Pipeline Wiring

• The journal has a wiki page, ttrng3.github.io/Trade-Journal/wiki.html (Wiki button in the journal header). Each trading day’s bot and manual trades are compiled into linked pages (days, mistakes, setups, what-ifs, rules); a one-page weekly report follows on Saturdays. First compiled day 07/10.
• How it runs: a Mac job collects the bots’ day files to Drive 09 Trading/Trade Journal/Raw Records/Bots/ at 10:30 Mac local time, Tue–Sat (section 5); the existing Trade Journal routine at 11:00 Hanoi then adds a wiki step after the Webull sync. The wiki step runs on every routine run (Tue–Sat), quiet sessions included, and a wiki failure never blocks or undoes the journal’s own data push (Trade-Journal docs/nightly-sync.md). No new routine and no schedule change.
• Trade-Journal #14 1a9a42e, #15 5864b32, #16 1aa87e9, #17 7092f70. Runbook: Trade-Journal docs/nightly-sync.md, “The wiki step”.

• The master backup now reaches Drive (Trade-Journal #13, merge 502ae8e). Until 03/10 it lived only as a GitHub release. A Mac job now copies it to Backups/ each month and moves synced CSVs out of Raw Records/. October’s copy and the first move were done by hand the same day (recorded in the Trade-Journal PR #13 description). It is a new Mac dependency; see section 5.

• The journal page flags rule breaks (Trade-Journal #12, merge commit 1ee5b0d, 03/10). The trade log carries a badge per broken rule and a “Rule breaks only” filter; Reports has a Rules card; Settings has an R field that stays in the browser and never reaches the repo. Page behaviour only: the schedule, routine, data and preview id are unchanged, and the preview was rebuilt from the merged page the same day.

• Schedule moved 01:07 → 04:00 UTC (11:00 Hanoi), Tue–Sat. The Webull export lands in Raw Records/ around 08:30–10:00 Hanoi; the 08:07 fire found nothing on 25/09 and again on 26/09, and a manual run on 26/09 brought the new fills in. “The page did not update” in this pipeline means: check the CSV’s modified time against the run’s fire time before anything else.
• The Cowork preview is a new artifact. The old one was deleted on 26/09 by reading “no artifact link” as “no artifact” — the same mistake as 23/09 — and Ty reversed it within the hour. Rebuilt from the repo (fragment, then the 50 data files), routine repointed, README now reads One address, one preview: the Pages URL is the only link, the preview must exist, its URL lives only in the routine prompt. The rule applies to every row above.
• The page keeps a per-browser overlay. A CSV dropped on ttrng3.github.io/Trade-Journal/ used to vanish on reload (the snapshot alone was loaded). Imports, notes and settings made on the page now persist in that browser on top of the nightly snapshot, and prune themselves once the sync commits the same fills. The drop zone is also a click/tap/keyboard control — it only listened for a drag before.
• Omni-Audit preview rebuilt too, same afternoon. Same misreading on 25/09, same fix: new artifact, Part B7 restored in the monthly run, fallback repointed, docs rewritten to One address, one preview. Section 3 counts nine previews and 25 artifacts again.
• Pipeline Wiring became the ninth pipeline (Ty, 26/09). A Sunday 16:00 UTC routine runs tools/collect_status.py against the nine public repos, checks the nine previews with the Artifact tool, commits data/status.json + a heartbeat, and refreshes this preview. The “Next”, “Last heartbeat” and preview cells in section 1 now render from that file; the prose stays hand-written.
• Routine renamed Trade Journal daily sync (Drive CSVs → GitHub → Pages, preview after); report now says whether the newest CSV arrived after the fire.

• Repos: 12 → 14. famrec does not exist in the account. passbook, pipeline-wiring and a restored signal-desk-collector were missing from the count.
• Artifacts: 20 → 25, now 24. New since then: Omni Next Level, GDSH Turnaround v2, Passbook Thirty Visitors and two called Passbook Review. “Vitals Journal” is now titled Passbook. Pipeline Wiring now has a Pages site, so it moves to the preview list.
• Two preview labels were swapped. The Ops-Dashboard preview is “Dashboard Giám sát Ngân sách & Dòng tiền — KSNB”; the Omni-Audit preview is “Ecosystem Audit”. The previews were right; the names were on the wrong rows.
• Omni-Audit schedule. The page said 1st 09:00. The live system is the monthly system run on the 1st at 03:00, plus a fallback on the 2nd at 01:00. None of the routines I could find has cron 0 9 1 * *.
• “0 touch your Mac” was false. See section 5.
• Signal Desk was listed as broken. Its repo is back and the routine created sessions on 25/09.
• ECOPM was said to be two weeks behind (newest week 2026-09-W01). Its summary index ran to 2026-09-W03, the same week as OMNI. What was wrong: the 22/09 run wrote no detail files for W02 and W03, because the M365 connector returns only the TH summary sheet of these ~30 MB workbooks. Fixed 25/09 W02 and W03 backfilled from the source workbooks, and W01 rebuilt — commit 8fa5121, preview republished. The parser is now in the repo as tools/parse_sitecheck.py and runbook step 4 no longer claims the connector reads every sheet. A request for a lighter weekly file (no photos) has been drafted for Ty to send, so future weeks do not need a manual download.
• The routine-connection diagram still showed KB Monthly Update and a Monday publisher. Both are retired; section 4 is redrawn.

Every ECOPM detail file built before 25/09 stored each row’s STT (sequence number) in the status field for most rows. The page shows that number as the row’s status pill, so the “Tình trạng” column and its filter are wrong for most historical rows. September W01–W03 are now correct. Fixing the older weeks means re-reading their source workbooks; not done, awaiting your call.

Ty, 25/09: leave them for now. They fire once each, 23/11 and 11/12, and still hold the remote-devices tools. Come back to this before the first one fires.

Its owner was KB Monthly Update, which was folded into Monthly system run (1st, 03:00). The merged prompt kept the republish as Part C step 4 — but the 25/09 test run skipped it, and skipped the Ecosystem Health db-write with it.

The cause is worth knowing, because it will recur elsewhere. The run probed for the Artifact tool with ToolSearch, which searches only deferred tools. An already-attached tool returns “No matching deferred tools found” — so a present tool looks exactly like a missing one. The run read that as unavailable and dropped both artifact steps without flagging it. Artifact and ArtifactData were in its allowed_tools the whole time.

The prompt now states the tools are attached, forbids the ToolSearch probe, marks both steps mandatory, and reports each outcome on its own notification line. This page has no Pages site — the artifact is its only address, so a skipped step shows up nowhere. Next real run: 1 Oct.

Trashing a file does not revoke a credential. On 24/09 two files needed action at the provider:

• routine_fire_token.txt — an Anthropic sk-ant-oat01-… session token. Revoke it in your account settings.
• .github-token — 94 bytes, PAT-shaped, created 20/05. Revoke it if it is not one you already revoked.

Closed on Ty’s word, 25/09: a new token has been generated. One caution kept on the record: generating a new token does not switch off an old one. If either old token was never revoked or set to expire, it still works until it does.

A past session log shows the routine renamed “⚠ DELETE ME — retired Mac-bound Trade Journal” had a trigger id that the routine API returns not found for on 25/09, so it has been deleted.

On 25/09 the list call returned 20 rows plus a next-page cursor the tool cannot follow. Five routines were missing from it and were found only by fetching ids one by one: TMDV (enabled), the two Japan one-shots (enabled), the retired KSNB weekly (disabled) and a one-off check from 09/09 (disabled). I fetched all 31 ids found in Drive notes, the repos and past session logs: those 5 exist, 26 no longer do. A routine whose id was never written down anywhere could still be missing, but no recorded id is unaccounted for.