sovereign compiler emits no wasm threads (atomic.rmw ABSENT-PROVEN over 23506 sources), so the isolation the router pays for on every response is spent by nothing until that rung lands |
| Matrix cell | Simulation off the render thread -- cross-origin isolation already paid for · _ABSENT_:pf_worker_sim · WATCHING data |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for _ABSENT_:pf_worker_sim; declare _ABSENT_:pf_worker_sim TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-03 measure MEASURED 2026-09-03 on the served beach (630629 B, stamp b1788396100): COOP same-origin, COEP require-corp, CORP same-origin, CSP worker-src self blob -- all present on the wire; SharedArrayBuffer 0, new Worker 0, crossOriginIsolated 0, instantiateStreaming 0, manifest 0, serviceWorker 0 in the page bytes. PF1 and PF13 are the two rungs that turn those headers into a faster first frame and a steadier one. |
dossiers=13 cells_joined=13 no_matrix_row=0 (sum 13) gates_recorded=0 gates_unrecorded=0 no_gate_declared=13 (sum 13 of cells_joined) evgates=UNSTAMPED
Capability comparisons
Compare the field, search individual capabilities and open their source and adoption evidence. Documented presence does not establish comparative quality.
Position map — centrality and distinctiveness
The four-quadrant map the field uses for brand strategy (Dawar and Bagga, HBR June 2015), re-derived from this matrix on every publish. Centrality is the share of the category's feature mass a player covers, each feature weighted by how many hold it; distinctiveness is the average lead over each rival on the rows the player holds; breadth is the depth-weighted share of the whole matrix (the bubble); depth is how deeply the rows held are held; momentum is the day-over-day move off the spine (green rising, red falling, grey until day two); the dashed path runs first day → previous day → today. Dividers are the category means. Axes are fitted to the field of play, so read the tick numerals, not the frame. Rival marks are documented presence, so a rival's position reads the record, never its quality. The picture grades its own readability below; the table beside it is the same data for a screen reader or a second method.
Rows. 18 feature rows mapped of 23 matrix rows; 5 substrate rows (kernels, bounds and instruments the plan's rungrole marks substrate) are the estate's scaffolding, not category features, and stay off the map
Views. 2D cut: centrality, distinctiveness · 3D cube: centrality, distinctiveness, breadth · axes registered: centrality, distinctiveness, breadth, depth, momentum · a board picks its own in
knowledge/compare/performance.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaultsbubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 200–1000, distinctiveness 0–500 of 0–1000 permil (the full range put every player in one corner)
readability of the 2D cut, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 1 (a fact of the data: two players that close are that close) · data spread 593 permil of the plot · quadrant words unseated 0
text contrast, measured with wcag2-ratio (floors from contrast.conf), light theme: labels 16.24 (floor 4.50) · notes 16.24 (floor 4.50) · quadrant words 5.89 (floor 4.50) · tick numerals 5.89 (floor 4.50) · axis titles 5.59 (floor 4.50) · dark theme: labels 13.78 (floor 4.50) · quadrant words 5.55 (floor 4.50) · axis titles 5.97 (floor 4.50) · dark classes under their floor 0 (one figure serves both themes: a dark shortfall is a token to fix, never a class to hide) · classes refused under their floor 0 (a refused class is not drawn; the scale classes are measured, never hidden) · export: SVG PNG (receipt, rendered by the estate's own rasteriser from this page)
readability of the 3D cube, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 2 · unresolved labels 0 · mark overlaps 3 (a fact of the data: two players that close are that close) · data spread 375 permil of the plot
10 panels over 5 registered axes, every pair once (the lower axis on x, the higher on y) · each panel fitted to its own field of play, first and last tick numerals shown · no labels in a small cut, the legend names the colours; the grade below reports the mark terms only (MM summed over the panels, spread averaged)
readability of the small multiples, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 12 (a fact of the data: two players that close are that close) · data spread 615 permil of the plot
| Player | Quadrant | centrality | distinctiveness | breadth | depth | momentum | Rows held | Days on spine | First seen |
|---|---|---|---|---|---|---|---|---|---|
| Nishi | Unconventional | 415 | 350 | 333 | 750 | +7 since 2026-09-23 | 8 | 10 | 2026-09-15 |
| three.js | Mainstream | 896 | 97 | 500 | 692 | -28 since 2026-09-23 | 13 | 10 | 2026-09-15 |
| Babylon.js | Mainstream | 896 | 97 | 500 | 692 | -28 since 2026-09-23 | 13 | 10 | 2026-09-15 |
| Unity WebGL | Mainstream | 961 | 168 | 592 | 711 | +2 since 2026-09-23 | 15 | 10 | 2026-09-15 |
| PlayCanvas | Mainstream | 935 | 104 | 518 | 666 | -17 since 2026-09-23 | 14 | 10 | 2026-09-15 |
| Unreal PixelStream | Aspirational | 896 | 219 | 592 | 761 | -11 since 2026-09-23 | 14 | 10 | 2026-09-15 |
| Day | Matrix rows reviewed | Players recorded |
|---|---|---|
| 2026-09-15 | 23 | 6 |
| 2026-09-16 | 23 | 6 |
| 2026-09-17 | 23 | 6 |
| 2026-09-18 | 23 | 6 |
| 2026-09-19 | 23 | 6 |
| 2026-09-20 | 23 | 6 |
| 2026-09-21 | 23 | 6 |
| 2026-09-22 | 23 | 6 |
| 2026-09-23 | 23 | 6 |
| 2026-10-08 | 18 | 6 |
players 6|matrix rows 18|feature mass 77|centrality mean 833|distinctiveness mean 172|axes 5|spine days 10 (shown 10)|rows written today 0|readability defects 2D 0 cube 2|spine knowledge/status/cdmap/performance.spine
How to read this evidence. A Nishi source mark records whether the named symbol is declared in the selected source file. It does not establish working behavior or accepted delivery. A watching tag names an expected source entry; discovering it updates source coverage only. A dark tag means the file exists but the named symbol was not found: inspect the source and contract before assuming the capability is missing or renaming code. Peer codes are authored assessments; unassessed means no comparison is established, not absence. Qualification, deployment identity, runtime observations and acceptance must carry their own evidence. Adoption records the catalog's observed path through source, build, promotion, registration and invocation; libraries use importer evidence and gates use declared execution surfaces. A full catalog path does not establish quality, freshness or a successful user outcome. Partial and unknown states retain their stated next action. Census stamps: importers asof 1790637268, gate census asof 1790104131 (unix seconds; -1 = census absent).
Capability matrix — measured against source
◉ authored stronger claim; source present● present◐ partial○ absent? unassessed; excluded from relative comparison · click any capability for its evidence
Peer cell coverage: authored assessed 115, unassessed 0. Authored assessments are not verified outcomes.
No capabilities match this filter or search.
| Capability | Nishi | three.js | Babylon.js | Unity WebGL | PlayCanvas | Unreal PixelStream |
|---|---|---|---|---|---|---|
Sovereign DEFLATE with lazy matching -- the compressor is ours, not the platform'sSource present; authored stronger claim:gz_deflate in runtime/_hdl_build/nx_gzip_lib.nx, verified at emit. MEASURED 2026-08-28 end to end. Adding zlib-style lazy matching (hold each match one position, prefer a strictly longer match at i+1) took beach.html from 252105 to 237977 bytes and the shipped wasm module from 107878 to 103082, both inflating BYTE-IDENTICAL under nx_gunzip with CRC32 verified and sha256 equal to the source. nx_contentdiff GREEN lost_from_live=0, KAT selftest rc=0, promoted sha 4efb80fd with prev banked. The chain depth was deliberately NOT raised: zlib L6 at max_chain 128 and L9 at 4096 land within 0.06 percent of each other on this data, so depth is not the lever and raising it would only cost time. Every rival ships the platform's gzip and owns no compressor, which is why this is an exceed rather than parity. ⚠DELIVERY STATUS MEASURED FROM THE WIRE 2026-08-29, because a promoted tool is not a deployed daemon: a raw Accept-Encoding gzip fetch of the live page served 269112 of 509018 bytes, ratio 0.5287 -- the OLD greedy compressor's 0.5290, not the lazy 0.4993. The serving daemon compiled nx_gzip_lib in before the change and has not been rebuilt, so the win is real in the promoted tool and NOT YET REACHING VISITORS. The daemon rebuild is a front-door deploy and is deliberately an operator decision. The same fetch is the board's second-method evidence: GNU gzip 1.13 decoded that wire body CRC-clean and byte-identical to the identity fetch Adoption: LIB-WIRED importers=4 nonval=2 — fully adopted (top of its ladder).Pros Source symbol found; stronger capability claim is authored; ahead of three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ◉ | ○ | ○ | ○ | ○ | ○ |
The estate serves what it measures -- the page compressor is the same code we can run by handMeasured:gz_compress exists in runtime/_hdl_build/nx_gzip_lib.nx, verified at emit. CONTROL 2026-08-28: nx_gzip run by hand on beach.html returned 252105 bytes against the 252724 the live edge actually served, a 0.25 percent gap. That agreement is what PROVES the served bytes come from this code rather than from a platform library, and it is why a change here is a change to every response the estate makes. Rivals cannot run their host's compressor as a tool and so cannot take this measurement at all Adoption: LIB-WIRED importers=4 nonval=2 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ | ○ |
Inline payload weight is VISIBLE to the Core Web Vitals rulerSource present; authored stronger claim:perf_inline_js in runtime/_hdl_build/nx_uiq_perf.nx, verified at emit. nx_uiq_perf returned issues=0 verdict=PASS on a 477701 byte page that is 275468 bytes of base64 wasm inside one inline script -- 56 percent of the document. The ruler was not wrong, its SUBJECT was narrower than its name: blocking_js counts script src and an inline script is never a src. A PASS from a ruler blind to the dominant cost reads as this page is fine. perf_inline_js publishes inline_js_bytes and inline_js_largest. SHIPPED AND RE-MEASURED FROM THE PROMOTED BINARY 2026-08-28: bytes=508023 inline_js_bytes=504766 inline_js_largest=503141 -- NINETY-NINE PERCENT of the document is ONE inline script element, and the organ still returns issues=0 verdict=PASS, which is the point: it now SEES the dominant cost and does not yet judge it. The verdict is DELIBERATELY unchanged until an estate-wide baseline exists, because a detector that goes red on every page is one everyone learns to ignore. Lighthouse measures transfer size but attributes no cost to inline script weight either Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Source symbol found; stronger capability claim is authored; ahead of three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStreamCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ◉ | ○ | ○ | ○ | ○ | ○ |
Adaptive quality driven by the device's OWN measured frame budgetMeasured:wc_framems exists in runtime/nx_wasm_craft.nx, verified at emit. The controller derives every line from ONE budget reported by the page rather than from a shipped millisecond: tolerate is 1.5x the measured display interval, the refine line is tolerate minus a sixth, and refine uses the renderer's own cost model where rays scale as one over q squared, so the predicted EMA at the finer level is FMA times q squared over q-1 squared. The two bands cannot touch, so it cannot oscillate, and gate T20 already caught a refine-on-the-same-call-as-coarsen bug. This is genuine parity with engine-side dynamic resolution, not an exceed rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=41 nonval=25 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal PixelStream; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ● | ● | ● | ● | ○ |
The page reports its real display interval to the organMeasured:target_ms exists in runtime/nx_wasm_craft.nx, verified at emit. The page averages the rAF interval over frames 6 to 45 and hands the organ a measured millisecond, so the controller tunes against THIS display instead of a constant. Correct in principle and the reason the quality ladder works at all. See the continuous re-measurement row for the defect in how often it runs rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=41 nonval=25 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ● | ● | ● | ● | ● |
One emitter, every world, one gated pipelineMeasured:gpe_build_s exists in runtime/nx_game_page_emit.nx, verified at emit. nx_craft_emit reads the wasm ONCE and builds every world page from that one buffer, so craft beach vale isle crags and ridge are the same engine with different data. ⚠FIGURES RE-MEASURED 2026-08-28 AFTER AN ADVERSARIAL AUDIT OF THIS BOARD CAUGHT IT CONTRADICTING ITSELF: this row carried 478476, 477701, 477693 and 477952 while another row on the same board, dated the same day, carried 508023 for the same page. Both were measured; the pages were re-emitted by other lanes between the two readings, and a board that publishes two sizes for one file is worth less than one that publishes neither. CURRENT, all read in one pass: craft 508798, beach 508023, vale 508015, isle 508274, crags 508707 -- spread 783 bytes, 0.15 percent, so the one-engine-many-identities claim SURVIVES on fresh numbers. The module is now 211523 bytes, up from the 202226 this board was opened with. ⚠AND A WORLD THIS BOARD NAMED DOES NOT EXIST: earlier text here said six worlds and listed ridge. knowledge/world_recipes.conf DECLARES a ridge row, but sites/nishifamily/world/ridge.html is ABSENT on disk -- I read the recipe plane and asserted a published page from it. FIVE world pages exist. That a declared world has never been emitted is itself an open question for whoever owns the recipe plane, and it is recorded here rather than quietly dropped. ★EVERY SIZE ON A LIVE BOARD IS A MOVING TARGET WHEN OTHER LANES EMIT THE SAME PAGES -- these are stamped with their measurement pass and will go stale again. The ship lane refuses on any red from the craft gate, the feel oracle or the wasm VM gate before a page is written. Rivals all have deterministic build pipelines so this is parity rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separateCons not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ● | ● | ● | ● | ● |
External cacheable wasm with streaming compilationOpen — watching_ABSENT_:gpe_wasm_external : gpe_wasm_external, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. THE LARGEST UNTAKEN WIN ON THIS BOARD. The module is inlined as a base64 string literal, which costs a MEASURED 29.6 percent on the wire even after gzip -- 65864 bytes against 50840 for the same 118512 bytes of module -- because base64 destroys byte alignment and cripples LZ77 matching. It also forbids WebAssembly.instantiateStreaming, which is ABSENT-PROVEN across the whole runtime corpus at 23320 files with corpus_complete=1, so compilation can never overlap download. And with Cache-Control no-cache the identical 206600 byte module is re-sent on every one of the six worlds, while crossing between worlds is the advertised headline feature of every one of them. The preconditions are already satisfied: wasm maps to application/wasm in five serving organs and the CSP already carries wasm-unsafe-eval. Contract: emit the module to a content-hashed file, load it with instantiateStreaming and an arrayBuffer fallback, and serve it immutable. It MUST also teach nx_wasm_vm_verify and nx_wasmpage_census to follow an external reference, because both anchor on the base64 magic in the page and would otherwise go silently blind rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js (leads), Babylon.js (leads), Unity WebGL (leads), PlayCanvas (leads), Unreal PixelStream; open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ◉ | ● |
Continuous device re-measurement under thermal throttleOpen — watching_ABSENT_:wc_budget_remeasure : wc_budget_remeasure, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. LIVE DEFECT, not a gap. The budget is latched exactly once per page load: the sampler is guarded by a not-already-set test and fires at frame 45, so 40 samples from one moment set the frame budget permanently. Everything that makes mobile hard happens after frame 45 -- sustained thermal throttling, a 120Hz panel dropping to 60Hz on battery saver, visibility changes, another app taking the GPU. The failure is asymmetric: since tolerate is 1.5x the latched budget, a budget captured at 120Hz on a display that later runs at 60Hz makes the controller believe it is permanently over budget, so it coarsens and never refines. That is exactly the recorded operator complaint about the page being incredibly pixelated on capable hardware. Contract: re-measure on a rolling window and on visibilitychange and resize instead of latching rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ◉ |
Temporal reconstruction -- trade resolution for reconstruction, not for blurOpen — watching_ABSENT_:gpe_temporal_reproject : gpe_temporal_reproject, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. THE PRIMITIVE IS ALREADY BUILT AND VERIFIED AND WIRED TO THE WRONG CONSUMER. Resolution scaling today is SPATIAL only: the organ picks a ray resolution and the browser upscales, which throws away exactly the information a temporal method keeps. The hard part of temporal upsampling is motion vectors, and /world/index.html already documents dense exact optical flow obtained by reprojecting every pixel's known 3D world point through the next camera, verified three ways including parallax at 4.9x between near and far terrain. It feeds the synthetic-CV data lane and not the renderer. This is the adoption-gap shape rather than a missing capability. Contract: a history buffer plus the existing reprojection, so a phone renders a quarter of the pixels and presents a reconstructed frame rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL (leads), PlayCanvas, Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ● | ● | ◉ | ● | ◉ |
Quarter-resolution volumetrics with temporal reprojectionOpen — watching_ABSENT_:wc_cloud_quarterres : wc_cloud_quarterres, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. The cloud pass is a 16-step raymarch with multi-octave noise per step, run at FULL resolution every frame, and its own source cites the HZD and Nubis lineage -- which renders clouds at reduced resolution with a temporal reprojection pattern and updates a fraction of pixels per frame. We adopted the technique and not its cost model. Contract: render the volumetric pass at quarter resolution into its own target with a temporal pattern and composite rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
Mip-mapped voxel cone steppingOpen — watching_ABSENT_:wc_voxel_mip : wc_voxel_mip, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. vmarch walks a fixed 120-cell DDA budget through a 128 by 48 by 128 usampler3D for the camera ray AND again for the shadow ray AND again per NPC fragment for the occlusion test. A mip pyramid over the same volume lets distant rays take exponentially larger steps for the same visual result, which is the standard answer for a voxel raymarcher and costs no new asset. Contract: build the mip chain at world commit and select a level from cone footprint rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
Foveated and variable-rate raymarchingOpen — watching_ABSENT_:wc_foveate : wc_foveate, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. March fewer rays in the periphery and interpolate. A DDA raymarcher is the easiest possible subject for this because ray count is an explicit loop bound rather than a hardware rasterisation rate, and the estate already grades a 90Hz VR verdict where fixed foveated rendering is the field standard. Contract: an importance field over the ray grid with a reconstruction pass rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unity WebGL, Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ○ | ○ | ● | ○ | ◉ |
Real-time media path for in-browser callsOpen — watching_ABSENT_:pf_rtc_budget : pf_rtc_budget, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. Named as a deliverable and NOT yet measured on this board, which is why it is a watch and not a figure. The synthroom lane already establishes that real-time generated video is 2D-only at current hardware, so the body renders native and generation is confined to an offline bake -- that is a performance decision already taken. What is missing here is the BUDGET: an encode and transport cost per frame stated beside the render cost, so a call and a world can be judged against one clock. Contract: a per-frame media budget published by the same instrument that publishes frame_ms rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ◉ |
Explode-view and heavy-mesh delivery on a phoneOpen — watching_ABSENT_:pf_mesh_tier : pf_mesh_tier, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. The anatomy explode views are shipped as pre-rendered PNGs rather than as an interactive mesh a phone can stream, and the world assets are served whole. The field answer is one supercompressed asset transcoded to the device's native GPU format -- ASTC on mobile, BC on desktop -- plus geometry compression, so the same artifact serves every tier without a per-device build. Contract: an asset tier that reports which transcode target it chose and refuses to ship a tier it cannot decode rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
Dynamic Huffman -- the measured remainder of the compressor gapSource present; authored stronger claim:gz_dh_lens in runtime/_hdl_build/nx_gzip_lib.nx, verified at emit. SHIPPED AND PROMOTED 2026-08-29 (sha 6f568346, prev banked), AND THE ROUND-TRIP ORACLE EARNED ITS KEEP THE SAME HOUR: the first build's litlen code came out OVER-SUBSCRIBED BY EXACTLY ONE KRAFT UNIT on /world/beach -- the draft counted over-deep LEAVES where zlib's gen_bitlen counts over-deep NODES INCLUDING INTERNALS -- caught before any promote because the page failed CRC round-trip while the module passed, bisected locally against zlib's strict inflater which NAMED the defect (invalid literal/lengths set), and fixed by looping on the MEASURED Kraft sum itself rather than transcribing any counter: a count copied without its counting rule is a different algorithm wearing the same name. PROVEN after the fix, same bytes for every figure: lazy+fixed 254216 vs lazy+dynamic 217388 on the 509018-byte page -- a 14.49 percent gain, floor was 8 -- with zlib -9 at 214548, so the sovereign DEFLATE now sits 1.3 percent from zlib where it was 21.7 percent behind two days ago. Verified four independent ways: nx_gunzip CRC-clean and sha-identical on page and module, the CLI selftest choosing the FIXED path on tiny input and reproducing the old encoder BYTE-IDENTICAL (the cost comparison working, not a mode flag), an adversarial stress set (incompressible, mega-runs, exponential-skew deep trees, zeros, empty, one byte) all identical under zlib-strict inflate, and GNU gzip 1.13 on another machine decoding the wire-fetched output to the exact source sha. The tool is ahead of the SERVING DAEMON, which still carries the pre-lazy lib and needs its own rebuild-and-deploy (an operator decision) before any of this reaches a visitor. OWED and named: a dedicated nx_gzip KAT gate so this proof is a rostered tooth rather than a session record. MEASURED, not estimated. Isolating the levers with zlib's Z_FIXED strategy on the real module bytes: lazy matching and chain depth together account for 7.4 percent and DYNAMIC HUFFMAN accounts for a further 11.3 percent. The compressor currently emits fixed Huffman blocks with a hardcoded code table (BTYPE=01); dynamic Huffman builds per-block optimal tables from actual symbol frequencies, which is worth most on binary payloads exactly like a wasm module. Contract: BTYPE=10 with a length-limited canonical code, the code-length alphabet RLE, and a smaller-of-the-two block choice, verified by gunzip round-trip. No rival owns its compressor so none can take this Adoption: LIB-WIRED importers=4 nonval=2 — fully adopted (top of its ladder).Pros Source symbol found; stronger capability claim is authored; ahead of three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ◉ | ○ | ○ | ○ | ○ | ○ |
Simulation off the render thread -- cross-origin isolation already paid forOpen — watching_ABSENT_:pf_worker_sim : pf_worker_sim, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. ABSENT-PROVEN 2026-08-28 by nx_absent across all 404 world pages at coverage_complete=1 corpus_complete=1 truncated=0: not one page constructs a Worker, so world simulation and rendering share the main thread. THE ISOLATION COST IS ALREADY PAID -- the page ships Cross-Origin-Opener-Policy same-origin AND Cross-Origin-Embedder-Policy require-corp, which is exactly the header pair that unlocks SharedArrayBuffer, and the CSP already allows worker-src self blob. So the precondition for moving the simulation off the render thread is met and the benefit is not taken. This is the pattern ForzaTech uses for Fable's thousand-NPC population: the GPU thread is left unobstructed to draw while town economy and long-range NPC routines run on low-priority background threads that can never delay a frame. Contract: the tick runs in a worker over a SharedArrayBuffer view of the organ arena and the frame loop only reads it rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL (leads), PlayCanvas, Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ● | ● | ◉ | ● | ◉ |
Named quality modes rather than one silent adaptive curveOpen — watching_ABSENT_:pf_quality_modes : pf_quality_modes, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. The controller silently chooses a ray resolution and the visitor has no way to say which trade they want. The console field ships DECLARED modes instead -- Fable offers a 30fps quality mode at native 4K with full ray tracing against a 60fps performance mode with dynamic resolution and screen-space GI -- so the trade is a stated contract rather than an invisible heuristic. We already have the mechanism: the quality ladder and the per-class pixel budget exist and the HUD already prints the resolved px. What is missing is the DECLARATION. Contract: named modes with a published target frame time each, the chosen mode reported in the HUD, and the adaptive controller operating INSIDE the chosen mode rather than across all of them rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
Stream the surplus after first paint instead of shipping it in the documentOpen — watching_ABSENT_:pf_poststream : pf_poststream, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. Everything arrives in one 477701 byte document and nothing streams afterwards. The ForzaTech insight is that a streaming budget sized for 200mph has enormous surplus at walking speed, and Fable spends that surplus on foliage and town density rather than banking it. The browser equivalent is that bandwidth after first paint is almost entirely unused: the visitor is reading the page, and nothing is being fetched. Contract: ship the smallest playable world in the document and stream density -- higher-detail meshes, textures and distant chunks -- on an idle callback after the first frame, with the tier that actually arrived reported so a slow link degrades visibly rather than silently rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js (leads), Babylon.js (leads), Unity WebGL (leads), PlayCanvas (leads), Unreal PixelStream (leads); open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ◉ | ◉ |
Bake at emit so the runtime budget goes only to what must be dynamicMeasured:gpe_hair_chain_bake exists in runtime/nx_game_page_emit.nx, verified at emit. ALREADY PRACTISED HERE AND WORTH NAMING because it is the same principle photogrammetry serves for Fable: bake the expensive detail so the runtime processor is freed for what genuinely must be simulated. nx_hairdyn_lib runs at EMIT time and the trajectory it produces is baked into the page as a table the page replays, with the emitter's own source stating the rule -- a second integrator in JavaScript would be the duplicate-ruler defect and could drift from the shipped plant with no wrong number anywhere to find. Every number in that bake is derived rather than chosen: segment count from the groom's own points-per-strand, frame count from one natural period of the shipped plant. The exceed is not the baking, which everyone does, but that the baked artifact and the simulator are provably ONE solver rather than two that must be kept in agreement rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separateCons not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ● | ● | ● | ● | ● |
No page is built around a module that cannot fit its own declared memoryMeasured:wf_guard exists in runtime/nx_wasmfit.nx, verified at emit. A delivery guard with a real incident behind it. On 2026-08-14 /craft shipped BLACK TWICE because the framebuffer had outgrown the module's own declared linear memory -- 22022528 bytes required against what was declared -- while every byte-level stage of the pipeline reported green and the native gate reported 59 of 59, because that gate mmaps its own arena and therefore never meets the wasm bound at all. wf_guard is the arithmetic that decides it, and it lives in a LIBRARY rather than inside one emitter precisely so a second emitter cannot ship without it. The emitter's own header records why that mattered: a census of all 62906 shipped pages found 25 carrying an embedded module and only 9 reaching the function that used to hold the guard, so a guard placed in the shared path was only as broad as the unmeasured claim that the path was shared. Parity rather than exceed -- every engine validates its own memory bounds -- but it belongs on a performance board because the failure it prevents is a page that downloads its whole payload and then renders nothing rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=96 nonval=96 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ● | ● | ● | ● | ● |
Quality is an INPUT to the quality controller, not only its victimOpen — watching_ABSENT_:wc_quality_term : wc_quality_term, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. THE MECHANICAL REASON PERFORMANCE WORK SHOWS UP AS PIXELS, measured exhaustively 2026-08-28. wc_framems decides on ONE input -- an exponential moving average of frame time -- and coarsens the ray pixel size when it exceeds tolerate. There is no quality term anywhere in the decision. The estate already owns both rulers and the renderer imports NEITHER: nx_quality_balance.nx encodes the operator's own law in its first line -- BALANCE quality against speed -- and nx_absent over the whole runtime at coverage_complete=1 corpus_complete=1 returns matches=2, one of which is that header line itself, leaving exactly ONE importer and it is nx_builder_quant. nx_quality_metric.nx carries qm_ssim_milli and qm_psnr_cdb and its importers are VIDEO CODEC gates. So the controller can only ever trade image quality away, because image quality is not something it can see. ⚠CORRECTED 2026-08-28, SAME DAY, BY A SIBLING LANE AND VERIFIED HERE BEFORE THE CORRECTION WAS ACCEPTED -- THE FIRST VERSION OF THIS ROW CALLED IT AN ADOPTION GAP AND SAID THE CONTRACT WAS TO SOURCE A QUALITY TERM FROM THE RULERS THAT ALREADY EXIST. THAT IS WRONG AND IT WOULD HAVE SENT SOMEONE TO BUILD A WIRING EDGE. nx_quality_balance is not a render ruler at all: read it and it speaks PERPLEXITY-PERMIL and BITS-PER-WEIGHT over fp16 Q8_0 Q6_K Q4_K_M Q2_K, its floor is a max acceptable perplexity loss of 20 permil, and its declared consumer is the quantisation BUILDER. Its quality is MODEL-WEIGHT FIDELITY. The controller's subject is IMAGE quality in a render. Same word, different quantity, and no conversion between them exists anywhere in the corpus -- so importing it would hand the controller a number it cannot interpret. AND THE SECOND RULER DOES NOT DROP IN EITHER: SSIM and PSNR are FULL-REFERENCE metrics that need a ground-truth image, and a live renderer at a coarse ray size has no fine-grained reference frame to compare against, so neither ruler can be consulted at runtime as written. The importer figure was also corrected 22 to 20 by that lane, two having been self-references. WHAT SURVIVES UNCHANGED is the finding itself: wc_framems decides on one input, an EMA of frame time, and there is no image-quality term anywhere in the decision. Contract: a quality term in the controller decision, expressed in the RENDER units defined by the paired record rung, NOT sourced from the quantisation policy and NOT claiming to enforce a floor nobody can evaluate -- claiming to enforce a floor you cannot actually evaluate would be worse than not enforcing it rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
One record joins what a frame COST to what it LOOKED LIKEOpen — watching_ABSENT_:pf_quality_cost_record : pf_quality_cost_record, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. A frame-time number published without a paired image number is how a performance campaign silently becomes a pixelation complaint, and that is the state today: the HUD prints the resolved px and the organ receives frame_ms, but nothing anywhere writes the two into one row, so no run can be asked whether it got faster by getting worse. This is the same join /compare/enginelab exists to make -- its thesis is that five capture formats on five clocks cannot attribute a spike to its cause, and the fix is ONE record layout on ONE clock. The quality level belongs IN that record. Split deliberately across two boards: enginelab owns the record and the ruler because it is the instrument board, performance owns the result because it is the delivery board, and neither can quietly grade its own homework. Contract: frame cost and the quality level that produced it emitted as one row on one clock, so every rung on this board is judged on both axes or on neither rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind three.js, Babylon.js, Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ● | ● | ● | ● | ● |
A real browser field harness for this board -- LCP, INP and a frame-time distribution from a live browser run of a published world page, on the browser the operator mandated (PF10)Open — watchingruntime/nx_field_harness.nx : pf_field_harness, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. ADDED 2026-09-03 because nx_compare_rank performance read PF10 UNMAPPED (no matrix row carried its symbol), so the Core Web Vitals bar could never stop reading UNPINNED through the ranker. Rival cells coded at INDEX LEVEL and UNPINNED for this row: three.js and Babylon.js are libraries with no field harness of their own; Unity WebGL, PlayCanvas and Unreal Pixel Streaming ship profiler or stats overlays rather than a field vitals harness -- to be re-coded from the mirrored pages before any of these cells is quoted. Contract in performance.plan PF10: the harness is NAMED (WebDriver BiDi on Gecko or in-page instrumentation), the run produces LCP and INP and a frame-time distribution for a published world page, and the row appends to the same time spine as PF11. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unity WebGL, PlayCanvas, Unreal PixelStream; open contract; expected source declaration not established | ○ | ○ | ○ | ◐ | ◐ | ◐ |
Rival-claim provenance. Every Yes, Best or Part mark in a rival column is a claim about someone else's product. A cited row carries a reference mark that resolves to a pinned mirror (the refs gate measures that); an uncited row is an observation read and is badged in its evidence until a reference lands. The badge is the receipt, never the proof: a mark proves a mirror exists, not that the mirror supports the code. The rows are counted below.
| Class | Count | Note |
|---|---|---|
| rows with rival marks | 19 | rows carrying a Yes, Best or Part in a rival column |
| cited | 0 | a reference mark resolves to a pinned mirror |
| uncited | 19 | an observation read, badged until a reference lands |
| rows, summed | 19 | RECONCILED: the parts sum to the total 19 |
Delivery and evidence
Inspect risks, technical debt, rendered observations, experiments and references. Read scope and limitations alongside every result.
Risk register
| Risk | Likelihood x impact | Mitigation |
|---|---|---|
| Changing the emitter collides with another seat working the same file | likely x medium | nx_game_page_emit.nx is actively edited by other lanes and CAS refusals are already being observed on it. Coordinate before PF1 and re-derive anchors under fresh compare-and-swap rather than retrying a stale edit |
| Externalising the module silently blinds the wasm verifiers | medium x high | nx_wasm_vm_verify and nx_wasmpage_census both anchor on the base64 magic inside the page. PF1 is not done until both follow an external reference. Shipping the emitter half alone would disarm the check that caught craft shipping black twice |
| A compressor defect corrupts every response the estate serves | low x severe | gz_deflate is imported by the serving router so a bad block breaks everything at once. Every change round-trips through nx_gunzip with CRC32 and a sha256 identity check before promote, and the lazy-matching rung was proven on the CLI copy before the library copy was touched |
| The frame work is optimised against a budget nobody re-measures | likely x medium | PF2 is deliberately first and cheapest. Every later frame rung is judged against the device budget so a latched budget would silently invalidate PF4 through PF7 |
Release history and work log
The worked plan, on the board. Every leg appends what it measured, landed, retracted, learned and left queued, per rung, so a crash leaves the next seat a ledger here rather than a transcript to mine. Kinds: measure, land, retract, lesson, queue. Newest last.
| When | Rung | Kind | Entry |
|---|---|---|---|
| 2026-09-03 | PF13 | measure | MEASURED 2026-09-03 on the served beach (630629 B, stamp b1788396100): COOP same-origin, COEP require-corp, CORP same-origin, CSP worker-src self blob -- all present on the wire; SharedArrayBuffer 0, new Worker 0, crossOriginIsolated 0, instantiateStreaming 0, manifest 0, serviceWorker 0 in the page bytes. PF1 and PF13 are the two rungs that turn those headers into a faster first frame and a steadier one. |
| 2026-09-09 | PF2 | measure | 2026-09-09: /releases/beach-sand-2e93c4ce.html served SHA256 2e93c4ce615672cd78ceb0251ea3e32f7f784ba7a17f1126237861663f05cec6 in both captures. Evidence C:/Users/elder/nishi-ops/browser-review/beach-freeroam-adapter-20260909-1 and beach-freeroam-adapter-20260909-nvidia-1; both receipts captured, no page errors, browsers closed. Chrome docs confirm Windows powerPreference limitation: https://developer.chrome.com/docs/web-platform/webgpu/troubleshooting-tips . Native emitter active-adapter telemetry built SHA9091b6de5e912d314381e6b351a93b6b165ab08534fdbc0b7c2799f95d03e6f4; canonical emit passed, not yet production reviewed. GPU task 1788970740 stays open. |
| 2026-09-09 | PF1 | measure | 2026-09-09 review.cjs now validates gpu_timing_mode=native or disabled-for-diagnostic and records native timer.stop intervention before/after state. Timer-off run confirmed stopped=true,inflight0,cancelled3, no cleanup errors; both dedicated browsers closed without reported capture failures. Receipts nishi-ops/browser-review/beach-timing-native-20260909-1 and beach-timing-disabled-20260909-1. These are explicit diagnostic counterfactuals, not shipped feature removal. GPU readback alone is insufficient explanation; broader scheduling/compositor/GPU attribution and representative user-browser reproduction remain open. |
| 2026-09-09 | PF1 | measure | 2026-09-09 beach-longframes-20260909-1:3600 W-held intervals p99=24.995ms,max37.505ms on this run. Bounded Long Animation Frame observer retained9 records,dropped0, callback accounting0.545ms (not total observer cost). Eight ended before walking began. Remaining132.175ms record starts27197.495ms and straddles sampling endpoint27201.195ms; its128.140ms unattributed FrameRequestCallback starts27201.450ms AFTER that endpoint. Do not classify it as a reproduced in-walk application stall or dismiss it as proven browser overhead. Evidence capture/serialization is a confound. Contexts closed; startup hitch verified, sustained user lag still open. Primary API scope reference https://developer.chrome.com/docs/web-platform/long-animation-frames . |
| 2026-09-09 | PF1 | measure | 2026-09-09 beach-startup-profile-20260909-1 captured real served document, startup.cpuprofile, bounded LoAF records and600 free-roam frames; no reported page failures; browser closed. Profile begins before navigation and stops at requested visitor readiness; sampling overhead uncalibrated, not an FPS acceptance run. Local receipt C:/Users/elder/nishi-ops/browser-review/beach-startup-profile-20260909-1/receipt.json retains artifact hashes. Browser adapter now accepts explicit profile_startup and uses caller evidence/deadline budgets. LoAF scope corrected: retain runtime-provided presentation fields where exposed; these are not an input-to-photon measurement. |
| 2026-09-09 | PF1 | land | 2026-09-09 Performance board delivery: September WebGPU targets and retained startup evidence published through scoped sovereign generator; public nishifamily.com/compare/performance returned200 and new startup attribution card was visually reviewed. Local review tooling now captures optional startup CPU profiles; real served Beach run and cleanup verified. This is a telemetry/record delivery, not a Beach frame-rate fix. Worker migration, sustained-walk stutter, competitive benchmarks and structured API parity for these narrative targets remain unfinished. LoAF research-bar correction: current browser records may expose presentation fields; retain actual supported fields, without treating them as measured input-to-photon latency. |
| 2026-09-09 | PF1 | land | 2026-09-09 Native garment-binding preparation and isolated browser-worker qualification added; corrected review module https://nishifamily.com/releases/nxa-garment-bind-c788f996.wasm is 15,247 bytes, SHA c788f9964d0452e51eaf068f365fbec503c2b25c5ff2dff761e2ee377d8a319d. Exact public module and resident asset verified in Chrome; native reference, original loader output and worker results agree. Earlier e62fd282 review module failed because the general compiler predates the staged string-data fix; retained as failed evidence. No compiler promotion, shared-memory verifier bypass or Beach FPS improvement claimed. Canonical loader integration and startup/walking A/B remain required; full goal stays open. |
| 2026-09-09 | PF1 | land | 2026-09-09 Scoped Beach review delivery at https://nishifamily.com/releases/beach-garment-808ee6a1.html?gpu=1&hud=1&w=beach&review=0 ; canonical native emitter, byte identity, actual free-roam and start/end images verified. Native garment preparation runs off-thread, preserving existing weight composition. First ab7bbcae review failed on blob-worker relative URL; owner corrected, failed evidence retained. Review tool now captures named character failure promptly, verified against failed page. Follow-up startup capture retained all 87 long-frame records: 244.315 ms loader continuation and 106.295 ms texture-related continuation remain, plus unattributed stalls. Worker integration is delivered for review; overall stutter, art quality, shared-WASM proof and general deployment remain unfinished. |
| 2026-09-09 | PF1 | measure | 2026-09-09 Presentation evidence retained in master worklog and local hashed receipts beach-gpu-scheduling-20260909-1 and beach-composition-diagnostic-20260909-1. Both captured current 808ee6a1 document, 300 W-held frames, no page failures and closed browser cleanup. Trace SHA 415522f8095629f001c93beb295fb5d42729af5ac710262ffa870b393a9b1c30. Optional browser backend inventory and isolated composition diagnostic now available in review adapter. No persistent hardware settings changed; no frame-rate fix, equal elapsed route, exact user-browser identity or competitive superiority claimed. Next work: join delivered intervals to GPU/compositor scheduling evidence and retain unsupported telemetry explicitly. |
| 2026-09-09 | PF1 | land | 2026-09-09 Diagnostic trace clock anchors added to the existing review adapter and verified in real Chrome against public Beach 808ee6a1, 300 W-held frames, failures=[] and cleanup=closed. Retained trace SHA e25e9bd7f8b39553d784d58b46d3459741c4c1c44fc6da8bb8510f475747dc22 and receipt SHA 3d115ba8d2b30030e3b603c8f5f63901cc65be0515ddb5e041653ea2483e013e. Existing presentation board was published and visually checked on nishifamily.com. This delivers diagnostic evidence, not a smoothness fix. Severe stutter, presentation root cause, startup work, NPC warping, photorealism and structured API parity remain unfinished. Local analysis: beach-presentation-clock-20260909-1/presentation-clock-analysis.json. |
| 2026-09-09 | PF1 | land | 2026-09-09 Native timing batching review served and walked at https://nishifamily.com/releases/beach-timing-4618858f.html?gpu=1&hud=1&w=beach&review=0 SHA4618858f003f0469da2cb674dc149182acd6c8be0d1762066a6597c87286ff15,782424B. Canonical build and real GPU qualification passed; actual public document identity, four visitor submissions, timing completion and start/end images reviewed. Windowed/headless mode now explicit in review receipts. Changed viewpoints invalidate a causal performance comparison; zero stalls above50ms in this sample does not meet sustained SOTA acceptance. No Emscripten dependency, general compiler promotion or shared-memory verifier bypass. Next: reproducible routes, submission/upload attribution and measured quality-preserving improvements. |
| 2026-09-09 | PF1 | land | 2026-09-09 Review adapter native-state observations and simulation-duration stopping verified on both public Beach releases. Positive runs captured with no page errors and closed browsers; one-frame negative control correctly reports simulation-window-incomplete. Endpoint screenshots reviewed: both ocean route, sky/ocean fidelity still unfinished. Evidence retained in beach-tick-route-pair-20260909.json and master contract. This is stronger comparison instrumentation, not a completed performance fix or forced camera. Old/new documents contain identical worldWasm and shader bytes. Native slot batching remains in4618858f; severe tails, wider-route stability, NPC warp, startup and shared-WASM proof remain open. |
| 2026-09-09 | PF1 | land | 2026-09-09 Native reservation-record component and gate staged and executed through sovereign APIs. nx_wasm_layout_gate_review returned28/28, ELF48952B SHA66136b4417e10fa1305468309f8f436ef264efe8d32bec068a3fc46919a0c7c3. Full build and gate output reviewed. Qualification uses representative compiler IR; real Beach packaging/pair acceptance remain unfinished. No compiler promotion, shared verifier bypass, Emscripten addition or frame-rate gain claimed. Reusable native memory ownership proof is required before further shared off-thread preparation. |
| 2026-09-09 | PF1 | measure | 2026-09-09 Operator research direction retained: multivariate experiments beyond published recipes, browsers beyond Chrome, and eventual Nishi-browser native GPU capabilities. RedCube README, package, license and test configuration inspected. These are research requirements and source findings, not implemented experiment automation, cross-browser passes or a Beach performance fix. Beach remains the delivery pilot; API parity and public visual review of these new board records remain pending. |
| 2026-09-09 | PF1 | land | 2026-09-09 Compiler measurement component qualified through sovereign build/run/hash APIs. nx_compile_wat staged759488B SHA36018e9d6240d6868ced3fd6c940576ea5b50337c2028f61e77c419026c78d1e; gate47240B SHA f628e20be2be3bfd56e4c895b576de993418147d051091eb0e5fce300d3d274f,10/10 run1788979868. Full logs reviewed; real module byte identity checked. Measurement and browser research records prepared for existing /compare/performance surface. No compiler promotion or Beach visual/FPS improvement claimed. Adapter preference change and public visual review pending. |
| 2026-09-09 | PF1 | land | 2026-09-09 Graphics control delivered on scoped Beach review and visually inspected at nishifamily.com. Native emitter build1788980345, SHA10acd927177c6e7afe3c04e548283ae5fd8a18915d24b3fb55f2d009e586c292,707976B; publish1788980437 verified via actual document hash and interaction. Receipt beach-gpu-choice-20260909.json retains requested/actual adapter distinction and closed-browser cleanup. Same Chrome chose Intel for both preference modes; no speedup claim. Shared verifier component24/24 run1788980250; complete proof pending. Existing Beach art, sustained stutter, NPC warping and broad production rollout remain unfinished. |
| 2026-09-09 | PF1 | land | 2026-09-09 Native real-artifact layout join built102040B SHA9ebe6ef93ef874f09c34d9da81fe264e0b17835a509075c5497ae49e1ab81e63; full build output reviewed. Gate runs1788980827 and1788980960 passed12/12 on actual compiled garment artifacts; missing-record control1788980887 returned RED. Budgeted shared Wasm SHA7a4e60cbdd202d599f3435a29c924b079cb3de82411a1f1f692b38d9941493c1 remains staged on NAS for runtime qualification. No Beach rendering or FPS change claimed. Next required evidence: shared workers, initialized-data persistence, output equality and provenance binding. Existing graphics-choice review remains the evaluation URL. |
| 2026-09-09 | PF1 | measure | 2026-09-09 Shared worker runtime receipt shared-garment-runtime-20260909-v2.json passed with no errors and browser closed. Output SHAe77e70ef0d6ed5be9eb3a3e5aaae3505097fe86c5df9113719155e70ce9d2e71. First failed receipt retained as test-instrument defect evidence. Shared artifact is public test-only; existing Beach worker configuration and general pair acceptance unchanged. Next: source/BSS provenance and state ownership before production shared execution; sustained GPU-specific Beach frame pacing remains open. |
| 2026-09-23 | PF12 | queue | INTAKE 1790172869 routed by nx_intake_route (nishi-review-via-codex, epoch 1790172936) from operator-via-codex: Multimodal reality measurement composer for the Nishi estate. Build one owned, reusable, versioned evaluation workflow that accepts a qualified asset or scene plus a declared style and domain (photore -- sponsor intake-1790172869 plus2 |
| 2026-10-01 | PF11 | land | THE LADDER TO SOTA IS DECLARED AND ATTESTED FOR OCTOBER 2026 (cross-board ladder seat, T0 job 1790857487): four sotabar rows (SB-WEBVITALS-2609, SB-FRAMEPACING-2609, SB-UPSCALE-2609, SB-VR90-2608), four sotatarget rows (best-in-class T-BIC-FIELD, T-BIC-PACING, T-BIC-RECON; frontier T-FR-VR90) and thirteen rungrole rows (substrate 5, arm 3, contender 5); performance.refs CREATED with fourteen rows (the board had no refs file): ten mirrors fetched through nx_research_fetch on 2026-10-01 and read from the mirrored bytes, four shared pinned mirrors re-hashed today. nx_ladder performance reads LADDERED (13 of 13 placed, no unclimbed target); nx_barfresh reads FRESH with 4 of 4 ATTESTED by attestfile from this seat's reads (nx_frontier_scan attest answered ATTEST-UNATTESTABLE for all four refs and its fan-out over 68 fetches read newer_sources=0 stale_bars=0); nx_compare_rank performance is not refused. |
| 2026-10-01 | PF11 | measure | TWO GRAMMAR DEFECTS IN THIS PLAN WERE HIDING THE BOARD FROM THE RANKER, both repaired in place under a size compare-and-swap (65380 bytes before and after each edit): (1) PF11's deps field read enginelab EL7 and nx_compare_rank answered RANK REFUSED: malformed, missing, duplicate or cyclic authored graph, GRAPH OFFENDER rung=PF11 rule=bad-dep-id; spelled enginelab:EL7 (the cross-board token nx_xdep_lib reads) the graph reads malformed=0 and XDEP edges=1 done=1 unresolved=0; (2) the four ms rows listed their rungs separated by spaces, so with the graph repaired the ranker read ALL DECLARED VERSIONS MET -- ranking by value alone on a board with 1 of 13 rungs done; with commas it reads TARGET VERSION 0.1 (milestone M0). PF13 is still in no milestone row: named, not changed. |
| 2026-10-01 | PF7 | measure | THE VR BUDGET IS A CITED NUMBER ON THIS BOARD NOW: Meta's Set Display Refresh Rates page (its own lastUpdated field 2026-08-27, mirrored whole at 716752 bytes and read) states the frame budget as 13.9 ms at 72 Hz, 11.1 ms at 90 Hz and 8.3 ms at 120 Hz and that an app which cannot hold its chosen rate fails store review; the WG2 answer row's 11.111 ms and 8.333 ms are the same budgets by arithmetic. WITNESS COUNT STATED, NOT CLAIMED: this bar has ONE vendor page with a number, the Khronos loop definition (no number) and two research listings; a second headset vendor's published budget and a standards-body or peer-reviewed figure are owed before it is a triangulated bar. SB-UPSCALE is three vendor feeds and is labelled as vendor claims on the bar itself. |
| 2026-10-01 | PF10 | queue | ABSTENTIONS AND OWED READS 2026-10-01: web.dev's vitals article carries dateModified 2024-10-31 and its thresholds are unchanged, so the month on SB-WEBVITALS is the web-vitals library feed's (v6.2.2, 2026-09-14); the W3C specifications behind the thresholds were not re-read this month (www.w3.org answered 403 to the estate fetcher on the gpu board's two attempts) and the Long Animation Frames draft is the gameengine mirror of 2026-09-04; export.arxiv.org answered 429 to this seat, so the foveated listing is the mirror of 2026-09-27. OUR side of the Core Web Vitals bar is still UNPINNED: no field LCP or INP exists for /world/beach until PF10 lands, and the frame-pacing bar's axes (displayed against presented, dropped, animation error, stalls attributed) are exactly what the 2026-09-09 beach stutter rows lack. |
On these two registers. Rows are declared in the domain's plan file and carry the debt id, which is the join key back to the sovereign debt plane — that plane, not this page, is the authority on state. Reconciling them automatically (the regen reading the plane and refreshing these rows) is a named, owed rung; until it lands, treat an id here as a pointer to look up, not a status to trust.
Evidence for every claim. This is not an opinion matrix: each capability we mark present is grounded in real source and proven by a live-forked gate. See the autonomously-generated evidence page → (per-axis source lines + executed self-test verdicts) or the machine-readable evidence.json.
Honest verdict. The coverage above records source-symbol presence. Stronger comparisons are authored claims, not measured runtime superiority or accepted outcomes. A watch contract names an expected source symbol; finding it changes source discovery only. Qualification, deployment and runtime acceptance require their own evidence.
Person · product · place — not yet measured for this domain
Every compare carries this layer. Declare
knowledge/compare/performance.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain performance, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).Exclusive and shared capabilities
One capability is graded on one board; every other board references it. The fleet duplicate census (nx_rivalcensus dupes, on the fleetdupes clock row) reads every board's matrix and names each capability graded as a row on two or more boards, with a home derived from evidence: rung when exactly one board's plan contracts the symbol (that board is starred), gates when no plan contracts it and exactly one board's executed gate map runs the organ's gate (that board is plussed), ambiguous when several boards carry the deciding evidence (a decision for their seats), and none when no evidence decides it, with the reason beside it: unmapped (no board of the duplicate declares a gates map; declare the maps) or unproven (a map exists and runs none of the organ's gates; gate the organ on its proving board, or declare it a shared library by policy). A home elsewhere means this board should carry an @xrow reference to the home board's row rather than a graded row of its own; a home here means the other boards should. The table is this board's slice of one estate-wide census: the shape is shared, the rows are the board's.
This board grades 23 capabilities: 21 exclusive to it and 2 shared with other boards. Shared rows on this page: 2 = home here 0 + home elsewhere 0 + ambiguous 0 + undecided 2 + refused 0 (partition 2); undecided = unmapped 0 + unproven 2 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=7019.
| Capability | Organ | Boards grading it | Home | Rule | Evidence | Standing on this board |
|---|---|---|---|---|---|---|
| gpe_build_s | nx_game_page_emit.nx | sharehub, performance | — | none | unproven | undecided |
| wf_guard | nx_wasmfit.nx | sharehub, performance | — | none | unproven | undecided |
Gaps from the record — what the estate does that no board carries
The record census (nx_goalmap record) reads the invoked-tool population and every plan queue row and files each organ or directive that NO matrix, plan or gates row names. A row here is a callout the boards missed: adjudicate it onto a board or declare it infrastructure. Census state STALE (age 1396124 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.
| kind | name | board | source | evidence |
|---|
rows shown 0|this board's directives 0|estate-wide un-boarded organs 452 (listed in full on /compare/ecosystem)|census rows 639|malformed 0 (counted, never rendered)
References
Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the matrix notes jump here. A dash means honestly absent, never assumed.
- [webvitals_releases_202610] GoogleChrome. web-vitals releases (Atom feed), fetched and read 2026-10-01: ten releases listed, newest v6.2.2 on 2026-09-14, then v6.2.1, v6.2.0, v6.1.1; no October release publisher · read in our library
knowledge/fetched/cmp_performance_webvitals_releases_202610.xml· pinh49f7172afbd8c4dd248b7286c599f2b4d40f6e5b5781911068d1c8808ea71a00· accessed 2026-10-01 · mirrored-indexGrounds: SB-WEBVITALS-2609 - [webdev_vitals_202610] Google. Web Vitals on web.dev (article metadata dateModified 2024-10-31), fetched and read 2026-10-01: LCP should occur within 2.5 seconds of when the page first starts loading, pages should have an INP of 200 milliseconds or less and maintain a CLS of 0.1 or less, and a good threshold to measure is the 75th percentile of page loads, segmented across mobile and desktop devices publisher · read in our library
knowledge/fetched/cmp_performance_webdev_vitals_202610.html· pinh4d1ca56cf8edbc68fb5ce4b7820a8d937abbd7de61c53d834fb70d7027335641· accessed 2026-10-01 · vendor-docGrounds: SB-WEBVITALS-2609 - [loaf_spec] W3C Web Performance Working Group. Long Animation Frames API, Editor's Draft (the gameengine mirror of 2026-09-04, pin re-verified 2026-10-01): a long animation frame refers to any of the listed occurrences whose duration exceeds 50ms publisher · read in our library
knowledge/fetched/cmp_gameengine_loaf-spec.html· pinh35187db56732baeb493f58241a90af3bed3ca9fc7ec5e01a5c2c025f3a94cc9f· accessed 2026-09-04 · published-standardGrounds: SB-WEBVITALS-2609 - [presentmon_releases_202610] Intel GameTechDev. PresentMon releases (Atom feed), fetched and read 2026-10-01: newest v2.6.0 on 2026-09-21 (PSO compilation metrics tracked via ETW and attributed into frame and CSV capture; OS DDI flip events for per-flip timestamps and completed or dropped flip reporting; a Game Experience overlay preset highlighting metrics that correlate with perceived motion and animation quality), then v2.5.1 and v2.5.0 publisher · read in our library
knowledge/fetched/cmp_performance_presentmon_releases_202610.xml· pinh48780d958fd76063bb1dc93dc327e1bf3ce546628b354700a459046ad85dad5a· accessed 2026-10-01 · vendor-docGrounds: SB-FRAMEPACING-2609 - [presentmon_capture] Intel GameTechDev. PresentMon capture application README (the gameengine mirror of 2026-09-04, pin re-verified 2026-10-01), the metric table: GPU Busy as how long the GPU spent working on this frame, Dropped Frames as an indication that the frame was not displayed, Displayed Time as how long this frame was displayed on screen, Animation Error (MsAnimationError) as the difference between the previous frame's CPU delta and display delta publisher · read in our library
knowledge/fetched/cmp_gameengine_presentmon-capture.md· pinhb8df8b3887156fb59b5c0a465e3a574ab47fe18416961d43abe1df97b9460813· accessed 2026-09-04 · vendor-docGrounds: SB-FRAMEPACING-2609 - [dlss_releases_202610] NVIDIA. DLSS SDK releases (Atom feed), fetched and read 2026-10-01: newest DLSS 310.9.1 SDK on 2026-09-08 (Added DLSS Ray Reconstruction Transformer Mode, Preset F; bug fixes and stability improvements), then 310.7.0 on 2026-06-23; a vendor claim publisher · read in our library
knowledge/fetched/cmp_performance_dlss_releases_202610.xml· pinh20f8f5e0ee8dfe2fbe23bf1e250f15cb8f6cb982c8099ab19b6ecb07f244290d· accessed 2026-10-01 · vendor-docGrounds: SB-UPSCALE-2609 - [fsr_sdk_releases_202610] AMD GPUOpen. FidelityFX-SDK releases (Atom feed), fetched and read 2026-10-01: newest FSR SDK v2.3.0 on 2026-07-03 (updates a number of AMD FSR Redstone DLL-based ML technologies; FSR Ray Regeneration 1.2.0 with optional ambient and specular occlusion denoising and checkerboard rendering support); a vendor claim publisher · read in our library
knowledge/fetched/cmp_performance_fsr_sdk_releases_202610.xml· pinh86ff646a059cc28a0cd16f8e25b042e6fccc2fb00b2b181ba5b7d5e21064869b· accessed 2026-10-01 · vendor-docGrounds: SB-UPSCALE-2609 - [xess_releases_202610] Intel. XeSS SDK releases (Atom feed), fetched and read 2026-10-01: newest XeSS SDK 3.0.2 on 2026-07-24 (updates XeLL to 1.3.2.10), and XeSS SDK 3.0.0 on 2026-03-09 (Added 3x and 4x Multi-Frame Generation for Intel Arc GPUs; improved frame generation models); a vendor claim publisher · read in our library
knowledge/fetched/cmp_performance_xess_releases_202610.xml· pinh0732832e4b0006c01314cca2c78dda7aed1039f75b035821e20c80e04ca7fc0c· accessed 2026-10-01 · vendor-docGrounds: SB-UPSCALE-2609 - [meta_refresh_rate_202610] Meta Horizon developer documentation. Set Display Refresh Rates (page data lastUpdated 2026-08-27), fetched and read 2026-10-01: the default display refresh rate for apps is 72 Hz; the frame budget at each rate is 13.9 ms at 72 Hz, 11.1 ms at 90 Hz and 8.3 ms at 120 Hz; apps that cannot consistently render at the desired display refresh rate will fail Meta Horizon Store review; a vendor document publisher · read in our library
knowledge/fetched/cmp_performance_meta_refresh_rate_202610.html· pinh38e8e7d56ce7c338aefc385097d865f5b68e25966f176db8cb9707caee19fee9· accessed 2026-10-01 · vendor-docGrounds: SB-VR90-2608 - [openxr_xrwaitframe] The Khronos Group. OpenXR 1.1 reference page xrWaitFrame: xrWaitFrame throttles the application frame loop in order to synchronize application frame submissions with the display and returns a predicted display time for the next time that the runtime predicts a composited frame will be displayed publisher · read in our library
knowledge/fetched/cmp_performance_openxr_xrwaitframe.html· pinh3bae647f13a19286ae2b43a19fc9a84b168009d8bf115fb6f628db20e630733d· accessed 2026-10-01 · published-standardGrounds: SB-VR90-2608 - [openxr_xrframestate] The Khronos Group. OpenXR 1.1 reference page XrFrameState: predictedDisplayPeriod is the XrDuration of the display period for the next application-generated frame, for use in predicting display times beyond the next one; shouldRender tells the application whether to render its layers publisher · read in our library
knowledge/fetched/cmp_performance_openxr_xrframestate.html· pinha8c6167955bc969c0595c8eba20fb279af55cac6eb71c04a52c2eed711f9a470· accessed 2026-10-01 · published-standardGrounds: SB-VR90-2608 - [openxr_sdk_releases_202610] The Khronos Group. OpenXR-SDK-Source releases (Atom feed, the mirror shared with the gpu board), fetched and read 2026-10-01: newest OpenXR SDK 1.1.63, dated 2026-09-01 in its notes and published 2026-09-02; 1.1.62 on 2026-08-03 publisher · read in our library
knowledge/fetched/cmp_gpu_openxr_sdk_releases_202610.xml· pinh15e0b3f9a59139cdd4bba4c5d63766eb5332bdb630d9d3a3dd7930fd20f8560f· accessed 2026-10-01 · mirrored-indexGrounds: SB-VR90-2608 - [arxiv_splat_202610] arXiv API listing, 3D Gaussian splatting AND (real-time OR level of detail OR streaming), 30 entries, newest 2026-09-30, no October entry (the estate attester's fetch of 2026-10-01 copied to a pinned mirror shared with the gpu board); abstract read on this board: Atlas 2609.02352, on-device city-scale splats in VR with temporal-aware level-of-detail search and stereo rasterisation, 18.5x over the GPU baseline with 92.4 percent energy savings publisher · read in our library
knowledge/fetched/cmp_gpu_arxiv_splat_202610.xml· pinh5f66d1ad896313fa13cc053aab351a6c054dbcc066c8f0914a6f976089e5dcfd· accessed 2026-10-01 · mirrored-indexGrounds: SB-VR90-2608 - [arxiv_vrsplat_202609] arXiv API listing held by gameengine (foveated or virtual reality or headset AND Gaussian splatting, 40 of 104 entries, feed stamped 2026-09-27), pin re-verified 2026-10-01 and read on this board for Hybrid Foveated Path Tracing with Peripheral Gaussians 2601.22026 and PowerGS; NOT re-fetched this month because export.arxiv.org answered 429 publisher · read in our library
knowledge/fetched/cmp_gameengine_arxiv_vrsplat_202609.xml· pinh183ba1236e69521184d4769369b1e04532bc9ec00ef134247618b1bcf154f6c1· accessed 2026-09-27 · mirrored-indexGrounds: SB-VR90-2608
generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/performance.matrix · source checks show implementation presence; runtime and user-outcome evidence are reported separately · JavaScript supports page controls