Methodology
This page describes how the Stocklytics Score, AI Picks, fair value, and signals are produced. The figures are models for education. They are not a forecast you should treat as fact, and they are not advice. Read the disclaimer.
Stocklytics Score
Each company gets three pillar scores from 0 to 100: Value, Quality, and Growth. The Stocklytics Score is the rounded average of those three pillars.
A pillar is the rounded average of the metric scores we can calculate. If a pillar has no usable metrics, that pillar is treated as 50 so the overall average can still be formed. Metrics are compared with a US large-cap benchmark. The middle of that benchmark lands near 50. A saturating curve squashes the extremes toward 0 or 100 instead of spacing every name in a straight line. If the spread of a metric is too tight to use, the score falls back to a percentile rank. Figures that are missing, or so extreme they are treated as broken data, are left out.
Value uses forward price to earnings, PEG, free-cash-flow yield, enterprise value to EBITDA, the discount to the all-time high, and the discount to the one-year high. Lower is scored as better for forward price to earnings, PEG, and enterprise value to EBITDA. A company with no trailing profit scores 0 on forward price to earnings. PEG is only scored inside a bounded range.
Quality uses free-cash-flow margin, debt to equity, operating margin, the change in operating margin, return on capital employed, and the current ratio. Lower debt to equity scores higher.
Growth uses the earnings forecast, earnings growth over the last three months, revenue growth over twelve months, earnings growth over twelve months, a five-year earnings growth rate, and the upside implied by the analyst price forecast.
Trailing price to earnings, price to book, price to sales, price to free cash flow, and dividend yield are scored with the same curve for the valuation rows on a stock page. They are not part of the Stocklytics Score average.
Fair value
Fair value is a ten-year model of free cash flow, not a quote. Years 1 to 5 keep the starting growth rate. Years 6 to 10 fade that rate in a straight line to a terminal growth rate of 3%. The starting growth rate comes from analyst revenue estimates when those exist, otherwise from the company's own revenue history, and it is capped between 0% and 20%.
The discount rate starts from a simple cost of equity: a 4.3% base rate plus a 5.5% equity premium times beta. Beta is kept between 0.8 and 2.0. The rate is capped at 12% and is kept at least 5 percentage points above the 3% terminal rate. If beta is missing or not positive, the discount rate stays at 10%.
The cash flow we discount is a three-year average when that history exists, or trailing cash flow when that is the more meaningful base. Growth spending above depreciation can be added back, and the page says when that happens. Net debt is taken off the result. We do not publish a cash-flow value for banks, because the model does not suit them. If free cash flow is negative or too unstable, we do not publish a cash-flow value.
Where we also have a sensible earnings figure and a typical price-to-earnings multiple (kept between 8 and 40), we average the cash-flow value and that earnings value. If only one of them exists, that one is the fair value. If the result sits more than 80% below the live price, we mark it not meaningful rather than show it as a target. A value that is mostly the terminal year is flagged as uncertain.
AI Picks
AI Picks are 8 to 12 US common stocks, redrawn on a schedule. A page view does not call the model. The pool is built from listed US shares that are actively trading, excluding funds and ETFs. It is mixed on purpose: the largest companies, lower-beta large caps (beta about 0.15 to 0.85 and a market value of at least $8 billion), dividend yields of about 1.5% to 10%, lower price-to-earnings large caps (about 4 to 16, and at least $8 billion), names at least 12% below the one-year high (at least $5 billion), faster revenue growth of about 12% or more, and large technology and communication-services names.
A language model then chooses the picks from that list only. It writes a short note and assigns a 1 to 10 rating and a confidence figure. The rating bands are 1–2 strong sell, 3–4 sell, 5–6 hold, 7–8 buy, and 9–10 strong buy. Those words are the model's label, not an instruction. The model is not allowed to invent a price, a score, or a growth rate. A check drops a pick that cites a number that was not on the candidate row, apart from the rating, the confidence, and a portfolio weight where a model portfolio is built the same way. The note stays as written. Later quote refreshes can update the price shown beside it.
Separate stock reports are written ahead of time from the figures already collected for that company. Opening a stock page does not call the model. A question in the research box is answered from the saved note, or it says there is no note yet.
Signals
Signals look for a price channel on 5-minute prices. The scan uses recent bars (about 180 bars inside a short window of history). A channel needs enough span, repeated touches of the upper and lower lines, most candle bodies inside the band (about 65%), and no long run of closes outside it. A buy setup is a tag of the lower line. A sell setup is a tag of the upper line. A 14-period relative strength index and a 14-period average true range sit beside the entry. The levels are a description of the pattern, not an order to trade.
Limits
Market data can be delayed or wrong. A missing filing, a one-off accounting item, or a thin stock can move a score without the business having changed in the way the number suggests. Models are not updated tick by tick. See the editorial policy for how we correct a mistake.