Methodology:how every analysis is made

The rating is not freely selected by the language model. Software sets it after the text is written, from the margin of safety, the distance between current price and estimated fair value, and the fundamentals score, so the same figures give the same rating in every language. There is one ladder and it applies to both horizons. It is checked in this order and the first match applies:

• Strong Underperform: a fundamentals score below 3, or a price at least twice the estimated fair value.
• Underperform: a fundamentals score below 5, or a price at least 50% above the estimated fair value.
• Strong Outperform: a margin of safety of at least 40% and a fundamentals score of at least 8.
• Outperform: a margin of safety of at least 20% and a fundamentals score of at least 7, or a margin of safety of at least 10% for the lower rung of the same family.
• Monitor: no published fair-value estimate, or a fundamentals score of at least 7 without enough margin of safety.
• Market Perform: everything else.

So a stock trading above estimated fair value, or one with no published fair-value estimate, never receives an Outperform-family rating. A strong competitive advantage is a quality input that has already lifted the score. It is not a discount on the margin of safety and it lowers no rung.

The rule describes the price against the estimated value and the fundamentals. It is not a forecast.

This has applied since this version of the methodology took effect. How ratings were set before it is stated under the rating distribution and in the change log.

Beyond EDGAR, an analysis may use:

• Public and free-tier price interfaces for equity prices and price history, together with a local end-of-day store built from the same data, and for the S&P 500 (SPY) series used as the benchmark. No market-data licence underlies these; see the Terms.
• FRED for government bond yields, ICE BofA high-yield spreads and currency rates.
• The SEC Financial Statement Data Sets, for filings not yet shown by its main interface, and the SEC frames interface, for where a company stands among all companies that file.
• For banks: the FDIC (BankFind) for the reports of insured banks and the list of bank failures, and the Federal Reserve (FR Y-9C and FR Y-9LP, through the FFIEC) for the statements of the group and of the parent company.
• The FDA Orange Book and Purple Book for the patents and exclusivity periods of marketed drugs, and the companies' own filings for revenue by product.
• Damodaran (NYU Stern) for sector and country betas and risk premia.
• Nasdaq Trader for the list of tradable symbols and Wikipedia for the S&P 500 constituents.
• Marketaux for per-instrument news headlines and for aggregated news sentiment, computed as the average of the scores the provider itself assigns to those articles.
• Web search through the Anthropic model for recent news and sector context. Management tone in the MD&A section is computed from the text of the filings themselves and not from an external vendor.

Macro indicators such as the VIX, bond yields and the S&P 500 trend are derived from the sources above.

Every analysis goes through the same stages in the same order:

1. 1Macro context and quantitative screening against ten quality and valuation criteria with sector-aware thresholds, producing a fundamentals score from 0 to 10. Six of the ten mean nothing for a bank (a bank has no gross margin, for one), so a bank is scored on the rest and compared with other banks.
2. 2Review of recent sector and news context.
3. 3Stock-level research covering valuation, peers, multi-year quality signals, insider transactions, institutional and ETF holdings, revenue by region, dilution from stock-based pay and buybacks, risk factors the latest 10-K added or removed, the SEC's questions, lawsuits and loss contingencies from the notes to the statements, recent 8-K filings and catalysts.
4. 4Final rating under the published scale and the applicable rules.
5. 5A second Devil's Advocate pass, a separate model call over a digest of the same analysis, examining the weakest assumption, an overlooked risk, the bear case and a thesis-invalidating event.

The estimated fair value is a weighted combination of methods, chosen by the kind of company (next part). For most companies there are two: a DCF, which brings the free cash flow the company is expected to produce back to today's value, and multiples, which price its earnings, sales and operating profit (EBITDA) at what similar companies trade for. The weights are not published.

• The multiples are the medians of the company's peer group, a narrow group of similar companies whose medians are worked out again every week from their filings and prices. Where the group has fewer than five companies with a figure, the medians of its sector are used, and where those are missing, a fixed table by sector.
• For a profitable company the sales multiple is scaled to its own net margin against the group's. The P/E is also adjusted for the company's revenue growth against the median growth of its peer group, and the sales multiple, when it is not scaled to the margin, for both its growth and its gross margin against its sector's average.
• When gains from outside the business, such as investment income or the sale of assets, carry a large part of the earnings, the P/E prices the operating profit instead. When a tax credit carries them, the earnings are counted at a normal tax rate. When operating profit itself includes large gains from a sale or a mark on investments, at least a tenth of it over the last twelve months, the P/E and the EBITDA multiple price the company without them, and the analysis says so.
• Earnings Power Value and the Graham Number assume no growth, so they are not part of the midpoint: they are shown apart, as a floor. Earnings Power Value taxes operating profit at the company's own rate, never below 15%, and it is not worked out for banks, insurers or mortgage REITs, whose operating profit is already after the cost of the money they lend or invest. They enter the midpoint only when no other method can be worked out, and for a bank the Graham Number does when its book-value method cannot.
• Where three or more methods are available, the one furthest from the others is left out of the midpoint. The methods built for one kind of company (the book-value method, P/FFO, the dividend model, net cash) are never the one left out.
• A method that leaves the shares no value, because the debt is larger than the value of the business, is kept at zero rather than dropped, and the certainty grade is then at most medium. When three or more methods are available it is the one left out of the midpoint, but it still sets the bottom of the range.
• The fair-value range runs from the lowest to the highest method value. With one method only, it is that value plus and minus 20%.

The output may include a fair-value range, bear, base and bull scenarios, a certainty grade and a reverse DCF, the growth today's price implies.

The methods depend on the kind of company:

• Most companies: the DCF and the peer group's multiples.
• Banks: no DCF, because a bank's cash flow does not describe its business. The midpoint combines the P/E with a book-value method: book value plus the return above the cost of equity the bank has earned on it, averaged over its last three years and fading over time. A bank earning less than its cost of equity is valued below book and is given no growth. Sales and enterprise-value multiples do not describe a bank and are not used, and the Graham Number is shown only as a floor. With no free cash flow to read, a bank's certainty grade is never high.
• Insurers that underwrite risk (life, accident and health, property and casualty, surety and title): the same book-value method, on five years of returns, with the multiples. The loss ratio, claims against the premiums earned, is shown as a check and does not enter the value. Insurance brokers and health plans take the general path.
• Mortgage REITs, which hold loans rather than buildings: the book-value method alone, when at least three years of returns are on record and their average is positive, and no fair value otherwise. With one method, their certainty grade is low.
• Other REITs: price to FFO (earnings before the depreciation of buildings, after preferred dividends) at a multiple set for their kind of property, beside the DCF and the multiples. AFFO, which also takes off three years' average spending on the properties, and the share of it paid out as dividends are shown but do not set the value.
• Utilities: a dividend model, discounted at the cost of equity, beside the DCF and the multiples. The dividend grows at its own pace per share of the last five years, which in our tests forecast utilities' dividends better than the growth the company can fund from the profit it keeps (its return on equity times the share of profit it does not pay out). When the two differ, the analysis says so.
• Companies that lose money but grow fast: a growth DCF. Revenue grows at the company's recent pace, held within a limit set for its sector, the margin moves toward its sector's mature level, and every year's growth has to be paid for with new investment, at the sales the company and its peers get from each dollar of capital. The result is then weighted by the chance the company fails first, which rises as its cash runway shortens (or is higher still when the share price and the debt imply it), with a failed company counted at half its book value.
• Drug and biotech companies with little or no revenue yet: net cash per share, beside whatever else can be worked out.
• Cyclical businesses: they are identified by industry, with at least seven years on record and a cycle visible in their own last decade of margins or revenue, leaving out companies whose margins have risen steadily or stepped up, unless they sell a commodity. They are valued on mid-cycle figures. Mid-cycle cash flow is the median margin of up to the last ten years applied to trend revenue. When today's cash flow is under half of it, the mid-cycle figure replaces it; otherwise today's converges to it over three years when revenue sits clearly above or below its trend. The multiples price mid-cycle earnings at the peer group's own price over mid-cycle earnings, not over the last year's, and the DCF grows at the revenue trend, not at the last year's growth. For a cyclical that is also a serial acquirer, the trend loses what the purchases added, as below.
• Shrinking companies (revenue falling by more than 2% a year over five years, down in at least three of them, no single year down more than a quarter, not cyclical, cash flow still positive): the DCF may start from a fall in cash flow, and it fades to no growth rather than to the long-run rate.
• Serial acquirers (buying businesses in at least three of the last five years, for a material share of their operating cash flow): the DCF grows at the growth the company did not buy, its reported growth less what the purchases added, never below 2% for a company that was growing.
• Makers and sellers with a lending arm that holds a material share of their assets, such as a carmaker financing its buyers: no DCF, because the arm's lending runs through the cash flow. The P/E, on mid-cycle earnings for a cyclical, and the book-value method on the consolidated accounts. Their certainty grade is at most medium.
• Parts of today's cash flow with a known end are modelled as ending: a drug's sales after its U.S. exclusivity ends, for companies whose revenue is concentrated in one or two molecules (dates from the FDA Orange Book and Purple Book and from generic settlements; cell and gene therapies, which no biosimilar has ever copied, get no end date), with a date further off counting less, since a successor becomes likelier (in full within ten years, not at all beyond twenty), and the drugs still in development valued at what their research cost, at most what the ends take away; and tax not paid while loss carryforwards last. For a drug company the scenarios then vary the date exclusivity ends instead of the growth rate.
• Digital-asset treasury companies: no fair value, because the shares move with the coins they hold. The value of their holdings per share at the filing date (mNAV) is shown instead.

No fair value is published when the filings are missing or too old; for royalty and energy trusts, whose reserves run out; for digital-asset treasury companies; for a mortgage REIT without the record above; for a maker with a lending arm and no earnings to price; when no method can be worked out; when the debt cannot be read reliably (no debt line beside large liabilities or an interest bill, or an interest bill the debt we read cannot explain); and when the fair value is at least four times the price, or more than twice the price with a low certainty grade.

The methods that value the whole business, such as the DCF and EV/EBITDA, reach the value of the shares the same way:

• subtracted: debt, including finance leases; preferred stock; the stakes outside holders have in subsidiaries (minority interests); obligations to retire assets, net of the funds set aside for them; accruals for environmental clean-up; and the deficit of pension plans, after tax (a surplus is not added);
• added: cash and short-term investments, long-term securities, and stakes in other companies at their value in the accounts.

Operating leases are not subtracted as debt, because the cash flow has already paid their rent. For banks, insurers, brokers and other financial companies, the only claims beyond debt and cash are preferred stock and minority interests: their securities and stakes are their business. The methods that price the shares directly, such as the P/E, the book-value method, the dividend model and P/FFO, need no bridge.

The interest earned on cash and securities is taken out of the cash flow, so the cash is not counted twice; where the company does not report it and the amount matters, it is estimated at the 10-year Treasury yield. Stock-based pay is deducted from free cash flow before the DCF, as the cost it is. The value is then divided by the diluted share count: for a profitable company, its own ratio of diluted to basic shares; for a loss-making one, the options, share units, warrants and convertible bonds worth exercising at today's price. A convertible counted as shares is taken out of the debt.

• The DCF starts from the company's own growth in cash flow and revenue, blended with its sector's, and it cannot grow faster than the company can fund: its return on invested capital times the share of profit it reinvests, plus a limited allowance for wider margins. It starts from a free cash flow that is half the last twelve months' and half the company's average free cash flow margin of its last three years applied to today's revenue, because a single year is a noisy starting point: in our tests the blend forecast the next years' margins better.
• That growth is held for five years for every company, then slows each year until year ten to a long-run rate, which is then held for ever. A longer window for companies with a decade of high returns on capital was tested and dropped, because they did not grow faster in the years it added. The long-run rate is the 10-year Treasury yield less 1.5 points, within a floor and a ceiling.
• Each year's cash flow is discounted from the middle of the year, not from its last day, because a company takes in its cash through the year.
• The cash flow is discounted at the weighted cost of capital. The cost of equity is the 10-year Treasury yield plus the company's beta, measured on five years of its weekly share prices against the S&P 500 and moved a third of the way toward 1 (its sector's beta when the price history is shorter or the shares trade thinly), times the equity risk premium that Damodaran (NYU Stern) works out from S&P 500 prices, with a country premium for where the business is: by where it sells when its filings give its revenue by country, otherwise by the country of its address. The premium updates itself each month from the figures he publishes, so the one in use is his latest monthly estimate, not the day's. The cost of debt is the rate the company pays, never below the 10-year yield plus one point. A rate under 2% or over 15% is treated as unreadable, and the cost of equity is used alone. The dividend model and the book-value method are discounted at the cost of equity.
• The reverse DCF solves the same model, with the same window and long-run rate, for the growth today's price implies.

The certainty grade measures how closely the valuation methods in the midpoint agree with each other, not how likely the estimate is to be right.

• High: the gap between the highest and the lowest method is at most 35% of the midpoint, and the company had positive free cash flow and positive earnings over the last twelve months.
• Medium: the gap is at most 60% of the midpoint.
• Low: a wider gap, or one method only.

It is then lowered for any of these reasons:

• a method leaves the shares no value: at most medium;
• the company rests on one or two drugs: at most medium, and low when a drug's share of revenue or the date its exclusivity ends could not be read;
• the growth today's price implies is six points or more away from the model's base case, or no growth rate the model can carry reproduces the price: at most medium;
• the fair value is more than twice the price and the no-growth floor does not agree with the other methods: at most medium, because a gap that large should not rest on growth alone;
• the company has a lending arm: at most medium;
• outside holders' stakes in its subsidiaries are larger than the company's own book equity: low;
• the estimate is under 40% of the price: low, because the market is paying for something the figures do not carry;
• the devil's advocate rates the thesis 5 out of 10 or lower: at most medium.

• Every value is an estimate built on the filings and on assumptions. It can be wrong, and it is most sensitive to growth and to the discount rate. It is not a forecast of the share price.
• For banks and insurers, the book-value method gives a price-to-book about a fifth below what the market pays, taken as a group (23% below for banks at the latest measurement, on 9 November 2026). The market pays for growth and returns the method does not carry. We state this as the model's reading rather than fit the model to market prices.
• The box "How common this is in the industry" counts companies with filings for the whole decade, so companies that failed or were bought are missing from it, and high growth can look more common than it was.
• For a company that reports in another currency, growth is measured in that currency while the value is in dollars. Where that currency's inflation has run more than two points a year away from the dollar's, the analysis says so. The value is not adjusted, because in our tests the adjustment did not forecast growth in dollars better.

If an estimate is not sufficiently reliable under the system's controls, it is withheld and the omission is explained on the analysis. For example: earnings driven mainly by an accounting tax credit, or a fair value that is a multiple of the price without sufficient certainty. We would rather show no estimate than a wrong one. Where two models disagree, the disagreement is published rather than a silent choice between them.

Prices are refreshed when the analysis runs and carry their own timestamp. Fundamentals reflect the latest available company filing. Caches retain data from a few minutes for prices to several weeks for structural data such as filing indexes, fund holdings and identifier crosswalks. The Market Radar is refreshed as a rule every Saturday, through a manual process rather than an automated schedule, and the list states its scan date. Open track-record entries are updated daily after market close. The watchlist calendar and the sector pages answer from each company's last entry, which may be the previous day's, while the day's is built behind, and the page says so for as long as that lasts.

Entry and exit rules are defined in advance and applied uniformly. A position closes on the first rule that fires: target reached, safety stop or time stop, with the per-horizon thresholds published alongside the track-record data. Returns are computed as price return: without dividends, without fees and costs, and without adjustment for corporate actions such as splits, until the licensed price provider is activated. The excess return is the position's return minus the S&P 500's (SPY) over the same holding period: a plain difference, on the same price-return basis, not adjusted for risk. Negative ratings are recorded on their own counter and are not blended into the returns of the positive picks.

We cover stocks that file with the SEC, so mainly US issuers. Companies without such filings cannot be analysed in depth. Analyses are produced with the assistance of large language models and may contain errors or omissions. Critical information should be verified against primary sources.

Under the title of each section of an analysis is one sentence saying in plain words what its figures show, with a coloured dot. These sentences are not written by a language model. They come from fixed rules over the same figures the section shows, for example legal liabilities as a share of the company's market value, or the change in the share count over five fiscal years. The dot says whether the picture on that measure is strong, mixed or weak, against fixed thresholds. It is not a rating of the stock and does not change the rating.

When someone analyses a company again, the analysis shows what changed since their own previous analysis of the same company: the rating, the scores, the price, the margin of safety and what the company filed with the SEC in between. The comparison is made by the software over the two stored analyses.

Around the analysis's text, the page shows a few boxes that no language model writes. The software works them out with fixed rules, and the analysis keeps the result as it was the first time it was opened, with its date:

• "How strict the model is here": among the listed companies of the same sector that our pages rank, how often the valuation model finds the price above fair value. When the sector has fewer than 15 companies, the comparison is with companies of a similar return on capital (ROIC). A company the model gives no fair value for counts neither way. It is measured again with every weekly scan.
• "How common this is in the industry": how many companies of the same industry and a similar size (revenue within four times larger or smaller) changed their free cash flow over the last decade at least as much as today's price assumes. The rate is measured between two three-year averages ten years apart, so one good or bad year does not decide it. When fewer than 20 companies share the size, the whole industry counts, and if that is too few as well, the box is not shown. It is not given for banks.
• "Quick check": up to eight tests, each with its limit written beside it: return on capital (ROIC) above 10%, net debt under 3 times EBITDA, positive free cash flow over 12 months, five-year average revenue growth above 0%, the same or fewer shares over five years, a price up to the top of the fair-value range, accruals under 5% of assets, and, for companies that pay a dividend, free cash flow at least 1.5 times the dividend. A test without data is not shown. The check is not the analysis's score and does not change it.
• "At a glance": when an analysis has several stocks, it puts them side by side on the same measures, with each one's price against its fair value on one shared scale. It describes the figures, it does not pick a stock.
• "New results since this analysis": if the company filed a new 10-Q, 10-K or 20-F, or an 8-K with results, with the SEC after the analysis's day, the page says so on the analysis, with a link to the filing. The analysis does not change by itself. Run it again and the new one shows what changed since the previous one.
• The "Disagree? Set your own growth" slider gives an approximation shaped like a ten-year DCF that passes through two points of the analysis: the price at the growth the price assumes, and the fair value at the base case. It is not a new estimate.

An ETF report does not end in a rating. It describes what a fund holds from its latest N-PORT filing with the SEC:

• its positions and their weights, and its largest stocks scored with the same quantitative screen every stock goes through,
• the same measures for the S&P 500, worked out the same way from SPY's holdings, for comparison,
• the fund's size, its net flows of money and its price path, with a note when net assets are under 50 million dollars,
• what changed in its positions since the previous filing, by share count, leaving out stock splits and beyond the fund's own flows,
• how it earns income when it does not hold only stocks: written call options, banks' structured notes, derivatives that raise its exposure, securities lending,
• its companies' dividends, the funds with the largest shared weight and, for bond funds, the yield to maturity, duration, credit quality and maturities of the bonds.

The yearly cost is from the fund's prospectus, for the ETF's own share class. The report's text is written by a language model over a digest of these figures, and any sentence with a figure that is not in the digest is removed before publication.

The free data pages do not go through this method and give no rating: the company pages (financial statements, valuation, results, dividends, ownership, filings), the sectors, the stock comparison, the DCF calculator, ETF overlap, the portfolio x-ray, dividends and bond ETFs. One figure is the exception: the Valuation tab of a company page can show our estimated fair value, worked out each day by the same valuation the analyses read, which this page describes, with its range, certainty grade and methods, and with no rating. They describe figures from companies' and funds' SEC filings, such as what a company stated in its latest 10-K, what an ETF holds, how well a dividend is covered or the yield to maturity of the bonds a fund holds, with no score and no ranking of ours. The DCF calculator starts from the filed figures and computes with the assumptions the reader enters. The portfolio x-ray adds up the ETF holdings and the stocks the reader enters, with their amounts, and shows which companies and sectors they end up in, which companies repeat and what the ETFs cost a year. It describes the composition, does not rate the positions, does not suggest changes, and the positions are not stored. When each of them changes is set out below, and each explains on its own page how every figure is arrived at. One block, “What the company measures itself”, is read by a language model: the Management's Discussion and Analysis of each 10-K is read once and the model copies the operating figures management reports (subscribers, deliveries, stores) with the sentence it took them from. A figure is published only when that sentence is in the report verbatim and carries the figure, when the company reports the metric in two consecutive years, and when the next report does not contradict it. Where one of the three is missing, the metric is not shown.

Each page changes at the pace of its source:

• Prices are read when a page is opened, or every few minutes in the background for the /stocks hub and the sectors, and kept for between a few minutes and an hour, so they run a little behind the market.
• A company's statements, results and filings are read from the SEC when its page is opened and kept for one to six hours.
• The /stocks hub, the sector pages and the watchlist calendar come from a build that runs in the background through the day. While a build runs they show the previous one, which may be the day before's, and the page says so.
• Insiders' trades back to 2006 and every manager's 13F positions come from data sets the SEC publishes each quarter, and the newest Forms 4 are read as they are filed. The notes to the statements come from a data set the SEC publishes each month.
• Segments, executive pay, say-on-pay votes and trading plans are read on Saturdays from the 10-Ks and proxy statements filed that week. What the company measures itself is read once from each 10-K.
• The dividends page, the bond ETF pages and the ETF overlap and holdings pages are rebuilt every Saturday, with the Market Radar. A fund's holdings change only when it files a new N-PORT, which the SEC makes public once a quarter.
• The DCF calculator and the comparison of companies are worked out when they are opened.

The company pages need no account, and one rule runs through every section: the recent is open, the history is on the plans. Without an account each section shows its last three fiscal years and four quarters (the last year of insiders' trades, the last four quarters of the managers' positions); a free account shows five years and eight quarters (three years of trades); the Investor and Pro plans show the full history, back to 2009 where the filings allow, and the Excel file of every page. The current figures and ratios, the latest year of every section and the filings list are open to everyone.

About 3,500 US-listed companies have a page, but not every block reaches all of them. Executive pay, say-on-pay votes, buybacks and trading plans are read for the companies of the S&P 500: today executive pay for 493 of them, say-on-pay for 447, buybacks for 458 and trading plans for 305. What the company measures itself has been read for 389 companies so far. Elsewhere these blocks are left off the page rather than shown empty.

The “Worth knowing” block on a company page lists, from its filing record, the events of the last three years a holder would want to be told of, each with its filing:

• 8-Ks that report a bankruptcy, a material cybersecurity incident, a notice from the exchange that a listing rule is not met, a change of auditor, or that earlier statements should no longer be relied on;
• notices that a 10-K or 10-Q will be late (the NT forms);
• a Form 25 that removes the shares from the exchange, or a Form 15 with which the company stops reporting, for 30 days after it is filed;
• doubt about the company's ability to continue as a going concern in a filing of the last two years, found through the SEC's full-text search. When the search cannot be read, the page says it was not checked, not that nothing was found;
• a merger, tender offer or going-private deal the company is a party to, from the filings of the last year;
• the SEC's litigation releases and administrative proceedings of the last 180 days. They are matched to the company by its name only, so one may concern another company with a similar name or people connected with it, and the SEC's release is linked so it can be checked.

The block lists what was looked for, so an empty list means none of these was found, not that nothing happened.

The Ownership tab reads the company's own filings and those of its insiders and holders:

• insiders' trades from their Forms 4 back to 2006, and what each insider holds after their latest Form 4;
• the trading plans (Rule 10b5-1) that officers and directors adopted or ended, as each 10-Q and 10-K has reported them since April 2023;
• the managers' positions by quarter from the 13F filings since 2013, with about thirty well-known managers marked by name. The list is ours and the same for every stock; being on it says nothing about the stock;
• the ETFs that hold the stock, from their N-PORT filings;
• executive pay against the share's performance, from the Pay versus Performance table of each proxy statement (DEF 14A) since 2023;
• the shareholders' vote on that pay (say-on-pay), from the 8-K the company files after its annual meeting (item 5.07);
• buybacks by quarter, from the repurchase table of each 10-Q and 10-K since 2021, each checked against the cash the cash-flow statement shows was paid for them.

Beside the statements, the Financials tab shows:

• revenue, and where the company gives it operating income, by segment or product and by country or region, from the XBRL of each 10-K, each year read from the latest report that carries it;
• from the notes to the statements (the SEC's Financial Statement and Notes data sets): when the debt falls due, year by year; what the company owes under leases; its pension plans, what they owe against what has been set aside; goodwill by segment; and the bridge from the US statutory tax rate to the rate the company actually paid.

The figures are as the company tagged them, one filing for each note and date. Where a company tags a note differently, or not at all, the card is left out.

A few figures on the pages combine the filings with the share price:

• on the Results tab, how the share moved on the first trading session after each results release, close to close, next to the S&P 500 on the same day. A release after the close is measured on the next session;
• on the comparison of companies, the total return: the price with each dividend reinvested on its ex-dividend day, beside the price change alone;
• on the Dividend tab, everything the company returned to shareholders over the last year, dividends and buybacks together, as a yield at today's price and per $1,000 of shares;
• on the Valuation tab, what today's price assumes: the DCF calculator's own model solved for growth instead of value, with ten years of the same free-cash-flow growth and 2.5% after, at a return of 7%, 9% and 11%. It is not shown for banks or in a year of negative free cash flow. For a REIT it works on free cash flow, while the analysis values a REIT on P/FFO, so the two can read differently.

None of these is a forecast or a rating.

The /stocks hub describes the market of the day, with no rating:

• the market map: the largest companies as tiles, each sized by the company's market value at today's price and coloured by the day's move;
• the day's largest moves, up and down, which can be narrowed to companies worth $2 billion or more;
• the week's results, out and due, with the date the company announced or, where it did not, our estimate from its earlier filings, marked as an estimate;
• new listings (IPOs): who filed to list (S-1 or F-1) and who priced (424B4), from EDGAR's daily index of filings, with blank-check companies (SPACs) marked;
• “What they filed with the SEC”: what every company with a page filed in the last week, worded as the follow emails word it.

The sector pages sum the same figures by sector, each company counted by its market value at today's price.

Market Radar is a weekly scan of about 6,300 US-listed stocks. It produces two lists of at most ten stocks and one of ETFs, and it gives no rating. A stock enters a list only by meeting every requirement of that list, and only then are the stocks that met them ranked. When fewer than ten meet them, fewer are shown.

• The medium-term list (about six months) is produced only while the S&P 500 itself is in an uptrend. It asks for growing earnings, a good return on capital, a minimum size and a sound balance sheet; a price in an uptrend, near its high and ahead of both the market and its own sector without having run too far; clean accounting checks, a valuation that is not extreme for the company's growth, no heavy selling by insiders and no recent disappointment in results. The stocks that pass are ranked on the price's rise over recent months against the market, and on earnings, quality and value.
• The long-term list (about twelve months) asks for a company under no financial strain, trading below our estimated fair value with a margin of safety, with a moat label (returns on capital and margins above its sector's for years) or a decade of good returns on capital, sound financial health, clean accounts, positive owner earnings and revenue that is not shrinking fast. The stocks that pass are ranked on the margin of safety, quality, standing against peers and the fundamentals score.
• The ETF list ranks equity ETFs on the fundamentals scores of the stocks they hold, weighted as each fund holds them, from their latest N-PORT filing, and only when enough of the fund can be scored.

The moat label describes the past. It is wide when the company's return on capital has been well above its sector's average in nearly every year on record and its gross margin clearly above its sector's, and narrow when its return on capital has been clearly above the sector's for several years, or above it with a clearly higher gross margin. Banks are not labelled. In our test on a sample of companies labelled each quarter since 2011, five years later 85% of those labelled wide still had a return on capital above their sector's median, against 72% of those labelled narrow and 41% of those without the label. Today's return on capital alone sorted that outcome better than the label does. The test covers only companies still listed today, so it overstates how many kept their returns, and it says nothing about the share price.

The thresholds and the weights of the ranking are not published. Each list's exit rules are set in advance, and how the earlier lists did is on the Radar's public record.

The watchlist calendar and the emails about the companies and sectors a reader follows are not ratings. They report events: results, dividend declarations with their ex-dividend and payment days, material 8-Ks, purchases in the market by officers and directors from Forms 4 (not those made under a trading plan), intended sales from Forms 144, and the days of the Fed's and the ECB's rate decisions and of US inflation, from the published schedules of those bodies. The dates and amounts come from the SEC filings and from the companies' announcements on their investor pages, each with a link. Expected results days are our estimate from the pace of earlier filings, are written as an estimate and change when the company announces. What the sources did not have is missing, and the page says so, rather than filled in.

There are two kinds of price alert, and neither is a rating either. On the Investor and Pro plans a reader can set levels of their own on a stock they follow, a price below or above a level or a dividend yield above one, and the daily email says when one has been crossed. On Pro, an analysis with an Outperform-family rating also sets an alert at the price level the analysis stated: the price is checked every 30 minutes, and an email goes out when it reaches that level.

For every published or disseminated analysis, we retain the version of the output as it appeared at dissemination time, together with its identifiers and timestamps. Depending on the output, the record contains: issuer and financial instrument, rating and direction, reference price, estimates and time horizon, methodology and rule version, data and source dates that accompanied the analysis, data-quality notes, automated-control results and the Devil's Advocate output. The stored dissemination version is not rewritten by any later stage.

During production, numeric claims capable of automated verification under the documented controls are compared with the data returned by the tools of the same analysis, and the result is recorded. The check covers figures carrying a unit, such as amounts, percentages, multiples and share counts, in defined fields of the text, with a tolerance of one unit in the last digit. The count and the kind of findings are recorded. The check does not itself withhold or correct a figure. This disclosure does not mean that every number or source can be automatically verified, or that errors are impossible.

For weekly Market Radar lists, each selection is recorded at scan time with the metrics that qualified it, such as score, value estimate and key quality inputs, and the week's list is archived before the next one replaces it. The original entry is retained, while subsequent status updates or corrections are recorded separately with their date and reason. Original rules are not silently or retrospectively changed.

Records are protected by access controls. The recommendation record is retained for five years and the access audit record for twelve months, in accordance with the retention policy and data-protection law. The record makes it possible to reproduce what was published and which data were available at production time. It is not a guarantee that the output was correct.