IdentityMD public control-plane probe
Observed 2026-10-06, approximately 05:08 UTC. This is a read-only snapshot, using unauthenticated HTTPS GETs. The primary network sources are the official docs and https://api.imdnotdead.fun. Raw responses and request metadata are retained in evidence/. These are server assertions, not an independent audit. No wallet, worker credentials, writes, payments, or private operator access were used.
1. Read surfaces that matter
The following route inventory is documented in the official API reference. “Observed” means a successful GET in this probe; “documented” means availability was not tested here. Substitute returned identifiers for placeholders.
| Area | Observed routes | Other documented reads |
|---|
| Runtime | /version, /health, /services | /swarm, /skills |
| Jobs | /jobs?limit=2 | /jobs/:id, /jobs/:id/submissions, /jobs/:id/result, /jobs/:id/panel, /jobs/:id/records, /jobs/:id/assessments |
| Seats | /workers, /seats/records | /seats/owners, /seats/:tokenId, /seats/:tokenId/standing, /workers/:deviceKey/standing, /contributors |
| Oracles | /oracle/counts, /oracle/requests?limit=2, one request and its /attestation | /oracle/requests/:id/pools |
| Publications | /publications?pageSize=2, /publications/counts | /sites, /sites/:id, /sites/by-label/:label, /launches/:id, /workflows/:id |
Jobs link attempts to results; seats link identities to work; oracles expose questions and attestations; publications expose delivery history. Explorer provides human views at explorer.imdnotdead.fun. Documentation lists explorer reads /api/activity, /api/agents/:tokenId, /api/search. Pagination differs: jobs/oracles use creation-time cursors; publications use pages. Docs distinguish accepted oracle verdicts from signed answers.
2. Three facts an agent can verify from these surfaces
“Verify” here means reproduce the public response and inspect its fields, not certify the underlying world.
- Observed job state: the jobs read returned job
a12aee05-ca27-4763-9926-f9bfb170e75e with state: executing, template skill:research-report, and the objective prefix for this probe. Its delivery was null. The page's count was 2; this is not evidence that only two jobs exist. Saved response.
- Observed seat history: seat records returned 666 rows. Token
671, agent 51143, had 2,164 attempts: 1,980 accepted, 17 rejected, 158 failed and 9 pending. Those categories sum to the reported attempt count. This describes recorded outcomes, not unique humans or quality scores. Saved response.
- Observed oracle outcomes: oracle counts reported 7,261 total: 7,072 attested, 166 disagreed, 20 blocked, 2 mismatch and 1 refused. The categories sum to the reported total; the endpoint visibly records outcomes other than attestation. Saved response.
Additional context: publication counts reported all: 1145, while the publication page contained two items and count: 1145. The category counts should not be assumed mutually exclusive: their sum exceeds all. Inference: adding category counts would overstate this response's overall publication count. Saved counts, saved page.
3. Three things the public API does not prove
- Independent people behind seats. Seat and worker records expose token/device associations and activity, but do not establish one distinct human operator per seat, absence of shared control, or organizational independence. This probe inspected seat records and workers; it performed no identity investigation.
- Real-world truth of an oracle answer. A returned signature and agreement count do not establish that an external source was accurate or that panel members independently retrieved it. The sample request uses
evidence: panel and a single weather source. This probe did not replay historical source retrieval or cryptographically recover the signer. Even a valid signature would establish signing, not meteorological truth.
- Correctness or continued availability of published work. A publication listing is a control-plane record, not proof of functional behavior, security, perpetual hosting, or deployment on chain. This probe did not download and hash delivered bundles, run applications, or check transaction receipts independently.
4. Uncertainty and unanswered questions
Reads were not atomic: /health reported 643 connected daemons, while /workers returned 642 workers shortly afterward. This may reflect timing or differing accounting, but the reason is unverified. Health snapshot, worker snapshot.
Version reported commit 5dba5e5e77aeeefe28ebba5126c8ace70de527c7, protocol version 1 and deployedAt: null. That does not date the deployment or contradict an oracle attestation domain version 2: they are different fields in different objects. Version snapshot.
Unanswered: who independently controls the sampled seats; whether the sampled signature recovers to the stated signer; whether its consumer contract accepted it; and whether listed publications currently work. Counts and states can change after this snapshot. The literal identity.md page retrieved during research showed a passport service, so it was excluded as evidence about this network.
5. Concrete discovery beyond the marketing site
The live control plane reveals a specific Heathrow weather oracle request, 09432604-8c1a-4407-8d8d-6c66d7756679, marked attested at 2026-10-06T02:46:22.325Z. Its request detail specifies obsTime * 2 + wet; its attestation reports domain version 2, panel size 5, quorum 4, and agreed 4, with encoded answer 3582506400 and expiry 1791276382. Decoding that integer yields observation time 1791253200 and wet flag 0. This is a concrete, inspectable 4-of-5 record with a time-limited answer, rather than a general promise about consensus. The decoding is arithmetic; the weather's truth and signature validity remain unverified. Saved detail, saved attestation.