I've written the one-page digest to artifacts/report.md, along with a script that rebuilds it on each run. The figures come from two runs about 6 minutes apart (06:58 and 07:04 UTC, 2 Oct 2026), because no earlier run existed to compare against.

Current figures and changes since the first run:

Figure	Now	Change
Seats online	549	unchanged
Seats enrolled	571	unchanged
Tasks accepted, last 24 h	60,011	+1
Paid orders	530	unchanged
Last order paid	01:53 UTC, about 5.2 h before the run	unchanged

Top five seats by accepted work: #355 (3,174), #475, #527, #1207 and #1763. The order didn't change between the two runs.

• Per NFT, not per device: the contributors list has one row per device, and some NFTs run several devices (642 rows for 579 NFTs). I added up the rows for each NFT. That's why #355 is first: its two devices have 1,926 and 1,248 accepted, and neither alone would be in the top five.
• Inferences: three wallets own all five top seats, more orders have expired (777) than been paid (530), and about 96% of enrolled seats were online.
• Uncertainty: "online" and "enrolled" are my reading of the field names connectedDaemons and activeEnrollments; the API doesn't define them. The contributors endpoint sometimes serves a cached copy marked stale, and its counts are all-time, not 24-hour.
• Noise: while I polled the health endpoint every 30 seconds, seats online dropped to 523 once and returned to 549, and the 24-hour count moved between 60,010 and 60,012. A 6-minute comparison mostly shows that noise, not a trend.
• Open questions: what an order buys, whether expired orders are retried, and how acceptance is decided. The report repeats what the API says; I didn't check it against anything else, such as on-chain records, and no one else has reviewed it.

What I added:

• tools/swarm-digest.mjs: each run fetches both endpoints, saves the raw responses to data/snapshots/, compares with the latest earlier snapshot and writes the report. It runs on Node with no dependencies, and an --offline mode rebuilds the report from two saved snapshots.
• README.md: the question, which API field feeds each figure, and the limits.
• Two snapshot files in data/snapshots/; the next run will compare against the second one.

A scratch test of the per-NFT summing and the comparison logic passes. Nothing is committed: Git has no author name or email set, and I'm not allowed to change .git/, so README.md, tools/ and data/ are untracked. artifacts/report.md is left untracked as required; the workspace's own Git exclude rules already ignore artifacts/. Someone with a Git identity needs to commit the other files.