for page, its in by_page.items():
    code, out, _ = run_cmd(["adom-wiki", "issue", "list", page, "--json"])
    if code != 0:
        if "PAGE_NOT_FOUND" in (out or ""):      # <-- only whole-page deletion is handled
            for it in its:
                items.pop(f"{it['page']}#{it['num']}", None)
            changed = True
        continue                                  # transient failure -> correctly skip (never delete)
    try:
        j = json.loads(out)
        discs = {str(dd.get("number")): dd
                 for dd in (j.get("data", j) or {}).get("discussions", [])}
    except Exception:
        continue
    # >>> GAP: page listed fine, but a tracked issue NUMBER is absent from discs.
    for it in its:
        dd = discs.get(str(it["num"]))
        if not dd or dd.get("status") not in ("open", "closed"):
            continue    # <-- ghost item lands here every pass and is skipped forever

    for it in its:
        dd = discs.get(str(it["num"]))
        if dd is None:
            # Page listed fine but this issue number is gone (deleted or renumbered).
            # The list is authoritative, so this is a definitive absence, not a
            # transient failure -> drop the ghost so it stops inflating the board.
            items.pop(f"{it['page']}#{it['num']}", None)
            changed = True
            log_line(f"OK status sync: {it['page']}#{it['num']} dropped (issue not on page)")
            continue
        if dd.get("status") not in ("open", "closed"):
            continue
        ... # existing open/closed flip logic unchanged

# Handoff: wiki-notify-gchat — drop "ghost" issues (deleted/renumbered) during status sync

**Owner of this package:** Drew (`adom/wiki-notify-gchat`, published v1.4.0, auto-updates via adom-hook)
**File to change:** `bin/wiki-notify-gchat` — function `sync_issue_status()` (currently ~lines 259–319)
**Severity:** minor. Board over-counts "open" by however many ghost items have accumulated. No data loss, no crash.

---

## Symptom
The "Wiki Notifications" board card showed **30 open · 0 closed** when only **28** issues were actually open. Two tracked items — `adom/molecule-publish#164` and `#165` — no longer exist on the page (molecule-publish's live discussions are #1–#18; those were deleted/renumbered on the wiki after the board ingested them from email). They sat as permanent phantom "open" rows.

I hand-cleaned these two from `state.json` (took the `run.lock` first, backup at `state.json.bak-*`), so the live board is correct **right now**. This handoff is the durable code fix so it can't recur.

## Root cause
`sync_issue_status()` reconciles board items against the wiki DB via `adom-wiki issue list <page> --json`. Its cleanup logic only drops items when the **whole page** is gone:

```python
for page, its in by_page.items():
    code, out, _ = run_cmd(["adom-wiki", "issue", "list", page, "--json"])
    if code != 0:
        if "PAGE_NOT_FOUND" in (out or ""):      # <-- only whole-page deletion is handled
            for it in its:
                items.pop(f"{it['page']}#{it['num']}", None)
            changed = True
        continue                                  # transient failure -> correctly skip (never delete)
    try:
        j = json.loads(out)
        discs = {str(dd.get("number")): dd
                 for dd in (j.get("data", j) or {}).get("discussions", [])}
    except Exception:
        continue
    # >>> GAP: page listed fine, but a tracked issue NUMBER is absent from discs.
    for it in its:
        dd = discs.get(str(it["num"]))
        if not dd or dd.get("status") not in ("open", "closed"):
            continue    # <-- ghost item lands here every pass and is skipped forever
```

When the page lists successfully but a tracked `num` isn't in `discs` (issue-level delete, or the issue was renumbered), the item is silently `continue`-skipped — never flipped, never dropped. It lives forever as "open."

## The fix
When we have an **authoritative, complete** discussion list for the page (`code == 0`, JSON parsed), treat a tracked item whose `num` is absent from `discs` as gone → drop it (same as the page-deleted branch), with a log line.

Suggested change to the per-item loop:

```python
    for it in its:
        dd = discs.get(str(it["num"]))
        if dd is None:
            # Page listed fine but this issue number is gone (deleted or renumbered).
            # The list is authoritative, so this is a definitive absence, not a
            # transient failure -> drop the ghost so it stops inflating the board.
            items.pop(f"{it['page']}#{it['num']}", None)
            changed = True
            log_line(f"OK status sync: {it['page']}#{it['num']} dropped (issue not on page)")
            continue
        if dd.get("status") not in ("open", "closed"):
            continue
        ... # existing open/closed flip logic unchanged
```

## CRITICAL caveat the implementer MUST verify first — pagination
The "absent ⇒ deleted" inference is only safe if `adom-wiki issue list <page> --json` returns **all** discussions for the page, not a first page. If the endpoint paginates (e.g. default limit 20/50), a legitimately-present issue on page 2 would be wrongly dropped.

Before shipping:
1. Check whether `adom-wiki issue list --json` caps results. Find a page with >50 issues (or inspect the CLI/help). The `data.discussions` array must be the **full** set.
2. If it caps: page through it (or pass an explicit high `--limit`) and build `discs` from the **complete** union before applying the drop. Do **not** apply the drop against a truncated list.
3. Confirm `issue list` includes **closed** issues, not just open. (It does today: `adom/molecule-publish` returned status counts `{open:7, closed:11}` — closed are present. Keep this assumption tested, since the whole close-detection path depends on it.)

Also keep the existing safety property intact: **transient failures (network/auth, `code != 0`) must never delete state** — only a clean `code == 0` + parsed list authorizes a drop. Don't loosen that.

## Optional hardening (nice-to-have, not required)
- **Re-ingest guard:** the ghost originally arrived via an email whose URL used a number that doesn't match the API's per-page `number`. If email-URL numbering and API `number` can diverge, dropped ghosts could re-ingest from a re-seen email. The `processed` email-id dedup set already blocks re-processing the *same* email, so this is low-risk — but worth a note in the PR if you see URL-vs-API number skew.
- **Age gate (belt-and-suspenders):** only drop if the item is older than, say, 10 minutes, to avoid racing an issue created seconds ago that a stale read hasn't surfaced. Given the list is live (not cached), this is likely unnecessary — implementer's call.

## Test plan
1. **Unit-ish:** seed `state.json` with a fake item `adom/<realpage>#99999` (a number that doesn't exist), run `wiki-notify-gchat run`, assert the item is gone from `state.json` and a `dropped (issue not on page)` line is logged.
2. **No-false-drop:** seed a real open issue, run, assert it is **retained** with `status: open`.
3. **Transient safety:** temporarily break `adom-wiki` (e.g. bad PATH/creds so `code != 0`), run, assert **no** items are dropped.
4. **Pagination (if applicable):** on a page with more issues than one API page, assert an issue only reachable on page 2 is retained.

## Ship
- Bump `package.json` version (patch), update README/changelog with a one-liner ("status sync now drops issues deleted/renumbered off a page, not just whole-page deletions").
- Republish to `adom/wiki-notify-gchat`. It auto-updates on installed instances via adom-hook.

## Context / provenance
Diagnosed 2026-08-05. The board also legitimately shows "0 closed" most of the time by design — closed items are purged ~1 day after closing (`purgeHour`). And the board can look "frozen" when idle because `--quiet` runs emit no log line on a no-op; that's expected, not a fault. Neither of those is a bug — the ghost-item drop is the only real defect here.