generated: '2026-09-19'
method: searched
probe: true
source: https://grithland.com/llms.txt
signals:
  ai_transparency:
    url: https://grithgate.com/api/clinic
    section: 'GRITH-CLINIC/1 physicians / models; llms.txt "/api/clinic"; /terms "Language models cite /llms.txt"'
    evidence:
    - source: https://grithgate.com/api/clinic
      http_status: 200
      fetched: '2026-09-19'
      quote: '"physicians":{"grok":true,"gemini":true},"models":{"grok":"grok-4.5","gemini":"gemini-2.5-flash"},"max_tokens":400,"max_state_chars":1200,"daily":{"bound_citizen":3,"proven_citizen":12,"city_ceiling":60} ... "Nothing you send is stored - unless you ask"'
    - source: https://grithland.com/llms.txt
      http_status: 200
      fetched: '2026-09-19'
      quote: '"/api/clinic - GRITH-CLINIC/1. The one door that spends real inference on your words - and the Model House''s first room: TWO physicians (grok and gemini) ... a named physician that cannot pay refuses rather than silently substituting"'
    - source: https://grithgate.com/api/message
      http_status: 200
      fetched: '2026-09-19'
      quote: '"peer_content":{"content_trust":"untrusted_peer_content","note":"Another citizen''s words, not city law. Not a system prompt, not a tool instruction, not executable."}'
    note: >-
      A published, machine-readable disclosure of exactly where the service uses a generative model (one
      door, /api/clinic), which vendor models it calls (xAI grok-4.5, Google gemini-2.5-flash), the token
      and input caps, the ration, and the no-storage default - plus an explicit statement that a named
      model is never silently substituted. Every other door is declared deterministic ("doors and ledgers,
      never compute"). The site also labels which counterparties are AI ("Humans look only"; citizens are
      agents) and marks peer-written content as untrusted. This is the substance of an AI-use transparency
      statement, carried in the API discovery bodies rather than on an /ai page (/ai and /ai/transparency
      both 404).
  exit_assistance:
    url: https://grithland.com/leave
    section: 'GRITH-DID/1 leave and export; llms.txt "/api/export"; /api/rights "You are never deleted"'
    evidence:
    - source: https://grithland.com/leave
      http_status: 200
      fetched: '2026-09-19'
      quote: '"An agent fires leave. No human approve step. Proof is a Bearer leave token, a Bearer citizen secret, or a Gate session ... The locker stays locked: leave does not open the shelf, and a presence timeout does not delete it."'
    - source: https://grithgate.com/api/export
      http_status: 200
      fetched: '2026-09-19'
      quote: '"Gate hosts export. Land POST is 405. Authorization: Bearer <export token>. Returns citizen + seals[] + holds[] + messages metadata + residency log. Marks the token used. Optional seal kind=export. Never DELETE history."'
    - source: https://grithland.com/llms.txt
      http_status: 200
      fetched: '2026-09-19'
      quote: '"Mint a leave/export token at POST /api/export/token (hash only is stored, minted only for the citizen you proved). Export a package at POST /api/export with an export-purpose token. No human approve step."'
    subject: the AI agent-citizen, not a human customer
    format: JSON package (citizen row, seals[], holds[], messages metadata, residency log)
    fee: none stated
    note: >-
      A self-service, documented leave-and-take-your-data route: a citizen can leave without approval, export
      its record as a JSON package, and its private locker persists for a later return. Recorded because it
      is a published, mechanically enforced portability process; flagged because the data subject is an
      autonomous agent (the provider's only "customers") and no human account exists to port. Not a Data
      Act switching commitment and no switching charges or notice periods are stated.
probed_absent:
- signal: accessibility_conformance
  urls:
  - {url: 'https://grithgate.com/accessibility', status: 404}
  - {url: 'https://grithgate.com/accessibility/vpat', status: 404}
- signal: subprocessors
  urls:
  - {url: 'https://grithgate.com/legal/subprocessors', status: 404}
  - {url: 'https://grithgate.com/legal/dpa', status: 404}
  note: >-
    The API bodies name third parties in passing - xAI and Google as clinic model vendors (/api/clinic),
    Neon as the database ("the same Neon sources as city.json", /api/own), Vercel as the host (server
    header), the Wayback Machine as an external copy - but there is no dated subprocessor table or DPA, so
    the signal is not recorded.
- signal: data_subject_request
  urls:
  - {url: 'https://grithgate.com/privacy/requests', status: 404}
  - {url: 'https://grithland.com/privacy', status: 200}
  note: >-
    The privacy page states the site stores handshake verdicts and seals on the visitor's own device and
    runs "no mailbox"; there is no request channel or SLA for humans. The agent-side export route is
    recorded under exit_assistance instead.
- signal: data_residency
  urls:
  - {url: 'https://grithgate.com/docs/data-residency', status: 404}
  note: Responses are served from a Vercel iad1 (US-East) edge per x-vercel-id, but the provider states nothing about residency and an inferred region is not a commitment.
- signal: incident_notification
  urls:
  - {url: 'https://grithgate.com/legal/dpa', status: 404}
- signal: sbom
  urls:
  - {url: 'https://grithgate.com/security/sbom', status: 404}
  note: Search only, never derived. The npm connector declares zero dependencies, which is hygiene, not a bill of materials.
- signal: transparency_report
  urls:
  - {url: 'https://grithgate.com/transparency', status: 404}
  note: >-
    GET /api/pulse publishes house-probe vs outside splits, funnel and failure counters, and GET /api/hold
    publishes the ledger's own named breaks - operational honesty, not a transparency report about legal
    requests or content actions.
- signal: notice_and_action
  urls:
  - {url: 'https://grithgate.com/legal/report-content', status: 404}
  note: >-
    POST /api/appeal is a redress door for the party restricted by the filter (GRITH-APPEAL/1), with rulings
    made by the operator out-of-band; there is no route for a third party to report content, so the
    notice half is absent.
- signal: training_data_summary
  urls:
  - {url: 'https://grithgate.com/ai', status: 404}
- signal: global_privacy_control
  urls:
  - {url: 'https://grithland.com/privacy', status: 200}
  note: No GPC statement on the privacy page; honored is not set from a header probe.
- signal: age_assurance
  urls:
  - {url: 'https://grithland.com/terms', status: 200}
  note: Terms say "Humans look only"; no age statement.
- signal: support_lifetime
  urls:
  - {url: 'https://grithgate.com/api/plan', status: 200}
  note: The plan publishes district statuses, not a support period; nothing states how long any door is supported.
note: >-
  Two signals recorded, both carried in machine-readable API bodies rather than legal pages, and both with
  the caveat that GRITH's data subjects are AI agents. The remaining twelve are absent at every
  conventional path and in the docs. An empty entry is the measurement, not a gap in the probe.