Fit

Fit

Does not fit the framework (P1 not met); contested: P4b

Does not fit the framework. Comcast clears the universe screens and trips no hard exclusion, and the dislocation-and-yield pillars are strong — a real 51% drawdown, an adjusted FCF yield above the 10% bar, a repurchase history that halved-then-cut the share count. But the year-10 durability gate (P1) does not hold at very-high conviction: the profit core, residential broadband, is losing subscribers to fiber and 5G fixed wireless, and that gate alone decides the answer. Confidence is medium — one non-load-bearing criterion (P4b) is contested and no probability spread exceeded 0.25. No exclusion hit; no watchlist_only flag; no prior-driven-risk flag.

The year-10 gate is binary by the framework's own construction: any proper doubt resolves to not met. Here the doubt is not stylistic — Comcast's own 10-K names the attackers and its own subscriber table quantifies the loss.

Universe and exclusions — unsoftened

Comcast passes both universe screens and trips no disqualifying exclusion. Stated plainly, so nothing hides behind the pillar arithmetic.

Universe — clears both. Class A common stock trades on the Nasdaq Global Select Market (CMCSA) — a US primary listing, not an ADR and not a Chinese issuer (U1, met). Market capitalization is roughly $82.7B — 3,709M diluted shares at the 2026-07-24 close of $22.30 [1] — about 8x the framework's $10B floor, even though that cap already reflects the halving from the 2024 peak (U2, met).

Exclusions — no hit. Comcast is a broadband, media and theme-park operator, not an automobile OEM (X1, not met). The promotional-CEO screen fails to fire: a 60-year-old business under founder-family control with a clean promise-versus-delivery record on capital return and Peacock profitability, against one slipping operating guide (X2, not met). The consensus-saturated-darling screen fails in the other direction — the stock is down 51% and trades near 1.4x EV/sales and 4.6x EV/EBITDA, the market pricing decline, not disruption (X4, not met). China dependence is immaterial: a single 30%-owned park (Universal Beijing) with non-recourse debt, inside a theme-parks segment that is 8% of revenue [2] (S1, not met — no sensitivity flag).

One dissent is worth recording without softening the result: on the structural-decline screen (X3), three jury seats read not met while one codex seat read met — residential video is in outright run-off (down 38% in customers over four years), but consolidated revenue has not fallen for three consecutive years, so the disqualifier does not fire. The verdict resolves not met; the dissent is the same broadband-and-video doubt that carries the gate below.

Pattern match

Comcast is closest to the framework's fourth setup — a quality monopoly/duopoly on a fear dip — but the match is partial. The dip is real (51% off the November-2024 peak, on a named, dated fear: broadband subscriber losses to fixed wireless and fiber), which is the pattern's entry signature. What the pattern also requires is that the fear be specific and testable and ultimately wrong — TikTok did not kill Instagram; one China quarter did not break Apple. Here the tested fear is not clearly wrong: the substitution is showing up in Comcast's own subscriber table, not just in a bear narrative. The setup therefore has the pattern's shape but not its resolution, which is exactly what the durability gate turns on.

It fits none of the other three patterns cleanly — it is not a large-bank cyclical at the bottom, not primarily a dividend-yield-carry name (the 5.9% dividend is a support, not the thesis), and not a healthcare/insurance one-year forecasting error that reprices back.

The pillar ledger

Source: deterministic tally (ruchir/fit_tally.json); reference lines from the framework spec. Verdicts are the framework's own, not scores.

Year-10 durability (gate) — P1, not met

This is the decisive point. The free-cash-flow leg of the gate is defensible — reported FCF was positive every year from FY2016 to FY2025, range-bound near $10.6B–$21.9B. The revenue leg is not high-conviction, and the gate requires both at very-high conviction. Domestic broadband customers fell 711,000 net in FY2025 to 31,255k, with penetration down to 47.6% from 49.8% [3], and the losses are accelerating (−66k in FY2023, −411k in FY2024, −711k in FY2025). Comcast's own 10-K names fiber overbuilders and 5G fixed-wireless carriers as direct broadband competitors [4], and T-Mobile has told investors it targets 18–19 million broadband subscribers by 2030. Four jury seats agreed on not met; the year-10 probability came in at 0.445 with a spread of just 0.04.

The strongest surviving counter-fact sits in the same breath: consolidated revenue held essentially flat ($121.4B in FY2022 to $123.7B in FY2025), and domestic broadband revenue actually rose despite the subscriber losses, because ARPU and fast-growing wireless offset volume. That is why the FCF leg holds — but pricing is a finite offset against a shrinking base, and that is the doubt the gate refuses to wave through. Full treatment: Durability.

FCF consistency — P2, met

Reported free cash flow has been positive and predictable for a decade with no negative or cyclical episode. The read is met, though one codex seat returned cannot_determine (cross-family agreement is therefore incomplete). The counter-fact is a data limitation stated plainly: the framework's preferred adjusted-FCF stability series is not_computable because stock-based compensation is missing from the structured feed for all ten years, so the consistency read rests on reported FCF rather than the SBC-and-acquisition-adjusted figure the framework specifies. See Yield and Durability.

Dislocation and yield — P3a, P3b, P3c, P3d, all met

The entry trigger is the framework's strongest region for Comcast. The drawdown is real and dated (P3a): 51.4% from a $45.14 peak on 2024-11-06 to a $21.92 trough on 2026-07-23, in legs anchored to broadband subscriber prints. Volume capitulation is present but modest (P3b): a 2.46x spike against the pre-peak median, clearing the ≥2x reference line but far below the double-digit multiples of true panic. The adjusted FCF yield clears the 10% moderate-balance-sheet bar on every basis (P3c): 24.2% current on FY2025 adjusted FCF of ~$20.0B, 19.4% on the three-year average. And consensus forward FCF clears the bar on every forecast year, so no mean-reversion underwrite is required (P3d, probability 0.88, spread 0.05).

The counter-facts travel with the numbers. The 51.4% depth stops short of the 60–70% forced-selling band the framework prizes, and roughly half the fall was a slow drift, not one capitulation. FY2025's headline yield is flattered by a low cash-tax year — income-tax payments were $755M versus $7,096M the year before [5] — so the run-rate normalizes nearer 15–17%. Still above the bar, but the current print overstates it. Full treatment: Dislocation, Yield, Damage Math.

Balance sheet and self-help — P4a met, P4b contested, P4c met

Comcast can comfortably outlast a multi-year problem without capital allocation being forced toward debt paydown (P4a): total debt of $98.9B [6] is 2.4x net-debt/EBITDA — moderate, inside the framework's sub-3.0x line — with near-$5B annual maturities against ~$21B of FCF and ~$21B of liquidity. The dividend is well covered (P4c): 5.9% yield at a ~22% payout, on a 17-year growth record, though the 2026 per-share cash rate was held flat for the first time in that streak.

The repurchase engine is the one contested criterion (P4b). The executed record is strong — roughly $62.7B of buybacks over 2016–2025 cut the share count 24% (4,875M to 3,709M), net of dilution, the opposite of a rising-share-count hard fail. But the buybacks were paused as of July 1, 2026 and are expected to stay paused through the ~one-year corporate separation [7], with $8.9B of authorization left unused [8]. Two claude seats read the executed habit as met; two codex seats read the live pause as not met — a 2–2 split resolving to contested. The flywheel is idle at exactly the moment a mid-teens yield would make it most valuable. Full treatment: Self-Help.

Diagnosis — P5, not met

The adversarial trial put the probability that the impairment is temporary at 0.38 (panel mean 0.44, spread 0.19, not contested), below the 0.65 line a fit would need and leaning toward a durable reset. A conservative permanent-reset FCFE perpetuity destroys ~$47B of value against ~$94B of price destroyed, so a real price-vs-value gap exists — but it survives at scale only in the temporary branch, which the trial rates the minority case. Note the nuance: at 0.38 with spread 0.19, P5 sits just above the framework's separate strongly-permanent cutoff (≤0.35 with spread ≤0.20), so P5 does not itself force the verdict — the P1 gate does.

In the same treatment, the temporary case has genuine, cited support: broadband losses narrowed 34,000 year-over-year in Q2 2026, the CFO dates the self-imposed repricing to lap in 2026, wireless hit a record quarter, and Peacock turned its first profit; reading the permanent brief first moved judges ~0.10 higher on temporary. Full treatment: Damage Math.

Instrument context — I1, not verifiable

Long-dated listed options on CMCSA extend to the January 2028 expiration (~18 months out) with ~31% 30-day implied volatility as of 2026-07-24 — inside the framework's up-to-50–55 reference band. The tally records I1 as not_verifiable, not "exists," because those facts come only from dated web sources (AlphaQuery, options-chain listings) with no in-corpus, PDF-citable filing to confirm them. Because the overall verdict is does_not_fit, the instrument line drives no watchlist_only overlay in any case. Facts only; see Clock.

What a 3x-in-3-years would require

The framework's target test is rendered from the tally's re-rating arithmetic — and here that arithmetic is unavailable: the tally records re_rating_math as null, because the applicable bar or normalized adjusted FCF needed to price the re-rating was not carried into the deterministic block. No target price is asserted where the inputs were not computed.

What can be stated is the base-rate context from this name's own history, framing only. An ~18-month full re-rating out of a 50%-plus drawdown has no precedent in Comcast's record: the only comparably deep episode (1999–2009, −68%) took roughly 12.7 years to reclaim its high, while the two moderate drawdowns that round-tripped in 10–15 months (2018 −29%, 2020 −32%) were half as deep as today's decline. Measured from the confirmed 2009 low the percentage recovery was fast — a double in ~1.8 years — but confirming the low is the precondition, and that is the durability question the gate leaves open. See Clock.

Contested and undetermined

Contested — P4b (repurchase engine). Both readings, with the split. The met reading: ~$62.7B of executed buybacks over 2016–2025 and a 24% share-count reduction, with the trend not rising — the framework's flywheel demonstrably real. The not-met reading: repurchases paused as of July 1, 2026 through the ~one-year separation, so the engine is switched off now, with $8.9B of authorization unused. Jury split 2–2 (claude a/b met; codex c/d not met); the name-masked seat resolved not met. The tally carries it as contested and, being non-load-bearing, it does not alter the gated verdict.

Undetermined. Nothing was marked cannot-determine. I1 returned not_verifiable (the options/IV facts exist only in web sources); it is recorded above and blocks no pillar verdict.

Provenance

Source: deterministic tally provenance block (ruchir/fit_tally.json) and trial tally (ruchir/trial/tally.json).

The verdict was pressed by two independent model families that agreed on the decisive gate and on the overall answer, disagreeing only on one non-load-bearing criterion (the buyback pause) — which is why confidence is medium rather than high. Re-running the trial with the briefs read in the opposite order moved the temporary probability only 0.10, and masking the company's name changed no gate criterion, so the answer is not an artifact of reading order or of the analyst's prior about the name.

The falsifier ledger

These are the standing what-would-change-this conditions carried from the tally — each a fact that, if it prints, cuts against the durability doubt (or, in the last several, would confirm it):

1. adjusted FCF or EBITDA declines where flat-or-better was underwritten
2. revenue declines for a third consecutive year
3. capital allocation pivots to debt paydown over repurchases
4. share count inflects upward
5. the industry repricing cycle fails to materialize where industry-wide mean reversion was underwritten
6. Aggregate convergence revenue (broadband+wireless) returns to YoY growth as free lines convert to paid — proving the offset actually offsets
7. Domestic broadband net adds turn positive without further ARPU decline across multiple quarters post-lap
8. Forward consensus FCF revised materially back up from ~$13.6B toward the $21.9B trailing level (the current cut is treated as durable, not one-year)
9. Connectivity & Platforms adjusted EBITDA grows YoY in H2 2026 after investments lap
10. Domestic broadband net additions turn positive or near-flat for four quarters after the 2026 investment lap without further ARPU decline.
11. Connectivity & Platforms EBITDA returns to sustained YoY growth and residential margins recover toward the FY2025 run rate.
12. Free wireless cohorts convert to paid relationships at scale so convergence revenue and ARPA grow without continued broadband price dilution.
13. Peacock and the post-separation media portfolio offset linear subscriber-fee decline without relying on temporary event spikes.
14. Broadband net losses re-accelerate past ~200k/quarter after the 2026 lap (Q4 FY2026-Q2 FY2027), showing the repricing bought no durable volume stabilization.
15. Annual Connectivity & Platforms adjusted EBITDA breaks below its stable ~$30B band in FY2026/FY2027 rather than reverting once transition spend laps.
16. FCF (excluding Versant removal and one-time transition investment) falls sustainably below the ~$20B run-rate, confirming the projected drop to ~$13.5B is structural not portfolio/timing.
17. Broadband ARPU keeps declining after the base is fully on new plans by end-2026, with convergence/wireless failing to turn residential revenue positive — a permanent revenue-ceiling reset.

Data gaps

What the run could not answer, from the tally's list:

• Adjusted FCF and adjusted-FCF yield are not_computable in fit_features because stock-based compensation is missing from the structured cash-flow feed for every year FY2016–FY2025; adjusted FCF (~$20.0B FY2025) and float-retirement (~4.1 yrs) were rebuilt from the filed 10-K (SBC $1,288M; 5-year-average acquisitions $562M), and fcf_stability could not be computed at all.
• balance_sheet_class returns unknown (the feed lacks a debt line); leverage (~2.4x net-debt/EBITDA, moderate) was computed from the 10-K debt footnote rather than the feature file.
• Live consensus tone / analyst-rating distribution could not be pulled (provider errors), so the darling screen is settled from corpus pricing and CapIQ-derived features rather than current sell-side sentiment.
• Official reported short interest, holder-base/13F flow, and forced-seller disclosures were unavailable, so the split between anchored/forced sellers and informed sellers is unresolved.
• Implied-volatility facts (level, rank, LEAP expiries, liquidity) are from web research, not PDF-citable filings — the reason I1 is recorded as not_verifiable.
• The precise asset perimeter of the announced NBCUniversal/Sky separation is described only qualitatively on the Q2 FY2026 call; no registration document is in the corpus, so post-separation leverage, dividend policy and buyback-resumption date cannot be quantified.
• Consensus does not isolate how much of the FY2025→FY2026 FCF step-down is one-time cash-tax/Versant removal versus the broadband operating hit, so the clean operating numerator is bounded, not point-estimated.
• The capitulation_gauge anchors "peak" to the 2024-11-06 local high ($45.14), not the 2021 all-time high ($61.75); the deeper −64.5% drawdown from the true peak is computed directly from the daily price record.