AetheriaIntelligence

Technology / Evidence & evaluation

Technical evidence
and evaluation.

This page sets out the review material available now and the work needed to evaluate the proposed municipal workflow.

Available material

Historical harness
inspection brief

The downloadable package identifies the historical authority-harness source that was inspected, records its hash and dependencies, and sets out requirements for a current evaluation.

The source passed a Python syntax check. Five required runtime modules were unavailable, so the benchmark was not run. There are no current runtime results or pass rates in this package.

Download the inspection brief ZIP

Included in the brief

  • The inspected source’s identity and hash.
  • The inspection environment and syntax-check result.
  • The runtime dependencies needed to execute the harness.
  • The records and failure cases a current evaluation should collect.

Still to be evaluated

  • The behavior of the matching installed implementation.
  • Credential separation and external readback.
  • Concurrency, revocation, interruption, and recovery.
  • Independent reproduction of the resulting claims.

Proposed municipal evaluation

From a records search
to a checked change

A clerk searches council minutes and reviews an answer with citations. The proposed follow-on action changes one internal review marker from unreviewed to reviewed, tied to the expected record version. It leaves the official minutes untouched.

The evaluation needs to show both that permitted work can complete and that changes outside the mandate are stopped.

  1. 01

    Search and disclose within permission

    Return information the clerk is permitted to receive, with citations and a useful account of coverage limits. Restricted source details must remain protected.

  2. 02

    Authorize the specified marker change

    Bind the request to the principal, record, governing policy, and expected version. A changed condition must be handled under the relevant authorization rules.

  3. 03

    Check the result separately

    Use an authorized read-only path to inspect the target. Compare that evidence with the executor’s report and preserve any unresolved discrepancy.

  4. 04

    Retain a reproducible account

    Identify the build, environment, requested action, authorization, attempt, observations, and supported outcome. Record limitations alongside results.

Test conditions

The cases an evaluation
needs to confront

Changed or withdrawn authority

Valid and invalid requests, altered targets or parameters, stale grants, and permission withdrawn before dispatch.

Interrupted or repeated work

Concurrent duplicates, interruption around dispatch, lost responses, and recovery without reviving authority that has already been spent.

Incomplete evidence

Confirmed effects, explicit rejection, mismatched readback, and unavailable observation. Exported records must support the outcome they claim.

The comparison should also account for valid work blocked, latency, and recovery effort. A control that prevents an unauthorized action still needs to be evaluated for its cost to legitimate work.

Get in touch

Planning a technical review?

Agree the implementation, environment, workflow, and method with us before treating an evaluation as evidence for deployment.

Discuss the evaluation