Methodology

• Fundamentals:SEC EDGAR — the official U.S. regulatory filing database. Annual 10-K filings provide the primary historical series. Quarterly 10-Q filings are used to compute trailing-twelve-month (TTM) figures for income statement metrics, cutting typical fundamental data lag from ~15 months to ~3 months. Balance sheet metrics use the most recent quarterly snapshot. XBRL concepts are sourced directly from SEC EDGAR machine-readable filings.
• Price / market data:Third-party market data provider, used as input to factor analysis. Data is refreshed periodically; it is not real-time and may be subject to occasional delays or interruptions. FactorForge does not redistribute live quotes and is not a market data vendor.
• ETF holdings:Sourced directly from issuer publications (iShares, SPDR, Invesco). Vanguard ETFs use an alias mapping to equivalent funds with machine-readable holdings.

Refresh cadence: universe and ETF holdings refresh weekly on Sunday (1am–5am Central Time); fundamental data caches are warmed daily. Most data points in the tool reflect the most recent completed refresh, not intraday movement.

Every holding is evaluated through six factor frameworks that run independently — they do not share results, and they can disagree.

Each metric is scored on a continuous scale and weighted by how reliably it contributes to that lens in historical data. Metric weights are tuned per lens.

Within each lens, individual metric scores sum to a weighted lens score. The score is then expressed as a ratio against the maximum possible for that lens (determined by which metrics had data available).

• Strong — ratio ≥ +30%
• Moderate — ratio between −30% and +30%
• Weak — ratio ≤ −30%

The ±30% threshold is calibrated from the distribution of scores across the covered universe — it is not derived from a formal statistical test. Thresholds may be refined as the model evolves.

Sector calibration. The same metric means different things in different industries. A P/E that signals undervaluation in technology would look stretched in utilities. Every threshold is sector-adjusted — a semiconductor is compared to other semiconductors, not to the broad market.

Each verdict ships with a confidence level based on how much of the lens's weighted data was available and how consistently those metrics pointed the same direction.

• HIGH — most weighted metrics are present and point the same direction.
• MEDIUM — signals conflict, data coverage is partial, or an edge case limits what the model can reliably compute.
• LOW — too few scoreable metrics to reach a reliable conclusion; common for companies with limited filing history or unusual financial structures.

Factor models have documented edges and documented blind spots. The material ones:

• Momentum can invert in narrative-driven markets. The momentum lens is calibrated for fundamentals-driven environments. When markets are driven by narrative or liquidity rather than earnings, price returns can run ahead of what the model tracks.
• EDGAR data lags. Fundamental data comes from filed 10-K and 10-Q reports. Quarterly 10-Q filings are incorporated as soon as they are filed, reducing lag to approximately 3 months for most metrics. Within-quarter developments — events that occur after the most recent 10-Q filing date — will not appear until the next quarterly filing.
• International ETF sector weights are approximated. For funds holding international equities, sector allocation is estimated rather than sourced from primary filings. Scores for these funds carry more uncertainty than domestic ETF scores.
• Regime sensitivity is historical correlation, not prediction. Regime signals describe how sector exposures have historically behaved in similar environments. They are not forecasts, and factor regimes do not repeat cleanly.
• Known sector misfits exist.Some business models don't map cleanly onto sector-calibrated thresholds — asset-light financials (payment networks, exchanges), slow-growth staples with elevated PEG, and companies with negative book equity from aggressive buybacks can all produce verdicts that feel off. Where the model detects these patterns it marks them as notes; for patterns it doesn't detect, the score should be read in context.
• The model scores what it can measure. It has no view on management quality, competitive moats, regulatory risk, or macro. A company with excellent financials and a strategy headed the wrong direction will score well until the financials catch up.
• Point-in-time only. The model reflects the portfolio as it is now — there is no time-series tracking of how factor scores, sector allocation, or verdicts change across runs. A snapshot-persistence layer is on the post-launch roadmap.

FactorForge is an educational factor analysis tool. It describes how a portfolio or individual holding aligns with well-established investment styles, using publicly available fundamental data.

FactorForge is not an investment advisor, broker-dealer, or market data vendor. It does not issue buy, sell, or hold recommendations. It does not perform personalized suitability analysis and does not consider your financial situation, goals, or risk tolerance. The same model is applied uniformly to every portfolio; no analysis is tailored to an individual user.

You are solely responsible for any investment decisions you make. Always consult a qualified financial advisor before acting on investment decisions.