Current engine: stockbox-analysis-engine-v2.7.0
How StockBox builds an equity analysis
The engine separates source facts, deterministic calculations, scoring and presentation. Missing or conflicting data can lower coverage and confidence or block an assessment.
Valuation
P/E, EV/EBITDA, EV/Sales, free-cash-flow yield and valuation-specific methods when compatible data is available.
Growth
Revenue, earnings, EPS and free-cash-flow growth are evaluated across compatible periods instead of relying on one growth number.
Profitability
Margins and capital returns such as ROIC/ROA measure the economics of the operating business.
Financial health
Debt, liquidity, cash/debt relationships and interest coverage are used only when the underlying values are reported and compatible.
Quality
Capital efficiency, cash conversion, stability and per-share discipline contribute to business and earnings quality.
Risk
Balance-sheet resilience and bounded market-risk context are separated from the company-quality assessment.
Coverage and confidence
A dimension must reach a minimum evidence threshold before it can contribute normally. Report-level confidence reflects data coverage, consistency and model constraints; it is not a probability of future returns.
Archetypes
Banks, insurers, REITs and other specialized businesses are not forced through identical industrial-company logic. Unsupported archetypes can return an insufficient-data research view instead of an unsuitable conclusion.
Profile-weighted, facts unchanged
Research profiles can change dimension weights and presentation emphasis. They do not rewrite company facts, source values or deterministic calculations and are not a personal suitability assessment.
Research view and valuation constraints
The overall research view requires sufficient score, confidence and data coverage. When evidence is insufficient, StockBox shows that limitation rather than forcing a directional investment conclusion.
Continuous improvement without self-modifying production
Batch QA, failure classification, provider diagnostics and regression suites feed calibration. Any scoring or data-logic change requires a new version and tests before production; the engine never autonomously rewrites its rules from user outcomes.
Static benchmarks are versioned fallbacks, not live peer comparisons. Research views are model-based research outputs, not guaranteed outcomes or individualized financial advice. Read about data sources.