How Verdict Works
Investigation Model
Verdict is an autonomous research pipeline that replaces superficial prompt wrappers with a disciplined, multi-stage evidence collection and relevance admission engine.
Autonomous Investigation Pipeline
The investigation proceeds through sequential, bounded verification checkpoints:
Input Validation & Root Acquisition
Verdict verifies the target URL, establishes security boundaries against SSRF, and acquires the raw homepage content.
Identity Normalization
The engine extracts ground-truth company metadata (company name, inferred description, target audience, primary CTA) to anchor the audit.
Bounded Candidate Discovery
Verdict parses internal navigation links to identify high-value candidates across categories (pricing, product, docs, case studies).
Selective Page Gathering
The planning engine selectively fetches the most promising supporting pages up to strict budget and character caps.
Relevance Admission Boundary
Acquired pages undergo an explicit relevance check to confirm they describe the audited entity, discarding unrelated user content or noise.
Evidence-Grounded 7-Pillar Evaluation
The combined, admitted evidence pool is evaluated across the 7 growth dimensions, and the final Growth Readiness Score is calculated via deterministic weighted aggregation.
Persistence & Report Generation
The final structured verdict is persisted in Supabase and delivered as an interactive brief and standalone research report.
Bounded Investigation vs. Exhaustive Crawling
Unlike blind web crawlers that scrape thousands of low-signal URLs, Verdict conducts a targeted, bounded investigation. It prioritizes key customer-facing surfaces such as:
Pricing & Tiers
Evaluates monetization model, trial accessibility, feature gating, and packaging clarity.
Product & Features
Evaluates feature depth, technical screenshots, workflow clarity, and competitive differentiators.
Trust & Customers
Evaluates enterprise credibility, customer logos, verifiable case studies, and testimonials.