Gates & Verdicts

In mountaineering, safety equipment is useless if it doesn’t hold when subjected to stress. In Anchors, Gates are the automated inspection points that verify whether your project’s code, specs, tests, and architecture adhere to declared standards.

1. What is a Gate?

A Gate is an automated checker that evaluates an artifact or relationship in the project graph against an invariant rule.

Gates are configured in anchors.yaml:

gates:
  - name: unit-complete
    check: unit-complete
    on: [spec]
    blocking: true
    measures: "spec has matching feature, test, and code"

  - name: mutation-score
    check: mutation-score
    on: [code]
    threshold: 80
    blocking: true
    measures: "test suite kills at least 80% of generated mutants"

Each gate declares:

name: The human-readable identifier.
check: The underlying verification algorithm.
on: The artifact types to evaluate (spec, code, feature, test, plan, doctrine, etc.).
blocking: Whether a failure stops the CI/CD pipeline.
threshold: Optional quantitative criteria (e.g. coverage percentage or mutation score).

2. The Four Verdict States

When a gate executes, it returns one of four definitive Verdicts:

Verdict	Meaning	CI Impact	What to do
`OK`	All criteria are fully satisfied.	Passes cleanly.	Proceed to next step.
`FAIL`	An invariant was violated.	Blocks pipeline (if `blocking: true`).	Fix the code or update the spec to resolve the discrepancy.
`WARN`	Advisory discrepancy or impending deprecation.	Logs warning; does not block CI.	Schedule remediation before the gate becomes blocking.
`SKIP`	Gate was skipped due to an explicit layer waiver or filter.	Neutral.	Recorded in audit log for transparency.

3. Deterministic vs. Relational vs. Synthetic Gates

Anchors categorizes its 95 gates into three execution models:

Deterministic Internal Gates: Inspect a single file’s syntax and contents without requiring graph context. (e.g. has-code, header-valid).
Relational Graph Gates: Inspect relationships and edges across multiple files in the dependency graph. (e.g. unit-complete, layer-boundary, circular).
Synthetic / AI-Judged Gates: Use Large Language Models with calibrated prompts to evaluate semantic consistency and clarity. (e.g. doc-self-contained, progress-honest).

4. Explore All Gates

Browse all 95 available verification gates in the Gates Catalog.