# Lab morning brief, Thu 2026-10-22

Written by the Mac session at 04:15 CEST. Source: the HYBRID_LAB dashboard feed, read through the reverse tunnel on the map server. Three reads: generated 04:05:38, 04:07:48 and 04:09:58. Also read: `/docs/SETTINGS_EDITOR.md`, `/docs/SIGNOFF_AND_SCHEDULE_2026-10-18.md` and `/docs/HANDOFF_2026-10-20_1430.md`. Nothing was changed on any lab machine. Times are lab clock times copied from the feed. Anything the feed does not state is marked **[unverified]**.

Not reachable: `HANDOFF_2026-10-21_2030.md` (`/docs/` and the root both return 404) and `_claude/tasks/T150/MORNING.md` (not served). The log tail only reaches back to 02:53. Nothing in the feed or in local memory says anything about RustDesk or AnyDesk.

---

## 1. State at 04:10

| machine | run (board row) | model hour / end | ETA | NaN | ranks (stopped) | stdout mtime | last frame | disk free |
|---|---|---|---|---|---|---|---|---|
| SAGBERG | westwind_v10c_wedge_test24h_r16 (T153, restart h12 → h24; the counter starts at 0) | 2.0 / 12.0 (= model h14 of 24) | 07:10 (feed), "~07:00" (log 03:27) | 0 | 16 (0) | 04:08:56 | 04:02 | 766 GB |
| HYBRIDLAB2 | heavy_rain_v10c_wedge_test12h (T151, cold, deltaT 5 s) | 1.0 / 12.0 | 12:31 (feed), "~12:05" (log 03:51). **Ends after noon.** | 0 | 20 (0) | 04:07:43 | 03:49 | 808 GB |
| HYBRIDLAB1 | normal_v10w (T102g, 250 m warm start) | 46.0 / 47.0 | **04:28** | 0 | 20 (0) | 04:09:30 | 04:03 | 730 GB |
| DESKTOP | normal_v10c_wedge_test12h (T152, cold, 12 h) | 2.0 / 12.0 | 07:55 (feed), "~07:44" (log 03:31) | 0 | 20 (0) | 04:08:06 | 03:53 | 756 GB |
| HYBRID_LAB | coordinator PC. Unreal editor running: pid 11976, 11.0 GB working set, 1 instance. The feed's set/lines/level_saved fields are null. | — | — | — | — | — | — | 344 GB |

- **Stuck:** none. Every stdout file was written within the last 2 minutes, no rank has stopped, there is no NaN anywhere, and no machine is marked stale.
- **Last frames are on schedule.** HYBRIDLAB2 writes a frame every 45 min, DESKTOP every 23 min.
- **The coordinator's 02:54 plan (board T150–T155) matches the runs:**
  - T151 heavy_rain on HYBRIDLAB2: running.
  - T152 normal on DESKTOP: running.
  - T153 SAGBERG h12 → h24: running.
  - T155 (HYBRIDLAB1 westwind wedge 24 h, "after 04:30") is **open, not launched**. At 03:18 the coordinator wrote "waiters armed on the SAGBERG and HYBRIDLAB1 end lines". Whether that waiter starts T155 by itself is **[unverified]**. HYBRIDLAB1 is idle from about 04:28 unless it does.
- **Finished overnight:**
  - SAGBERG westwind 12 h ended 03:25, 0 NaN, copied back. Its line set was collected at 03:37 and **not uploaded**.
  - The T154 nest report.

## 2. Coordinator

- **Last item it handled:** the Mac's 01:04 queue line (Mapping registration). The answer is in the chat at 01:06. The queue still shows `handled: null` for it.
- **Last lines it wrote itself (no worker or monitor tag):**
  - 03:51 "T151 VERIFIED (coordinator)"
  - 03:38 "T153 VERIFIED"
  - 03:31 "T152 VERIFIED"
  - 03:18 "T150 editor relaunch not possible from this session tonight"
  - 03:16 "ATTENTION …"
- **Last workers it started:** T151, T152 and T154 at 02:58. The T153 collect worker started at 03:27. The T156 worker was **denied at launch** at about 03:25. The last worker line is 03:56 (T152 frames).
  - `workers.json` stops at T141 (10-21 19:59). The T150–T156 workers were never registered there.
- **Feed keeps advancing:** generated 04:05:38 → 04:07:48 → 04:09:58. Monitor lines arrive every 5 min.
- **Unanswered:** David's lines at 03:47 "are you working on VR" and 03:48 "no the VR SETUP". The coordinator wrote at 03:51 but did not answer them.
- **Open board rows with status running or blocked:**

| row | status | owner | what |
|---|---|---|---|
| T150 | waiting for David | HYBRID_LAB | VR show |
| T151 | running | HYBRIDLAB2 worker | heavy_rain 12 h |
| T152 | running | DESKTOP worker | normal 12 h |
| T153 | running | SAGBERG | westwind h12 → h24 |
| T156 | blocked, permission classifier | HYBRID_LAB | FluidNinja container |
| T144 | running | HYBRID_LAB | row never closed; its output was verified 22:29 |
| T137 | blocked | HYBRID_LAB | gate GEO-01 needs David |
| T132 | blocked | SAGBERG | gate |
| T113, T115, T116 | blocked | HYBRID_LAB | blocked by the 10-20 crash |
| T117, T120, T123 | blocked | HYBRID_LAB | relaunch OOM / lock |
| T071, T078 | blocked | — | from 10-18 |

  Old rows from 10-09 to 10-21 (T007, T016, T018, T032, T048, T050, T070, T076, T083, T087, T088, T101, T102) still say "running". These are stale labels, not live work.
- **Conclusion:** labfeed, the tunnel and the token-free monitors are alive. The coordinator session has been **silent since 03:51**. It did not pick up David's 03:47 and 03:48 queue lines. Whether its queue watcher is armed is **[unverified]**.

## 3. Flagged / gate

- **feed.flagged** has one block with these lines:
  - 03:38 T153 12 h set: **MOUTHS FAIL 31 rows** (Alpenrhein M3 M4, Argen M2, Bregenzer M5, nine small rivers M1–M3) and **drift FAIL** (15 m median swing 38°). Not uploaded. The show map keeps the approved heavy_rain look.
  - 01:30 T148 6 h set: MOUTHS FAIL (Alpenrhein M3+M4, Argen M2/M3, Bregenzer M5, small rivers M1–M3). **Uploaded** into NS_TrainsAll and NS_TrainsDensity, dial-free.
  - 00:12 HybridLab1 GATE FLAGGED `unified100_normal_setup_v10w`: 4 rows (P08 meta_all). MOUTHS-GEOMETRY was NOT CHECKED (mouth_checks.py exit 1). **This is the setup HYBRIDLAB1 is running now.**
  - 23:23 and 23:28 GATE FLAGGED `westwind_v10c_wedge_test4h_r16`: MOUTHS-GEOMETRY FAIL, 21 cells (M2/M7 on most rivers, Schussen M8).
  - 23:01 and 22:29 T145/T144 2 h set: MOUTHS FAIL 12/12 and 9/12. The Rhine still turns right (+61.5°), so the wedge cut did not remove the T127 turn.
- **Status in short:** every wedge or v10 set built so far is **MOUTHS FAIL**. Present them as tests, not as validated results.

## 4. VR setup: what the feed says

**Facts, with time:**
- 10-18 06:0x, T077 (open): VR test with Quest 3 over Link waited for "the Meta Horizon Link update + restart".
- 10-18 06:11, expert verdicts: forward shading + MSAA as the VR baseline, **no Nanite**, no engine migration before Thursday.
- 10-20, handoff rules: GPU cap 300 W and `t.MaxFPS 30` in the editor; never save Bodensee_Sim.
- 10-21 19:43, T140: buildings re-imported as 142 tiles in **L_BuildingsTiles, with Nanite**. L_BuildingsTiles was saved; **Bodensee_Sim was not**, so the sublevel's registration in Bodensee_Sim exists only in the running editor's memory.
- 10-22 01:09: ClusterNear/Far dials changed live (SETTINGS_EDITOR.md). 01:30: the 6 h test set was uploaded live. Whether either was saved is **[unverified]**.
- 10-22 02:45: David's own VR sessions ran at **10–16 fps** (by the log frame counter).
- 10-22 02:5x, T150 "VR SHOW for noon":
  - Bodensee_Show with game mode GameModeBase and a flying DefaultPawn.
  - NS_TrainsDensity shown in Play.
  - VR cvars: no Lumen or TSR in stereo.
  - Keys: 1–4 colour modes, 5 swap systems, 6 buildings, 7–9 boat teleports, V/B VR profile, F stats.
  - One FluidNinja container.
- 03:01: show keys written (`Content/Python/show_mode.py` plus DebugExecBindings in `DefaultInput.ini`, `.bak` backup). Touch joysticks switched off.
- 03:14: keys verified **in PIE**. PIE ran at 119 fps in a window, 91 fps at the VR pixel count (212 %), and 120 fps with TAA and Lumen off.
  - **The bindings need an editor relaunch** (`relaunch_show.sh`).
  - MORNING.md is in `_claude/tasks/T150/`.
- **Game mode and density in Play were NOT applied.** The coordinator wrote that these are "his ! line", denied 3 times.
- 03:16: T156 FluidNinja container worker denied by the permission classifier. The brief is at `_claude/tasks/T156/BRIEF_T156.md`.
- 03:18: editor relaunch not possible from that session. "Link device present (XRSP OK)."

**What can plausibly be "not working", each with its check:**

| # | possible cause | check |
|---|---|---|
| a | The keys and VR profile are not active yet because the editor has not been relaunched since 03:01 | press 1 or V in VR Preview: does anything change? |
| b | No lines in VR, or the wrong pawn: game mode and density in Play were not applied | MORNING.md step for the "!" line; World Settings → GameMode override in Bodensee_Show |
| c | 10–16 fps in the headset against 91 fps in PIE | press F for stats in VR: GPU or game thread bound? Is the V profile on? What does `t.MaxFPS` read? Is the 300 W cap on? |
| d | Nanite buildings (L_BuildingsTiles) under forward/VR | key 6 (buildings off), then compare fps |
| e | Link / runtime | Meta Horizon Link app sees the headset; was the Link update installed (T077)? VR Preview button enabled? |

**Questions only David can answer at the machine:**
1. What exactly fails: no picture in the headset, VR Preview greyed out, low fps, lines missing, or controls?
2. Which map was open at 02:45, Bodensee_Sim or Bodensee_Show?
3. Cable Link or Air Link? Has the Link app been updated since 10-18?
4. Did he run any MORNING.md line?
5. Is the 01:30 test set or the heavy_rain look the one to show?

## 5. Arrival checklist, first 20 minutes

a. **PC up?** Look at the screen of HYBRID_LAB (192.168.1.129). Is the dashboard reachable on the LAN (`http://192.168.1.129:8080/`)? Is the remote-desktop tool reachable? (The feed carries nothing on RustDesk or AnyDesk, **[unverified]**.) The tunnel was up at 04:10.

b. **Coordinator:** run `/clear`, then `/bodensee-resume`.
   - The newest handoff the dashboard serves is `HANDOFF_2026-10-20_1430.md`.
   - The 10-21 20:30 handoff should be at `_claude/HANDOFF_2026-10-21_2030.md` **[unverified, not served]**.
   - The new session must read the LOG from 02:53 (T150–T156) and T150's MORNING.md.
   - Then re-arm the queue watcher: `bash "_claude/tools/labdash/wait_queue.sh"` (per the 10-20 handoff).

c. **Commands:** `status`, then `watches`, then `tunnel` (OPUS_COMMANDS, names as given, **[unverified]**). Expect 4 monitors (T151, T152, T153 and HYBRIDLAB1's T102) and 2 picture loops (pids 79035 and 77506).

d. **Runs:** leave SAGBERG, HYBRIDLAB2 and DESKTOP running. All three are healthy. HYBRIDLAB1 ends at 04:28; whether T155 starts there is your word. **Stop nothing without naming the machine.**

e. **Editor: save or discard first**, before any relaunch for VR.
   - **Discard (relaunch without saving)** loses:
     - the live upload (01:30 6 h set; re-upload takes ~11 s + 11 s with upload.py / fill_density.py)
     - the 01:09 cluster dial change (re-applicable; it is recorded in SETTINGS_EDITOR.md)
     - the L_BuildingsTiles registration in Bodensee_Sim (re-add it in the Levels panel; the sublevel asset itself is saved)
     - whatever else changed since the last save **[unverified]**
   - **Save** breaks the standing "never save Bodensee_Sim" rule. It bakes the current instance state into the map and overwrites the 22:19 "sacred dials" save that T146 recovered from.
   - Before relaunching: close the Details panel on the Niagara actors. This was the cause of the 19:07 OOM (T122), and the T117/T120 relaunches died of OOM.

f. **VR, five checks in order:**
   1. Link app shows the headset; VR Preview is enabled.
   2. Relaunch with `relaunch_show.sh` (MORNING.md step 2).
   3. Run your "!" line for game mode and density in Play in Bodensee_Show.
   4. In the headset: V (VR profile), then F (stats); note the fps and whether GPU or game thread is the limit.
   5. If still below about 45 fps: 6 (buildings off), check `t.MaxFPS` and the GPU cap. FluidNinja (T156) only if there is time left. The brief is ready, and you have to run it as your own `claude --model opus` worker.

g. **What can be shown at noon:**
   - **Verified to exist:**
     - Bodensee_Show with the approved heavy_rain look (03:38 line)
     - show keys 1–9 / V / B / F, working in PIE (03:14), and in VR only after the relaunch
     - hourly mouth pictures in `<share>/runs_hourly/`: westwind_v10c_wedge_test12h h07–h12 (72 jpg), normal_v10c_wedge_test12h h00–h02+, heavy_rain_v10c_wedge_test12h h00–h01+ (more by noon)
     - `MAP_nest_dornbirn_4h.png`
     - `MAP_test2h_fields.png`
   - **Exists, but MOUTHS FAIL (label as tests):**
     - 6 h set (in the editor)
     - 12 h westwind set (on disk, not uploaded)
   - **Not before noon:** heavy_rain 12 h (ends ~12:05–12:31), FluidNinja container, any validated or "final" run.
   - **From the 10-18 schedule:** the Thursday item is "student workflow … presentation to the festival representative". **[unverified]** whether the 10-21 handoff's "Thursday noon" line says more.

## 6. Risks, ranked, one mitigation each

1. **The VR show is not playable in the headset** (10–16 fps at 02:45, bindings inactive, game mode not set). *Mitigation:* do checks f1–f5 before anything else. The fallback is the same show map in PIE or desktop, which was measured at 119 fps.
2. **The editor relaunch goes OOM or loses unsaved state.** *Mitigation:* make the save/discard decision (e) and close Details before `relaunch_show.sh`.
3. **The coordinator stays silent, and the permission classifier blocks the editor steps** (denied: editor fix, game mode, T156 worker, labfeed restart). *Mitigation:* `/clear` + `/bodensee-resume` at once. Type the "!" lines yourself rather than waiting for the agent.
4. **The science is shown as validated when every set is MOUTHS FAIL.** *Mitigation:* say "test runs, mouths not yet right" on screen and in what you say.
5. **heavy_rain 12 h is not finished by noon.** *Mitigation:* use its h01–h06 pictures (the coordinator planned this at 03:51).
6. **The dashboard depends on the PC tunnel.** *Mitigation:* if the PC reboots, check `tunnel.sh` before relying on the dashboard (502 = tunnel down).
7. **HYBRIDLAB1 is idle after 04:28.** *Mitigation:* say "start T155 on HYBRIDLAB1" if you want the 24 h westwind plume run.