A critique that shows its work

A score with no breakdown is just a vibe. "3 out of 5" tells you a verdict but not why, and a verdict you can't interrogate is a verdict you can't act on. This release makes the PRD critique explain itself — and makes it harder to fool.

Scores you can interrogate

The headline 1–5 stays, but now it comes with a breakdown. Every critique shows how it scored across the dimensions that matter — Strategic Clarity, Engineering Readiness, User Flow Coverage, Domain & Risk, and Completeness — each with a short rationale and a readiness band (Excellent → Not Ready). When a PRD lands a 3, you can see which dimension dragged it down instead of guessing.

Deterministic structural checks

Some gaps don't need an AI to spot. The critique now runs cheap, deterministic checks before and alongside the model: user stories written for a generic "user" instead of a real persona, requirements with no MoSCoW priority, acceptance criteria that aren't testable Given/When/Then. These surface as anchored findings in your editor, the same as every other comment — fix them in place.

Rigorous mode

The most dangerous thing an AI writing assistant does is flatter you. Ask it to fill a PRD and it will happily invent a baseline metric, a stakeholder name, a design link — all plausible, all fabricated, all the kind of detail a reviewer assumes you verified. Turn on Rigorous mode when generating a draft and Thinkr does the opposite: anything the brief doesn't support becomes a visible [TBD: …] marker instead of a confident guess. Honest gaps you can close beat invented facts you'll never catch.

The critique respects this too — it treats a [TBD] as an honest gap and tells you how to gather the missing input, never to invent it.

Sharper everywhere else

The review also got more demanding under the hood: it now checks instrumentation events for noun_verb naming and metric mapping, and grounds its compliance lens with named regulations (UU PDP, OJK eKYC, GDPR, PCI-DSS, HIPAA) when a PRD touches sensitive data — so "consider compliance" becomes "name the regulation and the owner."

What's next

Per-org weighting for the score dimensions, and a deterministic pre-check you can run before spending a full critique. Tell us what your reviews keep missing at support@usethinkr.com.