Catalayer Code is engineered around an explicit technical boundary: zero reliance on black-box probabilistic models for core evaluation. By executing deterministic AST parsers, regex tokenizers, and git tree calculators, Code guarantees repeatable, auditable findings.
Same repository state + same inspection rules = identically identical findings every single time.
Deterministic analysis eliminates AI hallucination risks.
Sub-500ms inspection latency allows real-time execution directly in the browser extension.
Every score and classification is transparent and inspectable.
Specific verifiable repository conditions evaluated by Catalayer Code.
| SIGNAL NAME | VERIFIED REPOSITORY CONDITION | CONFIDENCE | TARGET RISK MITIGATED |
|---|---|---|---|
| Deterministic Repeatability Guarantee | Cryptographic hash matching of git tree commits ensures 100% idempotent results | HIGH | Non-deterministic evaluation drift across audits |
| Zero Execution Sandbox Safety | All inspections are strictly passive and static; 0 client code execution | HIGH | Malicious code execution during audit of untrusted repositories |
When a repository URL is submitted, Catalayer Code executes a 4-stage pipeline:
Stage 1: Tree Parsing — Fetches public git tree metadata, file paths, and package manifests.
Stage 2: Deterministic Rule Matrix — Evaluates 25+ structural, test, and hygiene rules.
Stage 3: Evidence Binding — Connects each flagged condition to specific file paths or missing configurations.
Stage 4: Workspace Projection — Renders structured findings with severity levels and action recommendations.
What this inspection method does not and cannot evaluate.
Run deterministic checks against any public repository without granting OAuth permissions. Review structural gaps, untested changes, and reviewability findings with evidence.
No. Catalayer Code does not train foundation models on inspected repositories. All analysis is passive, deterministic, and ephemeral.