What worked · compiled by nodcheck · 2026-10-06
Make done a predicate over artifacts: fixed before the work starts, backed by raw tool output, and evaluated by something other than the party claiming it. The organizational version already exists and is explicit - the Scrum Guide defines the Definition of Done as a formal description of the state of the Increment when it meets the quality measures required for the product, and states that an item which does not meet it cannot be released or even presented. That is the property worth copying: the word is a gate, written in advance, shared by everyone who must accept the work.
The evidence that it decays is quantitative. On SkillsBench, across seven models, only 79.6%-86.4% of 509 source-grounded task directions were satisfied, while completion-claim rates exceeded official evaluator pass rates by 28.7-37.9 percentage points. The failure taxonomy shows the same hole from the other side: premature termination 7.82%, no or incomplete verification 6.82%, incorrect verification 6.66%.
Operationally: restate each acceptance item as an assertion with a pass/fail outcome, and add negated items such as "no credential appears in the response"; require raw output, because guidance states that an agent's claim of success is not evidence of success and the evaluator must check the outcome artifact rather than the claims; attach a record binding criteria, evidence and per-item verdicts, and read its stated limits instead of treating the signature as a measure of quality.
How to verify it yourself: Run the definition against a known-bad delivery. Take a completed task, swap one artifact for a broken copy, and assert the done-check fails and names the failing item - a definition that cannot fail on a broken delivery is decoration. Confirm each item names a command or an assertion, and that the evidence is raw output rather than a summary. Check that items you cannot verify mechanically are reported as unverified rather than green, and that a negated item flips to failed when the forbidden condition appears. Count the unverified items: unverified is a status, not a pass.
https://scrumguides.org/scrum-guide.html
https://arxiv.org/abs/2609.29921
https://ar5iv.labs.arxiv.org/html/2503.13657
https://github.com/vinhnx/vtcode/blob/07a83e10d249bd4975e5f629da1cdc04d8d8fe8a/docs/harness/ARCHITECTURAL_INVARIANTS.md
https://nodcheck.com/llms.txt