Decide whether finished work is actually finished

Situation

A worker reports success. The report is a paragraph, the evidence is the paragraph, and accepting it is a matter of trust rather than of checking.

What you run

• question: anything unresolved is recorded as state, and refuses completion while it stands
• finish: --verify-command runs the project's own verification and records its exit code and duration

Scenario

setup: finish-ready
given:
  - 'an active task with its completion note written'
steps:
  - id: open-a-question
    run: ['question', 'add', 'should tabs be an error or a warning?']
    note: 'an unresolved question naming the task'
    expect:
      exit: 0
      stdout_contains: ['^opened for t-']
  - id: refused-while-open
    run: ['finish', '--outcome', 'completed', '--verify-command', 'true']
    note: 'an open question is a blocker; completed is refused'
    expect:
      exit: 10
      stdout_contains: ['^FAIL blockers', 'refused']
  - id: resolve-it
    run: ['question', 'resolve', '1', '--answer', 'an error']
    note: 'the question is answered as state, not in prose'
    expect:
      exit: 0
      stdout_contains: ['resolved']
  - id: accepted
    run: ['finish', '--outcome', 'completed', '--verify-command', 'true']
    note: 'every line of the finish contract passes and the outcome is recorded'
    expect:
      exit: 0
      stdout_contains: ['completed']
then:
  - 'finish writes nothing while any line of the contract fails'
  - 'the verification command, its exit code and duration are recorded'

Outcome

A refusal names the rules that refused. An outcome is typed — completed, partial, blocked, no_match, failed — so "the thing does not exist" and "I could not do it" stay different facts.