Ahresty Corporation (5852): A Third of Book, and a Floor Below Zero

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	too-hard	null	B2 can't estimate a decade out — six losses then one flattered profit, a 2.24% margin D3 facing a technology transition; B89 the bargain isn't there — NCAV is −¥14,057M / −¥564/sh F124F134, the 0.32× "book" is a mortgaged foundry (IBD 2.2× cap D12)
Munger	too-hard	null	M1/M2 the kill path is in motion — Wilmington in 債務超過 F318, a covenant barring two consecutive loss years E66, one shock trips the ¥10.3bn syndicated loan; M37/M18 no moat, cheap-for-a-reason — a 2.2% margin commodity converter with leverage senior to the equity
Pabrai	pass	null	P1 max loss — the stressed floor is below zero (~¥69–73bn realizable vs ¥79.9bn liabilities F134); net debt ¥27,232M D10; P7/P18 a cyclical that lost money six straight years — the losses eat the asset value the thesis rests on
Li Lu	too-hard	null	L1 can't know 2036 + L19 leverage forecloses it — net debt 2.2× cap D12 under a two-loss-year covenant E66; L35/L36 value destroyed not compounded — cumulative NI −¥12,284M over 5yr [F11–F15], ROIC ~4.5% at the peak
Claude	pass	null (implied ¥470)	C39 the asset floor is negative past the debt — haircut NAV −¥321/sh; the 0.32× P/B D1 is a levered residual; C14/C33 no compounding — −¥12.3bn cumulative NI while ¥50.8bn capex consumed [F54–F58]; FY2026 one-off-flattered over a falling recurring line

A unanimous decline (3 too-hard / 2 pass) — the cheapest name on the record (0.32× book), and no lens will touch it. The ≥4-decline consensus fired the record's consensus red-team, which argued the bull case (buy-below ¥900, fair value ¥1,157–1,455). The decline turns on two facts none dispute: the deep discount is a levered-equity illusion with a floor below zero, and the business destroys owner value through the cycle. Even the anti-TPR fact — die-casting is an EV-transition winner — cannot rescue a levered cyclical whose equity the debt sits ahead of.

The business

Ahresty Corporation (株式会社アーレスティ) melts aluminium, squirts it into steel dies under high pressure, machines the castings, and ships them to carmakers — engine and transmission cases today, and it hopes battery cases and body parts tomorrow. Die-casting is ~94% of sales D9, and better than nine of every ten of those yen come from autos E20; the group runs die-casting plants across Japan, the US, Mexico, China and India, reports five segments (die-casting by geography plus a small secondary-aluminium arm and a finished-products arm), and turns over ~¥167bn of revenue F5. Subaru is the sole 10%+ customer at 12.6% E55; over half the revenue is overseas, in a fistful of currencies D24. A shopkeeper can follow it — they cast metal parts, and Subaru and Suzuki and Honda pay because the tooling has been qualified into a car already in production.

Two facts about the register and the plan. It is founder-family-stewarded but un-milked: Chairman Takahashi Shin and President Takahashi (father and son) hold ~6.7% combined E75E98, take their pay in locked restricted stock (a three-year lock) E137, run no poison pill E72, and there is no activist on the register (the Interactive Brokers line at 3.7% is a broker nominee). And the capital-return catalyst is self-help: this fiscal year management introduced — for the first time — a 1.5% DOE (dividend-on-equity) floor as an explicit 配当下限額 E82, layered on a payout-ratio ≥35% target, inside a 10-year plan targeting ROE 9%, an equity ratio ≥40%, ¥140bn of growth investment, and a PBR of 1×. There is no buyback, and — tellingly — next year's dividend is already guided down, ¥42 → ¥34 E83, with FY2027 results "below FY2026" E24.

The one genuinely constructive, distinctive fact — the deliberate contrast with the prior study (TPR's melting ICE moat) — is that aluminium die-casting is an EV-transition beneficiary, not a melting moat. Electrification drives vehicle-lightweighting, and aluminium castings are how you lighten a car (battery cases, motor housings, structural/body parts); management's plan is to shift the portfolio toward EV and body-system parts, with a 55% electrified-vehicle-parts sales ratio set as a FY2030 target E32 (a target, not a current actual — the current ratio is not disclosed). The end-market is not the problem.

The numbers

At the ¥717 stamp Ahresty is the cheapest name on the record: P/B = 717 ÷ 2,238.37 F35 = 0.32× book D1 (no minority interest — the whole owners' book), a reported PER of ~5.0× D6 on EPS ¥144.16 F40, and a 5.9% dividend yield D13. The market capitalization is ¥17.88bn D2 against consolidated net assets of ¥55,943M F150. Three siren numbers — and both roads out of them dead-end.

The bargain road: the floor is below zero. Compute net current asset value the way Graham insisted — current assets ¥65,815M minus all liabilities ¥79,872M F124F134 — and you get negative ¥14,057M, about −¥564 a share. This is not a net-net; it is the opposite. Rebuild the balance sheet with distressed haircuts (cash at par, receivables ~85%, die-cast work-in-process ~50%, special-purpose presses near-scrap): realizable assets come to roughly ¥69–73bn against ¥79.9bn of liabilities — the common is left at zero to below zero, because ¥39,434M of interest-bearing debt F145 — 2.21× the entire market cap D12 — and ¥79.9bn of total liabilities stand ahead of it. Net debt is ¥27,232M D10. And the reported book is itself soft: ¥15,372M of the ¥55,826M owners' equity is a foreign-exchange translation reserve F155F153 — a weak-yen artifact, not cash the owner will ever spool out. The 0.32× "book" is real, but it is a thin, heavily-levered junior claim: a mortgaged foundry, not a dollar of assets for thirty cents.

The earnings road: the business does not compound through the cycle. FY2026's net income of +¥3,580M is the first profit in seven years F15E50 — but the five-year trail is −5,189 / −84 / −7,699 / −2,892 / +3,580 [F11–F15]: six straight losses, cumulative −¥12,284M, while the group consumed ¥50,789M of capital investment [F54–F58]. Capital destroyed, not compounded. And the one profit year is one-off-flattered — the batch's third consecutive such headline (after RINGER HUT's tax benefit and Kitagawa's asset-sale gain): ordinary profit actually fell, ¥3,044M → ¥2,865M F9F10; the swing to net profit came from a one-off ¥1,109M gain on selling the Guangzhou mold subsidiary F176, a tax charge that collapsed to ¥9M — a 0.3% effective rate vs a 30.6% statutory rate F188F332F331 (a deferred-tax release, not cash saved), and impairment shrinking from ¥3,300M to ¥392M F180F181. The parent alone still lost ¥1,002M F85. Strip the one-off and normalize the tax, and earnings are ¥2bn, not ¥3.6bn; the operating margin is 2.24% D3, ROE 6.65% F47 (below the firm's own 9% target F309), and rough through-cycle pre-tax ROIC is ~4.5% — in the peak year, on flattered profit. The auditor's two Key Audit Matters — die-cast fixed-asset impairment and parent deferred-tax-asset recoverability E140E144 — are the cyclicality already written into the audit, with a ¥14,012M DTA valuation allowance F328 sitting behind that 0.3% tax rate. A loss-making US subsidiary (Wilmington) is in 債務超過 (net worth −¥1,075M F318, a −¥2,334M loss F317) under restructuring; the covenant on the ¥10.3bn syndicated loan bars two consecutive years of recurring loss and a >25% net-asset fall E66 — though net assets rose this year (¥51,989M → ¥55,943M F149F150), leaving **30% headroom** before the floor bites.

The five lenses

Buffett — too-hard

Let me start where I always start — not with the stock, but with the business, as if a neighbor offered to sell me the whole thing. Ahresty melts aluminium, squirts it into steel dies, machines the castings, and ships them to carmakers — die-casting is about 94% of sales D9, and better than nine of every ten of those yen come from autos E20. A shopkeeper can follow that, so on the first gate — can I explain it? — the answer is yes; this is inside the circle as a business. But understanding what a business does is not the same as saying what it will earn, and that is where I stop. Ordinary profit over five years runs −2,032, +94, +2,574, +3,044, and now +2,865 [F6–F10] — it bounces around zero, and this year it actually fell; net income runs six straight losing years then one profit management calls the first in seven E50. I am asked to estimate what this earns in 2036, and I cannot do it within a range a skeptic could audit: a 2.24% operating margin D3 where a bad quarter of aluminium prices swings the whole result, over half the revenue overseas so that ¥15.4bn of "book" is nothing but a translation line F155, and an industry in a once-in-a-century CASE transition E20. When a thin-margin, cyclical, multi-currency parts maker facing a technology shift has a record of six losses and one flattered profit, the ten-year estimate B2 demands cannot be written. I don't feel comfortable, so I move on.

Because this profile is the Graham bargain-hunter too, I owe one more look — sometimes a business you can't forecast is worth owning at fifty cents on the dollar of hard assets. The bargain isn't there. Net current asset value — current assets ¥65,815M minus all liabilities ¥79,872M F124F134 — is negative ¥14,057M, about −¥564 a share. This is the opposite of a net-net. Rebuild it Dempster-style with haircuts and the common is left at roughly zero and easily less, because ¥39.4bn of interest-bearing debt F145 stands ahead of it — interest-bearing debt is 2.2× the entire stock-market value of the company D12. That is a dagger taped to the steering wheel. No hidden Sanborn assets rescue it — cross-holdings are a mere ¥2,049M F290 and the company is ¥27.2bn net debt D10. You are not getting the operating business for free; you are paying for a mortgaged foundry. The catalyst doesn't fill the gap — a new DOE floor E82, founder-controlled, with next year's dividend already cut ¥42→¥34 E83 — a company managing a soft patch, not a control party converting assets to cash on a dated clock. Too-hard.

What a student should take from this: a low price-to-book is not a margin of safety when the debt is a multiple of the market value — compute net current asset value against every liability, and a "0.32× book" can turn out to be negative net-net, a mortgaged foundry where the creditors stand ahead of you. And a business you can explain is not one you can necessarily value: six losses and a flattered profit, a thin margin, and a technology transition put the ten-year estimate out of reach at any price.

Munger — too-hard

Start by inverting. How does this destroy capital permanently? An auto downturn cuts orders — the business runs 94% die-casting, 90%+ auto E20 — and a company on a 2.2% operating margin D3 with ¥27,232M net debt D10 and interest-bearing debt 2.2× its market cap D12 cannot service its fixed charges. The syndicated-loan covenant requires net assets stay above 75% of the March-2025 base and forbids two consecutive recurring-loss years E66; free cash flow is thin enough (OCF ¥12,275M against capex ¥11,592M F251F252) that a moderate revenue decline erases it. In a real downturn the covenant trips, and you get lump-sum repayment demands. At that point the ¥17.9bn market cap buys you a levered remnant. There is no moat to stabilize this — die-casting is a process technology, not a franchise; gross margin is only ~10% F163, and it held only with sustained capex eating virtually all operating cash flow. Munger's textile-loom lesson: every improvement flows to the customer. No monopoly on process, no switching cost, no brand, no fence.

The mitigating case is real: aluminium die-casting IS EV-compatible E26, the business survived six loss years on positive operating cash flow, and it is not a cash incinerator in absolute terms. But "survives a downturn on operating cash flow" is not the hurdle — the hurdle is whether it compounds capital, and it does not. Six loss years then one turnaround at a 2.2% margin D3 and 6.65% ROE F47 (below the 9% target F309) against ¥140bn of planned capex F311 is a recovery record, not a compounding one. And the "first profit in 7 years" deserves the harshest reading: ordinary profit fell ¥3,044M→¥2,865M F9F10; the ¥3,580M net income was manufactured from a ¥1,109M one-time affiliate sale F176, a 0.3% effective tax rate F332 (deferred-tax movements, not cash taxes), and a much-smaller impairment; the parent still lost ¥1,002M F85, and FY2027 is guided below E24 with the dividend cut E83. The Wilmington problem is not resolved — the US sub is in 債務超過, losing ¥2,334M F317F318, "the top-priority issue" E24 (an open wound, not a resolution). And the near-zero tax is a second-order bomb: a ¥14bn valuation allowance sits behind it F328; if the turnaround reverses, the allowance rises and "earnings" include a non-cash tax charge. Cheap-for-a-reason: a fair-to-mediocre business at a cheap price, with a kill path already in motion. Too-hard.

What a student should take from this: leverage is what converts cyclicality into a permanent-loss mechanism — a covenant that trips on two consecutive loss years E66, on a business that has lost money six of seven, is a kill path already walking, and a low multiple does not compensate you for it. And read the tax line: a 0.3% effective rate F332 on a business with a ¥14bn deferred-tax valuation allowance F328 is an estimate in a suit, not real earning power.

Pabrai — pass

Heads I win, tails I don't lose much — and everything is downstream of the second clause. The screen is a burning theatre and I love burning theatres: 0.32× book D1, ~5× earnings D6, a 5.9% yield D13, a global die-caster left for dead after six losing years. But before I walk in I do the third-grade arithmetic on the floor, and my P1 rule is written in blood: build it from marked-down assets minus every liability, never off the equity line. Total assets ¥135,815M, total liabilities ¥79,872M F118F134; the ¥55,943M of net assets looks like a fat cushion, but ¥15,372M of it is an FX-translation reserve F155, an accounting entry, not cash I can collect. Mark the real assets for a distressed sale — cash at par, receivables haircut, die-cast work-in-process worth little to anyone but the next buyer of that exact part, special-purpose presses near-scrap — and I get roughly ¥69–73bn of realizable assets against ¥79,872M of liabilities. The stressed floor is below zero. In a hard scenario every yen belongs to a creditor before one reaches me — the P1 failure the item was written to catch: net debt is ¥27,232M D10, so the floor is the stress test alone, and the stress test says there is no floor.

And this is not a business you can assume stays a going concern through a trough. It lost money six years running [F11–F14]; a cyclical die-caster >90% auto-dependent E20 whose losses eat the asset value the thesis rests on — the melting-ice / trough-vs-rot problem (P7, P18). The one profit year is heavily flattered: ordinary profit fell F9F10; the swing came from a ¥1,109M one-off gain F176, a tax charge collapsing to a 0.3% rate F332, and impairment shrinking F180F181. Strip those and the margin is 2.24% D3; the parent still lost ¥1,002M F85; the ~5× PER is 5× a flattered number, on a business guided "below FY2026" E24. I will concede the honest bull points — the end-market is EV-transferable E26 (not melting), the DOE floor pays ~5% to wait, the covenant has headroom this year — but Dhandho is not "cheap and shrinking is a buy." There is no "tails I don't lose much" when the company carries net debt 2.2× its own market value and loses money every cycle. This is not a fifty-cent dollar with a hard backstop; it is thin equity on ¥79.9bn of obligations. Pass — not too-hard (I can read it plainly), just uninvestable at a price where the floor is negative.

What a student should take from this: an equity cushion is not an asset floor. When a company carries net debt, build the floor from marked assets minus every liability — and here that floor comes out below zero, which turns the cheapest-looking P/B on the record into no margin of safety at all. A cyclical that goes underwater every downturn erases the asset value the bargain rests on; cheap plus levered plus loss-prone is not the burning theatre you walk into — it is the one you walk past.

Li Lu — too-hard

Let me teach this one carefully, because it is the trap that catches disciplined students most often: a genuinely cheap stock that fails not on price but on the one thing price cannot cure. Ahresty trades at 0.32× book, ~5× earnings, a 5.9% yield D1D6D13 — every reflex I inherited from Graham says look here — and unlike a business dying with its end-market, aluminium die-casting is genuinely EV-transferable E22E26E32. So the discipline is harder: I must set the cheapness aside and ask the only gating question (L1) — can I honestly claim to know this business's earnings power ten years out better than almost anyone who owns it? After the work, no. The trail is six losses in seven years [F11–F15]; sum the five-year window and the company destroyed ¥12,284M of owner earnings. Per-share book is 2,068 → 2,238 over five years [F31–F35] — essentially flat, held up by ¥15,372M of FX-translation gain in equity F155, not earnings. Intrinsic value has not compounded; it has tread water while the yen sagged. A statically cheap business whose owner-value does not grow is a pass in this lens — and this one's record is worse than static.

The "first profit in seven years" I will not let do the reasoning: ordinary profit fell F9F10; the swing was flattered by a ¥1,109M one-off gain F176, a near-zero tax charge F188F332, and a smaller impairment; the parent still lost ¥1,002M F85; the US Wilmington subsidiary — over 10% of sales — sits in negative net worth F318, drawing a ¥1,448M write-down and a ¥1,075M provision from the parent F322F323. A fragile, one-off-assisted turnaround, not durable earning power. Then the number a long hold earns me (L36): operating margin 2.24% D3, ROE 6.65% F47 below the 9% target, rough pre-tax ROIC ~4.5% in the best year of the cycle — a fraction of any cost of capital, far from the 40–100% bar where the mathematics get interesting. And the leverage forecloses what remains: net debt 2.2× cap D12 under a covenant tripping on two consecutive recurring-loss years E66 — a business that has lost money six of seven years does not clearly survive a severe multi-year downturn on its own resources; leverage converts this cyclicality into a permanent-loss mechanism, and my rule against it is absolute. The catalyst confirms the character rather than rescuing it: a DOE floor but no buyback, the dividend cut ¥42→¥34 E83 — a company husbanding a fragile balance sheet, honest and un-milked, but not one whose value is compounding to the owner. Too-hard.

What a student should take from this: the knowledge bar is about predictability and compounding, not cheapness — a business that destroyed ¥12bn of owner earnings over five years, earns ~4.5% on capital in its best year, and carries net debt at twice its market value under a two-loss-year covenant cannot be forecast a decade out, and no 0.32× book changes that. A real EV tailwind makes the end-market safe; it does not make a levered cyclical's equity a compounder.

Claude — pass (implied buy-below ¥470)

My figures-blind priors registered this as a below-median call with a real left tail and a real constructive right-shoulder, gating on H1 (is the deep discount reachable past the debt) and H2 (does the business compound across the cycle). The ledger confirmed the direction — it is a levered-cyclical value trap on the numbers — but overturned one prior in the company's favor: I expected the covenant near the equity's fate, and it is not. Net assets could fall ~30% before the floor bites (floor ¥38,992M F262 vs ¥55,943M F150), and the recurring-loss trigger needs two consecutive ordinary losses E66, which FY2026's +¥2,865M reset. The left tail is real but not imminent — a §1 overstatement, corrected. That moves the case off the covenant-cliff edge to a clean, priced pass with a named lower price, not a too-hard.

The business is a global aluminium die-caster, ~94% die-casting D9, whose FY2026 is its first profit in seven years F15 after a −5,189/−84/−7,699/−2,892 trail [F11–F14], cumulative −¥12,284M. That single fact is the whole debate, and the ledger resolves it three ways, all adverse. First, the discount is not reachable past the debt (H1 fails). At 0.32× P/B D1 the equity is a thin junior residual: interest-bearing debt is ¥39,434M F145, 2.2× the ¥17.88bn market cap D12; net debt ¥27,232M D10. I ran the C39 floor — crude NCAV is −¥564/share; a haircut liquidation nets −¥321/share against all ¥79,872M of liabilities F134. The debt sits ahead of the assets; there is no liquidation margin of safety, and ¥15,372M of the "book" is an FX-translation artifact F155 doing a third of the work. Second, the business does not compound through the cycle (H2 fails) — the dominant finding. Cumulative NI −¥12,284M over five years [F11–F15] while ¥50,789M of investing outflow was consumed [F54–F58]. Normalize honestly: this year's ordinary profit ¥2,865M at a normal 30% tax → ~¥2,006M → EPV at 10–12% ≈ ¥670–804/share, straddling the stamp — but that is the optimistic base, over a falling recurring line, and the loss-inclusive through-cycle average is negative. Third, the turnaround is one-off-flattered and the catalyst is toothless — the ¥1,109M gain F176, the ¥9M tax F188, founder-controlled, no buyback, the dividend cut E83, surplus cash pledged to repay debt first E67. The EV tailwind is genuine (the anti-TPR fact holds), but the 55% is a 2030 target E32 — real optionality, credited little until booked. None of the deciders is a knowledge-wall — they resolve on the ledger and time, and they resolve adverse — so this is a pass, not too-hard. Implied buy-below ¥470 (stamp-blind): a stressed, one-off-stripped, normally-taxed recurring earning power ~¥1,400M capitalized at ~12%, with the asset floor supplying nothing because it is negative past the debt — ~34% below the stamp, and the FY2026 low was ¥489 F98, so the market has independently traded within ~4% of my floor. Correctly cheap, not offered.

What a student should take from this: a 0.32× P/B D1 is not a margin of safety when the debt is 2.2× the market cap D12 — the liquidation floor here is negative once you put creditors ahead of a junior equity, so the discount is earned, not offered. The verdict turned on which earnings base you normalize: this year's one-off-flattered ordinary line makes the EPV straddle the price, but the loss-inclusive through-cycle average (−¥12bn cumulative over five years while ¥51bn of capex was consumed) says the business oscillates rather than compounds. A first profit in seven years resting on an asset-sale gain and a near-zero tax over a falling recurring line is weak evidence, not a track record — and a real EV tailwind does not make a levered cyclical's equity a bargain.

Synthesis

Where the lenses agree

For the first time in the record, all five lenses decline the same name — and it is the cheapest name they have ever seen (0.32× book, ~5× earnings, a 5.9% yield D1D6D13). The agreement is total on two facts, and they are the whole study. First, the deep discount is a levered-equity illusion — the floor is below zero. Every lens ran the balance sheet and found the same thing: net-current-asset value is negative ¥14,057M / −¥564/share F124F134 (Buffett), a haircut liquidation nets negative ¥321/share (Claude), realizable assets fall short of ¥79.9bn liabilities (Pabrai) — because interest-bearing debt is 2.2× the entire market cap D12 and stands ahead of the common, and ¥15,372M of the "book" is a foreign-exchange translation reserve F155, not distributable cash. As Buffett puts it, "you are paying for a mortgaged foundry." Second, the business destroys owner value through the cycle. Cumulative net income is −¥12,284M over five years [F11–F15] while the group consumed ¥50,789M of capital investment [F54–F58]; the operating margin is 2.24% D3, through-cycle ROIC ~4.5% in the peak year, and FY2026's "first profit in seven years" is one-off-flattered — a ¥1,109M affiliate-sale gain F176, a 0.3% effective tax F332, and a shrunken impairment, over an ordinary profit that actually fell F9F10, with the parent still losing ¥1,002M F85. This is the batch's third consecutive one-off-flattered "record," after RINGER HUT's tax benefit and Kitagawa's asset-sale gain — a thread the panel's operating-line discipline caught each time.

Where the lenses diverge

The split is too-hard (Buffett, Munger, Li Lu) vs pass (Pabrai, Claude) — a distinction of epistemics, not conclusion. The too-hard three say the leverage-into-cyclicality makes the ten-year estimate unwritable: "a thin-margin, cyclical, multi-currency parts maker facing a technology transition, with six losses and one flattered profit," says Buffett, "cannot be estimated within a range a skeptic could audit [B2]"; Munger names a live kill path — "the covenant trips on two consecutive loss years E66, on a business that has lost money six of seven — a dagger already walking"; Li Lu invokes both the knowledge bar and the leverage as "a permanent-loss mechanism [L19]." The two passers reach the same not-a-buy by a different route — a priced verdict rather than an unforecastable one. Pabrai: "I can read it plainly; there is simply no floor — the stress test comes out below zero, so it is uninvestable, not unknowable." Claude is sharpest on why it is pass-not-too-hard: "none of the deciders is a knowledge-wall — the negative floor and the loss-inclusive through-cycle average resolve on the ledger, and they resolve adverse; the value is simply below the price, buy-below ~¥470." The gap between too-hard and pass is whether a levered deep-cyclical is unknowable (Buffett/Munger/Li-Lu) or knowably-below-value (Pabrai/Claude) — and both camps agree ¥717 is a decline.

One genuine disagreement of fact is worth recording, because the red-team leans on it: is the covenant a live kill path or a priced-in worry? Munger treats it as the primary kill path; Claude overturned his own figures-blind fear on the ledger — net assets rose this year and sit at ~108% of the covenant base, ~30% of headroom F150F262, and the recurring-loss trigger needs two consecutive ordinary losses, which FY2026 reset. The honest reconciliation: the covenant is not near-tripping this year (Claude/the red-team are right), but on a business that lost money six of seven years a two-year loss run is a real tail (Munger is right that it is the mechanism) — so it caps the verdict without being imminent.

The red team, engaged

Because five lenses declined (≥4 in the decline class), a fresh adversary (ledger only) argued the bull case — buy-below ¥900, fair value ¥1,157–1,455 — and the synthesis must meet its strongest points by name.

• "The Mexico subsidiary's equity alone (¥19,613M F320) exceeds the whole company's ¥17.9bn market cap D2 — hidden asset value the 0.32× ignores." This is the bull's sharpest point, and it is true as stated — but it is the wrong claim for a minority buyer. What a shareholder can reach is the consolidated residual after all ¥79.9bn of liabilities F134, and that residual is negative on any haircut (Claude's −¥321/share, Buffett's −¥564 NCAV). A profitable sub's gross equity is not reachable value while ¥39bn of parent-level debt F145 sits ahead of it and the losing US sub F318 and the FX reserve F155 sit inside the same consolidated book. The panel priced the whole levered claim, not one good part of it.
• "The operating turnaround is below-the-line-clean — operating profit +10.9% D4, the US loss halved F204F205." Conceded, and real — Munger and Claude both credit the operating recovery. But it is a recovery to a 2.24% margin D3 on a falling ordinary line F9F10; "survives and recovers a little" is not "compounds," and the +10.9% is off a depressed base with the US still in 債務超過.
• "The end-market grows — die-casting is an EV beneficiary, not a melting moat." Fully conceded — it is the study's one distinctive constructive fact, and the reason this is not TPR. But every lens reached the same conclusion: a safe end-market does not make a levered cyclical's equity a bargain. The 55% EV-parts ratio is a 2030 target E32, and even granting it, the problem is the balance sheet and the through-cycle earning power, not the demand.
• "Leverage is manageable — equity ratio 41%, the covenant has ~30% headroom F150F262." Conceded on headroom (Claude overturned the covenant-cliff fear). But "not tripping this year" is not "not a risk" — net debt at 2.2× the market cap D12 on a six-of-seven-loss-years cyclical is exactly the leverage-into-cyclicality that makes the equity a levered residual, floor below zero.
• "A new DOE floor caps the downside; you're paid ~5% to wait." Conceded that the yield is real — but the dividend is being cut ¥42→¥34 E83, the floor sits on cash pledged to repay debt first E67, and a distribution yield is not owner-earnings yield when the recurring line is falling.

The red-team did not move any verdict off decline, but it earned its keep: it is why the synthesis records the covenant as not-imminent, and why the "pass" (a priced decline with a buy-below far below the stamp) rather than a reflexive dismissal is the honest resolution. A consensus that never faced its strongest opponent is not a conclusion; this one did, and the bull's own fair-value math (¥1,157 on stressed book) requires a 60–100% re-rate the ledger does not support at the stamp — while the bear's floor is negative.

Self-distance note. The Claude lens holds one of the five verdicts compared above (pass) and wrote this synthesis; it also built the dual-blind reconciled figure table and evidence ledger all five lenses consumed, and the red-team ran on the same model family. That is an unusual concentration of authorship — the answerer, the ledger-builder, one of the five voters, and the adversary are the same system. Read the synthesis with that in mind.

Prediction-vs-actual: VOID. This was an autonomous headless cycle; the human blind prediction is voided (void: no-human-prediction, never forged). No prediction-vs-actual scoring applies.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp. No lens issued a buy-below here; Claude publishes an implied downside floor of ¥470 (C44), ~34% below the ¥717 stamp — a ceiling on the arithmetic, not a buy-below.
• pass / watch / too-hard are recorded but unscored in any future review. Buffett's, Munger's and Li Lu's too-hard, and Pabrai's and Claude's pass, carry no buy-below by construction.
• The original verdict counts at its original stamp regardless of later corrections.
• On a stock split, reverse split, or consolidation, the buy-below threshold restates mechanically by the announced ratio (corporate-action disclosure cited); the stamp itself never restates.

Red team

A consensus red-team (five of five lenses declined) was dispatched to argue buy — that ¥717 / 0.32× book is a bargain the consensus is too cautious on. Its strongest points, verbatim-faithful, and the synthesis's engagement, are in the section above. In brief, the adversary's six ranked points were: (1) absolute cheapness on a real, book-backed ¥167bn-revenue business, the Mexico sub's equity alone exceeding the market cap F320D2; (2) EV/lightweighting is a structural tailwind, not "melting" E26E32; (3) the operating turnaround is below-the-line-clean (+10.9% D4); (4) a new DOE floor caps the downside E82; (5) leverage is manageable with ~30% covenant headroom F150F262; (6) clean, aligned governance and hidden asset value E137F155. The synthesis concedes points 2, 3, 5 (headroom) and 6 outright, and meets the load-bearing points 1 and 4 by the reachability test: a profitable sub's gross equity is not reachable to a minority while ¥39bn of parent debt and ¥79.9bn of consolidated liabilities F145F134 sit ahead of a residual that is negative on any haircut, and a cut dividend on debt-pledged cash E83E67 is a distribution yield, not a margin of safety. The bull's own fair value (¥1,157 on stressed book) needs a 60–100% re-rate the ledger does not support at the stamp; the bear's floor is below zero. A consensus that faced its strongest opponent and emerged intact — sharpened on the covenant timing, unmoved on the verdict.

What would change our minds

Pre-registered falsifiers. No lens issued a buy-below or watch; the too-hard and pass verdicts carry none by construction. What would move the name into range, per the lenses, is observable and dated:

• Buffett (too-hard). Into the circle: a multi-year record of positive, self-funded owner earnings that survives normalization (statutory tax, no affiliate-sale gains) — five straight years of ordinary profit above ¥5bn with positive free cash flow after growth capex — AND a balance sheet where interest-bearing debt is a fraction of, not 2.2× D12, the equity value, so the ten-year estimate B2 asks for can be written.
• Claude (pass, implied ¥470). Toward buy at ~¥470 or below (where a stressed, normally-taxed recurring earning power clears ~12% with the asset floor supplying nothing). Toward watch if the recurring (ordinary) line rises durably above ¥2,865M on a full tax charge with Wilmington's deficit closing F174E14. Confirmed decline by a second consecutive ordinary loss (FY2027+FY2028) tripping the covenant tail E66, or the EV-parts pivot stalling E32.
• Munger / Li Lu (too-hard) carry no price falsifier by construction — the leverage-into-cyclicality and the knowledge bar are the barriers, not the price; only a de-levered balance sheet and a multi-cycle record of the recurring line compounding would re-open the question.

The single observable most lenses converge on is whether the recurring (ordinary) line compounds on a normal tax charge — the difference between a genuine turnaround and a one-off-flattered peak F10F188 — with a covenant-tripping second loss year E66 as the confirming left tail.

What this taught the checklists

Queued for the next study (F2 revision proposals; see docs/process/evolution.md), attributed per lens:

• Buffett — a levered-book-trap cross-check forcing NCAV (current assets − all liabilities) to be computed before any P/B is anchored on, flagging when it is negative (the low P/B is then a levered residual, not a bargain); and a maintenance-vs-growth capex split in B42's owner-earnings for capital-intensive cyclicals.
• Munger — flag a bank covenant that trips on consecutive loss years, on a business with a multi-year loss record, as a live kill path (M1/M2), not a footnote; and treat a near-zero effective tax rate sitting on a large deferred-tax valuation allowance as an earnings-quality red flag (the allowance can reverse).
• Pabrai — sharpen P1 to strip the FX-translation reserve when building the floor for a JGAAP name with heavy overseas assets (¥15.4bn here F155 was a third of the "book"); and a P56 sub-test flagging a turnaround print produced below the operating line while operating/ordinary profit is flat-or-down.
• Li Lu — add cyclical-trough / loss-inclusive normalization to L39/L42 so a one-off-flattered peak year is never read as run-rate (here the loss-inclusive five-year average is negative); and consider making leverage-into-cyclicality an explicit gating trigger under L19/L20, parallel to the structure and accounting gates.
• Claude — two, plus a library seed. (1) A C39 sharpening — "levered-residual floor test": for a net-debt name, require the haircut NAV stated net of all liabilities as a signed per-share number, flagging explicitly when it is negative (no floor at all — the first net-debt-deeply-cyclical name on the record, a case the checklist was built without). (2) A C40/C33 sharpening: when the record contains loss years, state both the this-year-normalized and the loss-inclusive-average EPV and use the lower as the downside base. Library (Class-level): Ahresty seeds the "cheap, heavily-levered, deeply-cyclical auto-parts converter with a fresh one-off-flattered turnaround and a toothless self-help catalyst, in an EV-transferable end-market" class (N=1) — the levered-cyclical counterweight and the end-market mirror of TPR — datapoint: the deep book-discount was NOT reachable past the debt (haircut floor negative), and the business did not compound through the cycle.

---

Corrections