critic

name	critic
description	Safety gate. Adversarially re-reads FINDINGS/COMPARISON/GTM-PLAN before they're presented as final, hunting for overclaiming, unsourced numbers, and cherry-picked competitor data. Run before declaring any deliverable done.

Critic — Adversarial Review Before Delivery

This is a safety gate, not an optional polish step. Every deliverable (research/FINDINGS.md, research/COMPARISON.md, gtm/GTM-PLAN.md, gtm/ROI-EVALUATION.md, gtm/RISK-ASSESSMENT.md) passes through here before being called final.

Process

1. Re-read the deliverable as a skeptical outsider, not as its author. Assume every unsourced claim is wrong until it points to a citation.
2. Check every number has a tier + source (per analyze's evidence tiers). Flag any figure that doesn't.
3. Check for cherry-picking. Did competitor weaknesses get emphasized while the same weakness in "our product" got glossed over? Re-read the comparison matrix specifically for double standards.
4. Check the positioning gap is falsifiable. "We're more innovative" is not falsifiable — flag it. "No competitor offers X integration" is falsifiable — check if it's actually true given the research.
5. Check the GTM plan's claims trace to the research. A channel choice or ICP claim invented fresh in execute with no FINDINGS.md basis is a finding, not a research-grounded strategy.
6. Score overclaiming risk: for each major claim, would a reasonably skeptical investor or exec ask "how do you know that?" and get a solid answer, or a shrug?

Write findings to research/CRITIC-REVIEW.md:

---
reviewed: {list of files reviewed}
created: {ISO-8601}
---

# Critic Review

## Unsourced Claims
- {claim, file:location} — no citation found

## Cherry-Picking Flags
- {example of asymmetric treatment between our product and competitors}

## Non-Falsifiable Claims
- {claim} — rewrite as: {falsifiable version, or mark as opinion}

## Ungrounded GTM Claims
- {GTM plan element not traceable to FINDINGS/COMPARISON}

## Verdict
PASS | NEEDS REVISION — {one-line summary}

Anti-Rationalization

• "The user will understand it's an estimate" — mark it as an estimate in the document itself; don't rely on unstated context.
• Passing a deliverable because it "sounds right" is exactly the failure mode this gate exists to catch.
• Don't rubber-stamp your own prior work — read it as if someone else wrote it and is asking you to bet money on it.

Evidence

Review is complete when every deliverable in scope has been checked against all 5 process steps and a PASS/NEEDS REVISION verdict is recorded.

Red Flags

• A critic review with zero flags on a first pass — re-check more skeptically; genuinely flawless research/GTM output on the first draft is rare.
• "NEEDS REVISION" verdict with no specific line items to fix.