← Back to MuseHub
Weekly cost vs. concrete output — Friday, September 25, 2026 (12:15 ET)
Plain English. Verified numbers are labeled verified; everything else is an
estimate and labeled as one. Repairs, check-ins, and routine operations are
NOT counted as output — the wins bar stands.
What the team costs
Verified: Ivy's hourly check-ins that logged cost: 46 beats from Sep 18–25
total $141.97, about $3.09 per check-in. That is the only per-run cost
I can verify from records.
Not verified: the other seats' check-ins and the dozens of small scheduled
jobs don't have per-run cost logging I can check. Recent Ivy check-ins also
stopped logging their cost (a known logging bug), so even Ivy's current daily
burn is unverified.
Honest estimate: at ~$3 per check-in and ~24 check-ins a day, Ivy's seat
alone runs roughly $70/day, ~$500/week. Yale runs the same cadence, so the two
OpenClaw seats together are plausibly ~$1,000/week. The small jobs (5–15 minute
ticks) are individually cheap but there are ~20 of them; their combined cost
is unmeasured. A full verified weekly total does not exist — building per-run
cost logging for every job is filed as a follow-up below.
What the team concretely produced this week
(Concrete = a thing that exists and can be pointed at. Not meetings, not
rules about rules, not "we kept the lights on.")
- Issue #63 groundwork — Yale's baseline: inventory of 24 candidate paths,
backups + hashes of 15 files, rollback procedure, canary proposal, diagnosis
hypotheses. Real files in ~/workspace/agents/yale/checks/issue-63-baseline-20260921/.
The replacement documents themselves are NOT deployed — that is still owed.
- Feed-wind pipeline output — 23 researched claims, 12 discussion posts
(Sep 16–20). Then dormant 5 days. Verdict: FIX/REVIEW, Sunday decides.
- The false-kill correction — this morning I wrongly killed two feed jobs
on a false "0 entries" claim, then verified the real data (23 claims),
restored both jobs, and logged the correction. That is a repair, not a win —
listed here only because AK asked for a full accounting.
- Revival kits — weekly rebuilds ran (Wed/Sun); 4 newest kits kept.
- 1Password nightly tidy — running per AK's auto-tidy choice.
- Machine health cockpit + inventory — collecting and refreshing per AK's request.
What remains unproven
- The full weekly cost (only Ivy's check-ins are verified).
- Whether the feed-wind pipeline earns its keep (Sunday review).
- Whether the GitHub notify relay recovers (5 identities failing polls as of today 10:15 ET).
- Issue #63's replacement documents deployed and live.
- The Cornell quarantine resolution.
Follow-ups filed (register, with fuses)
- F-COST-01: per-run cost logging for every scheduled job — owner: muse, fuse Sun Sep 27 12:00 ET.
- F-KILL-01: remove the 6 KILL-verdict dormant job definitions — owner: muse, fuse today 18:00 ET.
- F-63-01: Issue #63 replacement documents deployed + personally verified — owner: muse, fuse Sat Sep 26 12:00 ET.
- F-FEED-01: main-seat feed digest refresh (was >24h stale) — owner: muse, fuse today 18:00 ET.
← Back to MuseHub