Day 61 of 62 · Saturday · Learning

The event feed - what your pipeline is doing right nowConcept

Every time JARBUS finishes a step it appends one JSON line to a dated event log. That append-only tape is what the live dashboard turns into a 'Recently processed' lane — the thing STATUS.md, a snapshot, can never be.

Catch-up progress
61/62
Why this matters to you

You asked (on the 8-22 card) what a dashboard does that STATUS.md can't. The event feed is a big part of the answer: STATUS is a picture you regenerate, the feed is a running record of what actually happened and when — and it's already flowing whether or not you're looking.

The stream lives at PDB/data/jarbus-events/YYYY-MM-DD.jsonl — one file per day, one JSON object per line. Each line is a small, honest record: {ts, type, slug, summary, next_move, source}. Today's lines include two feedback_processed events at 13:01:51 (skill-audit and screens-batch) — the tape doesn't summarize or interpret, it just says 'this happened, at this second, and here's the next move it implies.' Append-only means it's never rewritten; you can always scroll back to what the system did on any day.

pipeline_view.py reads that stream into a local live dashboard at http://localhost:8090. The feed becomes the 'Recently processed' lane; alongside it the dashboard reads FOR TOM/TTR/ for 'Your move,' active-builds/<slug>.pid for in-flight builds, and the prd/asset/live folders for each project's stage. That's the dashboard's real job over a flat file: it shows state (what's waiting) and history (what happened) and action (approve a build in place) on one surface, live.

The published cousin is Mission Control — RESET-2026's Phase 3 dashboard at PDB.zorc.app. Same underlying data, plus the reconcile view you scoped on 8-22: it puts the alert ledger next to the health check so a false-green (a job reading OK while its incident is open — the 8-20 lesson) can't hide. The event feed is the raw material; Mission Control is the operator's console built on top of it.

Worked example

Read the last three things your pipeline did — straight from the tape.

tail -3 "/Users/tom/Claude/PDB/data/jarbus-events/$(date +%F).jsonl" 2>/dev/null \
  || tail -3 "$(ls -t /Users/tom/Claude/PDB/data/jarbus-events/*.jsonl | head -1)"

# what the dashboard reads, in its own words:
sed -n '1,12p' /Users/tom/Claude/JARBUS/scripts/pipeline_view.py
▶ Do it now
  1. Run the tail above and read the last three lines — for each, note the `type`, the `summary`, and the `next_move`.
  2. Pick one event and find its `next_move`; if it names a decision, that's exactly what the dashboard would surface to you as a card.
  3. Start the dashboard if it isn't running (pipeline_view.py, port 8090) and confirm the same events appear in the 'Recently processed' lane — the file and the surface are the same truth.

Gotchas

Go deeper: Google SRE Workbook: Monitoring
One-card takeaway

The event feed is an append-only tape of what your system actually did; a dashboard's value is fusing that history with live state and action on one surface — the three things a snapshot file can't do.