docs: mirror the streamed ceiling curve and the ray-charge law from docs/04 The wiki carried the single-tree budget tables only, so a reader of the status page could not see that across region cells the same ceiling moves coverage by 11x to 56x while the share of surface kept does not move at all, nor the rule that a ceiling below max_steps + 1 draws nothing by construction.
docs: OQ-51's spike re-measured on current main - 199 tests clean, agreement unchanged, one additive CI conflict
docs: front page claimed no culling, no WGSL and no compaction; all three exist and are measured Reading the panel end to end rather than grepping it found more than the pattern search had: the phase status, the open-question count, three capabilities listed as missing that have since landed, a fourth-device phrasing that predates the gate naming four machines, and a WASM parity figure quoted as current when it describes an older 13-line boot. The block is 18 lines and hashes to db71f8a0... today, which the evidence manifest publishes and CI reproduces. The standing lesson, and the reason for the read-instead-of-grep: docs that agree with other docs are not verified. This page was internally consistent and publicly wrong.
docs: the public front page said 118 tests, 48 questions and no spikes run; all three are now wrong no longer
docs: OQ-60 is a first-impression-versus-predictability trade, not a performance one
docs: P1.6's repair half landed in the binary, with the walk cost measured (P1.6, OQ-46, ADR-028)
docs: what the GPU spike measured in CI per runner, and why none of it can be R1 (S9, R1, ADR-026)
docs: the device command's opt-in ISA axis, and why it stays opt-in (P1.5, S3)
docs: the determinism axes CI covers and the vacuity rules each comparison enforces (P1.5, S3, AS2)
docs: x86 codegen level shown not to move the boot digest or asset bytes, with the vacuous-hash lesson (P1.5, S3, AS2)
docs: SIMD-invariance checked per architecture, with the vacuous-check bug that finding exposed (P1.5, S3)
docs: the WASM SIMD axis measured inert on the boot block - 17 lines, one sha256, native included (P1.5, S3)
docs: P1.4's pop-in measured on the arriving region's pixels - 0 of 70 in frame one, order is 2x there
docs: OQ-60's settle curve, and the Safari/ARM64 leg that closes P1.5's browser clause short of the named machine
docs: step 9b performs the carry - 813 frames, zero extra charge, and the uncollected free rays (R3, ADR-024, OQ-60)
docs: the scalar-hash CI leg - one axis of the N2600 target removed ahead of the device (P1.5, S3, ADR-017)
docs: OQ-59 - a streamed frame can be cut but not carried, plus the sub-ray ceiling law
docs: the per-region descent, its two-orders table, and the demo's own numbers (ADR-024, R3)
docs: OQ-58's thresholds measured - distance-invariant frame, 2^24 placement limits, three architectures agreeing
docs: the far region cell that rounded geometry away - R5's measurement, OQ-56's correction, OQ-58
docs: pop-in counted in frames - 5 vs 9 by scan order alone (R3, P1.4)
docs: 1080p measured, derived and ceilinged - what an AS5 ceiling actually buys (R1, AS5, S1)
docs: AS5's compile-time ceiling exists, calibrated 27.3x loose, and the threshold stays the owner's call
docs: the demo spans three region cells; OQ-57 was an assertion that could not be true (ADR-024, R4)
docs: region streaming on the CPU path - it flattens the bound test, not the march (ADR-029, OQ-55)
docs: the demo draws under a ceiling - half the work, 99% of the scene, and the cut verified across engines (ADR-024, R1)
docs: the budget walk order measured - same ceiling, 0% or 25% of the scene (ADR-024, R1)
docs: the frame work budget pass - a ceiling that holds, priced as a curve (ADR-024, R1)
docs: the demo grew a spread scene, so ADR-024 pays 26.4% inside the shipped binary
docs: OQ-54 closed - the licence was earned, the saving was a k-aware arm (ADR-024)