Answer each question out loud before opening it. If you miss more than three in the checkpoint, revisit the lessons linked at the end.
A decision by named people that residual risk is acceptable, based on evidence; not a proof of correctness.
A set agreed at planning: evidence per feature, coverage decided, proofs, planted-bug detection, bug trend with effort, open bugs decided, final regression, reviews.
When closing it costs more than its risk, with a specific, justified, owned, approved waiver and a follow-up.
Answer with residual risk: known gaps, waivers, assumptions, an estimate of unknown bugs with a range, and scope.
A package in advance, criteria walked one by one, open items and waivers with owners, independent questions, a recorded decision.
Task. You are asked to sign off a DMA engine. Coverage is 97% with 40 excluded bins, all regressions pass, no critical bugs are open, formal was not used, nobody has run planted bugs, and the bug curve found 3 bugs last week. Write what you would ask for, what you would conclude, and the decision you would recommend.
Ask for: the plan with evidence per feature (which features lack evidence?); the 40 exclusions with reasons and approvers (unreachable proved, or waived?); the remaining 3% holes by cause; a planted-bug or mutation run to show the checks detect bugs; formal on the descriptor state machine and arbitration, or a justification; the bug curve with effort beside it and what technique found last week's 3 bugs; open major bugs and their triage; changes in the last month and how each was re-verified. Conclude: the evidence shows exercise but not detection, exclusions are unexamined and bugs are still arriving, so the residual risk is not stated. Recommend: not yet; conditions to reach go are planted-bug detection on all interfaces, reviewed exclusions, a flattening curve over two more weeks with sustained effort, and a written residual-risk statement.
The one that states what was not verified.
Planted bugs or mutations caught.
A falling rate can come from falling effort.
What, why not closed, evidence, consequence if wrong, owner and approvers, follow-up.
0.953 flits per cycle: one idle cycle between back-to-back packets, because a new grant follows the tail by a cycle.
A specification correction after the architects confirmed the bandwidth budget, plus a silicon validation item.
Known gaps, accepted waivers, unchecked assumptions, unknown bugs.
Go, go with conditions, not yet.
BUG_16 and BUG_23.
Go with conditions: the bandwidth budget confirmed and the saturation loops added or waived; silicon validation starting with throughput, counter saturation and sleep and wake.