# i4h scissors->cystic artery verify receipt -- 2026-11-06 (lead re-read of raw metrics)
Worker: claude sonnet subagent; out t2/artifacts/i4h-scissors-artery/ (RESULT.md ends I4H-ARTERY-DONE, patch.diff 62 lines).
Lead check = re-read baseline/ and patched/ metrics.json + run.log:
| run | steps | physics Hz | step p50 ms | peak_blood_alive | peak_deleted tets | finite |
| baseline | 2400 | 56.0 | 17.9 | 0 | 128 | True |
| patched | 2400 | 10.4 | 95.9 | 4096 | 2495 | True |
Blood first appears between steps 200 and 400 (blood=1970 at step 400, cutnodes=2, srcs=3). By step 1800 blood holds at 4096 = saturated (likely the pool cap, not checked in code; treat as a lower bound). sealed=0 throughout.
Open findings: (1) patched deletes 19x more tets (2495 vs 128) -- the scissors path is removing a large tissue volume, not making a clean artery cut; check visually before any claim. (2) 5.4x slowdown not isolated. (3) No render of this run yet -> no media claim.
Verdict: bleeding mechanism DONE (engine numbers); visual/realism NOT verified.

## Render check 2026-11-05 23:1x EST (t2/artifacts/i4h-artery-render/, Slang, 400 frames, 0 dropped, 11.3 Hz)
Lead look at the side-by-side at t=10.0 s (patched | unpatched): in the patched run the gallbladder is almost gone (only a small white piece is left behind the tools), so a large tissue region was removed. **No red blood is visible** in any frame.
The render run's metrics.json has no systems_fire block, so its blood count is unverified. The renderer does draw fluid pools (render/fluid.py), which leaves 3 candidates: blood is not alive in this run, the pool is drawn behind tissue, or it is off-view.
Verdict: ARTERY-RENDER-PARTIAL. Engine bleeding fires; visible bleeding is NOT shown; the cut is too large (confirms open finding 1). Nothing goes to the Mac.

## r2 (t2/artifacts/i4h-artery-r2/, patch-r2.diff), 2026-11-05 23:5x EST
Lead re-read render_final/counters.log summary: steps 2400, first_blood_step 255, peak_blood_alive 4096, peak_cutnodes 1, sources 2, deleted 205 (baseline 128, r1 2495); render and counters from the SAME run; mp4 401 packets.
Root causes (worker, from logs): oversized cut spheres in the GB wall (~700 tets/step) + 60-step dwell per snip + stretch_breaking cascade (thr 12). Fix: cutting closure_delta 0.01, radius_scale 0.05, margin 0; tear threshold 30; one snip at closest approach. Blood invisible in r1 = ~95 % of particles hidden behind the GB + low-contrast colour; render-yaml-only tweak: colour [0.8,0.02,0.03], render_radius 0.03.
Lead look (sbs t=14, t=26): GB intact; bright red patch at the cystic pedicle and red dots on the floor/flap = visible blood; but pale ring/disc sprites cover the cavity floor (irrigation particles) and the blood reads as sprites, not a wet stream/pool. Run-to-run deleted 205-262.
GLM gate (glm-5.3-flash, attempt 4 after 3x429; review/media-20261106-i4h-artery-r2/glm-verdict.md): **VERDICT claim=YES realism=4 show_owner=OK-with-label**. Bleeding visible at the cystic pedicle from t=8 s; gallbladder intact in all frames. Blood reads as static decals and hard-edged disc sprites, with no stream, no pulsation and no wetness; the volume barely grows from 8 s to 26 s (particle cap); translucent grey disc particles float in the field. Agrees with the lead look.
Status: DONE as a labelled step (artery cut + visible bleeding in i4h); realism 4/10 -> the blood/fluid render is the next best-backend item (D-S15b).