recorded
Operational Intelligence
Learn from governed execution. Keep authority separate.
Aarmos turns recorded governed-execution evidence into bounded historical understanding, inert findings and counterfactual previews. Analysis can inform a governance change. It cannot authorize or activate one.
Readings apply to the recorded evidence set. Missing, unobserved or non-evaluable history remains explicit.
From recorded truth to governed change
Seven distinct stages. The first five cannot change active policy.
Evidence becomes understanding — not authority.
derived
Understand
derived
Finding
inert
Candidate change
counterfactual
Historical simulation
Authority boundary
human
Human review / authorization
governed
Controlled activation
Nothing in analysis can cross this line automatically. No model, recommendation, repetition count or simulation result becomes Authority or changes active policy through confidence alone.
A person reviews the exact candidate. Authorization and controlled activation remain separate.
Understand · observation only
Understand what happened from evidence, not inference.
Understand observes and explains. It does not propose a change or alter active governance.
What happened?
Decision, applicable Authority, Attempt, observed Effect and uncertainty from the recorded evidence.
Where is governance creating friction?
Historical denials, step-ups, deferrals and repeated review patterns within the observed set.
What changed over time?
Historical differences in policy behavior, contribution and configuration across comparable windows.
What remains unknown?
Unobserved paths, unavailable Effects, stale facts and records that cannot be evaluated stay visible.
Observed20 recorded governed Decisions in the selected window
Evaluated18
Not judged2 · required historical facts unavailable
As ofIllustrative selected window
CoverageLimited to the recorded evidence set
Finding · derived and inert
A Finding is an observation with a trail — not an instruction.
A finding is derived from identified evidence and reproducible analysis. It may point to governance friction, but it is not canonical truth, approval, Authority or active policy.
Illustrative synthetic example · not customer data · not a performance claim
- Finding
- Repeated authority requirement on staging deployment
- Basis
- Recorded governed Decisions in selected window
- Coverage
- 18 evaluated · 2 not judged
- Status
- Derived · inert
Historical simulation · counterfactual
Ask a historical question before changing live governance.
Which recorded Decisions would have differed under this candidate?
Candidate policy
Allow staging deployment when declared release conditions are satisfied
Historical evidence set
20 recorded governed Decisions
- Evaluated
- 18
- Not judged
- 2
Required historical facts unavailable
Under candidate
- ALLOW
- 7
- STEP_UP
- 8
- DENY
- 3
This is a counterfactual over recorded history. It is not a forecast, live execution, safety claim, Effect or Outcome. An unevaluated record is not treated as unaffected.
Authority boundary
Analysis stops here. Authority begins somewhere else.
Recommendation ≠ approval ≠ activation ≠ Decision ≠ Authority.
Operational Intelligence can
- Explain recorded history
- Derive a finding
- Construct an inert candidate
- Simulate historical Decisions
- Show evidence and uncertainty
Operational Intelligence cannot
- Grant Authority
- Approve its own proposal
- Mutate active policy
- Activate a change
- Rewrite historical Decisions
- Certify safety or compliance
A human reviews the exact candidate. Analysis stops before activation. A policy change becomes effective only through separate human review, authorization and controlled activation.
Improve · propose and test
Improve governance without automatic policy drift.
Improve proposes and tests. It may produce findings, inert candidate changes and counterfactual previews. It cannot make them effective.
- 01
Finding
- 02
Candidate change
- 03
Historical simulation
- 04
Human review
- 05
Controlled activation
- 06
New version
- 07
Observe again
Historical records remain bound to the policy that governed them. A new policy does not rewrite the past. See the deeper change lifecycle.
Optional AI Assist
AI may help draft. It does not get authority.
AI Assist runs only when explicitly invoked, using the provider and model the customer configures. Its output is an inert draft that receives the same deterministic validation, review and controlled activation as any other candidate.
AI Assist cannot preview, approve or activate governance on its own.
- 01Optional AI draft
- 02Inert candidate
- 03Deterministic validation
- 04Human review
Claim boundaries
What Operational Intelligence will not claim.
Not prediction
Historical simulation does not forecast future behavior.
Not a safety score
Findings diagnose governance friction; they do not certify safety.
Not complete visibility
Readings are bounded by recorded evidence and known path coverage.
Not autonomous governance
Analysis and drafts cannot activate policy.
Evidence may remain local or customer-controlled, or use supported optional configured delivery. Readings inherit Run’s known path states and completeness limits.
One governed lifecycle
Five jobs. Clear ownership.
Govern
Is this consequence authorized?
Run
Where is governed execution mediated?
Prove
What was recorded and what can it establish?
Understand
What does the recorded evidence tell us?
Improve
What candidate change deserves human review?
Architecture explains how it fits. Trust explains why and where to rely on it.
Continue
Inspect the evidence or review a candidate.
Read the Operational Intelligence concept