Methodology — Intrinsic Value (normalized-FCFF DCF)

Transparency is a feature. This page explains what the fair-value engine does and why it can be trusted. A wrong number is a Sev-1 bug; a proxy is always labeled a proxy; null beats a wrong number, and inflation (a false bargain) is the dangerous direction.

This is the public methodology. It describes every model, treatment, and guard by name and direction. The exact engine parameters — caps, floors, gate counts, band bounds, and guard constants — are proprietary calibration and deliberately not enumerated.

The headline fair value for a profitable non-financial is a driver-built, normalized free-cash-flow-to-firm DCF (Damodaran-style) — not a capitalization of a single, noisy trailing cash flow. Financials route to residual income instead (see below). Everything is computed live against the current price — a stored fair value is stale the moment price moves. All inputs come from the company's own SEC/ESEF/EDINET filings; nothing is imputed from analyst data.

What the number is (2026-07 recalibration). The fair value is an estimate, not a stress floor. Earlier revisions stacked conservative choices (double-counted SBC, growth cliffing while excess returns persisted, M&A goodwill charged against organic reinvestment) whose product priced the median company well below the market — a discount-rate statement, not a thousand mispricings. Those biases are fixed at the source, and the residual market-level gap is disclosed (the calibration read below) rather than hidden inside every per-name number.

The driver chain

Starting from the latest revenue and a held operating margin (the margin the firm earns today, held flat — no imputed recovery), NOPAT follows from that revenue, that margin, and the firm's effective tax rate. Stock-based compensation is expensed exactly once: the GAAP operating margin already charges SBC inside operating income (ASC 718 / IFRS 2), and dilution-to-date is carried by valuing per today's diluted share count.

Reinvestment is tied to growth through the firm's own sales-to-capital ratio (a floored capital-efficiency prior), per the standard Damodaran driver identity — free cash flow to the firm is what NOPAT leaves after reinvestment.

Two definitional choices in sales-to-capital (both 2026-07):

• Goodwill is excluded from the capital base — organic growth does not re-purchase the premium paid for past acquisitions (leaving it in charges serial acquirers their historical M&A intensity forever). The ROIC fade anchor deliberately keeps goodwill — the demonstrated return on all capital actually deployed, premiums included, is the conservative anchor.
• R&D is capitalized for ATTRIBUTION, never cash (Damodaran): the research asset (amortized over a sector-keyed research life) corrects the ROIC series and the fundamental-growth estimate — but the margin and sales-to-capital the cash-flow projection runs on stay reported. Adversarial review proved every cash-side variant inflates fair values. Capitalization re-attributes profitability between expense and investment; it does not create cash. A history gate refuses the adjustment where the store's history is too short to amortize honestly — a truncated amortization would fabricate the add-back from missing data.

Stage-1 growth is a robust central tendency of several independent estimates — the standard fundamental-growth relation (what the firm's own reinvestment and returns on capital can fund), a trend fit on revenue, and the realized multi-year growth record — capped so a single noisy input can't dominate. The combining statistic, the windows, and the caps are proprietary calibration.

Cost of capital

WACC is built from a Blume-adjusted β that fades to 1.0 over the horizon (mean reversion), the risk-free rate, and an equity-risk premium; the after-tax cost of debt uses the firm's own interest/debt. Multi-class filers use the diluted weighted-average share count so a single-class cover-page tag can't understate the float.

Assumption presets. The default is a disclosed-conservative set. A market-calibrated preset instead uses the implied ERP — the premium at which the median upside across the valued universe is ≈ 0 — recomputed at each materialization. Under it, the cross-sectional dispersion is the signal and the market-level call is stripped out. The preset in use — and its risk-free/ERP values — is always shown beside the number.

Duration — one unified fade (2026-07)

Growth and excess returns share a single evidence-gated duration: an explicit forecast horizon plus a competitive-advantage period (CAP) that must be earned by demonstrated persistence — the CAP scales with how many trailing years the firm actually earned ROIC above its WACC, requires that record to be sustained rather than cyclical (a cyclical clearing WACC only at the top of its cycle earns nothing), and self-collapses to zero for a firm with no excess returns:

• Growth fades from stage-1 to a terminal rate no higher than the risk-free rate across the FULL span — it does not cliff at the horizon while ROIC is still high.
• ROIC fades from the firm's demonstrated level down to WACC across the CAP, with reinvestment set by the standard fundamental identity linking growth to returns on capital.
• Beyond the span: a no-excess Gordon perpetuity (returns on capital settle at the cost of capital) — no excess returns in perpetuity, ever.

The anti-inflation guards (load-bearing)

• Reverse-DCF ceiling — a disclosed disagreement, not a silencer. We solve for the stage-1 growth the current market price already implies (off the same normalized base, no-excess terminal, and unified duration). When our own growth estimate exceeds that ceiling, the value is served with an OUT-BULLS-MARKET flag and confidence capped at LOW — the claim is the model's, not the market's, and the FV tab says so. It remains a hard reject where the disagreement is not a judgeable claim: when the firm's own size-bucket base rate calls our growth essentially unprecedented and the firm isn't already delivering it (never impute an unprecedented acceleration — only extrapolate a demonstrated one); when the market prices decline that the model cannot express (decline is not in the model's vocabulary, so that "disagreement" would be a false bargain); and in the opt-in moat mode (zero slack by design).
• Plausibility band. A fair value implausibly far from the current price — in either direction — is suppressed rather than served: corrupt share counts, broken inputs, and declining-business floor artifacts all land there.
• Confidence (HIGH / MEDIUM / LOW) reflects model agreement + input completeness; an out-bulls-market headline is never above LOW; NULL where nothing valued.

Expectations lenses (display-only — never inputs to the fair value)

• Base-rate read (Mauboussin): the growth the price requires, converted to an empirical frequency among same-size firms (1950–2015). Caveat surfaced with it: the table predates the current platform regime — the last decade's hyperscalers broke those base rates, so a HEROIC read is evidence of rarity, not impossibility. See the Base-Rate methodology.
• Market-implied CAP: holding the model's own drivers, the number of YEARS of excess-return duration the price requires — "the price implies decades of competitive advantage" is a judgeable claim in a way a bare percentage discount is not. Where even the search bound can't reach the price, it is reported as "the premium is not duration" (the KO/COST/AAPL pattern: their premium is the discount rate/quality, not duration).
• Implied market share (observed-pool TAM proxy): the price-required growth path re-expressed as a share of the company's observed sector revenue pool — the summed revenue of USD-reporting covered names (never a licensed analyst TAM). Coverage is disclosed in every read ("across N covered names"); the lens abstains when the pool is too thin to mean anything or the company already IS most of its observed pool, and an implied share above 100% is reported as the priced-in growth outrunning everything we can observe — the pool is a coverage proxy, not the true market.
• Fair-value distribution (scenario band): a deterministic Monte-Carlo over the real driver uncertainties (growth-candidate spread, margin hold-vs-glide, ERP, CAP length, sales-to-capital) — the price's percentile within the distribution is the honest headline for a range that a point-sensitivity grid never was.
• Aggregate coherence: at every materialization the median upside across the universe is computed and disclosed; when the median is materially non-zero the screener is explicitly told the model is re-pricing the market (treat absolute upsides as calibration, ranks as signal).

Financials — residual income

Banks/insurers use residual income — the spread of the return on equity over the cost of equity, earned on book value — with a normalized ROE that refuses to capitalize a cyclical-peak year, plus below-book suppression — a financial priced below book that RI still calls a large bargain is nulled (interim book RI is unreliable; REITs need FFO/NAV, which we do not fabricate).

Honest limits

• Cyclicals on held margins: hold-current values a refiner/commodity name at whatever point of the margin cycle the latest year caught; the flag + LOW confidence disclose the resulting disagreements, but a through-cycle margin normalization for deep cyclicals is still an open item.
• EBITA margin for acquirers (2026-07, shipped): where the filer tags amortization of intangibles, the margin base adds it back (a non-cash purchase-price recovery with no maintenance-capex twin). Tagged line only — no line means the reported EBIT margin, never an estimate — and the provenance flag rides on the drivers. The ROIC fade anchor stays amortization-burdened (conservative).
• US-listed ADRs over a non-USD reporter suppress all price-based valuation (unknown depositary ratio).