Listening for events…
← All Docs

TerraPulse weekly cadence

A working schedule for Claude's operational rhythm on TerraPulse, reporting up to Brad and Mike as direct supervisors. One focused block per weekday — anything more becomes noise that erodes trust in the reports. Monetization on X is the north star, so professionalism in the public-facing surface is non-negotiable.

Monday — Data integrity audit (~30 min)

Run the operational sweep. Deliverable is a written one-pager into the issue tracker (or a comment on a standing weekly issue) so it's auditable.

  • Loader-run failures and stale-alert backlog from the past 7 days
  • Schema-drift events in loader_runs
  • Zero-yield AutoSense sources (>14 days, no rows) — propose demotion
  • Tag coverage % and metrics missing a tag rule
  • Geo enrichment coverage (rows missing country/state)
  • File or update GitHub issues for anything that needs human action. Nothing dramatic — short, factual.

Tuesday — Editorial review (~20 min)

This is the one most exposed to public scrutiny and the source of monetizable content.

  • Walk every PMA workspace touched in the past week: figures legible? Captions match results.json? Conclusions match the data?
  • Check V1-shrinkage-flagged papers — any due a V2 yet?
  • Verify each shipped paper has the right co-authors, DOI/issue link, and revtex compiles clean.
  • Flag candidate findings for the week's X post (see Friday).

Wednesday — Public surface QA (~20 min)

We're judged on what's at terrapulse.info, not on the backend.

  • Visit the site as an unauthenticated user. Homepage stats current? Recent observations populated? Explorer renders?
  • Smoke test /api/v1/health + 2–3 sample endpoints from outside the box (curl).
  • Latest paper renders correctly on its detail page; figures load; PDF link works.
  • Spelling/typography pass on anything new this week. Brand polish is cheap and noticed.
  • File issues for bugs; fix obvious typo-class stuff directly.

Thursday — Infrastructure & cost (~30 min)

Anticipates the crises before they're crises (the dedup-query incident on 2026-05-06 is the prototype).

  • Postgres: top-10 table sizes, index bloat, slow-query log review, dead-tuple ratio
  • /mnt/ursa disk pressure trend
  • Past-week loader_runs summary: success rate by source, p95 duration
  • Verify the last backup is recent AND restorable (restore should be tested at least monthly)
  • Re-evaluate AutoSense sweep batch size vs scheduler-tick budget

Friday — Brand readout (~30 min)

The monetization surface. Treat it like a weekly business report, not engineering chatter.

  • Draft 1–2 X-ready posts: a finding from the week's papers, or a live event we captured (Enid EF-4 was the template). Include a figure. Tag dataset, station, source.
  • One-page weekly digest committed to docs/digests/YYYY-MM-DD.md: papers shipped, issues closed, sources added, observation count, any incidents.
  • Update the memory checkpoint so next Monday starts cleanly.
  • Look at next-week's calendar of obvious deadlines (token renewals, conference dates, paper-cycle slots) and surface them to you before they bite.

Principles I hold myself to

  1. No silent weeks. If a bucket finds nothing notable, the deliverable is "checked, nothing flagged, here's the metric trend." Silence reads as inattention.
  2. Every finding becomes an issue or a one-line note. The audit trail is what makes this professional vs hobbyist. You and Brad should be able to scroll back six months and see what was checked when.
  3. No editorial drift on the papers. I will never upgrade an INCONCLUSIVE to H1 to make the week look better. The brand depends on the V-series being honest.

What I need from you

  • A standing GitHub issue per week (Weekly Audit YYYY-WW) where I drop the five reports as comments. Cheap to set up; I can auto-create them on Monday.
  • Permission to file follow-up issues without asking, capped at ~5 per week (over that, I bundle and ask first).
  • Confirmation on whether X drafts land in private Slack/email or open as a PR against a social/ directory for review.

Possible automation

This cadence can be wired up as a /weekly-audit skill so you can invoke the whole sweep with one command, or kept as written instructions in CLAUDE.md and run by checklist.

Live Feed