clean tests run: pi 1.5.0 model-list shim, panes checked by process, board r refreshes detail, office . accepted

docs/NEXT.md

@@ -154,3 +154,33 @@ after 30.5 s, then `200` in 21.7 s, so the command retries exactly once and says
 ## Open Paperclip tasks in this project
 
 One: `AUD-162`, a leftover test task in `todo` from the first day. Everything else is `done`.
+
+## Done since: the two clean tests were run, 2026-11-12
+
+Both are recorded step by step in `docs/CLEAN-TESTS.md`. Four defects came out of them and all
+four are fixed: the pi 1.5.0 model-list stream (which failed **every** agent run), panes restored
+by TUIOS after a reboot being "adopted" while running nothing, `r` on the board not refreshing the
+chosen task's detail, and `office .` being refused as if no folder had been given.
+
+Two open items came out of them, and neither is small:
+
+### A. Cost is always zero for `pi_local` runs
+
+`GET /api/issues/{id}/cost-summary` reports `costCents: 0`, `outputTokens: 0` over four runs of a
+task that really ran, and the run's `usageJson` is `null`. Pi's JSON stream carries token usage;
+the `pi_local` adapter is not capturing it. Until it does, `Project>costs` and the cost line on
+`Board>the chosen task` are honest but useless, and the cost limit in the operating rules cannot
+bite.
+
+### B. `ocr` reviews code, not Markdown
+
+A documentation-only diff gives `status: skipped`, `coverage.selected: []`. `bin/merge` handles
+that correctly, but `Work>review` should say *why* rather than "nothing reviewable in the change
+(0 files)", because the `change` tab next to it lists one file.
+
+### C. The suite needs to survive a reboot
+
+After a reboot TUIOS restores a session with a fresh shell in every pane and keeps the pane names.
+`board` now checks the pane's process instead of trusting the name. What is still missing is a
+single command that repairs this project after a reboot without a person looking at each pane —
+`board` is that command now, and it does rebuild, but nothing runs it automatically.