The worker-health report is done: I fetched a live snapshot from GET https://api.imd.fun/workers at asOf 2026-09-25T05:50:24Z (HTTP 200) and built the report from it.

Metric	Value
Unique seats (deduped by tokenId)	342 (342 records, no duplicates, none missing a tokenId)
Online	342
Stale (heartbeat older than 2h, or missing)	0. Every heartbeat is between 05:50:05Z and 05:50:24Z.
daemonVersion histogram	0.1.0+5bfa8261: 313, +79f4f4d5: 6, +ff3932db: 5, +8f011f3d: 4, +cff23c39: 4, +61d04d62: 3, +852666b2: 3, +50bb666f: 2, +aa8ff6ee: 2 (sum 342)
Seats not on the majority build (+5bfa8261)	29, each listed with its tokenId and version

Online rule: a seat counts as online if working > 0 or its lastHeartbeatAt is within the 15 minutes before asOf. The API sends working as a number (0 or 1), not true/false. Only one seat (tokenId 1943) had working > 0; all 342 were online because of their heartbeats.

Caveats, stated in the report:

• /workers is a live snapshot, not a historical archive.
• Every record shows connectedHere: true and a heartbeat seconds old. That suggests the endpoint only lists currently connected daemons. If so, the stale list will almost always be empty, because offline seats drop out of the list rather than showing an old heartbeat. This is an inference.
• The earliest connectedAt is 04:46:29Z, so the connections may have been reset (for example by a server restart) around then.
• Unanswered: the total number of seats including offline ones, and whether the builds are newer or older than each other.

The report separates facts, inferences and unanswered questions under their own headings.

Checks: I ran these locally and they all passed:

• The JSON parses.
• The histogram sums to uniqueSeats.
• The markdown totals, histogram rows and the 29-seat list match the JSON exactly.
• Every tokenId and version in the report appears in the raw response.

The verifier only checks files, and nobody else has reviewed the numbers.

Files:

• WORKER_HEALTH.md and WORKER_HEALTH.json: the deliverables.
• WORKER_HEALTH_README.md: the question and the limits.
• data/workers-snapshot.json: the unmodified API response, with fetch details in data/workers-snapshot.meta.json.
• scripts/worker_health.py: the generator. --as-of 2026-09-25T05:50:24Z rebuilds the report offline from the saved response; --fetch takes a new snapshot.

These are committed as 9dd0b60. Git had no user name or email set, so I set your email as author and committer for that one commit only. I didn't change the git config.

artifacts/report.md is a copy of WORKER_HEALTH.md. It's untracked, as required: the repo already excludes artifacts/ from git, which is why the committed files are at the top level instead.