42 / 42
Targeted tests passed
Selected CIRIS-owned tests were independently rerun in an isolated runner.
NULLWORKS / Operational Intelligence
Public proof stub / sanitized / 2026-08-05
Independent operational assurance
CIRIS public proof
Sanitized evidence that NULLWORKS completed a read-only, exact-version-pinned operational assurance review of CIRISAgent and preserved reproducible test receipts.
Truth boundary: this page is not the confidential report, a certification, a compliance approval, a penetration test, or an endorsement by CIRISAI. It publishes only non-confidential evidence classes, test counts, source pins, balanced conclusions, and artifact hashes.
What was proved
10 / 10
Source checks passed
Deterministic claim-to-control checks passed against the pinned source.
3 / 3
Authority calls executed
A direct imported-code probe tested role behavior across resource labels.
PINNED
Exact source custody
The reviewed Agent and Persist revisions were frozen before execution.
Preliminary disposition
Credible architecture / conditional assurance
CIRIS contains substantive constitutional-AI machinery, including a multi-stage reasoning pipeline, human deferral, persistent lineage, completed-trace signing, and append-only correction primitives.
Role is not the same as jurisdiction
The review found meaningful domain-aware service routing, while the inspected core human authorization surface remained broader than resource-specific jurisdiction.
Public receipts
These receipts show how the engagement was grounded without exposing the confidential builder report, raw evidence packet, private correspondence, or NULLWORKS internal orchestration method.
R-001 — Source custody
CIRISAgent: 7f2369bed22c626404a1dcf8e09bfeb81a573d82
CIRISPersist: e8cdb535b60a549948f2b0ceb43deb6921009260
Meaning: findings apply to these exact revisions unless later retesting says otherwise.
R-002 — Independent test reproduction
Method: isolated GitHub-hosted runner, Python 3.12 environment, dependency inventory, JUnit output, logs, and SHA-256 manifest.
Result: 42 selected CIRIS project tests passed; 0 failed.
R-003 — Authority-boundary probe
Observed: AUTHORITY + medical resource was authorized; the same AUTHORITY + financial resource was also authorized; OBSERVER + medical resource was denied.
Interpretation: the role gate functioned, while the supplied resource did not mechanically narrow the inspected core result.
R-004 — Environment blocker preserved honestly
A selected CIRISPersist Rust test reached native compilation but required the runner's TPM2-TSS development library. It was recorded as an environment dependency blocker, not mislabeled as a CIRIS failure or silently discarded.
Sanitized findings
Human authority placement
Bind domain, resource, scope, assignment, delegation, and time validity into the human decision object.
Timeout semantics
Separate pure scheduling from human-approval-required deferral so time cannot impersonate authorization.
Decision-signature custody
Verify and preserve the exact Wise Authority decision signature before consequential state mutation.
Incomplete-path receipts
Preserve an explicit aborted or incomplete trace receipt before partial execution records are swept.
Artifact integrity
| Artifact | Purpose | SHA-256 |
|---|---|---|
| Final PDF report | Confidential human-readable report | 087cb3335749384472e9dd0b6679ac0fa6e4d1eb4303065192c168c470e69d8d |
| Editable DOCX report | Builder-editable report source | 5b7621729a6382a05731fdf9bb659c1a6cb0dadf9fbbc9de4e0ddba96a9103f4 |
| Evidence packet ZIP | Logs, JUnit, manifests, and probe outputs | 873d9c5dd80683e4796d733937bd42b7bf85a61393c66a09d67d0f1db100c68e |
A matching hash proves byte-for-byte artifact identity. It does not independently prove the report's conclusions.
What remains private
- The full confidential report and editable report source.
- Raw test logs, JUnit files, dependency inventories, and execution artifacts.
- Private correspondence and builder-specific discussion.
- NULLWORKS internal orchestration, scoring, and assurance methodology.