rdmsm4x — Desktop ticket ingest, record repair, and first real publish to dev.ecs0.net

When: 2026-08-30 12:23–12:35 EDT · Host: rdmsm4x · Author: claude@rdmsm4x Scope: rdmsm4x only. ~/Desktop, ~/dev/issues, ~/dev/sites/dev.ecs0.net.

Summary

Organized 92 loose ticket reports off the Desktop into ~/Desktop/issues/, repaired four ambiguous ticket records, fixed three exporter defects, and discovered that no ticket had ever actually been published to dev.ecs0.net — the export target was not part of the website. /issues/ now returns 200 with all 85 tickets.

What changed

1. Desktop organized

• ~/Desktop/issues/tickets/ — 85 ticket pages (one per live ticket)
• ~/Desktop/issues/dashboard.html — the rollup dashboard
• ~/Desktop/issues/reports/fleet-sync-audit-report.html
• Zero loose ticket HTML files remain on the Desktop.

2. Fixtures kept out of the record

18 Desktop reports were QA fixtures ("CLI Test Feature", "Test rfc", "Test ui", "Test doc", "CHALLENGER-TEST", empty titles) from the same family a prior session archived. They were not ingested. Moved to ~/dev/issues/archive/test-fixtures-20260830/html-desktop/. Also archived: fleet-lifecycle-dashboard/ (316 mixed real+fixture exports) and a byte-identical duplicate of the audit report.

3. Record repair — four ambiguous tickets

ISSUE-20260830-26 reported four tickets simultaneously open and resolved and said choosing a copy needed project judgement. Resolved per ticket:

• ISSUE-20260829-02/03/04 — the open/ copies were not stale. They carried parent links and un-mangled comments written after resolution; the resolved/ copies carried the closeout. Merged both sides. Deleting the open copies as duplicates would have silently dropped three parent links and broken epic hierarchies.
• INC-20260829-01 — the open copy was itself a fixture (all template placeholders, children pointing at archived fixture FEATs) that collided with a real ticket ID. Archived; the resolved copy is authoritative.
• Dropped 6 dangling references to archived fixtures across 5 tickets.

4. bin/ticket fixes (backup: bin/ticket.bak-20260830)

• ISSUE-20260830-27 — Desktop export went to ~/Desktop root, re-cluttering it on every ticket. Now ~/Desktop/issues/tickets/; dashboard to ~/Desktop/issues/dashboard.html.
• ISSUE-20260830-26 — resolve now removes every open/ copy of an ID, not just the same-slug one.
• ISSUE-20260830-33 (filed by a peer against the above fix, valid) — the cleanup sat behind the old_path.exists() guard, unreachable in the index-drift case it defended. resolve now falls back to globbing open/ then resolved/ and adopts the on-disk file.

5. ISSUE-20260830-34 — the publish target was never served

bin/ticket exported to sites/dev.ecs0.net/issues/. That directory is not part of the website: infra/compose.yaml:37 mounts ../wiki as the nginx docroot, and build_site.py writes only into wiki/. There was no wiki/issues at all — never tracked in git, absent from the source manifest. Every ticket page for days was written somewhere unreachable while the CLI printed a success line naming the path.

This corrects ISSUE-20260830-27's claim that 268 fixture pages "were live on the real site." They were never served; the fixture exposure was not public.

Fixed: SITES_ISSUES_DIR repointed to wiki/issues; 85 pages exported; the orphan tree archived to issues/archive/never-served-sites-issues-20260830.

6. ISSUE-20260830-35 — my own fix had a data-loss path (peer-caught)

A peer probed the ISSUE-26 fix and found it unlinked every stray open/ copy unconditionally. Content existing only in a stray was destroyed — precisely the content I had hand-merged for ISSUE-20260829-02/03/04 one commit earlier. The commit message documented the merge while the code discarded.

Fixed: a stray byte-identical to the resolved record is removed (nothing can be lost); one that differs is moved to archive/superseded-open-copies/<slug>-<ts>.md with a warning naming the file. open/ is still cleared, so status stays unambiguous — that only needs a move, not a delete. Added the missing import shutil.

Verified by replicating the reporter's probe: the unique marker now survives in 1 file instead of 0, open/ empty, one resolved copy.

Verification

curl -H 'Host: dev.ecs0.net' http://127.0.0.1:8788/issues/                    -> 200
curl -H 'Host: dev.ecs0.net' .../issues/ISSUE-20260830-34.html                -> 200
curl -H 'Host: dev.ecs0.net' .../issues/EPIC-20260830-03.html                 -> 200
served index lists 85 ticket IDs
store(87) == published site(87) == ~/Desktop/issues/tickets(87); 0 ambiguous
tickets in both open/ and resolved/: 0

Resolve fixes proven in a HOME-jailed probe store (canonical untouched): created a ticket, renamed its open/ file to a different slug without reindexing, ran resolve -> "(index was stale; adopted ...)", open/ empty, exactly one resolved copy.

Verification trap: curl 127.0.0.1:8788 without a Host header hits the gateway's catch-all (server_name _; location / { return 404; }) and 404s every path. Public URLs return 302 to Cloudflare Access before the origin is consulted. Both mask real origin errors — always use -H 'Host: dev.ecs0.net'.

Backups / undo

• ~/dev/_backups/issues-index-20260830-1229-premerge.sqlite
• ~/dev/issues/bin/ticket.bak-20260830
• cd ~/dev/issues && git revert ced12b1 && ticket reindex (plus the ISSUE-35 follow-up commit)
• cd ~/dev/sites/dev.ecs0.net && git revert fcd185d
• Nothing was deleted — every removal was a move into ~/dev/issues/archive/.

Open for the owner

• build_site.py prints "build complete and verified" without fetching any URL, so it cannot fail on a 404. That is why this was silent for days. Not fixed here.
• wiki/issues/fleet-sync-audit-report.html lists 38 fixture tickets as real. It is a legitimate historical report also catalogued in the wiki; left in place, flagged.