Fit

Fit — Plains GP Holdings (PAGP)

Outside the framework's universe (U2 not met); contested: P2

Outside the framework's universe. PAGP fails the market-cap universe test, so the framework does not reach the pillars: the listed Class A equity is worth $5.21B on the 197,904,124 shares outstanding, or $6.13B on the fully-exchanged basis, against a $10B floor [1]. The confidence tier is high: two model families agreed, the trial was order-stable, and load-bearing spreads were at most 0.15. No exclusion fired and the name-mask probe raised no prior-driven risk. One pillar is contested — P2, FCF consistency, which cannot be tested on the framework's own adjusted-FCF basis.

The verdict here is the deterministic tally's; this tab renders it and cannot soften or extend it.

Universe and exclusions — unsoftened

Here is the decisive point. What trades under the ticker PAGP is not the pipeline. It is a holding vehicle whose listed Class A equity is worth far less than $10B, and that alone puts the company outside the universe the framework will consider.

The arithmetic: 197,904,124 Class A shares were outstanding at 20 February 2026 [2]; at the $26.33 close on 31 July 2026 that is $5.21B, 52% of the line. On the wider, fully-exchanged basis — AAP's approximately 233.0 million PAA common units, the count once the Legacy Owners' stapled interests convert one-for-one — the value is $6.13B, 61% of the line, still $3.87B short [3]. The feature file could not compute market cap ("no positive annual period-end or outstanding share count"), so this is built from the filed share count and the dated close, not from the feature.

The counter-fact sits in the same breath, and it is a large one: the underlying operating partnership clears the line comfortably. PAA's common units were worth $17.33B on 31 July 2026, and PAGP consolidates $31.3B of assets and $10.93B of net debt against $2,809M of LTM Adjusted EBITDA attributable to PAA [4]. The enterprise is large and essential. The listed instrument a Class A buyer actually owns is not, and the framework's universe line is drawn on the listed instrument.

U1 — geography — is met. PAGP is a Delaware limited partnership, primary-listed on the Nasdaq Global Select Market as symbol PAGP; it is not an ADR and has no Chinese connection [5]. One qualifier belongs on the record: what lists is a partner interest in a wrapper that owns no operating assets and draws all its cash from an indirect limited partner interest in PAA [6], and because PAGP elected corporate tax treatment it issues a Form 1099 rather than a Schedule K-1 [7].

No hard exclusion fired. Each was checked against the record:

• Car company (X1): not an auto manufacturer — crude midstream and merchant logistics across 20,405 system miles, zero automotive revenue.
• Promotional CEO (X2): not triggered — across a five-promise sample from 2022 to 2025 management kept its leverage, cost-programme and distribution-cadence commitments while missing two (its Permian volume forecast and its own narrowed FY2025 EBITDA range), and officers and directors own roughly 57.8 million units worth about $1.3B with hedging and pledging prohibited [8].
• Structural decline (X3): not present — the three-year high-single-digit revenue-decline disqualifier is false (longest run is two years), and the fee/services line rose 49% since FY2016.
• Consensus darling (X4): not triggered on valuation — PAA trades at 0.39x sales, the cheapest in its own disclosed peer group, on a split analyst book with a mean target below the market. The counter-fact: the total-return chart has run bottom-left-to-top-right for five years — the Class A index reached 327.17 against 195.98 for the S and P 500 over 2021–2025 — so the entry point is not fear either [9].
• China (S1): zero — FY2025 revenue and long-lived assets are entirely US and Canada; the single "China" reference in the 10-K is a risk factor naming it as a global consumption market [10].

Pattern match

Of the framework's four setups, PAGP most resembles the second — high dividend yield plus high FCF yield, business not going away. It carries a 6.34% distribution yield on a business that is not disappearing. But the pattern's own checks do not clear: the yield case in that setup turns on a high FCF yield and on the distribution never having been cut, and here the adjusted look-through FCF yield is 4.67% (below every bar), while PAA cut its distribution 48% in the 2020 downturn, from $1.38 to $0.72 per unit [11]. It is not a cyclical-bank bottom, not a healthcare forecasting error, and not a tech monopoly on a fear dip. As an actionable setup it fits none of the four patterns — and in any case the universe gate has already closed the question.

The pillar ledger

The gate rule applied is U2 not_met → out_of_universe; the pillars below were still adjudicated by the jury and are recorded for completeness. Reference lines, not grades.

Source: deterministic fit tally (ruchir/fit_tally.json); per-criterion evidence in the pillar treatments below.

Year-10 gate (P1) — not met

The gate is binary by construction, and any proper doubt fails it. It fails on both legs. Roughly 96% of FY2025 revenue ($42,408M of $44,262M) is crude bought and resold at the prevailing WTI price, so "year-10 revenue higher" is a commodity-price forecast, not an earning-power judgement [12]. On the cash leg, Crude Oil Segment Adjusted EBITDA per tariff barrel has fallen from $1.140 (FY2019) to $0.663 (FY2025) — a 42% compression — even as tariff volumes rose 46%; the company's own Item 1 has disclosed for five straight years that Permian overbuild "puts downward pressure on tariffs and margins," and Q1 FY2026 confirmed certain long-haul rates reset to market in 2025 [13]. This is a contested oligopoly of six named competitors, not a monopoly or duopoly — the P1 probability sits at 0.5 with a 0.13 spread.

The strongest surviving counter-fact, in the same treatment: segment Adjusted EBITDA has still risen every year, from $1,909M (FY2021) to $2,344M (FY2025), guided to $2,640M–$2,700M for FY2026, and the contracted book is 76% larger than at end-2021 ($3,430M versus $1,949M of remaining performance obligations) [14] [15]. Volume growth and acquisitions have outrun the per-barrel compression so far. "So far," at the very-high-conviction bar the gate demands, is not enough. Full treatment: Durability.

FCF consistency (P2) — contested

This is the contested pillar. The two Claude jurors read it met; the two Codex jurors returned cannot determine — a split across families that the tally records as contested. The reason is a data limitation, not a disagreement about the business: the framework's rolling five-year average adjusted FCF cannot be computed here, because fit_features.fcf_stability returns an empty series and fit_features.adjusted_fcf carries stock-based compensation as null for every year FY2016–FY2025 and treats acquisitions as an implicit zero [16].

On reported FCF the rolling five-year average rose in every window, from $783M (FY2016–2020) to $2,048M (FY2021–2025), with one negative year, FY2016, driven by a growth-capex build rather than an operating loss. But the framework's adjusted basis would deduct the acquisition spend the feature omits — $2,651M in FY2025 alone and over $5.7B cumulatively since 2016 — which takes a large bite out of that reported series [17]. The pillar is contested because the test as constructed cannot be run, not because the answer is known and disputed. Full treatment: Durability.

Dislocation and yield (P3) — the entry trigger does not stand today

A real dislocation happened: the shares fell 23.9% from $22.13 (30 Jan 2025) to $16.85 (10 Oct 2025), 83% of it in four sessions on the April 2025 tariff-and-OPEC+ crude shock — P3a is met. But three of the trigger's four conditions do not clear:

• P3b — no capitulation. Volume peaked at 1.59x the pre-peak median, and it peaked in the April event window nearly six months before the low; the 10 October trough itself printed on 0.86x median volume. Peak fear did not mark the bottom.
• P3c — adjusted yield far below the bar. FY2025 adjusted FCF of $1,537M falls to $865M after the preferred and joint-venture-minority claims that rank ahead of the common — $1.229 per look-through unit, a 4.67% yield at $26.33. The balance sheet is levered (net debt $10.93B / $2,833M Adjusted EBITDA attributable to PAA = 3.86x) [18], so the applicable reference line is 25%. The yield sits 2,033 basis points below it, and below the 10% default and 8–9% fortress lines as well.
• P3d — consensus does not clear the bar and there is nothing to mean-revert. Capital IQ consensus free cash flow of $1,580M (FY2026) rising to $2,043M (FY2029) is 4.97%–7.46% on a look-through basis. Consensus is rising, guidance was raised to $2,880M in May 2026, and the price already sits above its pre-drawdown peak [19] — the usual "buy the cut, underwrite the reversion" argument is unavailable because there is no cut to reverse.

The counter-fact worth carrying: the FY2025 5-year acquisition average is inflated by a single year; on the pre-2025 run rate the FY2025 yield is 7.30% rather than 4.67% — a 263 bps improvement that still leaves 1,770 bps to the levered bar. Full treatment: Yield and Dislocation.

Balance sheet and self-help (P4) — the flywheel does not run

The point of the framework's dislocation is a buyback flywheel, and it is absent here. Cash spent on common repurchases was $8M across FY2023–FY2025 combined, against $2,709M of common distributions over the same years; $190M of a November-2020 authorisation sits unused; and the unit count has risen 1.07% over three years on roughly $50M a year of equity-indexed compensation with no repurchase offset — the framework's rising-share-count condition (P4b, not met). Capital allocation is explicitly sequenced with debt paydown first and repurchases fourth of four, gated on leverage reaching the bottom of the 3.25x–3.75x range (P4a, not met) [20]. At today's price, retiring the entire common float out of adjusted FCF takes 12.1 years, or 21.3 years of the FCF that actually reaches common holders — against the framework's absurdity marker of roughly three.

The counter-facts, fairly stated: the acquisitions were funded with debt and divestiture proceeds, not equity — the 10-K states the company does not plan to issue common equity to fund spending [21] — the unit count is still ~21M below its 2020 level, and prices paid on the units it did buy ($8.03 and $9.86) were well below today's $26.33. And the balance sheet can plainly outlast the problem: investment-grade, $2.0B of liquidity, no senior-note maturity in 2027 or 2028, and $3.3B of NGL-sale cash received in May 2026. The debt paydown is transitional. But willingness to repurchase at the moment of maximum yield — the mechanism the framework needs — is not on the stated plan. Dividend safety (P4c) is 160% covered on the company's own distributable-cash-flow measure yet only 0.82x on adjusted FCF available to common, and the payout was cut roughly in half in 2020 [22]. Full treatment: Self-Help.

Diagnosis (P5) — met; the damage was temporary

The adversarial trial — two opposing cited briefs, three blind judges — ruled the impairment temporary with probability 0.71 (per-judge 0.71 / 0.68 / 0.78; order-stable, gap 0.02). The one durable drag — Permian long-haul recontracting — is a bounded ~$235M/year step that management's own 2025→2026 bridge more than offsets with $250M of Cactus III EBITDA, $50M of efficiency and $50M of optimization, so net run-rate EBITDA rose from $2,833M to $2,880M [23]. P5 being met does not rescue the case: it establishes that the 2025 fall was a re-rating rather than a permanent impairment, but the re-rating has already been paid back — at the trough the price fell about $1.04B against a plausible permanent NPV hit near $0.66B, and at $26.33 that gap is closed. The strongest counter-fact the trial preserved: the self-correction relied on $2.9B of acquisition capital, and the 10-K's "overbuild" language is genuine evidence that a further recontracting drag could recur. Full treatment: Damage Math.

Instrument context (I1) — not verifiable from the corpus

Facts only, and unverifiable here: listed PAGP options run out to 21 January 2028 (17.6 months from 3 August 2026) with 17,294 contracts of open interest, and 30-day implied volatility was 19.68% on 31 July 2026 — well below the framework's ~50–55 reference line, but the longest expiry falls 0.4 months short of the 18-plus months the framework prefers, and daily turnover in that January 2028 line is thin. The evidence is web-only (a published option chain and volatility statistics); the corpus holds no PDF page for either, so I1 is recorded as not verifiable. Full treatment: Clock.

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

The tally records the re-rating arithmetic as unavailable: "Re-rating math unavailable because the applicable bar or normalized adjusted FCF is missing." The feature file could not compute market cap, adjusted FCF, the yield baseline or the consensus forward yields, so the framework's standard bar-yield price cannot be struck from it.

The target test can still be stated as arithmetic from the surviving yield claims. To reach the 25% levered yield the balance sheet selects would take roughly $4.65B of adjusted FCF to the common — about five times FY2025 and more than PAGP's entire consolidated Adjusted EBITDA. To reach even the 10% default line takes about $1.86B to the common, roughly 58% above the FY2029 consensus mean, or a price near $12.30 — a 53% fall from here. The growth mechanism that would lift cash flow is already inside the rising consensus path, so the yield does not re-rate upward from a depressed base; there is no depressed base [24].

Base-rate context from this name's own history reinforces the point rather than softening it. Across twelve drawdowns of 20% or deeper since the 2013 listing, every round trip completed inside 24 months came from an episode shallower than 40%; the two genuinely deep recoveries took 26.5 and 82.6 months, and three deep episodes never regained their peak. The 2025 drawdown was a shallow 23.9% and behaved exactly to that base rate, recovering in 386 days. The deep capitulation the framework hunts has, in this name, always taken years — and it is not on offer today. See the episode set in Clock.

Contested and undetermined

Contested: one pillar, P2 (FCF consistency), split across model families — the two Claude seats read it met, the two Codex seats returned cannot-determine — because the framework's rolling five-year adjusted-FCF series is not computable from the feature file (stock-based compensation is null for every year FY2016–FY2025 and acquisitions are treated as an implicit zero). The reported-FCF series is stable and rising; the adjusted series the framework requires simply cannot be built here.

Undetermined (cannot-determine on the framework's basis): none of the gate criteria. The only not-verifiable item is I1, instrument context, where the option-chain and implied-volatility figures are web-only with no corpus page to cite.

Provenance

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

Two model families sat the jury and agreed on every gate. The verdict was pressed hard: fifty claims were triaged, sixteen taken through the full skeptic protocol where the arithmetic is recomputed from cited pages, and only one weakened (a secondary sensitivity note on the consensus yield claim) while none were refuted; the single unverifiable item is the web-only option data. Because the out-of-universe result rests on a filed share count and a dated price rather than a modelled judgement, the confidence tier is high and the spread that matters most (P1 at 0.13, P5 at 0.10) never widened enough to put the gate in doubt.

The falsifier ledger — standing what-would-change-this conditions

The framework's template falsifiers, applied to this name:

• adjusted FCF or EBITDA declines where flat-or-better was underwritten
• revenue declines for a third consecutive year
• capital allocation pivots to debt paydown over repurchases
• share count inflects upward
• the industry repricing cycle fails to materialize where industry-wide mean reversion was underwritten

The name-specific falsifiers, with thresholds, directions and windows (verbatim, as recorded in the tally):

Data gaps

The run could not answer several things the framework normally computes:

• fit_features could not compute market_cap, adjusted_fcf, adjusted_fcf_yield, yield_baseline, balance_sheet_class, fcf_stability, share_count_trend, float_retirement_years or the consensus forward yields — every one is flagged not_computable, with market cap failing on "no positive annual period-end or outstanding share count." The U2 arithmetic here is built from the filed Class A share count and the dated close, not the feature.
• Stock-based compensation is null for every year FY2016–FY2025, and acquisitions are treated as an implicit zero, so no adjusted-FCF series or rolling five-year stability could be built from the feature file; the figures in this report are rebuilt from the filed cash-flow statements.
• The structured balance-sheet feed reports only $1,099M of long-term debt for FY2025 (the term-loan line), irreconcilable with the ~$10.7B debt stack in the footnote; the filed figure is used for the balance-sheet class.
• No PAGP annual report earlier than FY2021 is in the corpus, and the pre-2016 daily-price series is internally inconsistent (2013–2014 closes of $58–$85), so long-run absolute-dollar chart work is unreliable; base rates here are expressed as percentage depths and calendar durations, which are scale-invariant.
• Reported short interest is unavailable for PAGP (FINRA returned no rows), so any short-driven component of the 2025 selling is invisible; and the option-chain and implied-volatility figures are web-only, with no corpus page.
• PAGP-level per-share cash flow after the 12 May 2026 Keyera NGL-sale close is not yet in the corpus; the FY2025 10-K predates completion and the Q1 FY2026 10-Q still reports the Canadian NGL business as discontinued operations held for sale.