We do not test whether components look valid.
We test whether the entire causal chain is valid. immo.quick tests the complete normative source-to-receipt chain as a coherent system rather than validating individual components in isolation.
The three outcome classes are reported separately. A simulated violation counts as a passed negative test only when the validator actually detects the violation.
A source, specification, proof, build or deployment can each be individually valid. Execution remains invalid if those states do not belong to one coherent, authorized provenance chain.
Named at the architecture level, not the construction level.
An expected failure passes only when the violation is actually detected.
The current internal run contains 128 tests. 94 positive tests passed. No expected-negative test was confirmed as a correctly detected violation. 34 outcomes are therefore classified as unexpected_failures and treated as real technical findings. This includes eight CROSS_JURISDICTION_SIMULATED cases whose simulated violations were not detected by the validator.
Projection Boundary
ZERO_DISCLOSURE and external evidence projections are tested against indirect identifiers, timestamp granularity, rare categories and combined quasi-identifiers.
Signature Claim Authority
Signature schemes may produce only the trust claims supported by their actual key, provenance, revocation and temporal boundaries.
Stable Finding Identity
stable_finding_id, deduplication and derived_from preserve finding identity and lineage across verification runs.
Deterministic Release Gate
Open CRITICAL or HIGH findings and insufficient regression coverage deterministically block release.
The verification architecture must not reclassify an undetected violation as a successful negative test. Internal verification only, not external certification or a general security guarantee.
Security claims should be tested, not just stated.
Fortress does not treat architectural claims as marketing statements. Defined invariants are continuously exercised against adversarial state combinations and regression scenarios.
What is deliberately not shown here:
- Concrete manipulation values and hash examples
- Internal IDs and invariant IDs
- Exact tested field relationships and baseline-validation mechanics
- Capability-state structures and Execution Rights Graph node logic
- Timing/race and cross-chain construction detail
The suite demonstrates the behavior of the architecture against the defined adversarial scenarios. It does not claim that no unknown attack class can exist.
For auditors and enterprise due diligence, controlled review can disclose test ID, test objective, expected and actual result and the associated invariant/control class. Complete reproduction steps, test fixtures, proof artifacts and regression logs remain part of the NDA-governed security review.
Four formal boundaries. 16 internal adversarial tests. 12 expected violations correctly detected. 0 unexpected failures.
Explore Forensic Evidence Governance →