How the TickerWhale numbers are built, in plain English. Every figure is computed by deterministic code from filed fundamentals, market data, and deduplicated news — the AI only ever explains computed numbers, it never invents one. Models are versioned (research_v4, fairvalue_v12), so every score can be traced to the exact configuration that produced it.
The headline number blends five factor families. Each fundamental metric is z-scored against the company's peer group — industry first (minimum 5 members), then sector, then the whole covered universe — on outlier-winsorized samples, so a name is judged against its real competitors, never a global average. 70+ is strong, 40–69 mixed, below 40 weak. It is not a price target.
Margins, return on equity, and revenue/earnings growth, each compared to the company's own peer group (industry first, sector next, market last). Durable growth counts; a one-off earnings spike above +50% YoY is excluded from PEG rather than treated as proof of cheapness.
Trend, momentum persistence (1/3/6/12-month), relative strength vs the S&P 500, volatility, drawdown, and liquidity. This family is also shown on its own as Technical Pulse for timing context.
P/E, EV/EBITDA, EV/Sales and PEG versus the peer group — cheaper than peers scores higher. A P/E that has spiked past 3x the peer median is treated as a trough-earnings artifact and excluded instead of flooring the score.
Leverage, realized volatility, and drawdown behavior. High risk drags the score even when other families look strong.
Machine-read, deduplicated news clusters. Entity resolution is confidence-gated, so a story is never attached to the wrong ticker. A name with no qualifying news simply carries no news score; the news weight is never silently redistributed.
When a family's data is missing the score degrades honestly and its confidence level drops — it never fills gaps with invented values.
A modeled fair-value range with a confidence level, built in four explicit stages:
The engine picks the right toolkit per business: DCF + earnings-power + peer multiples for operating companies, a multi-stage growth DCF for fast growers (using the company's own cash-flow margin, not a fixed assumption), justified P/B for banks and insurers, FFO multiples for REITs. Pre-profit stories get no number at all.
The central estimate is the median of the sub-models, and the range is their interquartile band — a single extreme model can never drag the target. Outliers are winsorized, not deleted.
When the cash-flow models and the market-comparable models disagree wildly, the target is anchored to the street-like comparable and dropped to low confidence — but the central value itself is never clamped to a round boundary. Extreme-but-corroborated targets keep their real value for ranking; the app simply prints anything beyond ±70% as a '±70%+' bound instead of a false-precision number.
No number is shown when sub-models are uncorroborated, when the band is wider than the estimate is meaningful, or when the data basis is inconsistent (for example a share-class mismatch). The app says 'no reliable estimate' instead of guessing.
Score and target answer different questions — quality/momentum vs price. The scorecard labels the combination explicitly ("strong profile · rich price" and its converse) so the two never blur.