Verify the Reviewer's Premise, Not Only Its Verdict

Findings and Remedies Fail Independently

A reviewer hands you a finding and a suggested fix. The natural response is to treat them as one unit: if the finding is real, do the fix. They are two claims that fail separately, and the combinations need three different responses rather than one.

The instructive case is real finding, wrong remedy. One round raised a genuine defect, and the fix it recommended rested on a claim about observable behavior — that a particular operation would warn the reader. Measurement refuted it: the operation emitted nothing and behaved correctly. Following the remedy would have restructured a lesson around raw machinery the lesson was deliberately not about. The right move was to accept the finding, measure the premise, diverge on the fix, and log the divergence, so a future reader can see that the reviewer was heeded and where.

The Rules Also Fail Independently From the Verdicts

A subtler version shows up when a reviewer cites a rule. In one round a reviewer stated an orthography rule correctly and applied it wrongly; in the next, it stated a different rule wrongly and reached the right verdict anyway. Same reviewer, two rounds, both halves failing in opposite directions.

The practical consequence is that you cannot delegate the check to whether the reviewer sounded authoritative. A rule and a verdict are separate objects, and "the rule is right so the application is right" is a step you have to actually take rather than assume.

The Reviewer's Blind Spot Is Whatever You Withheld

If the review package is pre-extracted — prose without code bodies, changed fields without their surrounding context — the reviewer will confidently raise findings about the omitted material. One round raised a well-argued blocker that a referenced script did not exist anywhere in the work; it was sitting in a code block the package deliberately strips.

That is not a bad reviewer. It is a reviewer answering the question your package posed. The fix is to name the omissions in the request, along with anything already raised and declined in earlier rounds, so each round spends its attention on new ground.

Accept the finding, measure the premise, diverge on the fix, and record the divergence. A review log that only records agreements teaches a future reader that the reviewer was always right, which is both false and the exact belief that makes the next wrong remedy expensive.

Code

Four cases, three correct responses·text

FINDING   REMEDY    WHAT TO DO
--------  --------  --------------------------------------------
real      correct   apply it. the common case; move on.

real      WRONG     accept the finding. measure the premise the
                    remedy rests on. fix it your own way. LOG the
                    divergence, with the measurement.
                    <- the case that costs the most when missed

wrong     n/a       stet, with a reason. "declined: measured X"
                    is a real log entry; "declined" alone is not.
                    a stet is a POSITION, so expect it to return
                    in a later round wearing different words -
                    and re-judge it, because sometimes the second
                    phrasing is right where the first was wrong.

wrong     n/a       ...but the reviewer could not have known,
(package)           because your package withheld the evidence.
                    name the omission in the next request.


WHAT A GOOD DIVERGENCE LOOKS LIKE IN THE RECORD
  "R2-03 accepted, remedy declined. The finding is right - the
   lesson does over-promise. The remedy assumed the operation
   emits a warning; measured in a clean environment, it emits
   nothing and defaults correctly. Fixed by narrowing the claim
   instead of restructuring around the machinery."

Exercise

Take the last review comment you acted on and separate it into the finding and the remedy. Write down the factual claim the remedy depends on, then check it. Do this for five comments and count how many remedies rested on something nobody verified. That count is how much of your review process is running on the reviewer's confidence rather than on evidence.

Hint

The premise is usually a claim about behavior — what a tool does, what a user sees, what happens by default. Those are cheap to check and almost never checked, because the reviewer stating them sounds like someone who already did.

Findings and Remedies Fail Independently

The Rules Also Fail Independently From the Verdicts

The Reviewer's Blind Spot Is Whatever You Withheld

Code

External links

Exercise

Progress

Comments 0