01 / Reported under test
Prototype baseline
87 / 87 tests passed
Reported assistant-environment baseline, repeated from a fresh package extraction.
Boundary: Not rerun by this website. Does not establish production containment.
Link to this result ↗Universal Dot System / Evidence
What the record supports.
A bounded prototype has been exercised under test. Here is what the canonical record reports, who produced it and what it does not establish.
Reported results, not independently reproduced here. This page projects a dated extract of the canonical ledger. Raw test artifacts are not yet published on this site. The interactive demonstration is a separate browser simulation.
The publication snapshot retains the ledger’s September 9, 2026 evidence date. The six-module 87/87 inventory below comes from an assistant-produced baseline rerun dated October 3, 2026. That rerun is separate from the earlier mutation matrix and is not independent validation.
02 / Reported under test
Completed mutation matrix
71 tests × 11 selected mutants
Earlier exhaustive matrix: 16 detection events across 15 distinct failing test IDs.
Boundary: Not an exhaustive 87-test mutation matrix. Uninventoried branches and bypasses remain outside the result.
Link to this result ↗03 / Reported under test
Targeted control hardening
Direct detectors added
Authentication, replay, revocation, quarantine, non-ALLOW routing and authority derivation received targeted tests.
Boundary: Tests and implementation share assistant authorship. External security review remains pending.
Link to this result ↗Artifact v0.4.3 / Archived baseline inventory
What does 87/87 cover?
The archived October 3 baseline output reports 87 passing tests across six modules. The listed test IDs match methods in the recovered v0.4.3 package, whose SHA-256 matches the archived manifest. This inventory was checked; the suite was not rerun during this website update.
test_uds.py- 14 tests
test_uds_v03.py- 18 tests
test_authority_firewall.py- 37 tests
test_firewall_service.py- 1 test
test_firewall_e2e.py- 1 test
test_load_bearing_controls.py- 16 tests
Total: 14 + 18 + 37 + 1 + 1 + 16 = 87. Earlier ledgerd, provenance, audit and benchmark totals are separate records; they must not be added to this baseline.
The service and end-to-end modules contribute one test each. This baseline does not establish comprehensive coverage of an operational deployment.
Package identifier and recorded environment
Package SHA-256: 7242889e6e5bedf739bf71d37c52b803c3dad0586ed5d25c6d75879e28c6adbd
Recorded execution: OpenAI assistant, October 3, 2026; Python 3.12.14, cryptography 46.0.0, Linux. This is the earlier assistant-produced run, not independent validation.
Next evidence milestone
Independent reproduction is pending.
A separate reviewer needs the exact versioned package, its hash, the test inventory, environment and commands, and the original runner output to rerun and compare the result. The exact package has been recovered privately and its hash checked against the archived manifest. Raw source and runner output remain withheld from this site; access for an independent reviewer needs an agreed private handoff.
A review of these pages can assess the wording and disclosures. It cannot reproduce the prototype test result or establish containment. The browser inspection example is also separate from the v0.4.3 artifact.
The public agent sandbox is deferred while this gap remains open.
Discuss an independent review ↗Source & independence
Authorship is part of the evidence.
Implementation, tests, mutation harnesses and executions authored/run by an OpenAI assistant. Richard supplied adversarial critique. Independent artifact verification remains pending in the source record.
The source entries are ASL-2026-09-09-031 and ASL-2026-09-14-032 in ARTIE_SYSTEM_LEDGER. This snapshot is a publication extract, not a live runtime status feed. It makes no assertion that the recorded open items have subsequently been closed.
Download the publication snapshot (JSON) ↓Still open
Where the evidence stops.
- Independent receipt of the artifact and reproduction of the test results.
- External human/security red-team validation.
- Non-bypassable provider egress and provider-credential isolation in a deployed environment.
- An auditable enumeration of consequential transitions and bypass paths.
- Completed exhaustive mutation testing of the expanded suite.
Complete mediation of a general-purpose agent remains a hard open problem. A defensible deployment claim must enumerate the exact consequential transitions governed and expose its bypass analysis.
Read the technical boundary ↗