Read the result

Start with the overall status, then read the individual clauses, source locations, diagnostics, and unmodeled files. A green test run and a satisfied change contract answer different questions.

Status	What it means	What to do next
PASS	The declared, supported obligations are satisfied.	Continue ordinary code review and your existing checks.
INCOMPLETE	A required fact or test scenario is missing, skipped, or unsuccessful.	Implement the obligation or supply the required successful execution.
DRIFT	A boundary, preservation, scope, or intent constraint is violated.	Inspect the finding. Fix the change, or explicitly review revised intent.
UNKNOWN	Evidence is absent, stale, ambiguous, or outside supported analysis.	Obtain usable evidence or record the remaining manual review.
ERROR	Input is invalid or a command could not complete.	Fix the command or inputs before interpreting a result.

Exit codes#

0 means PASS. 1 means INCOMPLETE or DRIFT. 2 means UNKNOWN or an input/command error. Failures can combine in the order INCOMPLETE, DRIFT, UNKNOWN; UNKNOWN takes exit-code precedence. Do not treat exit 2 as a warning you can silently ignore.

A concrete finding#

In the drifting newsletter PR, all three behavior tests pass. ChangeClause still reports:

DRIFT — behavior test exit 0, verification exit 1
DRIFT      no-auth-boundary: Forbidden fact observed in head.

The contract forbids a runtime dependency from src/newsletter.ts to src/auth.ts. Static analysis observes that dependency at src/newsletter.ts:1. The helper is harmless; the finding is an architectural boundary violation, not evidence of a security flaw.

What PASS does not establish#

PASS does not approve a PR or prove that its behavior is safe. It only covers the selected obligations within implemented capabilities. Unspecified behavior, unsupported source, test quality, deployment behavior, and runtime effects remain outside that conclusion.

The JSON assessment separates policy, static-structure, test-definition and test-execution checks and lists unresolved clauses. It always records intentInterpretation: not-performed, testAdequacy: requires-human-review, and behavioralCompleteness: not-assessed. These are assurance limits, not additional passing checks. Text reports print the same limits and manual review actions.

changedScenarioFiles lists changed files referenced by explicit evidence paths or recognized test definitions. It is a review aid, not complete detection of every test/configuration change. Read the full file inventory and diff. Removing assertions while retaining scenario names can still pass execution clauses. Resolve adequacy through meaningful assertions, deliberate defect tests and human review, as recorded in the pilot evaluation record.

Evidence and trust#

Static facts describe extracted code structure. They do not establish execution or reachability. Imported Vitest reports record test execution and are bound to the selected source and contract, but remain self-attested.

Read the evidence and trust reference for source binding, stale results, contract selection, and unsupported attestations. The CI examples provide complete reports and the exact runs behind the displayed results.