Scoring & Enforcement
VERDICT does not guess your score. It operates on a strict, mathematically constrained rubric enforced by JSON Schema.
The Rubric
For each of the first 6 pillars (Positioning, Messaging, UX, Conversion, Trust, Defensibility), the engine assigns a score from 0 to 10 based on the following criteria:
- 0-3 (Critical Failure): The startup completely lacks this element. (e.g., No target audience identified, extremely confusing messaging, zero social proof).
- 4-6 (Weak/Generic): The element exists but is poorly executed. (e.g., "We help teams work better", fake-sounding testimonials).
- 7-8 (Competent): Good, standard execution. (e.g., Clear headline, decent onboarding).
- 9-10 (World-Class): Exceptional execution that creates a massive competitive advantage. (e.g., Undeniable ROI, deep network effects, viral onboarding loop).
Enforcing the Rubric (Beating Positivity Bias)
Standard LLMs suffer from Positivity Bias—they desperately want to tell you that you are doing a good job. Left unchecked, an LLM will give every startup a 9/10 for "Messaging" just because the website has words on it.
To counteract this, VERDICT uses:
- Aggressive Persona Prompting: The system prompt strictly commands the model to act as a cynical, hyper-critical auditor who actively searches for flaws.
- Schema Constraints: We use structured output (JSON schema) to force the model to justify its low scores before it outputs the number.
- Hard Limits: We instruct the model that scores of 9 or 10 must be incredibly rare and require overwhelming evidence.
The Overall Score (0-100)
The final Growth Readiness Scoreis a calculated weighted average of the 6 pillars, mapped to a 0-100 scale. It represents the company's overall readiness to scale and deploy capital into growth channels.