Controlled Pilot Candidate · 1,423 passed · 2 skipped · 0 failed · higher-readiness claims RED / LOCKED pending Constitution lifecycle evidence
AI governance · Python · authority · evidence · controlled execution

Fresh Direction /
Sovereign AI Governor

A provider-neutral governance layer designed to keep AI proposals separate from execution authority. The engineering focus is explicit authority, evidence, human approval where required, hard execution boundaries, single-use permits, recovery and forensic evidence.

Controlled Pilot Candidate
1,423 pass · 2 skip · 0 fail
Numeric claim withheld
Gates 5–9 evidenced
Evidence rule: the previous public TRL 6 reassessment is no longer asserted as the current whole-product claim. The latest readiness audit keeps higher-readiness claims RED / LOCKED until authoritative Constitution activation, reconstruction and rollback evidence is available. This is not a certification, third-party assurance result or production-readiness claim.
Architecture

Authority before action.

The design treats the AI as a proposer, not the authority. Permission to act is carried through a controlled path and bound to the exact governed action.

Constitution Engine→Authority & Evidence Decision→Human Authority when required→Hard Execution Boundary→Evidence & Recovery→Change Certification

Execution controls

  • Provider & Action Normalisation: external actions are translated into a governed form before authority is evaluated.
  • Current Operating Envelope: runtime authority is tied to the active operating constraints rather than an informal prompt.
  • Single-use permits: execution authority is bound to the intended action and relevant Constitution/policy/Operating Envelope state.
  • Hard Tool Gateway: the controlled execution path remains the authority boundary; browser or UI approval alone is not execution authority.

Agent identity

  • Agent Passport: enforceable identity/scope is checked at action admission.
  • Lifecycle persistence: passport state is persisted and restored with fail-closed integrity checks.
  • Revocation: lifecycle authority can revoke a passport without granting execution authority.
  • External-action scope: the bounded GitHub read requires github:repository:read; older active passport state must be re-provisioned or recreated to carry that scope.
Latest verified work

What the current evidence supports.

The public description stays at property level: what was demonstrated, what boundary was exercised and what remains unproven.

07

Receipt persistence and restart integrity

Execution receipts and their integrity state were hardened so recovery does not depend on one process lifetime. Receipt verification remains fail-closed after restart and state reconstruction.

Important finding retained: replacing the master secret surfaced as a receipt-integrity failure rather than a separate security-state-binding failure. The result still denied execution; the classification remains visible.
● Gate evidence retained
08

Durable review context

The Decision Centre restart path was hardened so a consequential action waiting for human review can retain the context needed to validate the later decision rather than relying on ephemeral process state.

Why it matters: an approval after restart must still refer to the same governed action, authority and review context. A stale or unverifiable review context must fail closed.
● Restart/recovery evidence retained
09

Governed external action in Linux OCI

The external-network/container evidence gap was exercised with a bounded GitHub repository read through the governed path inside a Linux OCI container.

Boundary: the action was constrained to the governed repository-read scope and evidence was re-checked against stable GitHub response fields and the recorded commit. This does not generalise Linux OCI proof to Windows-specific storage adapters.
● Gate 9 evidence retained
NOW

Controlled Pilot Candidate

The latest readiness audit supports bounded controlled-pilot work, with GREEN-only operation and AMBER/RED fail-closed behaviour in the accepted pilot posture.

Current boundary: authoritative Constitution activation, reconstruction and rollback evidence remains unresolved, so higher-readiness claims stay RED / LOCKED. Gate 10 integrated operational-governance proof remains pending.
● Current bounded status
Evidence chain

Selected current implementation areas.

The underlying Governor repository remains private; these are the implementation concerns the current assurance work is built around.

Engineering concernCurrent evidenceWhat it supports
Governance lifecycleConstitution, authority/evidence decision and governed-action lifecycleSeparates recommendation from authority and gives unknown/missing evidence explicit fail-closed outcomes.
Human authorityDecision Centre review context and restart/recovery testsConsequential actions can remain AWAITING_HUMAN without approval becoming detached from the governed action after restart.
Hard execution boundaryDecision/execution bridge, Hard Tool Gateway and permit controlsApproval is not itself execution; a separate controlled path authorises the exact action.
Receipt integrityPersistent receipt key/state, verification, reconstruction and replay controlsExecution evidence survives restart and rejects corrupted or unverifiable state.
Agent PassportPersistent lifecycle, reconstruction, revocation and action-admission evidenceAgent identity/scope is an enforceable runtime input rather than descriptive metadata.
External actionLinux OCI governed GitHub repository readDemonstrates the governance path around a real bounded external network action in a relevant containerised environment.
VerifierSeparately-run verification process with trusted local key custodyProvides a second verification pass over retained evidence; it is not claimed as third-party, externally anchored or independent-key-custody assurance.
Regression evidence

The latest full safe suite is green.

The current published checkpoint comes from the latest full safe-suite run rather than an older frozen count.

Latest verified safe suite: 1,425 collected · 1,423 passed · 2 skipped · 0 failed. Passing automated tests do not by themselves establish production readiness or close the Constitution lifecycle evidence gap.
Current limitations

What the current pilot status does not mean.

These constraints are part of the evidence, not footnotes to hide.

Constitution lifecycle gap remains open

The authoritative production Constitution activation, reconstruction and rollback path is not yet evidenced to the level required for a higher-readiness claim.

Backup / restore is not claimed complete

The current accepted pilot posture does not claim a completed production backup-and-restore capability.

External dependency remains bounded but real

The GitHub demonstration depends on external API/network availability. The bounded Gate 9 path did not use credentials.

Gate 9 did not exercise every subsystem

Gate 9 proves its scoped external-action path; it does not by itself re-prove every Governor component or the complete Constitution lifecycle.

Verifier independence has limits

The verifier is a separately-run process using trusted local key custody. It is not third-party verification, externally anchored verification or independent-key-custody verification.

Cross-platform evidence is specific

Local Windows storage adapters still require a Windows execution host. Linux OCI evidence should not be presented as proof that Windows-specific adapters run in Linux.

Readiness boundary

Higher-readiness claims remain locked until the evidence closes the gap.

The automated suite is strong, but test volume is not a substitute for the missing Constitution lifecycle proof.

Required proof before a higher claim

  • Authoritative Constitution activation path.
  • Constitution reconstruction after restart/recovery.
  • Rollback evidence that preserves the authority boundary.
  • Gate 10 integrated operational-governance proof.

Still separate from pilot evidence

  • Production backup and restore.
  • External evidence/audit anchoring.
  • Production observability and operational hardening.
  • Independent re-audit and stronger independent key custody.
Claims deliberately not made: current whole-product TRL 6, production-ready, independently certified, third-party verified, externally anchored, or proven safe for unrestricted autonomous execution.