plan: record the outstanding unit/soak work as the only remainder

docs/superpowers/plans/2026-10-30-vt1200-video-streaming.md

@@ -251,3 +251,26 @@ git commit -m "feat(tiles): bad apple tile stream end to end"
 ## Execution handoff
 
 Plan 1's unit measurements (fps for `L,XY` and `B,XY`, heap headroom, `OS_TOUCH`) size this plan's expectations; nothing here depends on them being good, only on them existing.
+
+---
+
+## Outstanding: unit and soak (the only remaining work)
+
+Everything in this plan and in `2026-10-30-vt1200-stream-transport-and-row-strips.md`
+is implemented and verified **in MAME** (145 host tests, `alloc`/`rx`/`draw`/`ring`/
+`dict`/`tiles`/`delta`/`stream`/`replay` modes green, baseline `samples=198`
+unmoved, ROM md5 `b46a3e442ccb86134d2356bd70d22fd2`). What remains needs the physical
+terminal, which is not attached to this machine:
+
+1. **Plan 1, Task 7 step 5-6** — `sudo ./test-stream.sh <iface> 08:00:2b:16:4b:5a 128 8`
+   for 300 frames, then the same with the blit forced to `L,XY`, then a 6,300-frame
+   soak. Record fps for both, delivery, drops, `OS_TOUCH`.
+2. **This plan, Task 6 step 5** — `sudo ./test-stream.sh <iface> <mac> --tiles` for 300
+   frames and the 6,300-frame soak; record fps, drops, `OS_TOUCH`, `decode_worst`.
+3. **The real clip, not the synthetic one** — `make_tiles.py` against actual Bad Apple
+   frames (the ffmpeg recipe in `make_tiles`' docstring, untested here because ffmpeg is
+   not installed), then the replay and the unit run above. The synthetic clip's numbers
+   (1,280 B full frame, ~36 B delta, PSNR 29.47 dB) are a floor, not a prediction.
+
+The `malloc` context (debugger entry vs a `create_process` child) is also unproven on
+hardware; `test-stream.sh`'s preflight is what will surface it.