harness: faster, narrower Waiting for model… shimmer (plan #669)

[btipling]

Summary

Retune the Busy-row Waiting for model… shimmer (plan #669) — paint-only LUT in the Wasm harness, fully contained to text_wave.zig + its tests + one docs row. No protocol, host, rect-spinner, or cap changes.

• Double speed — SPEED = 2 inside headPosition (multiplied in u32 so phase 255*2 can't wrap u8). N*STEPS/SPEED = 27 ticks → ~2.7 s full loop at the locked HARNESS_BUSY_TICK_HZ = 10. STEPS stays 3 (no sub-char smoothing loss).
• Narrow, localized comet — replaces the N-relative full-ring gradient (cyclicDistance, now deleted) with a directed trail behind the traveling head: HEAD_SPAN=1 acid accent head, MUTED_SPAN=2 warm_muted, DARK_SPAN=1 warm_border. The rest of the line stays warm_accent — dark occupancy drops from ~6/18 to 1/18. colorStep is now absolute-scalar and never returns 3.
• Reduced motion / phase 0 — solid ramp[0] fast-path kept (a comet at head 0 would otherwise paint a dim trailing-edge tail); clock suffix stays solid warm_accent.

Scope: native/harness/src/{text_wave.zig,text_wave.test.zig}, docs/harness-limits.md. Host tick, STEPS, rect spinner, and protocol untouched.

Verification (Zig gates, run on host in-sandbox)

Gate	Result
zig fmt --check text_wave.zig text_wave.test.zig	✅
zig build test-rich	✅ exit 0 — text_wave test artifact green
zig build test-rich-invariants	✅ exit 0
zig build harness -Doptimize=Debug	✅ wasm32 Debug smoke

Note (environmental, pre-existing, unrelated to this PR): two test-rich child steps model_picker_layout and transcript_split_layout print failed command in this sandbox because the dvui testing backend can't load the external DejaVu Sans Symbols font. Those tests live in files this PR does not touch; they fail identically without my changes. The authoritative Wasm gate is build-harness CI on the self-hosted runner, which compiles the wasm32-freestanding release build (waiting below).

Vitest delta: None — this is a Zig-only change; no TypeScript code changed, so no vitest run is indicated (per implement-plan TS-vs-Zig scope table). Correctness gate for this layer is the Zig compile + build-harness.

Tests

text_wave.test.zig rewritten to the comet/SPEED model (+ plan tests 1–10):

• headPosition period: N=18 returns to the same head at phase 27 (full loop = 27 ticks = ~2.7 s at 10 Hz)
• phase * 2: phase 7 / N=10 → 4.667 (updated); phase 255 / N=20 → 10.0 via the u32 path (updated)
• trailDistance: 0 at the head, increases behind the travel direction; rings around at the string end
• colorStep spans: head accent / 2 muted / 1 dark / accent rest
• far-from-head trail (≥4) is step 0, not dark
• colorStep never returns 3 (sweep to trail 256 + boundary pin)
• N=18 comet occupancy count: exactly 1 dark, 2 muted, 15 accent
• N=0/N=1 → 0
• STEPS == 3, SPEED == 2
• phase-0 fast-path domain rationale (dim comet tail at the trailing edge)
• kept: UTF-8 multi-byte / emoji scalar counting + countScalars pins

Operator shimmer/reduced-motion smoke (plan tests 11–12) are gated on the build-harness Preview, per the plan's DoD.

Fixes #669

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
invincible	Ignored		Aug 18, 2026 9:58pm

[btipling]

Implements plan #669 (HANDOFF-READY).

• Speed: SPEED=2 in headPosition (u32 multiply, no u8 overflow) → 27-tick loop ≈ 2.7 s at the locked 10 Hz host tick; STEPS stays 3, HARNESS_BUSY_TICK_HZ untouched.
• Occupancy: deleted cyclicDistance; directed comet behind the head (HEAD_SPAN=1 / MUTED_SPAN=2 / DARK_SPAN=1), rest of the line warm_accent — dark drops from ~6/18 to 1/18. colorStep never returns 3 (surface).
• Reduced motion: phase-0 solid fast-path + solid clock suffix kept.

Gates: zig fmt --check ✓ · zig build test-rich ✓ (text_wave artifact green) · test-rich-invariants ✓ · zig build harness -Doptimize=Debug ✓. build-harness CI on the self-hosted runner is green (wasm32-freestanding release + rich host unit tests — run 32156606219).

This is a Zig-only change (no TS), so no vitest delta is indicated per the TS-vs-Zig scope table.

Docs (docs/harness-limits.md) shipped in the same PR (timeless, no issue numbers). Operator shimmer smoke (plan tests 11–12) is gated on the build-harness Preview, per the plan's DoD.

Not merged per implement-plan; next steps are adversarial-review, then an explicit merge request.

[btipling]

Adversarial review — PR #671

Verdict: PASS WITH NOTES
Repo: btipling/invincible
Scope: main ← plan/waiting-shimmer-tune · 3 files · paint-only LUT retune (text_wave.zig + tests + one docs row)
Lenses run: L1, L3, L4, L5, L6, L8, L9 (skip L2: no API / secrets / runner / Wasm-memory trust boundary; skip L7: no config/host seams)
AGENTS.md read: yes · docs/feature-divide.md read · SECURITY.md skipped (diff does not touch workflows, secrets, runner, or API)

Findings

Sev	Lens	Finding	Break scenario	Refutation attempt	Confidence
Minor	L1+L9	Localized comet makes the pre-existing busy_tick 255→1 wrap a visible teleport. text_wave.zig headPosition / bridge.zig inv_set_busy_tick	Start any Busy turn ≥ 25.6 s (tool-using agent turns do this constantly). Host phase maps 1..255 then wraps to 1 (never 0). For N=18: headPosition(255, 18) = (510 % 54)/3 = 8.0; next tick headPosition(1, 18) = 2/3 ≈ 0.667. The 4-scalar comet jumps ~7 glyphs in one 100 ms frame — from around for back to Wa. Old full-ring cyclicDistance gradient hid the same skip as a phase hitch; a sprite does not. Module header also claims this wrap is “after 12.8 s” (SPEED confused with wrap period); bridge.zig and docs/harness-limits.md still correctly say 25.6 s. No test pins headPosition(255, 18) vs headPosition(1, 18).	Defender: wrap is pre-existing, protocol/host is out of scope, plan residual already said “ring-wrap has no dark spike.” Counter: they analyzed wrap for occupancy (still 1 border scalar — correct) and missed that deleting the full-ring gradient turns the same skip into a traveling-head pop. In-module math cannot make 255→1 continuous: period 54 does not divide 255*SPEED. Real fix is a follow-up on the u8 remap (wrap at a multiple of N*STEPS, or a wider phase), not a merge block on this paint PR. 12.8 s is comment-only and does not change runtime.	high
Nit	L8	Options.ramp comment still documents the old 4-stop layout [head, trail1, trail2, rest] (text_wave.zig · Options)	A later caller (module advertises “generic reusable text wave”) puts rest ink in ramp[3] per the comment. colorStep now paints rest as ramp[0]; ramp[3] is never indexed. Rest of the line silently uses head color.	Only in-tree caller is busy_row.zig with WARM_RAMP, where [0] is already the desired rest (warm_accent). No current user-visible miss. Comment rot, not a live break.	high

Residual risk

u8 phase 1..255 will always be incommensurate with the 27-tick / 54-raw comet period, so a long turn will keep one discontinuous frame every 25.6 s until the remap changes. SPEED=2 on STEPS=3 also dwells the hard-edged head 2 frames / 1 frame / 2 frames… on successive scalars (100–200 ms) — below the bar for a finding, but the 10 Hz step the plan already called aesthetic-only. Operator smoke (plan tests 11–12) is Preview-gated and was not independently exercised here. warm_border on teal_bg is still one hard-to-read glyph; that is the requested dark stop, not a regression.

Merge guidance

• PASS WITH NOTES: safe to merge from this attack; nits optional.
• Do not treat the wrap teleport as a #669 blocker. If it bothers operators on long turns, file a follow-up on inv_set_busy_tick / phase width — not another text_wave occupancy tweak.
• Fix the 12.8 s header (and the Options.ramp comment) whenever the file is next touched.

What was not attacked

Live Preview operator shimmer / reduced-motion eyeball (plan 11–12). Production Gateway. Self-hosted runner beyond reading build-harness.yml path filters + confirming run 32156606219 is green. Host setBusyTick / HARNESS_BUSY_TICK_HZ (untouched). rect_spinner LUT. Vitest (Zig-only diff; DI/cost gate N/A).