Fit

The Answer

Does not fit the framework (P1 not met)

Alcoa clears the universe screen and trips no exclusion, but the framework's lone gate — year-10 durability — is not met, and nothing offsets a gate. Confidence is low: the name-mask probe left a load-bearing probability gap of 0.21, above the 0.20 line, so a prior-driven-risk flag stands. There is no watchlist-only flag, no exclusion hit, and nothing the jury recorded as contested.

Here is the decisive point. Alcoa is a price-taker at the bottom of a fragmented, globally traded commodity chain — a 3.1% share of world primary aluminum and 6.4% of world alumina, at prices set on the LME and the Alumina Price Index [1]. Its larger segment by capacity sold below cash cost in the first half of 2026, at a realized $329 per tonne against a $352 operating cost [2]. The framework's gate asks for very high conviction that both year-10 revenue and year-10 adjusted free cash flow will be higher than today's; on a business with no market-structure moat and a segment currently unprofitable, that conviction is absent.

Sources: market cap and yield derived from the FY2025 Statement of Consolidated Cash Flows [3] and the 31 July 2026 close; drawdown from data/prices/daily.json; temporary probability from the blind adversarial trial (ruchir/trial/tally.json).

This tab renders the deterministic tally (ruchir/fit_tally.json). It reports the machinery's verdict and the arithmetic behind it; it does not re-adjudicate the case. Each evidence tab linked below carries the full treatment.

Universe and Exclusions

Universe — both tests clear. Alcoa Corporation is a Delaware corporation whose primary listing is the New York Stock Exchange under "AA", trading in U.S. dollars — not a Chinese company, a Chinese ADR, or an ADR of any kind [4]. Market capitalization is $11.81 billion — 261 million shares at the $45.26 close of 31 July 2026 — 18.1% above the $10 billion line, with a break-even at $38.31 [5]. Both universe criteria are met; see Business.

Exclusions — no hit, one signal worth stating anyway. Every hard exclusion was checked and none fired:

• Auto OEM (X1) — not a hit. Alcoa manufactures no vehicles; it is an upstream producer of bauxite, alumina and primary aluminum ingot, two steps removed from any automaker [6].
• Promotion pattern (X2) — not a hit. The exclusion needs both a repeated promise-versus-delivery gap and weak insider ownership. Only the ownership prong is present: insiders hold about 0.25% of shares, but the stock-ownership guidelines (6x salary for the CEO) are real and met, and the delivery record is mixed rather than promotional — of six commitments, two were kept or beaten, one half-kept, one late, two missed, with both misses disclosed by management in the same numbers used to set the targets [7]. One prong is not two; see Self-Help.
• Structural decline (X3) — not a hit, but the counter-fact is the reason P1 fails. The framework's exclusion targets a secular, non-mean-reverting decline diagnosed as such; the jury did not classify Alcoa there, and the numeric disqualifier for the gate — high-single-digit revenue decline for three straight years — does not fire (fit_features.revenue_trajectory records zero consecutive decline years). The counter-fact stands in the same breath: the physical base has shrunk even as priced revenue rose. Alumina production fell 27.3% and consolidated smelting capacity 22.2% over the decade [8]. That erosion did not trip the exclusion, but it is exactly what denies the year-10 gate its conviction; see Durability.
• Consensus-saturated story (X4) — not a hit. Alcoa trades at 0.92x FY2025 sales with a sawtooth price history since the 2016 separation — the opposite of the extreme-multiple, consensus-owned growth narrative the exclusion targets.
• China dependence (S1) — no sensitivity flag. Disclosed China revenue and China long-lived assets are effectively zero, so the framework raises no China sensitivity flag. The honest counter-fact: the exposure runs through price, not the balance sheet — Chinese and Indonesian capacity sets the alumina price that has driven Alcoa's larger segment to negative EBITDA. This is a price channel, not a China revenue dependence, and the fact test scores it clean; see Business.

Pattern Match

Ruchir's system recognizes four setups. Alcoa's live 46% drawdown, dated triggers and cheap headline yield make it look like a dislocation entry, but on the pattern's own checks it fits none of the four cleanly.

It is not a cyclical-bank bottom (setup 1) and not a high-dividend-yield case (setup 2 — the dividend yields 0.88%). The closest analogue is setup 3, the healthcare/insurance forecasting error — an industry-wide misforecast that reprices 1:1 and then mean-reverts as the book readjusts. The mechanism differs in the way that matters: an insurer's premiums reprice on a regulated schedule, whereas Alcoa's recovery depends on an undated LME aluminum price, not a repricing calendar with regulatory friction. It is also not a quality tech monopoly on a fear dip (setup 4): the monopoly/duopoly structure that setup relies on is absent — Alcoa is a single-digit-share price-taker [9]. The setup the price advertises is a commodity-price bet, not the industry-wide repricing the framework's dislocation pattern hunts.

The Pillar Ledger

The criteria and where each sits against the framework's reference lines:

Source: ruchir/fit_tally.json; deciding numbers from the surviving claims cited in the sections below.

Year-10 gate (P1) — not met

This is the criterion that decides the verdict. The jury voted not met on all four seats. Conviction in the framework comes from market structure, regulatory entry barriers, capital intensity, essential products and long history; only one of those applies cleanly to Alcoa. Market share, FY2025: primary aluminum 2,319 kmt of a 74,520 kmt world, 3.11%; alumina 9,640 kmt of 150,490 kmt, 6.41% — and even the "outside of China" leadership Alcoa claims covers only about 41% of the market [10] [11]. The Alumina segment's unit economics inverted: a realized-price-less-cost spread of +$163 per tonne in FY2024 fell to +$98 in FY2025 and to −$23 in the first half of 2026, and management wrote the segment's goodwill to zero [12] [13].

The strongest surviving counter-fact sits in the same treatment. Capital intensity is a genuine barrier — Western greenfield smelter capex runs $7,500–$9,100 per tonne against roughly $1,150 in China, and Alcoa has no greenfield plans anywhere because no site clears its return hurdle [14]. Management's own thesis is that the world will need roughly 7 million more tonnes of ex-China aluminum and 18 million more tonnes of alumina over the decade, and that new supply is hard to build [15]. That is a real barrier to entry; it is not the same as a monopoly's pricing power, and it did not stop 13.4 million tonnes of Chinese capacity being added in a decade. Under the gate's rule, genuine doubt resolves to not met — and here the doubt is not merely genuine, it is the majority reading. Full treatment in Durability.

FCF consistency (P2) — not met

The framework wants a stable rolling five-year average of adjusted FCF. That series cannot be computed for Alcoa — the deterministic feature file returns not_computable for every year because stock-based compensation is absent from the structured cash-flow feed — so the reported-FCF substitute stands in, and it is unpredictable. Reported FCF ran −$715M, $819M, $49M, $307M, $41M, $530M, $342M, −$440M, $42M, $567M across FY2016–FY2025; the rolling five-year average moves between $100M and $349M, a coefficient of variation of 0.45, with negative years in FY2016 and FY2023 [16]. The counter-fact: SBC is small when Alcoa's own filings disclose it ($41M in FY2025), so the illustrative FY2025 adjusted figure is close to reported — but the negatives here are price and restructuring events, not the every-5-to-8-year underwriting cycle the framework treats as healthy. See Yield.

Dislocation and yield (P3) — event yes, capitulation no, yield short

P3a (met). The drawdown is real and dated: −46% from an $83.79 close on 2 June 2026 to $45.26 on 31 July, with two primary-document triggers — the 30 June South32 acquisition and the 16 July Q2 print. Those two events explain only about 19.5% of the fall; roughly 82% was an undated June retreat in the aluminum price that management itself attributes to sentiment, saying the metal "has gone to pre-Iran conflict levels" though the underlying supply disruption remained [17]. See Dislocation.

P3b (not met). The fear gauge does not show capitulation. The volume-spike multiple is 1.23x — on both the feature-file leg and the live June–July leg — against the framework's 2x reference line, and no session reached the 4.8x–6.7x band of this stock's ten largest historical volume days. This was a price slide, not an emotional flush.

P3c (not met). Adjusted FCF yield is 4.45% on FY2025 figures and 0.16% on the three-year average, against the 8% fortress reference line the FY2025 balance sheet selects — 355 basis points short on the better of the two readings [18]. The counter-fact: net debt of $842M against $1,940M of FY2025 segment EBITDA is 0.43x, genuinely fortress-class [19] — but a fortress balance sheet sets a higher yield bar, not a lower one, so the strong balance sheet makes the shortfall wider, not narrower.

P3d (not met; probability 0.19, spread 0.24). Consensus forward free cash flow clears the bar on its face — 13.25% for FY2027, 16.09% for FY2028 on today's market cap — but falls to 7.18% and 9.85% once the framework's SBC and acquisition deductions are applied, and the announced $3.1 billion South32 purchase reinstates a $620M-a-year acquisition deduction that pulls the adjusted figure back below the line [20]. The probability of clearing the applicable 10% bar in any single year FY2027–FY2029 is put at roughly 15%. This is the widest spread on the board (0.24); the jury still agreed on the direction. See Yield.

Balance sheet and self-help (P4) — outlasts, but the engine is off

P4a (not met). The balance sheet can comfortably outlast a multi-year trough: after the May 2026 redemption of the 2028 notes there is no bond maturity before 2029, against roughly $3.0 billion of liquidity, and both revolver covenants clear by wide margins [21]. That is the supporting fact. Against it: capital allocation has been ranked with debt repayment first on every call from July 2023 to April 2026, and the balance resolved on 30 June 2026 in favour of a $3.1 billion cash acquisition sized "not to exceed a leverage ratio of 2.0x" [22] — the framework's own falsifier of a pivot to debt paydown and growth at the moment repurchases would matter most.

P4b (not met — the hard-fail condition). Share count is rising, and the driver is acquisitions settled in stock: 178 million weighted-average shares in FY2023 to 261 million in FY2025, +46.6%, with about 17 million more committed to South32. No shares were repurchased in FY2023, FY2024, FY2025 or the first half of FY2026 [23]. A rising count on serial acquisition is the condition the framework treats as disqualifying on its own; the buyback flywheel the dislocation is supposed to unlock is not turning. See Self-Help.

P4c (not applicable). The dividend is $0.40 per share annualized — a 0.88% yield at $45.26, far below the roughly 4% level at which the framework's dividend-safety test engages. For the record, it is covered about 5x by FY2025 adjusted FCF and has been held at $0.10 a quarter since November 2021, through the FY2023 loss year [24].

Diagnosis (P5) — temporary lean, meaningful permanent tail

The blind adversarial trial — two opposing cited briefs, three independent judges reading in different orders — put the probability the impairment is temporary at 0.66 (per-judge 0.57, 0.66, 0.68; mean 0.64; spread 0.11; recorded not contested). The arithmetic behind it: on the most defensible pairing, a pre-spike price base against a permanent reading, price damage of about $4.9 billion exceeds value damage of about $2.0 billion — a gap near $2.9 billion, but not the two-thirds-cut-with-NPV-intact pattern of the framework's Centene precedent. Half the drawdown is the give-back of a two-week geopolitical price spike; the line that actually broke is the Alumina segment, and part of that impairment is structural [25]. P5 leaning temporary does not rescue the case: the gate (P1) is upstream of it and nothing offsets a gate. See Damage Math.

Instrument context (I1) — not verifiable

The tally records I1 as not verifiable. Listed options on Alcoa do run to 21 January 2028 — about 17.6 months, clearing the 12-month line and just short of the 18-month target — and at-the-money implied volatility on that expiry is near 59% against reference lines of up to about 55 acceptable and 60–70 elevated. But the I1 fact test asks for open interest and bid-ask spreads from a dated, citable source, and the spreads could not be sourced from any document in the corpus; the option facts came from third-party web pages with no filing page. The criterion therefore resolves to not verifiable rather than exists. See Clock.

What a 3x-in-3-Years Would Require

The framework's target test cannot be rendered from the tally as arithmetic: re_rating_math is null, with the note "Re-rating math unavailable because the applicable bar or normalized adjusted FCF is missing." The reason is the same one that dogs P3c — the adjusted-FCF series the bar would price is not_computable in the feature file, so no price-at-bar can be struck cleanly.

What the run can say about the re-rating mechanism it points to instead: restoring the roughly $10.1 billion of market value lost since 2 June 2026, at the ~4.2x EV/EBITDA that fall implies, would take about $2.4 billion of additional EBITDA — equivalent to a +$1,008 to +$1,154 per tonne move in the LME aluminum price, 21% to 24% above the $4,752 per tonne Alcoa realized in the second quarter [26]. No dated corporate event sets that price. On this name's own nine-and-three-quarter-year record, an eighteen-month recovery to the prior high happened about 22% of the time — roughly one chance in five, with a median eighteen-month forward return of 0.0% from comparably depressed sessions. The base rate is in Clock; it is context for the target test, not a recommendation.

Contested and Undetermined

Nothing was contested; nothing was undetermined. No criterion carries a contested verdict, and none resolved to cannot-determine — the tally's flags.contested list is empty and no seat returned a missing-datapoint verdict. The two non-standard resolutions are neither: P4c is not applicable (the dividend is too small to trigger the test) and I1 is not verifiable (option spreads could not be sourced). The widest genuine disagreement was P3d's 0.24 probability spread, but the four seats still agreed on its not-met direction.

Provenance

Source: ruchir/fit_tally.json (provenance) and ruchir/refutations.json.

Two model families sat the jury and agreed on both the gate and the overall verdict, which is why the verdict itself is not in doubt; confidence is nonetheless low because the name-mask probe — which re-runs a seat blind to the company's identity — left a load-bearing probability gap of 0.21, just above the 0.20 line, and that raises a prior_driven_risk flag: it means a model's prior about the name may have moved a probability more than the evidence alone should. The skeptic pass pressed 51 claims and refuted none, so no cited claim on this tab was overturned in review.

The Falsifier Ledger

These are the standing conditions that would break, or confirm, the framework read — the first five are the framework's own templates, the rest are the name-specific trip-wires the tally recorded with their thresholds, directions and windows. Reproduced verbatim from ruchir/fit_tally.json:

Data Gaps

The run could not close several gaps, from the tally's list. The framework's own yield basis is the largest: fit_features.adjusted_fcf, adjusted_fcf_yield, yield_baseline, fcf_stability, float_retirement_years and balance_sheet_class are all not_computable because stock-based compensation is missing from the structured cash-flow feed for every year FY2016–FY2025 (though the filings themselves disclose it for FY2019 onward), so every adjusted figure on this report was rebuilt from the filed statements rather than taken from the feature file. Beyond that: no pro-forma financials for the South32 AliGroup assets exist in the corpus, so the post-close year-10 revenue and adjusted-FCF base cannot be computed; no LME or alumina index price series is in the structured data, so the price path driving the whole re-rating cannot be independently charted; short interest is entirely unavailable (FINRA returned no rows), and the insider-transaction and beneficial-ownership records both pre-date the June 2026 fall, so the identity of the seller cannot be evidenced; destination-basis China revenue is not disclosed, so the S1 quantification is an upper bound on a point-of-sale basis only; and the capitulation-gauge feature measures the already-healed 2024–25 drawdown rather than the live 2026 leg, which the Dislocation tab re-derives.