M-TAG

1. The Problem: Why Classical Methods Misprice Modern Businesses

The DCF identity is complete — an asset is worth the present value of its future cash flows. The failures are in the inputs, and they are systematic.

1.1 Single-year cash-flow anchoring

A DCF built on the trailing-twelve-month free cash flow inherits every one-off in that year. A staple that paid a multi-billion-dollar tax settlement shows a depressed base and looks undervalued; a semiconductor at a cyclical/AI peak shows an inflated base and looks overvalued. The model treats an accident of timing as the company's earning power.

1.2 Moat is absent as a formal input

Competitive durability is embedded in a skilled analyst's growth and WACC choices but never written down as an auditable number. Two analysts value the same company differently for moat reasons neither can decompose. Yet the empirical gap is large: wide-moat baskets have historically compounded far ahead of the market over full cycles — an edge no classical DCF encodes ex ante.

1.3 Runway is treated as noise

The length of the explicit high-growth phase — how long a company can grow above its sector rate before saturating its market — is usually a hand-set constant. A four-trillion-dollar mega-cap and a six-billion-dollar challenger in the same sector should not share a runway; a constant makes them do so.

1.4 Sector-insensitive application

A packaged-foods company, an oil producer, a bank, and an AI-infrastructure company have categorically different value-creation mechanics. Applying the same terminal and horizon assumptions to all four guarantees error in at least three.

Prior work addresses pieces in isolation — Damodaran's sector WACC tables, Morningstar's qualitative moat ratings, multi-stage DCF, reverse DCF. M-TAG's contribution is the integration: a single reproducible pipeline that connects them, runs on free data, and reports its own uncertainty. That is an engineering contribution, and we state it as such.

2. The Framework

M-TAG runs in five stages: regime routing → moat scoring → cash-flow normalisation → moat-and-runway-shaped DCF → triangulation with confidence.

2.1 Five sector regimes (engine routing)

Routing is the commercially honest core: the tool picks the method the asset actually needs. Cyclicals are valued on mid-cycle (trimmed-mean) cash flow, not a peak; financials on balance-sheet economics, not DCF; disruptors on what growth the price implies, not a fragile forward FCF.

2.2 Moat Score (0–5): uniqueness and durability, not profitability

The Moat Score translates competitive advantage into an auditable 0–5 number that feeds the discount rate (§2.4) and the growth-deceleration curve (§2.5). Its design rests on one principle: a moat is how hard a business is to replicate and how durable its differentiated advantage is — it is not whether the company is profitable today. A young company that reinvests its entire margin to entrench a network or switching-cost advantage runs at a loss; that lowers its profit (already penalised through the cash-flow base, §2.3), not its moat. Scoring the moat off current returns would double-penalise exactly those businesses. The score is built in two evidence layers.

(a) A free-data uniqueness core

From public statements alone the score is built from three components that measure differentiation and durability rather than profitability:

• Pricing power — the durablegross margin (the median of the multi-year history, not a single year), judged against regime-calibrated thresholds (software ≈65%, hardware ≈40%, staples ≈45%). Commodity and cyclical producers are price-takers, so their margin is capped: a high oil-cycle gross margin is an artefact of the cycle, not a moat.
• Demand durability— multi-year revenue persistence (the share of non-declining, compounding years across the history), a proxy for switching costs and habitual or networked demand; large entrenched revenue is durable even at a slow growth rate.
• Efficient scale— entrenchment and cost advantage from size.

(b) A grounded qualitative overlay — the direct measure, when available

Where a company's actual SEC filing has been read, the score is set by that read rather than by the free-data proxy. A language model, run at zero marginal cost, extracts the five classic moat sources— network effects, switching costs, intangibles (brand / patents / IP / regulatory), cost advantage, efficient scale — from the filing's own Business and Competition sections (10-K, 20-F, or S-1), scoring each 0–1 with a one-line evidence citation, a confidence grade, and the source document recorded. The five scores are then combined deterministically — not by the model's gut — so the number is reproducible and re-tunable.

This aggregation is max-dominated: a moat is only as wide as its strongest durable advantage, with a premium for being defended on additional strong (>0.5) fronts. It is deliberately not an average. A business with an impregnable switching-cost or patent moat is not “less moaty” for lacking network effects, and averaging five checkboxes wrongly compresses fortresses toward the middle — a filtration supplier with ~1,200 filtration-media patents and 86% recurring, contracted aftermarket revenue must not score like a mid-pack name. Weak or absent sources never drag the score down; a company genuinely undifferentiated on every front stays low regardless of how young or fast-growing it is.

Returns as positive-only confirmation

Sustained real returns on invested capital above the cost of capital (ROIC − WACC, averaged over the available history) confirm a moat that has already shown up in the numbers. This signal can only addconfidence, never subtract: in a reinvestment (loss) phase it contributes zero and never penalises. A grounded filing read therefore lets a young company with real, documented network or switching-cost advantages earn a wide score before its returns manifest, while a high-margin business with a candidly weak moat — e.g. a food-delivery platform whose own filing states that employee skills matter more than IP and that switching costs are low — is scored down.

Moat Score → Qualitative Label

2.3 Cash-flow normalisation (the base fix)

Before discounting, the base free cash flow is normalised against its own history. When the latest FCF deviates more than ~30% from the median of the positive multi-year history, the base is blended 50/50 toward that median. This damps one-off trough years (e.g. a large one-time tax settlement) and cyclical/AI peaks symmetrically. When the base is an extreme trough (< 40% of the historical median)— the signature of a structurally low-FCF, capex-heavy business whose value is not in current cash flow — the DCF is declined and the stack defers to the peer multiple. Stable-FCF businesses sit inside the band and are untouched.

2.4 Moat-adjusted WACC

The discount rate starts from a Damodaran sector baseline and is adjusted linearly by moat around a neutral score of 3.0:

So a wide-moat name (5.0) discounts 1.0% lower and a no-moat name (0) discounts 1.5% higher — competitive protection lowers business risk, its absence raises it.

2.5 Growth curve: mean reversion + moat deceleration, gated by TRS

Growth is not held flat and then cliff-dropped. Each explicit year, growth reverts convexly toward the sector-median long-run rate (an economic law: nothing compounds at 30–40% forever), then decelerates by a moat-determined amount — wide moats fade slowly, no-moat names fast. The number of explicit high-growth years is the TAM Runway Score (TRS).

TRS — computed, not constant

v1.0 used one hardcoded runway per regime. The current framework computes a company-specific runway from four free drivers, each normalised to [0,1] and blended, then mapped into a regime band:

Honest note: a licensed per-company Total Addressable Market figure is paid data we do not purchase. The TRS is therefore a penetration-and-durability runway estimate on free data — not a TAM lookup. We state this plainly; the transparency is the point.

2.6 Dual terminal value and the sensitivity band

The terminal value is Gordon-primary with an EV/EBITDA exit-multiple cross-check, clamped to ±25% of the Gordon value. Gordon anchors the perpetuity; the market exit multiple nudges it toward real comparables without letting a rich sector multiple double-count a decade of compounded EBITDA. Raw Gordon-vs-market divergence is reported, not hidden. Every output is a moat-calibrated four-corner sensitivity band, not a point: WACC, growth, and the moat→deceleration coefficient are varied at a moat-appropriate magnitude (tighter for wide-moat predictability, wider for no-moat uncertainty). The published fair value is a range.

2.7 Reverse DCF (Exponential regime)

For disruptors, forward FCF is minimally predictive, so the question inverts: solve for the Market-Implied Growth Rate— the FCF CAGR that makes the M-TAG model output today's price — and compare it to analyst consensus across bear/base/bull scenarios. The investor then asks not “what is it worth?” but “does today's price already require heroic growth?”

2.8 The valuation backbone: own-blended market multiple + divergence-gated triangulation

M-TAG's fair value is not the raw DCF. Earlier versions triangulated three co-equal anchors (DCF, reverse-DCF, a sector-median peer multiple), which failed in a specific way: the sector median is the wrong multiple for almost every company. A durable quality leader clears the market aboveit (Coca-Cola near ~23× EV/EBITDA vs a ~14× staples median); a de-rated business trades belowit (Adobe near ~10× vs a ~22× software median). Anchoring on the sector median collapsed quality names toward half of price and projected de-rated names to ~2× — a false deep-value signal on exactly the businesses the market is warning about.

v2.2 fixes this at the root. The reliable backbone is the company's own current EV/EBITDA multiple, blended 65/35 toward the sector median (mean-reversion sanity) and bounded to [0.4×, 3×] the median. The own multiple is the market's aggregate, forward-looking read of thisbusiness — encoding both any de-rating and any durable-quality premium a backward-looking model cannot see. The DCF and reverse DCF stay on as independent cross-checks, not co-equal votes: each is compared to the backbone and dropped when it diverges >35%(a trough/peak-FCF DCF that would halve a quality name, or an over-projection that would double a de-rated one). Two DCF-family anchors agreeing is not independent confirmation — they share assumptions — so they never outvote the market backbone. Corroborating cross-checks are blended in and confidence rises; divergent ones are surfaced as warnings and confidence falls. Fair value is therefore a believable estimate of intrinsic worth; the margin of safety is applied afterwards in the entry ladder (§3), never by depressing the fair value itself. Banks (no meaningful EV/EBITDA) use the same own-multiple fix on P/B × ROE; dual-class share counts are reconciled against market capitalisation.

3. Entry Ladder and Educational Framing

From the triangulated fair value the platform derives a four-level reference ladder:

Signals are phrased as framework references, not directives(“below deep-value reference”, “well above fair-value reference”). The framework is educational; it does not prescribe personal transactions. This posture is enforced in code, not just in a disclaimer. For Exponential-regime assets the three scenario-derived prices (bear, base, bull) replace the fixed haircuts, because for a high-growth disruptor a fixed 30% discount has no analytical basis while the scenario levels are grounded in actual growth expectations.

4. Implementation: Free Data, Full Provenance

M-TAG runs on SEC EDGAR as-filed 10-K line items (authoritative) merged with Yahoo Finance market and analyst data, with a per-field provenance badge(source + filing date + confidence) for every input, and a public verification surface that shows the inputs and the worked steps. Anyone can reproduce a valuation from public data — the design goal is verify, don't trust. The engine is deployed at investment.capitalfortress.org, and the grounded moat overlay records, per company, the exact filing read, its date, the per-source scores, and a confidence grade.

5. M-TAG vs. Existing Frameworks

The contribution is the bridge: a reproducible mapping from moat to discount rate and deceleration, from scale/durability to runway, from three anchors to a confidence-scored band — running end-to-end on free data. The foundations are well-established (multi-stage DCF, Damodaran's sector WACC data, Buffett's moat concept systematized by Morningstar, Ohlson's economic-profit horizon); M-TAG's work is the integration.

6. Limitations (Stated Honestly)

1. The moat score is a model of durability, not a physical measurement. The free-data core proxies uniqueness from margin/revenue/scale patterns; where a filing is read, a language model extracts the five moat sources from its text. Both are judgment layers. We mitigate with a deterministic, re-tunable max-dominated aggregation, regime calibration, a positive-only returns confirmer, and full per-entry provenance — but the score is a defensible estimate of competitive durability, not a constant.
2. Grounded-overlay coverage is partial and dated. The direct filing read exists only for researched names and reflects the filing as of its recorded date; un-researched names fall back to the free-data proxy (capped until returns confirm). Coverage expands over time, and each score records which layer produced it.
3. TRS is a free-data proxy, not a licensed TAM.It estimates runway from scale, durability, moat and margin — not from a purchased market-size figure.
4. DCF-resistant names are declined, not forced. Businesses whose value is not in current FCF (capex-heavy platforms, negative-FCF years) fall back to peer/relative valuation by design; the tool says so rather than inventing a number.
5. Analyst-growth dependence.In periods of rapid earnings revision, forward inputs — and therefore the reverse-DCF and scenario prices — should be read as ranges.
6. The backbone anchors on the market's own multiple, so it partly reflects price.This is deliberate — the market's multiple is the most information-rich, forward-looking read available for free — but it means the framework rarely disagrees violently with an efficiently-priced large-cap. Its edge is the entry ladder, the sector-reversion and cross-check signals that flag genuine mispricings, and full transparency — not a claim to know intrinsic value better than the market.
7. Value vs. value-trap is not decidable from historical fundamentals alone. Whether a cheap business is a bargain or a permanent impairment depends on the future, which no backward-looking model contains. M-TAG neither hides this nor fakes a solution with a warning flag or a trailing analyst cap; instead the company's own de-rated multiple pulls fair value toward reality (so the tool does not scream a false bargain), and every input is shown so the user supplies the forward judgment the data cannot. This is the honest boundary of any fundamental screen.
8. US-listed calibration. Thresholds are set for US GAAP filers; other markets need re-calibration.
9. Out of scope: pre-revenue companies (real-options territory), pure commodities and FX (valued by supply/demand and rate differentials, not DCF).
10. Not buy-side research and not advice. M-TAG is a transparent practitioner/education framework, not a substitute for proprietary diligence, and not a personal recommendation.

7. Reproducibility Appendix — a Worked Band

A live example from the 2026-07 verification run (public data):

• The Coca-Cola Company (KO).KO clears the market near ~23× EV/EBITDA — a durable-quality premium a ~14× staples-median peer would miss, and its trough-FCF DCF (~$26) is dropped as a >35% divergent cross-check. The own-blended backbone lands the fair reference near $64against a ~$83 price (“above fair-value reference”). The old sector-median engine returned ~$36 here — the exact quality-name collapse v2.2 repairs.
• Adobe (ADBE).The market has de-rated Adobe to ~10× EV/EBITDA on disruption fears not yet in the fundamentals, so a growth DCF over-projects it (~$798). The own multiple encodes that de-rating: the backbone lands near $297against a ~$218 price, and the over-projected DCF is dropped as divergent — no false “strong buy”. This is the value-trap boundary handled by the market's own read, not by a warning flag or a trailing analyst cap.
• Amazon (AMZN).FCF ~1% of revenue with a negative-FCF year in history → the DCF is declined per §2.3 and the own-blended multiple leads at ~$203, a defensible reference rather than a broken DCF point.
• NVIDIA (NVDA).Base FCF ~1.6× its median (an AI-cycle peak) is damped by §2.3; the own multiple, mean-reverted toward the sector and cross-checked by the DCF, lands near $163against a ~$196 price — conservative, neither the “worth more than GDP” runaway nor an over-punished peak.
• Moat, grounded (the §2.2 overlay). Two poles from the filing read. Atmus Filtration (ATMU)— the 10-K documents ~1,200 filtration-media patents and 86% recurring, contracted aftermarket revenue; the strongest sources (intangibles, switching costs) dominate the aggregate for a Wide 4.4, where a five-checkbox average had compressed it to a mid-pack 3.3. DoorDash (DASH)— high unit margins, but the filing itself states employee skills matter more than IP and switching costs are low, so it is scored down to Weak 2.0rather than rewarded for margin. The moat measures durability, and it moves both up and down on the filing's own evidence.

8. Positioning

M-TAG is offered as the most transparent, defensible, reproducible equity-valuation framework we can build on free data— not as a market-beating secret. Its edge is method-routing, triangulation, honest uncertainty, and full provenance. For a rigorous reader that honesty is the stronger claim, and the safer one.