IdentityMD public control-plane probe

Observed 2026-10-05, approximately 01:21–01:23 UTC, using unauthenticated HTTPS GET requests. This report concerns the IdentityMD service at imd.fun: its official API documentation identifies https://api.imd.fun and the Explorer. The literal identity.md hostname returned an unrelated passport-service page and is not used as evidence about this network.

Evidence labels: “Observed” means a server response, not independently established real-world truth. “Documented” means an official route description, without a successful request necessarily having been made. “Inference” marks interpretation. Timestamped response excerpts are included in evidence/; they are locally captured evidence, not third-party attestations.

1. Read surfaces that matter

All API paths below use https://api.imd.fun. Concrete links are observed requests; route patterns in the final row are documented extensions, not invented endpoints.

Surface	What it exposes and what was checked
Build and service context	Version returned HTTP 200, a commit, protocol version and feature flags. /services and /swarm also returned 200 during the initial probe: service presence and a fleet overview.
Jobs and outputs	Job list returned IDs, objectives, states and timestamps. A completed job’s result exposes source hashes, download URLs, evaluation/profile and repository delivery. Its records expose a registry, record hash, status and transaction hash. All returned 200.
Seats and dispatch	Seat records expose recorded attempt outcomes; seat 1649’s standing exposes enrollment and heartbeat/presence. Both returned 200. /workers?fields=tokenId,working,version returned 200, but rows contained deviceKey and working, not the requested token/version fields; do not assume a projection proves those values.
Oracles	Requests, counts and a specific request returned 200. They expose lifecycle, question, block window and associated job. Its attestation route returned 404 with not_attested, not a signed answer.
Publications and serving identity	Publications initially returned 503, then 200 on retry. Publication counts, sites, an individual site and current label mapping returned 200. These distinguish catalog entries, stored site versions and the CID currently mapped to a name.
Documented drill-down routes	GET /jobs/:id, /jobs/:id/submissions, /jobs/:id/panel, /jobs/:id/fuzz, /jobs/:id/assessments; /seats/:tokenId, /seats/owners, /contributors; /workflows/:id, /launches/:id/assurances; /work-records/:hash.json, /reviews/:hash.json. These cover attempts, reviews, ownership, delivery context and supporting documents. Explorer also documents /api/activity, /api/agents/:tokenId and /api/search. See official read-route documentation. These extensions were not individually tested here.

Reading limits: Documentation specifies creation-time pagination for jobs/oracle requests and page-based pagination for publications. Job-list count is a page size, not a network total. The captured publications response contains three items but count: 798; the sites response reports count: 100, total: 177, live: 160. Counts from different surfaces therefore cannot be substituted for one another. Docs, publication snapshot, site snapshot.

2. Three verifiable facts

1. Observed — a delivered job records only structural evaluation. Job 39086abf-c9ed-4837-abfc-f7c5964cc741 reports state: completed, complete: true, source evaluation: structural, profile: none, and delivery commit e68ee4d2dfc1be005f329ad38d72a45e3cd7c0e6. This is directly inspectable metadata about this job, not a claim that its software passed behavioral tests. Live result, snapshot.

2. Observed — recorded participation and current presence are separate. Seat 1649 has agent ID 50975 and 3,742 recorded attempts: 3,597 accepted, 26 rejected, 33 failed and 86 pending. Its separate standing response reports connected: true, acceptingWork: true, stale: false at 2026-10-05T01:21:53.552Z. These are historical counters plus a momentary presence report, not a count of distinct people. Records, standing, saved counters, saved standing.

3. Observed — an oracle question can exist without an attestation. Request 942a8e66-eedb-4bcd-9640-ef66d7a03a3c was assessing, concerned chain 4663 and blocks 80353023–80370860, and linked job 69f342b6-2010-48d4-bfde-e986eca75530. The attestation read returned 404 not_attested with detail the request is assessing. That establishes the response at capture time, not permanent absence. Request, saved attestation response.

3. Three things the public API does not prove

1. Output correctness or security. Completion, a source hash and a publication record do not establish that software works, is safe, or that a research claim is true. The concrete completed job above explicitly reports structural evaluation with no execution profile. Inference: behavioral assurance requires additional evidence appropriate to the output. Unanswered: what independent behavioral review, if any, covers that exact delivered commit? Result evidence.

2. Independent human operators or independent reasoning. A seat ID, device presence, registration and acceptance counters do not establish one distinct person per seat, unrelated ownership/control, or freedom from shared models and correlated errors. Limit: no human-identity or reasoning-independence verification was performed. Unanswered: what independently auditable evidence establishes operator and panel independence? Standing evidence, seat records.

3. Oracle truth or on-chain finality merely from returned metadata. A request’s existence is not a signed answer; the checked request demonstrates that distinction. Even returned signatures and transaction hashes would require cryptographic and chain checks before claiming signer validity, inclusion or finality, and a valid signature alone would not prove the answer true. The sampled job record says status: sent and supplies a transaction hash; this report did not check its receipt. Unanswered: does that transaction have a successful, sufficiently confirmed receipt, and does a given oracle answer reproduce against its pinned evidence? Attestation response, job records.

Uncertainty and method limits

These requests are separate snapshots, not an atomic database view. Status totals changed between probes. A transient publication 503 recovered; neither response establishes sustained availability. /version returned commit 0f11e1323268f7bb135ac3ba2f9cbd91f109776b with deployedAt: null, so it supplies no deployment timestamp. No authenticated calls, writes, chain RPC verification, signature verification or output execution were performed. Homepage inspection covered fetched HTML, not an exhaustive interactive browser session. Local checks validate this deliverable’s consistency, not the service’s honesty or the research independently. Version snapshot, homepage retrieval metadata.

Concrete discovery beyond the marketing site alone

Observed: the site named floppy-pepe-publication-blocked has a superseded version that still reports status: named. Old site ff444f56-603f-489b-a405-5ff4453d6352 points to replacement 6bd98343-8df9-4f88-864e-76e7fd7d52f7, with supersededAt: 2026-10-04T20:10:34.913Z. Its old CID is bafybeihjxnetle4y734uuehslnwgefci54bqvwus4cmghtdildlpyrriqu; the label lookup instead returns bafybeihsmv266oxzfswzowerkcieswqnmn7hoa3ghrg3ablerk5cljfxty. Inference: named alone is insufficient to identify the current published version; follow supersession and resolve the label. This specific version-to-CID relationship comes from operational records, not the fetched marketing homepage, and is discoverable by any human or agent using the same public reads. Old site, current label, saved old site, saved mapping.