Skip to main content

Applied review assessment: AR1–AR6

Score each gate separately: 0 missing or wrong, 1 works only with prompts or copied reasoning, 2 independently correct with reproducible evidence and stated limits, 3 also handles a fresh variant and compares alternatives. Every gate must reach 2. Quiz percentages and strong prose cannot compensate for a missing regression or unsafe recovery.

  • AR1 — Contract and scope: state matching, snapshot, ownership and trusted-caller assumptions; distinguish a verified defect from an open requirement question.
  • AR2 — Review findings: supply location, trigger, impact and verification for tenant filtering and cache scope; distinguish blockers from optional suggestions. No comment quota.
  • AR3 — Regression evidence: retain a behavioral failure before repair and pass afterward, including overlapping ids and a miss-then-hit sequence. Demonstrate that restoring one relevant fault makes your regression fail.
  • AR4 — Ownership and repair: preserve the input snapshot and fresh results, pass checks, and explain a changed sequence unaided. Accept an uncached design if it meets the contract.
  • AR5 — Decision and handoff: compare two viable designs, record costs and a reconsideration trigger, separate bug fix from cleanup, and propose recovery that preserves isolation. Explain why a revert cannot undo disclosure.
  • AR6 — Independent transfer and integrity: after at least a day, reproduce a changed case without solution guidance; record assistance, actual verdict and limitations. Label self-review or simulated feedback honestly.

Keep code, commands, outputs, review notes and score/repair/recheck dates together. If a gate fails, repair that gap and attempt a different case; do not restart unrelated completed work. Use the calibration key only after your attempt. The separate HCI/accessibility gates remain mandatory.

Practice mapping​

Use the review workbench: P01–P05 support AR1–AR2; P06–P10 and E1 support AR3–AR4; P11–P15 and E2 support AR2/AR5; P20 and E3 support AR6. P16–P19 reuse the separate required human-centered/accessibility evidence and cannot be replaced by lookup tests. Practice calibration covers the three original review samples as well. Original module hours and delayed checks are unchanged.