Hugging Face's logo

Datasets:
csoai
/
index-presence

License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
as_of: timestamp[s]
rekor: list<item: struct<log_index: string, http: int64, api_url: string, search_ui: string, uuid: string,  (... 178 chars omitted)
  child 0, item: struct<log_index: string, http: int64, api_url: string, search_ui: string, uuid: string, integratedT (... 166 chars omitted)
      child 0, log_index: string
      child 1, http: int64
      child 2, api_url: string
      child 3, search_ui: string
      child 4, uuid: string
      child 5, integratedTime: int64
      child 6, integrated_utc: timestamp[s]
      child 7, logID: string
      child 8, has_inclusion_proof: bool
      child 9, tree_size_at_proof: int64
      child 10, kind: string
      child 11, api_version: string
      child 12, spec_digest: string
dois: list<item: struct<doi: string, doi_org_url: string, http: int64, resolves_to: string, landing_excerp (... 153 chars omitted)
  child 0, item: struct<doi: string, doi_org_url: string, http: int64, resolves_to: string, landing_excerpt: string,  (... 141 chars omitted)
      child 0, doi: string
      child 1, doi_org_url: string
      child 2, http: int64
      child 3, resolves_to: string
      child 4, landing_excerpt: string
      child 5, mentions_us: bool
      child 6, datacite_http: int64
      child 7, title: string
      child 8, publisher: string
      child 9, published: string
      child 10, state: string
      child 11, url: string
      child 12, registered: string
ots: struct<block_966712_hash_http: int64, block_966712_hash: string, block_966712_time_utc: timestamp[s] (... 42 chars omitted)
  child 0, block_966712_hash_http: int64
  child 1, block_966712_hash: string
  child 2, block_966712_time_utc: timestamp[s]
  child 3, block_966712_exists: bool
  child 4, note: string
probes: list<item: struct<probe: string, reported_total: int64, rows_walked: int64, pages: int64, stopped: n (... 89 chars omitted)
  child 0, item: struct<probe: string, reported_total: int64, rows_walked: int64, pages: int64, stopped: null, distin (... 77 chars omitted)
      child 0, probe: string
      child 1, reported_total: int64
      child 2, rows_walked: int64
      child 3, pages: int64
      child 4, stopped: null
      child 5, distinct_councilof_ai_urls: int64
      child 6, walk_complete: bool
      child 7, sample: list<item: string>
          child 0, item: string
finding: string
api: string
to
{'as_of': Value('timestamp[s]'), 'api': Value('string'), 'probes': List({'probe': Value('string'), 'reported_total': Value('int64'), 'rows_walked': Value('int64'), 'pages': Value('int64'), 'stopped': Value('null'), 'distinct_councilof_ai_urls': Value('int64'), 'walk_complete': Value('bool'), 'sample': List(Value('string'))}), 'finding': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              as_of: timestamp[s]
              rekor: list<item: struct<log_index: string, http: int64, api_url: string, search_ui: string, uuid: string,  (... 178 chars omitted)
                child 0, item: struct<log_index: string, http: int64, api_url: string, search_ui: string, uuid: string, integratedT (... 166 chars omitted)
                    child 0, log_index: string
                    child 1, http: int64
                    child 2, api_url: string
                    child 3, search_ui: string
                    child 4, uuid: string
                    child 5, integratedTime: int64
                    child 6, integrated_utc: timestamp[s]
                    child 7, logID: string
                    child 8, has_inclusion_proof: bool
                    child 9, tree_size_at_proof: int64
                    child 10, kind: string
                    child 11, api_version: string
                    child 12, spec_digest: string
              dois: list<item: struct<doi: string, doi_org_url: string, http: int64, resolves_to: string, landing_excerp (... 153 chars omitted)
                child 0, item: struct<doi: string, doi_org_url: string, http: int64, resolves_to: string, landing_excerpt: string,  (... 141 chars omitted)
                    child 0, doi: string
                    child 1, doi_org_url: string
                    child 2, http: int64
                    child 3, resolves_to: string
                    child 4, landing_excerpt: string
                    child 5, mentions_us: bool
                    child 6, datacite_http: int64
                    child 7, title: string
                    child 8, publisher: string
                    child 9, published: string
                    child 10, state: string
                    child 11, url: string
                    child 12, registered: string
              ots: struct<block_966712_hash_http: int64, block_966712_hash: string, block_966712_time_utc: timestamp[s] (... 42 chars omitted)
                child 0, block_966712_hash_http: int64
                child 1, block_966712_hash: string
                child 2, block_966712_time_utc: timestamp[s]
                child 3, block_966712_exists: bool
                child 4, note: string
              probes: list<item: struct<probe: string, reported_total: int64, rows_walked: int64, pages: int64, stopped: n (... 89 chars omitted)
                child 0, item: struct<probe: string, reported_total: int64, rows_walked: int64, pages: int64, stopped: null, distin (... 77 chars omitted)
                    child 0, probe: string
                    child 1, reported_total: int64
                    child 2, rows_walked: int64
                    child 3, pages: int64
                    child 4, stopped: null
                    child 5, distinct_councilof_ai_urls: int64
                    child 6, walk_complete: bool
                    child 7, sample: list<item: string>
                        child 0, item: string
              finding: string
              api: string
              to
              {'as_of': Value('timestamp[s]'), 'api': Value('string'), 'probes': List({'probe': Value('string'), 'reported_total': Value('int64'), 'rows_walked': Value('int64'), 'pages': Value('int64'), 'stopped': Value('null'), 'distinct_councilof_ai_urls': Value('int64'), 'walk_complete': Value('bool'), 'sample': List(Value('string'))}), 'finding': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Index-presence board — measured 2026-09-22 from the pod (rp-3090-now)

Every row here was measured today from an unrestricted network, by asking each service's own index or search route and recording what it rendered or returned. Nothing was promoted from a ledger note, a prior audit, or a guessed URL.

This file is the only source the home-page rebuild is allowed to publish from. A row in §1 has an evidence URL and a quote. A row in §2 is a service that answered and said no. A row in §3 could not be settled and says why. Never publish from §2 or §3.

Four states, never conflated. SUBMITTED (we sent it), ACCEPTED (the service took it), APPEARING (a third party shows it), PRESENT (it still does). Most of what follows is APPEARING/PRESENT, verified by rendering. Where only submission or acceptance was observed, the row says exactly that.

Raw artifacts: /workspace/lanes/out/index-presence/ on the pod (18 JSON files, 58 screenshots), mirrored to ~/clawd/csoai-reach-pack-01Sep2026/ and to Hugging Face.

0. The finding that reframes the board

Every repository this estate publishes is invisible to anyone who is not logged in as us. https://api.github.com/users/CSOAI-ORG → 404. https://github.com/CSOAI-ORG → 404. git ls-remote on councilof-ai.git → 401. Authenticated with our own token, the same API reports 755 public repositories, none archived, private: false.

Control, from the same pod seconds apart: torvalds 200, anthropics 200, modelcontextprotocol 200, github.com/modelcontextprotocol/servers 200. The network is fine. The account is flagged (day 20 of a flag first seen 2 September). CouncilofAI-CSOAI is also 404, anonymously and authenticated — the "clean business account" is not reachable under that name either.

This one fact explains most of what the board used to call separate defects:

Consequence Measured today
Software Heritage cannot archive us from GitHub save request 2492509 → save_task_status: failed, visit_status: not_found
"npm/PyPI packages point at dead links" 7 of 7 dead URLs are github.com/CSOAI-ORG/*; all three target repos exist and the paths are correct
Published surfaces pointing at a 404 14,121 references — 13,428 in public/, 693 in client/src

It is one owner action, and it is the highest-value item on this board by a wide margin.

1. VERIFIED PRESENT TODAY — publishable

1.1 MCP directories (rendered in a browser, entry URL captured)

Directory Our entry Rendered evidence
Glama https://glama.ai/mcp/servers/CSOAI-ORG/councilof-ai Page titled "Council of AI GSPC by CSOAI-ORG | Glama"; carries 50+ outbound links to councilof.ai including every tool and a live inspector. Search page: "Matching MCP Connectors: agent-orchestrator-mcp io.github.CSOAI-ORG…"
PulseMCP https://www.pulsemcp.com/servers/csoai-org-csoai-governance-crosswalk-mcp "Showing 1 - 42 of 221 servers for 'csoai' … CSOAI Governance Crosswalk, CSOAI-ORG … EST VISITORS (WEEK) 47" — a third party's own traffic estimate
LobeHub https://lobehub.com/mcp/csoai-org-ecommerce-ai-mcp (+ 20 more csoai-org-*) "E-Commerce AI MCP Server — CSOAI-ORG — Unvalidated … Sep 22, 2026"
Smithery https://smithery.ai/servers/csoai/gspc "Council of AI GSPC — csoai/gspc — Independent AI-governance measurement over MCP. Canonical remote: POST https://councilof.ai/mcp … Stdio package: csoai-gspc-mcp@0.2.2" (also smithery.ai/servers/csgaglobal/csoai-compliance-council)
mcpservers.org https://mcpservers.org/servers/councilof-ai-mcp "Showing 1-1 of 1 servers — Council of AI GSPC MCP — Independent AI governance measurement MCP. 11 HTTP tools (7 free + 4 x402)." ⚠️ stale: "Board 22·22·0 · Hub 885/885/0"
MCP Toplist https://mcptoplist.com/server/io.github.CSOAI-ORG%2Fcsoai "SERVER Councilof AI — Rank #3,412 of 132,046 servers tracked — io.github.CSOAI-ORG/csoai"; and …%2Fcsoai-governance-crosswalk-mcp at "Rank #16,651". Org page: "CSOAI-ORG (368 servers)"
mcpverify.ai https://mcpverify.ai/archive Two dated observations: councilof.ai/mcp 2026-09-17 OK 19, and councilof.ai/api/assess 2026-08-29 HTTP_4XX 10. Site is self-declared "Pre-launch build"; the archive is live and carries us
A2A Registry https://a2aregistry.org/?q=csoai "1 matching agents of 387 … Council of AI — Measurement Agent, CSOAI Ltd, 1.0, 100.0% uptime, Reachable, Last checked 7 days ago"

1.2 The crown jewel — an independent curated list

EthicalML / awesome-artificial-intelligence-regulation (UK Institute for Ethical AI & ML) — https://github.com/EthicalML/awesome-artificial-intelligence-regulation. PRESENT. Exact current wording, rendered 2026-09-22:

Council of AI - Independent AI-governance measurement board (not certification). Live ruling 22 measured of 22 via https://councilof.ai/api/gspc; signed datasets DOI 10.57967/hf/10114; MCP registry io.github.CSOAI-ORG/gspc 1.0.3.

Two links to us on the page: https://councilof.ai/ and https://councilof.ai/api/gspc.

It is stale in our favour on two counts, and that is a correction we owe them, not a boast:

  1. "22 measured of 22" — the live board is 23 measured of 23 (read from /api/gspc at 16:42Z today: axes: 23, measured_axes: 23, unmeasured_axes: 0, since the effect-binding flip on 22 September).
  2. "io.github.CSOAI-ORG/gspc 1.0.3" — that version exists but is not isLatest, and its remotes are https://councilof.ai/api/assess (answers 400, not an MCP endpoint) and a personal *.workers.dev URL. The current latest is 1.4.2, remote https://councilof.ai/mcp (verified alive). A third party is publishing a pointer to a broken endpoint of ours, on our own numbers.

1.3 MCP registry — full walk, no stall

The prior census stalled at page 3 on a 20 s timeout. Walked to the end today with backoff. Window 2026-09-22T15:39:42Z → 16:32:08Z, page size 100:

  • 1,127 pages read, walk complete, no stop condition.
  • 112,606 version rows (one row per published version — a row is not a server).
  • 34,873 distinct server names.
  • 354 of them ours (io.github.CSOAI-ORG/*), from 1,342 of our version rows.

MCP Toplist independently counts 368 for CSOAI-ORG. Different windows and definitions; both are right about their own read, and they are not reconciled here.

1.4 402 Index — the "50 rows" correction, with the walk that proves it

Probe Reported total Rows walked Distinct councilof.ai URLs Walk complete
q=councilof.ai 73 73 (2 pages) 23 ✅
q=csoai 60 60 12 ✅
domain=councilof.ai 111,293 10,000 (capped) 22 ❌
provider=councilof.ai 111,293 10,000 (capped) 22 ❌

domain=, provider= and search= are silently ignored — each returns the entire 111,293-row index. That is almost certainly where "50 rows" came from: a partial read of the whole index, totalled as ours.

The true figure is 23 distinct councilof.ai URLs, all enumerated in 402index-walk-2026-09-22.json, and it is not the same number as the query's total. (The coordinator measured total=122 earlier today; I measured 73 at 16:29Z. The total moves; the 23 did not.)

1.5 Anchors — pure third-party proof

Anchor State Evidence
Rekor 2810720876 ✅ in the log, inclusion proof present kind: rekord, integrated 2026-09-12T19:21:59Z, uuid 108e9186e8c5677a16d7…, tree size 2,787,973,462, sha256 e2ce9569… — https://search.sigstore.dev/?logIndex=2810720876
Rekor 2837824756 ✅ in the log, inclusion proof present kind: rekord, integrated 2026-09-15T01:31:25Z, uuid 108e9186e8c5677a6b5c…, sha256 708a3763…
DOI 10.5281/zenodo.22293340 ❌ Zenodo record unavailable since 29 Sep 2026: account blocked by Zenodo; appeal pending. (re-read 2026-09-29; the row above this date said it resolved) status: https://councilof.ai/interop/zenodo-status.json
DOI 10.5281/zenodo.21991104 ❌ Zenodo record unavailable since 29 Sep 2026: account blocked by Zenodo; appeal pending. (re-read 2026-09-29; the row above this date said it resolved) status: https://councilof.ai/interop/zenodo-status.json
DOI 10.57967/hf/10114 ✅ resolves, findable → https://huggingface.co/datasets/csoai/gspc-board — the DOI the EthicalML entry cites
Bitcoin block 966712 exists, mined 2026-09-12T19:37:55Z, hash 00000000000000000000016cb3408982b0e136f878b0dc60668b773472c29689 ⚠️ A block existing is not an anchoring claim. Whether any of our .ots files carries an attestation into that block is decided by the proof bytes, per file, and is not established by this row. Say submitted unless the upgraded proof verifies.

1.6 Archival permanence — Software Heritage

The save endpoint needs no token (measured: POST with no Authorization → HTTP 200, save_request_status: accepted). GitHub is unusable as an origin (§0), so the public Hugging Face git origins were archived instead. 8 origins now carry a permanent SWHID:

Origin SWHID
csoai/councilof-ai-source (the repo's own mirror) swh:1:snp:67e55b8a4389b8c0d1c685d0201e155567ea6eb8
csoai/councilof-ai-mirror swh:1:snp:8953e67a1cf1c01f9e59011d86d46cfaebba2d02
csoai/gspc-board swh:1:snp:78a68e71ce0e56c77c481e6e0039710474097cca
csoai/gspc-boards swh:1:snp:7aa4da09bf716bbd3e36b7122dae04235e16e7c8
csoai/gspc-hub-cards swh:1:snp:0b81b7a475de3bfcfb8a16ec275f78b3c0bf6dba
csoai/gspc-swarm swh:1:snp:136000dcd7ccbef9a080ef524185647d5a95e06e
csoai/hub-queue swh:1:snp:c3d73c2f12b3796373484053ea136d48da7116cb
csoai/x402-bazaar-conformance swh:1:snp:939cf08f92985856c6064737503b9fbb344660dc

Browse any of them at https://archive.softwareheritage.org/browse/origin/directory/?origin_url=<origin>.

Measured constraint: X-Ratelimit-Limit: 10 per window, anonymous. The remaining ~106 public origins are queued through a standing loop at 8/run (≈14 days). A SWHID says these bytes existed on this date. It does not say anyone reads them.

1.7 Archival permanence — Wayback Machine

councilof.ai is already in the Wayback Machine: 386 captures across the site, the earliest 2026-01-08. Per-URL, as of 16:40Z:

URL Captures Latest Read back
/ 8 20260922152706 38,867 bytes ✅
/api/gspc 7 20260922081158 16,887 ✅
/api/corrections 3 20260922081158 30,730 ✅
/.well-known/x402.json 4 20260917192224 3,056 ✅
/.well-known/did.json 3 20260917192248 1,922 ✅
/root.json 3 20260917192240 13,624 ✅
/dashboard/?tab=board 3 20260922152736 ✅ (this is where /gspc lands)
/memberships/ 1 20260922160829 ✅ (created today, was zero)
/pay 1 20260917192227 6,680 ✅
/gspc 0 — 308 to /dashboard/?tab=board; SPN archives the destination, not the hop
/memberships 0 — 308 to /memberships/

Save Page Now works anonymously, and lies about it in both directions. Measured: GET https://web.archive.org/save/<url> answers HTTP 500 with the ordinary Internet Archive error page — and captures anyway: /memberships/ had zero CDX captures, was requested at 16:07:01Z, and appeared as capture 20260922160829. Meanwhile POST https://web.archive.org/save answers 401 {"message":"You need to be logged in to use Save Page Now."} for the same URL the GET route then captured. And it does not always take: a GET /save for /root.json at 15:34Z produced no new capture at all.

Therefore a save request's HTTP status is worthless as evidence, in both directions. The standing loop treats it as such: it requests, waits, and records a capture only when CDX confirms it and the archived bytes read back.

2. VERIFIED ABSENT — a service that answered, and said no

Surface Evidence (rendered)
mcp.so "Search results for 'csoai' — 0 results — No servers match 'csoai'". The term appearing on the page is the query echo, not a listing. Also checked councilof, gspc.
Agentic.Market Settled against the site's own search API: api.agentic.market/v1/search/services?q=councilof → {"services":[], "total":0}; same for csoai and gspc. Directory holds 2,858 services and runs ~$21k/day in x402 volume. We are not in it.
x402.org "Results For 'councilof' — Sorry, no results were found." Not on /members/ either — we are not an x402 Foundation member.
x402scan.com No match for councilof on its index.
Enterprise DNA (MCP) mcp.enterprisedna.co → NXDOMAIN. enterprisedna.co/?s=csoai (the company site) has no mention. Any board row citing this is citing a host that does not resolve.
AgentNDX Category error, not an absence. agentndx.com is an AI-sales-chat-widget SaaS — "Your AI sales team, live on the web" — with no directory, index or submission route. There is nothing here to be listed in. Remove the row rather than chasing it.

3. UNCHECKABLE — and exactly why

Surface Why
Bing (site:councilof.ai) "One last step — Please solve the challenge below to continue"
DuckDuckGo HTTP 202, "Unfortunately, bots use DuckDuckGo too. Please complete the following challenge… Select all squares containing a duck"
Mojeek "Verification required — Please complete the challenge to continue. Protected by ALTCHA"
Yandex "Please confirm that you are not a robot… Yandex SmartCaptcha"
Marginalia Answered cleanly: "No search results found." Weak evidence — it is a small independent index — but it is a real measurement, recorded as such.
mcpservers.org /category/all HTTP 403 behind a Cloudflare Turnstile (cf-turnstile-response input). The site's search route worked and is in §1.

So: APPEARANCE in a general search index cannot be verified from the pod at all. Every major engine walls automated queries. We can observe SUBMITTED and ACCEPTED; we cannot observe APPEARING or PRESENT for Bing, Google or Yandex without a credentialed webmaster console. Nothing published should claim otherwise.

The OTS row in §1.5 is also deliberately incomplete: a per-file audit of the served .ots bytes is needed before any of them is described as anchored.

4. Package URLs — root-caused, and it is not what the board said

A prior audit called "npm/PyPI packages pointing at dead links" the highest-ROI repair. Every declared URL of every package on public/interop/footprint-packages.json was fetched anonymously from the pod. 7 of 20 are dead — and all 7 are the same fault.

Registry Package Field Declared URL HTTP
pypi csoai-gspc Board mirror github.com/CSOAI-ORG/gspc-board 404
pypi csoai-gspc Source github.com/CSOAI-ORG/councilof-ai/tree/master/scripts/spray/pypi/csoai-gspc 404
pypi inspect-signed-receipt Repository github.com/CSOAI-ORG/inspect-receipts 404
npm csoai-gspc-mcp repository github.com/CSOAI-ORG/councilof-ai 404
npm csoai-gspc-mcp bugs …/councilof-ai/issues 404
npm csoai-governance-mcp repository github.com/CSOAI-ORG/councilof-ai 404
npm csoai-governance-mcp bugs …/councilof-ai/issues 404

Checked authenticated: gspc-board, councilof-ai and inspect-receipts all exist, all private: false, none archived, and scripts/spray/pypi/csoai-gspc/LICENSE is there. The declared URLs are correct. They 404 because the account is invisible (§0).

There is no correct URL to move to, so there is nothing to republish. The repair is the GitHub flag. All 13 live URLs (councilof.ai, csoai.org, huggingface.co/csoai, the PyPI project pages) answer 200.

A separate, real gap: three packages declare almost nothing. claimguard 0.1.0 and agent-incident-reporter-mcp 1.0.3 have no project URLs at all; csoai 0.2.2 has only Homepage. That one does need a release to fix.

5. MCP listings pointing at a dead host — 39, and one DNS record fixes them

Of our 354 registry servers, 40 declare a remote on their isLatest version. (The other 314 are stdio/PyPI packages with no remote — not a defect.)

  • 1 alive: https://councilof.ai/mcp → MCP initialize → 200.
  • 39 dead: every one on api.meok.ai, which has no DNS record at all (NXDOMAIN — not a timeout, not a 5xx). meok.ai apex answers 200; only the api. subdomain is missing.

Two ways to fix, and the cheaper one is not the one the board assumed:

(a) One DNS record. Point api.meok.ai at whatever serves those 39 A2A paths. Zero republishes, all 39 fixed at once. Requires that something still serves /v1/a2a/* — verify before creating the record, or 39 listings move from NXDOMAIN to 404, which is worse because it looks deliberate.

(b) 39 republishes. Registry versions are immutable; a listing is corrected by publishing a new version of the same name. The exact payload per listing is written out in mcp-remote-health-2026-09-22.json → broken[]: current version, bumped, with remotes: [{"type":"streamable-http","url":"https://councilof.ai/mcp"}].

This is an owner decision, not a lane one, because it crosses the CSOAI/MEOK boundary: CSOAI measures, MEOK hosts. Route (b) moves 39 MEOK-branded endpoints onto a CSOAI host.

A 40th, separate: io.github.CSOAI-ORG/gspc 1.0.3 points at councilof.ai/api/assess (answers 400) and a personal *.workers.dev URL. It is not isLatest — but the EthicalML list and mcpverify.ai both publish it, so a stale version is doing live damage (§1.2).

6. IndexNow — submission discipline, not setup

Setup was already done and I nearly duplicated it. councilof.ai has three key files committed under public/, and all three answer 200 live. The current one is 97a1aa3163534fae954108d8941eb361 (committed 2026-09-12, ab8a035a4) — the only one any script references and the one the 2026-09-21 receipt verified. The other two (1ab9ec14…, 529017b8…, both 2026-08-19) stay served so that anything submitted under them remains verifiable. No fourth key was minted.

What was missing was the discipline. The submitter derived its list from the sitemap and refused dead URLs, but sent every live URL on every run — 552 of them. The protocol asks publishers to notify meaningful changes, and the 21 September receipt for this host says in as many words "do not re-ping without a material public URL change".

--changed-only now keeps a per-URL content fingerprint and sends only what moved. Three byte ranges are masked, each measured today by fetching the same URL twice seconds apart and diffing — never masked on suspicion:

  • Cloudflare's rotating email-obfuscation XOR key (the only difference between two fetches of the home page 3 s apart);
  • a response's own generation time (ts, as_of, <pubDate>, and the guid derived from it). Before this mask, /api/feed.xml, /api/regulator-findings and /api/reported were reported as changed on every single run — which is exactly the resubmission the protocol asks us not to make.

Measured after the mask, on two consecutive runs: 553 live, 0 new, 0 changed, 0 submitted.

First run through a fresh state seeds and submits nothing, because announcing 553 URLs at once is a judgement the owner makes, not a script. Every result line ends indexing NOT_VERIFIED, including the runs with nothing to do — an absent log line reads like an absent run.

Also found and reported rather than papered over: https://councilof.ai/merge-me/ is in the sitemap and answers 404.

7. Standing loops installed (pod, registered in the scheduler)

Loop Slot What its receipt says
indexnow-ping.sh 06:30Z live=N new=N changed=N unchanged=N submitted=N accepted=N; indexing NOT_VERIFIED — verifies the key file is LIVE first, because a ping against a missing key file is a silent failure (the endpoint answers 200 and discards the batch at validation)
swh-archive.sh 07:30Z due=N submitted=N succeeded=N rate_limited=N; ledger holds N origins with a SWHID — 8 per run against the measured limit of 10
wayback-save.sh on demand targets=N due=N requested=N verified=N — verified means CDX-confirmed and read back, never "SPN answered 200"

All three take --now, none stamps itself, all three write a receipt on a run with nothing to do, and all three hold a lock. Scheduler backed up before each edit (scheduler.sh.pre-indexnow-ping-*, scheduler.sh.pre-swh-archive-*); scheduler verified running (pid 954782).

8. Needs the owner — exact action, in order of value

  1. Lift the GitHub account flag on CSOAI-ORG (and establish what happened to CouncilofAI-CSOAI, which is 404 even to our own token). GitHub Support → Account → appeal the flag; day 20, 5 escalations already logged. Fixes 7 dead package URLs, 14,121 published references and the Software Heritage route, with zero republishes. Nothing else on this board is close.
  2. Decide the api.meok.ai route (§5): create the DNS record only if something still serves /v1/a2a/*, verified first, or authorise 39 registry republishes onto https://councilof.ai/mcp. 39 listings are unreachable either way until this is decided.
  3. Open a correction PR on EthicalML/awesome-artificial-intelligence-regulation — 22 measured of 22 → 23 measured of 23, and io.github.CSOAI-ORG/gspc 1.0.3 → 1.4.2. We are the ones who should catch this; it currently points readers at an endpoint of ours that answers 400.
  4. Ask mcpservers.org to refresh Board 22·22·0 · Hub 885/885/0, same reason.
  5. Decide whether to announce the 553-URL IndexNow baseline once. One command, owner's call: python3 scripts/monitoring/indexnow_submit.py --changed-only --state <state> without --seed-only. The loop will never do it on its own initiative.
  6. An archive.org account would help but is not required. The anonymous SPN route works (§1.7). A logged-in account or S3 keys (https://archive.org/account/s3.php) would make it reliable rather than intermittent. Currently no archive.org credential exists anywhere in the estate — grep of the pod secrets, ~/.config, ~/.csoai-keys* found none.
  7. Submit to Agentic.Market (§2) if we want the x402 door discovered — it is a live directory with 2,858 services and ~$21k/day, and we are not in it. agentic.market/validate is its own route. Outward action, so it waits on the owner.
  8. Register csoai and claimguard project URLs on the next PyPI release (§4). No urgency.

9. What this board does NOT say

  • It does not say anyone reads our archived bytes, uses our MCP servers, or clicks a directory entry. Presence is presence.
  • It does not say our pages are indexed by Bing, Google or Yandex. That is §3, and it is unverifiable from here.
  • It does not reconcile 354 (our registry walk) with 368 (MCP Toplist) or 221 (PulseMCP). Three counters, three windows, three definitions. Each is right about its own read.
  • It does not describe any .ots file as anchored. That needs the proof bytes.

Measured 2026-09-22 by the index-presence lane, from rp-3090-now. Every number here came from a live request made today; re-derive before quoting, because these go stale in days.

Licence note

Files written by CSOAI in this repository (the measurement JSON files and PRESENCE-BOARD.md) are CC-BY-4.0, the licence councilof.ai states for its data. 2026-09-22/mcp-registry-servers.jsonl is a verbatim copy of third-party MCP registry records; it keeps the terms of its upstream source and is not relicensed by CSOAI.

Objections, contact and corrections

To object to a row, ask for a re-check, request a correction or ask for a record to be withdrawn, email nicholas@csoai.org or use the appeals and dispute route at https://councilof.ai/dispute/. Corrections are listed in the corrections ledger at https://councilof.ai/api/corrections, with what changed and when. CSOAI Ltd (company no. 16939677, England and Wales) is the accountable publisher.

How to cite

CSOAI Ltd (Council of AI). Index-presence board — measured 2026-09-22 from the pod (rp-3090-now). 2026. Hugging Face dataset csoai/index-presence. https://huggingface.co/datasets/csoai/index-presence

@misc{csoai_index_presence,
  title        = {Index-presence board — measured 2026-09-22 from the pod (`rp-3090-now`)},
  author       = {{CSOAI Ltd}},
  year         = {2026},
  howpublished = {Hugging Face dataset, https://huggingface.co/datasets/csoai/index-presence},
  note         = {Corrections: https://councilof.ai/api/corrections}
}

Licence: mixed. See the Licence note in this card.

Corrections and verification

Downloads last month
157
Total file size:
129 MB