How to push back on a critique finding

A critique that only agrees with you is noise dressed as feedback. But a critique you can't argue with isn't much better — it's just a different kind of black box.

Push back closes that gap. When a finding is wrong, you say why, and Thinkr rules on it: upheld or withdrawn. The exchange stays attached to the finding, so "we considered this and rejected it" is a thing you can show someone, not a thing you have to remember.

Where it lives

Open a PRD, go to the Findings tab in the left panel, and run a critique if you haven't. Every finding card now has three actions: Push back, Resolve, Delete.

• Resolve means I fixed this.
• Delete means get this off my screen.
• Push back means you're wrong, and here's why — the only one of the three that gets an answer.

Click Push back, type your argument, and send it. Thinkr re-reads your PRD as it stands right now (not the version it critiqued) and rules.

What actually wins an argument

Thinkr's default is to hold its position. It backs down for exactly one reason: the answer is already in your PRD and it missed it. And it has to prove that by quoting the passage, word for word.

This isn't a stylistic preference — it's enforced in code. If Thinkr decides to concede but can't produce a verbatim quote from your document, the concession is rejected and the finding stands. You'll see it say so.

So:

This wins:

"The metric owner is named — the Success Metrics section says it's owned by the growth PM."

This does not:

"I don't think that's a real problem."

"We already talked about this in standup."

"Our team doesn't work that way."

None of those are wrong, exactly. They're just not in the document — and a PRD is judged as read, not as intended. If the context lives in someone's head, the finding is right that it's missing.

The three things that can happen

Finding upheld. Thinkr explains why your rebuttal doesn't change what the PRD says. The finding stays open. You can still resolve or delete it — pushing back is not a permission request.

Finding upheld, severity changed. Your argument didn't kill the finding but it did change how much it matters. The card shows a Severity → badge. A good argument should be able to win something even when it doesn't win outright.

Withdrawn by Thinkr. The finding was wrong, and Thinkr quotes the passage that proves it. The card is marked Withdrawn by Thinkr and drops out of your open count — deliberately labelled differently from Resolved, because a withdrawal isn't a fix and the record shouldn't pretend it was.

Deterministic findings answer instantly

Some findings don't come from a model at all — generic user-story roles, missing MoSCoW priorities, untestable acceptance criteria, non-noun_verb event names. Those are structural checks that run the same way every time.

Push back on one and there's nothing to debate: Thinkr re-runs the check against your current draft. Still fires, the finding stands. No longer fires, it's withdrawn on the spot. You'll see "Deterministic check — re-run, not re-judged" on the ruling. It's instant and costs you nothing.

Two rounds, then stop

You get two pushes per finding. That's not a paywall — it's a guardrail. If two rounds haven't settled it, a third won't either, and the honest move at that point is to resolve it, leave it open, or go change the doc.

Why it works this way

The whole product rests on one bet: the scarce skill isn't writing PRDs, it's reviewing them — and the review is only worth something if it will disagree with you.

That's easy to say and hard to build, because a model asked "the user disagrees, were you wrong?" will almost always fold. Being agreeable is the default failure mode of every AI writing tool, and it's exactly what makes a flattering draft ship a blocker.

The verbatim-quote rule is the fix. Thinkr isn't allowed to be persuaded — only to be shown. Argue well and it'll say so. Argue loudly and it won't care.

Ready to try it? Open Thinkr, run a critique, and pick a finding you think is wrong.