Report anatomy

The handoff is assembled deterministically from the run's evidence. Models propose; the assembler only includes what the evidence supports. This page walks its sections in the order they appear.

Summary line

One line: the overall outcome, the count of findings by severity, and the coverage achieved. A P1 is never buried below the fold; if one exists, the summary leads with it.

Coverage

Which states the diff touched, which of them were walked, and under what authority (contract-backed or requested-goal-only). If a scoped state could not be reached, it is listed here as blocked with the reason, not silently dropped.

Findings

Each finding carries:

• Severity (P1 blocks a core journey or risks money or data; P4 is cosmetic) and tier (self-evident, internal contradiction, or declared expectation).
• The claim, stated concretely: not "checkout seems broken" but "the order total shows $102.60 where the line items sum to $97.20".
• Evidence: the screenshots and observation snapshots that support the claim, with the walk step each came from. Every claim links to its proof; a claim with no proof does not appear.
• A source pointer when the code locator found a supportable one: file and region most likely responsible. When no location is supportable, the finding says so rather than pointing at a plausible-looking innocent file.

Goal verdicts

Every goal the run pursued, each with its verdict (completed, defect, blocked, inconclusive) and its action journal: the committed steps, with before and after observations. Blocked goals name what stood in the way. This section is the raw material behind the summary, and it is complete: goals are never omitted.

Run provenance

The exact configuration the run executed under: base and head SHAs, workflow revision, manifest and contract hashes, model routes used per role, and the network events observed. Provenance makes a handoff reproducible and auditable months later.