Writing

WritingProtocolOct 7, 2026Yasuhiro Shinsho

Review Is Not Approval

A clean review is often read as permission to proceed, and afterwards nobody can say who decided. Review asks whether the work is sound; approval asks whether it may proceed and who answers for that. This protocol keeps the two apart, binds each to the exact version it covered, and says what makes either one lapse.

4 min read7 core pointsBilingual
ProtocolHuman reviewReview cycleResponsibility boundariesExecution conditions

Article

Review Is Not Approval

The problem, in one sentence: a clean review gets treated as permission to proceed, and afterwards nobody can say who decided.

The principle

Review answers one question: is this sound, and what is still open?

Approval answers a different one: may this proceed, and who answers for that?

They are usually answered by different people, at different moments, from different authority. A clean answer to the first is never an answer to the second.

The structure

ReviewApproval
The questionIs it sound? What is still open?May it proceed, and who answers for that?
Who answerssomeone other than the authorsomeone who holds the authority for this kind of decision
What it leavesfindings and open questionsa recorded decision
What it is bound tothe exact version that was readthe exact version that was approved
What makes it lapseany change to what was reviewedany change to what was approved, or to the conditions it assumed

The order is fixed, and each step leaves its own record:

  1. The work is prepared — by a person, a vendor, or AI.
  2. It is reviewed — findings and open questions, on one exact version.
  3. It is approved, or not — a decision on that same version, by whoever holds the authority.
  4. It is carried out — by whoever performs the change, who checks that 2 and 3 still hold and adds no judgment of their own.
  5. The result is read back — the state that now exists is compared with the one that was approved.

When to use it

  • A reviewer's "no issues found" is being read as a go-ahead.
  • AI or a vendor prepares the work, and a person is said to "check" it.
  • Approvals are recorded against a document's name rather than its version.
  • Nobody can say what would make an existing approval stop applying.
  • The person who holds the authority is also the only one who can carry out the change.

The common failures

Review taken as approval. "It was reviewed" closes a question nobody actually asked. The reviewer answered whether it was sound; nobody answered whether it should go ahead.

Approval of a moving target. The work changes after it is approved — a fix, a late edit, a new version from the vendor — and the approval carries over silently, because it was attached to a name and not to what was read.

Authority merged into execution. The person with the authority to decide is also the one who must carry out every change, so the two get merged to save time. The decision stops being recorded separately, and the record shows only that something happened.

Implementation notes

  • Keep two records, not one: the review's findings and the approval decision.
  • Bind each record to the exact version it covered. A change to that version invalidates both, and the review is done again on the new one.
  • When it is unclear whether an approval still applies, treat it as not approved. Fail closed.
  • Let whoever holds the authority approve without having to perform the change, and let whoever performs it check the conditions mechanically rather than decide again.
  • Read the result back after the change, and record that the state reached is the one that was approved.

None of this needs a new system. It can usually be done with the documents, tickets, and approval steps already in place — what changes is what each record is allowed to mean.

Related

The Practice Note AI Delegation Reveals How Work Was Designed looks at the acceptance conditions this protocol assumes already exist. Whether a workflow's AI output has defined acceptance conditions and reviewers is one of the questions the AI Operating Model Diagnostic Kit works through directly.

What Fragment Practice works on

Fragment Practice works on where review, approval, and execution sit in a real piece of work: who reviews which version, who holds the authority to let it proceed, and what makes an approval stop applying. It does not take over the approval, and it does not carry out the change.

Practical entry points

When this theme becomes practical.

If the note feels close to your situation, choose the next entry point: reusable working material, context-specific support, or similar cases.

Reusable material

Structure this inside your own organization

If the job is to structure this inside your own organization first, AI Operating Model Diagnostic Kit is the reusable working material for this theme.

See the AI Operating Model Diagnostic Kit

Context-specific support

Structure a context-specific issue

Use Services when an active issue needs context-specific structuring around AI governance, security governance, decision material, review points, and responsibility boundaries.

Explore Services

Similar situations

Compare similar situations

Use Cases to compare this theme with common situations where AI adoption, governance, review points, or responsibility boundaries needed structure.

Explore Cases

Related notes

Continue with nearby themes

These notes sit close to the same theme or practical line of thought.

Sep 7, 2026

Practice Note

6 min read

Faster Tasks Do Not Mean the Work Moves On

AI can produce minutes and drafts immediately, but if people still connect the judgment, verification, records, and handoff every time, the way the wo…

Practice NoteAI-nativeWork continuityWork design

Jul 14, 2026

Practice Note

4 min read

AI Delegation Reveals How Work Was Designed

Delegating work to AI does not create a new problem so much as make an old one impossible to ignore: the scope, completion conditions, and acceptance…

Practice NoteAI delegationWork designAcceptance conditions

Next entry point

From public notes to practical work.

Writing captures the thinking behind decision-ready material. When a theme becomes practical, Products, Services, and Cases provide the next entry points.