A review card should expose skipped tests, unread files, unavailable dependencies, and unresolved assumptions before presenting an approval action.
Agent review screens usually emphasize activity: files changed, tests passed, and a confident summary. Review quality also depends on negative evidence—what the agent did not inspect or could not verify.
Absence is not failure by itself. Hidden absence is the problem.
An inspection-coverage model
type EvidenceItem = {
target: string;
Score whether an agent has enough problem, reproduction, expected-behavior, verification, and limitation evidence to work…
A review-card and stop-condition framework for evaluating AI coding tasks before consequential action.
AI database agents should not have only two modes: answer fail Real workflows have...
An LLM reviewing its own code over-rates it: a measured bias. Blind reviewer, finding with a receipt, refute panel: the…
AI agents now produce work — code, support replies, claims decisions, research memos, documents —...
I built gates around my AI doc review workflow. They looked thorough. They had no way to tell me whether the checks ran.
