Finish the spec status update (previous commit only applied partially)

The prior commit's edit to the open-questions section silently no-opped
on a mismatched target, so the spec still carried the pre-implementation
questions and claimed the duplicate "Yard work" task was outstanding
after it had been deleted.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AwnALZL6bD4VBtATDVv3B

1 files changed, 32 insertions, 9 deletions

diff --git a/docs/superpowers/specs/2026-09-02-pull-based-punch-list-design.md b/docs/superpowers/specs/2026-09-02-pull-based-punch-list-design.md
index f9d2fb7..5ad5beb 100644
--- a/docs/superpowers/specs/2026-09-02-pull-based-punch-list-design.md
+++ b/docs/superpowers/specs/2026-09-02-pull-based-punch-list-design.md
@@ -29,10 +29,9 @@ Three defects and one design flaw:
 3. **Buckets never got a pool.** One bucket exists ("Yard Work", cycle 14d,
    pick 3) holding exactly 3 items. "Pick 3 of 3" makes the rotation a no-op —
    everything is always active. A duplicate free-standing "Yard work" task does
-   the same job in parallel. *Pool populated 2026-09-03 to 22 items (3 active +
-   19 dormant), so the rotation is real. The duplicate free-standing "Yard work"
-   recurring task is still there — deleting it ends its series, so it needs an
-   explicit call.*
+   the same job in parallel. *Fixed 2026-09-03: pool populated to 22 items, cycle
+   retuned to pick 2 every 4 days, and the duplicate free-standing "Yard work"
+   recurring task deleted.*
 
 4. **`due_date` is doing two incompatible jobs** — the design flaw underneath the
    complaint. It means both *"this is a hard deadline"* (Bills, meds, T-ball
@@ -108,8 +107,32 @@ back to the backlog. Carry-over count is the honest signal for over-committing.
 - Existing dated free-standing tasks: unchanged, still push. Most are real
   deadlines already.
 
-## Open questions
-
-- Should a bucket's active items auto-commit at cycle start, or always be
-  pulled? Leaning pulled, for consistency with the one rule.
-- Does the widget mirror TODAY exactly? Probably yes — the point is one list.
+## What shipped (1477ee5, 2026-09-03)
+
+- `committed_date` on `native_tasks` (migration 031), with
+  Set/Clear/GetCommittedTasks in the store.
+- Chain steps no longer stamped with a due date on unlock, and excluded from the
+  undated list so they don't fall through to it instead. Migration 031 cleared
+  the auto-stamped dates from the 8 open steps.
+- `BuildTimeline` includes tasks committed to today, deduped against rows already
+  present for another reason. The widget inherits this — it renders from
+  `BuildTimeline`.
+- "+ today" / "on today ×" controls on chain cards and active bucket items.
+
+Verified against production after deploy: 0 open chain steps still carry a due
+date; an uncommitted chain head is absent from the widget payload; committing it
+makes it appear; clearing it removes it again.
+
+## Still open
+
+- **Carry-over is unbuilt.** A task committed yesterday and not finished simply
+  stops appearing rather than showing as carried. Needs a decision: auto-roll to
+  today, or surface a "yesterday's unfinished" prompt during review.
+- **No dedicated review surface.** Pulling happens from the Chains and Buckets
+  sections of the Tasks tab, which works but isn't the single "plan the day"
+  screen described above.
+- **Buckets still push.** Active bucket items keep `due_date = now + cycle_days`,
+  so they arrive on the day's list on their own rather than by being pulled —
+  inconsistent with the one rule, though mild at 2 items per 4 days.
+- **No commit control on ordinary undated backlog tasks** — only chain steps and
+  active bucket items have one so far.