3 · Improve it.

You don't know what to change.

NeuralRing names each gap with its reason, proposes what would pass your policy, and re-issues your posture — and its certificate — as it improves.

  • Policy gaps: what fails your sovereignty policy and exactly why, with a code per exclusion.
  • Remediation: what to change.
  • Sovereign alternatives: providers and models already available that would pass, each with its evidence level.
  • An updated posture and an updated AI Supply-Chain Certificate every time the state improves.
Sample — a fictional organisation with illustrative values. Nothing here is a real record.

policy jurisdiction EU/EEA · min_tier 1 · min_assurance verified · exclude parent US · max_retention 30 d · custody customer/eu_operator · transfer none_needed_eu · no training use · evaluated 2026-09-14 09:00 UTC · valid until 2026-09-21 09:00 UTC

Gaps · why it fails
Atlantic Models· ie-dub · atlantic/orbit-4fails · 3
parent_jurisdiction_excludedultimate parent US is excluded by the stated policy
retention_above_maxretention 90 d exceeds max_retention 30 d
key_custody_not_allowedkey custody foreign_operator is not among the accepted custodies
Alternatives · what would pass
Rhône Compute · fr-par · rhone/ligne-8bpassesyour declared upstream
Provider
unknown
Infrastructure
unknown
Model
closed API
tier · assurance
0 · certified · 6 named dimensions at ≥ verified
differs on
parentJurisdiction, retentionDays, keyCustody
Meridian Cloud · nl-ams · meridian/delta-70bpassescatalogue
Provider
unknown
Infrastructure
unknown
Model
closed API
tier · assurance
1 · verified · 5 named dimensions at ≥ verified
differs on
parentJurisdiction, retentionDays, keyCustody
Posture now → after the suggested changes
Provider
4 of 5 lines→4 of 5 lines
Infrastructure
5 of 5 lines→5 of 5 lines
Model
5 of 5 lines→5 of 5 lines

The level is visible on every plan; the providers that get you there are revealed on Basic and Scale. Re-issue the certificate after you change.

Adopting a substitute discharges no obligation; the manifest is a recommendation, never a verdict.

The steps inside
3.1 Read the manifest and the gaps · instant
You
Read each line's check — pass or fail, with the reason code — and the substitutes the same filter would admit.
NeuralRing
Names what fails and why, lists what would pass instead, and shows which framework fields still lack evidence.
You get
A repair list you act on; nothing is switched for you.

The manifest is one table: for every provider, region and model you use — its origin (declared, observed, or both), its facts at their assurance, how often it was observed, and the check under your policy. A failing line names its reason codes and the substitutes: endpoints in the catalogue or among your own declared providers that the same filter admits, sorted by how closely they fit your policy — never by price and never by our interest. The gaps page lists, per framework you care about, which fields have no evidence yet and why it is absent. It is vocabulary for your own assessment, never a finding that an obligation is unmet.

3.2 Request the certificate · instant
You
Request it once the manifest reads the way you want.
NeuralRing
Issues a signed, dated, revocable statement of what was verified, at what assurance, over what coverage — and revokes it automatically if a covered fact moves.
You get
A certificate your auditor verifies offline, without NeuralRing.

The certificate is computed from records that already exist: the manifest, the check, the coverage, the gap counts and the external anchors bracketing its records. It is valid for one claim-watch interval; a re-evaluation is a new certificate, never an extension. If a fact behind any covered line changes, the certificate is revoked and stays retrievable with its reason; the reissue carries the difference. The bundle contains the document, the records, the profiles and the verifier script, so anyone can check it without trusting us.

In detail