Quality tests + price. Every assumption visible.

Earnings resilience

This financial-history screen does not establish an investor’s personal circle of competence. At least 7 annual periods, using up to ten years, few revenue declines or loss years, and stable operating margins. Higher variation can pass when all margins are positive and no year has a net loss, with either a nondecreasing rise or an upward linear trend explaining at least 90% of variation. The trend must rise in both halves of the window, have residual variation no higher than 0.35 of the mean, and show no latest decline. This exception needs complete consecutive annual observations. A single positive-margin dip can also pass when it fully recovers the next year and stays recovered, and holding just that year at its preceding margin brings variation to 0.35 or less. This path requires the same complete, loss-free history. Raw variation still marks a volatile business as cyclical in valuation. Since method 3.8.0 this screen is shown as evidence and is not required for Buy. Banks need positive earnings in nine of ten years; insurers need eight. Their sector tests measure capital returns, book value plus dividends, share dilution and accounting warnings.

Moat

Operating companies need a median return on invested capital of at least 15%. ROIC divides after-tax operating profit by invested capital. Years with nonpositive invested capital use equity plus debt and leases. Positive earnings with no positive capital base are represented as above 100%. Gross-margin resilience compares the median of seven to ten consecutive annual margins with the lower of the latest margin and the last-three-year median; a decline greater than four percentage points fails. A fully recovered historical dip does not impose a permanent veto. Banks use return on tangible common equity. Property and casualty insurers use their underwriting record, or tangible common returns when that record cannot decide the test. Other insurers use return on common equity. Nonpositive common capital fails the financial quality test.

Economics

The current model starts with net income, adds depreciation and amortisation, then subtracts estimated maintenance capital spending. Stock compensation is already expensed in net income and is charged once. Operating cash flow adds it back and does require a deduction. The cash-conversion test divides the sum of five years of owner earnings by the sum of the same years of net income and requires at least 0.80. The annual median and latest annual ratio are context, not that aggregate. Valuation uses its separately disclosed normalised earnings base. When depreciation detail is absent, the conservative cash-flow calculation deducts all capital spending. Some businesses replace what they depreciate inside operating cash flow rather than through capital spending: lease-to-own merchandise, streaming content, rental fleets bought as stock. When capital spending is under a quarter of depreciation and operating cash flow exceeds net income by less than a quarter of depreciation in at least three of five years, owner earnings are operating cash flow less capital spending and stock compensation, never net income plus that depreciation. A complete cash-flow statement with no capital-spending line counts as zero spending only for a business without material fixed assets; otherwise capital spending is unknown and no value is shown.

Management

Operating companies are judged on per-share economic progress, measured by earnings per share between three-year endpoint medians with at least seven consecutive observations and a positive ending level. The historical market-dollar retained-earnings calculation is shown only as context, because market rerating can obscure underlying progress. Share issuance, buyback timing, acquisition spending, ROIC stability and capital preservation remain separate allocation checks. Financial companies continue to use book-value and capital tests appropriate to their balance sheets.

Accounting

Accruals compare accounting profit with operating cash flow relative to assets. Receivables, stock compensation, restructuring and acquisition events provide additional checks. Each test requires its core measures. Supporting checks apply where their inputs can be established.

Price and value

Every operating company is valued at its current scale: the median owner-earnings margin of five consecutive annual observations times latest annual revenue, or complete newer TTM revenue. Loss years remain in the median; a nonpositive median or insufficient history cannot establish positive earning power. Revenue decline still sets growth to zero, and cyclical businesses keep their lower value estimate; this is a valuation input, not a Buy requirement. Owner earnings are projected for ten years plus a terminal value, discounted at a required return 10% a year (or local 10-year bond + 4 points if higher), divided by diluted shares, with only cash above 2% of revenue added. When minority shareholders own a material part of a consolidated subsidiary, only the parent’s share of its cash, depreciation and capital spending counts: by the parent’s share of consolidated profit, or, when no profit split is reported, of book equity. Lenders and settlement businesses valued as operating companies get no cash credit, because their cash is funded by deposits, borrowings or client settlements. A balance sheet a hundred times larger or smaller than the one before it, or with cash or equity above total assets, is treated as a unit error and the previous balance sheet is used. A share split after a Japanese annual report is applied to its share count when the report’s own year-end price (P/E × EPS) is a whole multiple of the split-restated price history. When a company’s vendor figures contradict its own filings in a way no rule corrects yet, its value is withheld and it cannot be a Buy until the figures are corrected. Owner earnings are after interest, so debt is reflected in the risk discount and leverage flag rather than deducted again. Banks and insurers use tangible book value and a ten-year median return on tangible equity capped at 25%. Expected return is the IRR of the same perpetual distributable earnings used to value the financial business, including the fixed payout haircut for the 4× book cap. Retained-earnings growth is capped at 6%. A Buy needs the quality tests and a price at or below the central value; there is no extra discount and no separate return hurdle, because at the central value the expected return equals the required return by construction. Yields refresh daily and moves of 0.1 percentage point trigger revaluation. Implausible yields fall back to the 30-day median, or a documented country default if the series is broken; unavailable local yields leave value unavailable. Companies passing all five tests with median ROIC excluding goodwill at least 20%, median owner-earnings return on total capital including goodwill and acquired intangibles at least 15%, and a full positive decade of owner earnings per share use their own winsorised growth, capped at 12% and fading to 3% over ten years. The two return medians use the last ten fiscal years with at least eight valid observations. Investment holding companies use reported NAV per share with dividends reinvested over ten years; total-return CAGR is capped at 12%. The ten-year NAV realization is discounted at 10%, and expected return is the IRR of that same realization at today’s price. Their central value is that discounted realization value, and a buy requires price at or below it. Other operating businesses retain the conservative 8% growth cap. Terminal growth represents a separate mature-economy assumption and can exceed a depressed near-term growth rate; this assumption is explicit, not a forecast. Where a lower-rate scenario would approach perpetual growth, its rate adjustment is reduced to keep the scenario finite. If the central required return does not exceed perpetual growth, value is unavailable. Low, mid and high values are scenarios, not statistical confidence intervals. Equal scenarios are labelled together.

Method changes

Rule changes are versioned and dated, with a reason recorded before their next publication. We change rules to correct errors or improve the economic reasoning, never to improve past returns. Earlier forward snapshots keep their original decisions and method version.

• 3.8.3 · 12 Oct 2026 — Data correctness. Hyperinflation: amounts in a hyperinflationary currency (IAS 29: Argentina from July 2018, Türkiye from June 2022, others on the IAS 29 list) are not comparable across years, and the vendor even takes one year's income statement and balance sheet from different filings, each restated into a different year-end currency. Such a company is now analysed in US dollars: each statement is placed in the year-end currency of the annual filing whose restated figures it equals and translated at that date's rate (IAS 21.42); a statement no filing places uses the latest 20-F's restated figures where reported and is otherwise unknown. Transportadora de Gas del Sur's earnings per share went from ARS 0.64 to ARS 1,456 in nominal pesos; in dollars it is US$0.38 to US$1.96. A year that mixes currencies now loses only the statement in the wrong currency, so PDD, Autohome, Embraer, Cementos Pacasmayo, PagSeguro, StoneCo and Tencent Music keep their full history. Petrobras, YPF and JBS, whose vendor rows mix the currencies of their 20-F, are built from the 20-F in US dollars, on its share count of every class; Petrobras is valued again (BRL 59.57 a share against BRL 63.06; it fails two quality tests). A one-year vendor currency label that the filing corrects no longer cuts the history before it (Credicorp, Intercorp, Dingdong, Stantec). The vendor's minority-interest line now gives the parent's share of consolidated profit, which allocates subsidiaries' depreciation and capital spending as well as cash (Orion Holdings' central value KRW 67,723 to 51,365, still a Buy). A company that fails a test Buy requires is published as not a Buy even when another test still lacks data. At publication, a newer vendor balance sheet on another currency scale is not used, and statements from a 20-F or measured in US dollars are never re-valued from vendor balance sheets. Nightly price histories that only apply share splits the released history had missed are no longer rejected; genuine conflicts keep the released older months and still refresh the latest ones. Buys are unchanged at 173.
• 3.8.2 · 12 Oct 2026 — Data correctness: every figure is shown in the currency it was reported in. The vendor labelled Petrobras's reais as US dollars (FY2025 sales BRL 497.5 billion shown as USD 497.5 billion, 5.6 times the US$89.2 billion in its 20-F), and mixed dollar and real statements inside single years. Vendor amounts are now checked against the annual filing: amounts equal to the filing in another currency, or all differing from it by one exchange rate, prove the label wrong. A year is relabelled, amounts unchanged, only when the filing points to the company's home currency; a year that mixes currencies or cannot be placed is left out, and the history restarts after it. Without a filing, a balance sheet or income statement that alone jumps by an exchange rate between years, and newer balance sheets or trailing figures on another currency scale, are not used. A valuation or price result from mislabelled years is not carried forward. Petrobras sales now read BRL 497.5 billion and its history starts in FY2024. Sixteen published companies had years in the wrong currency, among them ZKH, YPF, PDD Holdings, Budweiser APAC and Transportadora de Gas del Sur, whose FY2025 cash flows were dollars under a peso label; without that year it fails the moat and economics tests and leaves the Buy list (174 to 173 Buys).
• 3.8.1 · 12 Oct 2026 — Data correctness: value outliers came from bad inputs, not cheap companies. PROG Holdings topped the Buy list at a 154% expected return because the vendor adds depreciation of lease merchandise (US$1.6bn a year in the 10-K) to depreciation and amortisation, while buying that merchandise is an operating cash flow, not capital spending; net income plus that depreciation counted it as free cash. When capital spending is under a quarter of depreciation and operating cash flow exceeds net income by less than a quarter of depreciation in at least three of five years, owner earnings now use operating cash flow less capital spending and stock compensation (lease-to-own, streaming content, rental fleets). Japanese annual reports that tag capital spending with the issuer’s own cash-flow concept, or only in the capital expenditure overview, are now read (14 published companies had zero capital spending, NTT among them). Japanese share counts come from the annual report, so a split after it left the share count on the old basis while the price moved to the new one (Tokio Marine 15:1, Mitsui Kinzoku 10:1, Sumitomo Corporation 4:1 and 12 others): the report’s own year-end price (P/E × EPS) against the price history’s split-restated close now gives the split factor. Further input checks: a missing capital-spending line is unknown, not zero, for a business with material fixed assets, so no value is shown; a balance sheet a hundred times off its predecessor, or with cash or equity above total assets, is a unit error and the previous one is used; consolidated cash owned by minority shareholders is excluded using their share of book equity when no profit split is reported; lenders and settlement businesses valued as operating companies (credit services, mortgage finance, mortgage REITs, financial conglomerates) get no cash credit, because their cash is funded by deposits, borrowings or client settlements. Companies whose inputs contradict their primary filings in ways no rule corrects yet carry a reviewed data-quality flag: their value is withheld and they cannot be a Buy. Each publication logs, for review, every company whose central value exceeds 4× price, whose expected return exceeds 40% a year or whose owner earnings exceed half the price. The 3.8.0 Buy rule is unchanged.
• 3.8.0 · 12 Oct 2026 — Owner decision after the best-1 study: Buy means the company passes the quality tests and trades at or below our central estimate of value. The quality tests Buy requires are the ones best-1 tested: moat, economics, management and accounting, plus a profit record (at most two loss years in ten, a non-negative average operating margin outside banks and insurers, positive three-year average owner earnings per share). Earnings resilience is still computed, shown as evidence and used in valuation as before; it is no longer a Buy gate. There is no margin-of-safety discount and no separate return hurdle: at the central value the expected return equals the required return by construction. A quality company above its central value is shown as above value; the old in-between label is retired. Why: in best-1, buying every eligible company at or below value with equal money and holding it beat the index by +3.1, +6.3, +4.0 and +4.0 points a year in the four test periods, while the ranked three-name selection did not. That was a side check, not a pre-registered test, so the rule is forward-tested from today with a frozen buy list. Valuation math, the bank, insurer and NAV models, and every data guard are unchanged.
• 3.7.1 · 8 Oct 2026 — Buy verdicts follow the data: no approval manifest, and an unapproved Buy change no longer keeps the old published record. Corrects vendor statement inputs against issuer filings (IPS.PA, 000786.SHE, SIQ.AU, 300760.SHE, 600519.SHG, 0700.HK). Reviewed distinct issuers use the SEC ticker-map CIK, and another issuer's filing text or XBRL facts are excluded and flagged instead of failing the company. Calibration, evidence coverage, identity, share and publication-receipt guards are unchanged. Every verdict change is logged with its cause. The 3.7.0 rules are unchanged. These fixes were reviewed as 3.6.1, which never published.
• 3.7.0 · 8 Oct 2026 — Buffett/Munger evolution review from primary sources: rename the numeric understandability screen as earnings resilience because circle of competence is investor-specific; replace the operating-company $1 retained-earnings verdict with per-share economic progress, keeping the old market-dollar calculation only as context. The risk-free discount-rate candidate was tested and not shipped because it breached the owner performance veto; valuation rates and safety discounts remain as model choices.
• 3.6.0 · 7 Oct 2026 — Owner-authorized correctness override: charge stock compensation once, and value every operating business at its five-year median owner-earnings margin times current annual or complete TTM revenue. Keep loss years, revenue-decline override and cyclical discounts; financial and NAV methods remain separate. TTM maintenance uses annual capacity context. Historical returns are reported per correction, not a release veto; unapproved Buy flips retain complete published records. Cash-conversion presentation remains based on five-year totals.
• 3.5.0 · 6 Oct 2026 — Judge the record, not one bad year: one fully recovered positive operating-margin dip can meet the existing variation bar; gross-margin resilience compares recent and typical margins instead of permanently anchoring to FY2023. Raw variability, capital-return thresholds, financial tests, valuation and price gates remain unchanged.
• 3.4.0 · 5 Oct 2026 — Understandable margins may improve steadily: positive, loss-free histories can pass with a monotone rise or a dominant upward trend. Other quality and price thresholds, financial tests, verdict freezes and LTM confirmation are unchanged.
• 3.3.0 · 4 Oct 2026 — Quarterly LTM; two-quarter confirmation.
• 3.2.0 · 1 Oct 2026 — Expected return solves the same valuation cash flows; applied quality rules are explicit; capex flags require lagging earnings and goodwill uses assets.
• 3.1.0 · 1 Oct 2026 — Evidence-backed judgement (e.g. growth capex), business section, materiality-sized red and green flags.
• 3.0.0 · 1 Oct 2026 — Live index universe, valuation v2, financial tests, completeness requirements and thesis check.

The rules were not tuned to backtest returns, but several were shaped by in-sample cases. The honest test is the forward record.

Hindsight simulation, not a track record

The method was designed in 2026, after these outcomes were known. The accounts can contain later restatements. These sources of hindsight can flatter the results.

History covers analysed companies from today’s selected indexes, with coverage varying by quarter. Held companies outside those indexes are not included. Today’s index membership and restated accounts create survivorship and hindsight bias.

The fair-price count includes every buy-qualified company in that quarter and market scope. Headline gains are equal-weight average price changes for those picks and for all analysed index companies with available returns, including failed quality tests. Companies without a measured price change are excluded from averages. The headline follows the market switch, not the other list filters. Returns exclude dividends and are not annualised; overlapping cohorts are not an investable portfolio.

This table compares median price gains for buy-qualified picks with all analysed index companies with available returns in the same year-end quarter and market scope.