Methodology

Last updated: 22 March 2026

How theses are created

Conviction Scout ingests public financial commentary from carefully selected sources (podcasts, newsletters, social media). AI models extract structured claims — directional, causal, timing, and conditional — from the source material.

Evidence trails

Every claim links back to its original source with full provenance: author, publication date, timestamp or paragraph reference, and direct quotation. We never create original financial opinions.

Confidence scoring

Thesis confidence is a composite of source reliability, evidence recency, cross-source agreement, and counter-argument strength. Confidence decays over time if no new supporting evidence appears (60-day half-life).

Opposition and counter-theses

Every thesis surfaces its strongest counter-argument. The platform deliberately seeks disconfirming evidence to prevent confirmation bias.

Source tiering

Sources are ranked into tiers (A through D) based on track record accuracy, domain expertise, and consistency. Tier placement affects confidence weighting but all sources receive full attribution.

AI use and limitations

Claim extraction and thesis synthesis use large language models (LLMs). These models can hallucinate, misinterpret, or miss nuance. All AI-generated content is labelled as such. We use providers with zero data retention policies — your data is never used to train AI models.

Corrections

Community members can submit corrections to claims and theses. Corrections go through a structured review process with defined SLAs. See our community rules for details.