Scoring Engine How conviction, AI resilience, and financial health scores are computed

Platform Overview →

The pipeline uses a multi-layered scoring framework to evaluate investment opportunities. Every score is deterministic (reproducible from the same inputs), except AI Resilience which uses LLM-generated assessments calibrated against reference anchors.

How to read this guide

Each module page shows the components, formulas, thresholds, and data sources. Scores are 0–100 unless noted otherwise. All thresholds and weights are configurable via the score_config table.

Core Modules

Conviction Score

The primary investment signal. Four weighted dimensions: Fundamental, Thematic, Valuation, Catalyst. 0–100 composite with label thresholds.

AI Resilience

Five-dimension LLM-scored framework assessing how a company is positioned for AI disruption. Revenue Catalyst, Moat, OpLev, Pricing, Obsolescence.

Financial Strength

Balance sheet snapshot + three YoY trend signals. Sector-aware industry adjustments. 8 component ratios from quarterly financial data.

Signals & Trends

Estimate revisions, PE momentum, theme momentum, rating momentum, peer rankings, thesis consistency. TTL-based unified signal store.

Architecture Overview

conviction_scores (composite output)
  |
  +-- Fundamental (25%)
  |     beat_rate, surprise_quality, sentiment,
  |     estimate_trajectory, guidance, financial_strength
  |
  +-- Thematic (25%)
  |     ai_resilience, theme_health, peer_standing
  |
  +-- Valuation (25%)
  |     pe_percentile, peg, price_strength,
  |     net_debt_ocf (TTM, sector-aware),
  |     ev_fcf (TTM, sector-aware), pe_momentum
  |
  +-- Catalyst (25%)
        thesis_conviction (calibrated + penalty),
        rating_momentum, thesis_consistency

Data Flow

Raw data flows through three layers before reaching conviction scores:

Layer	Source	Output
Ingestion	FMP, yfinance, SEC XBRL, Reddit, news APIs	earnings_events, financial_health, companies, news_articles, analyst_ratings
Signals	Ingested tables + LLM analysis	ticker_signals, ai_scores, theme_momentum, peer_rankings, estimate_revisions
Scoring	Signals + company data	conviction_scores, investment_theses, earnings_previews

Configuration

All weights, thresholds, and labels are tunable at runtime via the score_config table without code changes:

Config Type	Key Settings
conviction	Dimension weights (default 25% each), label thresholds (Strong Buy ≥80, Buy ≥65, Hold ≥50), macro modifier toggle
ai_resilience	Dimension weights (RevCat 25%, Moat 25%, OpLev 15%, Pricing 20%, Obsol 15%), label thresholds
financial_strength	Industry adjustment factors (Energy 1.15, Utilities 1.15, Industrials 1.10)