Yorozu Corporation (7294): A Third of Book, Two-Thirds One Customer

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	too-hard	null	B2 — can't estimate earnings without forecasting Nissan; B89 — NCAV < 0, net debt, so no asset floor
Munger	too-hard	null	M1/M18 — Nissan 64% and declining + no supplier moat; M90 — Nissan trajectory unjudgeable; M2/M42 — two impairment cycles = permanent loss
Pabrai	pass	null	P12 — Nissan 64% disqualifying; P1 — soft floor post-impairment + net debt; P61 — guided down 44–47%
Li Lu	too-hard	null	L1 — can't know 10 years through Nissan/EV/cycle; L35/L47 — book round-trips, FX-inflated, twice-impaired = melting; L22 — ROE 3.9%
Claude	too-hard	implied ¥520	C98 — Nissan trajectory unknowable; C39 — floor soft, FX-inflated, control-locked; C9/C40 — margin at cycle top

The business

Yorozu makes the steel skeleton that sits under a car. Its products are automotive mechanism parts — suspension and chassis components: front and rear suspension members, lower arms, rear beams, radius rods, lateral rods — stamped and welded from automotive steel plate and sold to automakers who bolt them onto vehicles E1E83E37. It is an 81-year-old, TSE-listed Tier-1 supplier. The group comprises the parent and 19 consolidated subsidiaries E1, and it reports on a geographic basis in three segments — 日本 (Japan), 米州 (the Americas — its largest), and アジア (Asia) E2E3. FY2026/3 segment sales were Japan ¥54,479M, Americas ¥87,298M, Asia ¥34,551M E4F270F272F274. The manufacturing footprint is Americas-heavy: subsidiaries operate in the U.S. (Tennessee, Alabama), Mexico (Mexicana, Guanajuato), Brazil, Thailand, China (two 51%-owned Wanbaojing JVs), India, and Indonesia E6E7E8, and about 69% of sales are overseas E36. Two Americas subsidiaries — Yorozu Automotive Tennessee (85.01%-owned) and Yorozu Mexicana (92.35%-owned) — each individually exceed 10% of consolidated sales E7.

The one fact that governs everything is the customer. Yorozu's products are drivetrain-agnostic — suspension parts are, in management's words, "indispensable even in EVs," and lightweighting is a stated tailwind E5E34 — but its revenue is not customer-agnostic. In FY2026/3, sales to the Nissan group were ¥112,443M — 63.8% of consolidated revenue E11D5, to an automaker whose own trajectory is the subject of public distress. Toyota-group (¥24,177M, +10.5%) and Honda-group (¥19,782M) sales are real and Toyota is growing, but together they are ~25% of revenue against Nissan's near two-thirds E11F330F331. The keiretsu ties run deep: the President & CEO is a former head of Nissan's No.1 Procurement Department, an outside director ran Calsonic Kansei/Marelli, and Nissan Trading sits in the top-10 shareholders E71E70E52. Notably, Nissan Motor does not hold Yorozu shares — the dependence runs one way E57.

The numbers

Revenue was ¥176,330M in FY2026/3 (−1.2% YoY) F5E18. The five-year earnings record is the story: net income attributable to owners ran +¥876M, +¥1,422M, then −¥3,926M (FY2024) and −¥13,448M (FY2025), then a recovery to +¥2,075M (FY2026) F11F12F13F14F15. The two loss years were impairment years — write-downs of ¥11,800M in FY2024 and ¥9,145M in FY2025 (by segment: Americas ¥5,269M, Asia ¥4,080M, Japan ¥64M) — after which impairment collapsed to ¥105M in FY2026 F327F324E23F326. The Americas write-down was explicitly caused by "large variation in customer demand accompanying unexpected market-conditions changes" E88. Operating income recovered to ¥3,980M, ~13× the cyber-attack-depressed prior year — but that is a 2.26% operating margin F217D2E19. Comprehensive income swung −¥13,641M then +¥7,381M across the two years F19F20.

The balance sheet is not a fortress. Net assets are ¥64,053M, equity ratio 39.2%, and against ¥29,767M of cash & deposits the company carries ¥39,105M of borrowings — net debt ≈ ¥9,338M F25F35F126F411D1. Owners' equity (自己資本) is ¥54,154M and BPS is ¥2,629.49 F207F45; at the ¥858 stamp that is P/B ≈ 0.33× (858 ÷ 2,629.49). Cash & deposits alone (¥29,767M F126) are 1.68× the ¥17.67bn market cap. Treasury stock stands at 17.8% of issued after a large FY2026 buyback (756,859 → 4,460,459 shares) D6F369F368. Accumulated depreciation against ¥45,898M of net PP&E is ¥194,705M — the plants are heavily depreciated F151F361. Intangibles are ¥209M; there is almost no goodwill to write down F152.

Customer concentration is disclosed on two bases, both real and not in conflict: the MD&A states Nissan-group sales of ¥112,443M / 63.8% E11D5; the formal ≥10% major-customer note lists only the two Nissan-Americas entities — Nissan Mexicana ¥28,170M (16.0%) + Nissan North America ¥22,234M (12.6%) = ¥50,404M / 28.6% — because Nissan Japan (¥18,163M / 10.2% in FY2025) slipped below the 10% threshold in FY2026 and dropped out of the note E12E13F340F338F339. And 46.1% of trade receivables sit with major customers E16.

Then the guidance. Management guides FY2027/3 down: revenue ¥166,000M (−5.9%), operating income ¥3,300M (−17.1%), ordinary income ¥2,100M (−44.4%), net income ¥1,100M (−47.0%) F404F405F406F407. The dividend holds at ¥33/share (3.85% at ¥858) but the forecast payout ratio rises to 62.3% F409F410E43.

The five lenses

Buffett — too-hard

Let me tell you what this company does before I say a word about the stock. Yorozu stamps steel into the suspension and chassis parts that sit under a car — front members, lower arms, rear beams, radius rods E83. That part I understand fine; it's a metal-bashing business, the kind Charlie and I have looked at a hundred times, and B1's circle-of-competence test is not where this one dies. It dies on the one fact that towers over everything. When I buy the whole business — the only way I know how to think about a stock — I am buying its customers, and this company's customers are, to a startling degree, one customer: Nissan, at 63.8% of revenue E11D5. When I owned American Express through the salad-oil mess, the franchise sat with millions of cardholders and no single one could hurt it. Here it is the reverse — if Nissan sneezes, Yorozu catches pneumonia, and Nissan has been coughing for years E15E16.

Whether that rope frays turns entirely on Nissan's fortunes, and I cannot make a defensible ten-year earnings estimate for Yorozu without first making one for Nissan — and I cannot make one for Nissan. That is B2: "if you don't feel comfortable making a rough estimate of the asset's future earnings, just forget it and move on." The wreckage confirms the instinct — two loss years out of five F13F14, from writing down plants the customer stopped filling ("the customer changed its mind, and Yorozu's plant became worth less" E88). The recovery is real but thin — a 2.26% operating margin D2 with capex starved 46% below plan and next year guided down 44–47% E45F406F407. And the balance sheet does not cover me while I wait: net current asset value (current assets ¥81,649M less all liabilities ¥74,205M less minorities ¥9,406M F142F184F201) is below zero, and there is ¥9.3bn of net debt D1. At 0.33× book it is plainly not expensive — but cheapness is a trap, not an invitation, when the outcome hinges on a variable outside my competence. Into the too-hard pile it goes, and I sleep well having never swung.

What a student should take from this: circle of competence is not about whether you understand the widget — it's about whether you can forecast the thing that drives the earnings. A 0.33× book multiple is not a margin of safety when the assets themselves are hostage to the same customer's demand; a discount one impairment cycle can erase is bait, not protection. Verdict: too-hard.

Munger — too-hard

Invert first. Write the obituary before the prospectus. How does a suspension-parts maker like this one die? It dies when its one big customer stops buying — and here the one big customer is Nissan, 63.8% of sales D5E11, already shrinking (Nissan-group −1.6%, Asia revenue −11.2% on China production cuts E11E4E21). That is not a hypothetical kill path in a risk-factor section; it is in motion in this year's numbers. The economics are a commodity's: gross margin 11.4%, operating margin 2.26% in the good year and 0.17% the year before D4D2D3, ROE −6.1% then −22.8% then +3.9% F38F39F40. Name the moat mechanism honestly and it is thin — a designed-in tier-1 position the customer holds a knife over; Yorozu just cut its capex plan 46% because "key customers revised model-launch timing" E45. That is the fingerprint of the buyer's power, not the supplier's moat (M18).

I will weigh the other side fairly. The EV threat is milder here — suspension is drivetrain-agnostic and lightweighting is a genuine tailwind E5E34. And buying back 14.8%-authorized stock at a third of book, funded by trimming cross-holdings, is closer to Singleton than to mechanical dilution-offset E47E59. But they spent ~1.8× the year's net income on it while carrying net debt into a −47% year F264F15F407, thinning the very cushion a customer-concentrated cyclical needs — defensible, not obviously wise. The incentives are clean (bonus on operating margin, stock on mid-term operating profit, no one paid ≥¥100M E73E74E79); the anti-takeover plan's 20% trigger is the governance flag that blocks the one clean way value gets realized E64. Here is where I land: cheapness isn't even the deciding issue. The decisive variable (M90) is Nissan's ten-year volume trajectory, and I cannot judge it from this ledger — Nissan's own financials aren't in the corpus. Two impairment cycles totaling ¥20.9bn in two years F327F324 already prove this business converts a demand shock into permanent equity destruction. When the one variable that governs the outcome is unknowable and inversion finds the kill path already moving, the answer is not a lower price. It is the too-hard basket. Next.

What a student should take from this: cheap is not a thesis; it is a starting price for a business you must still be able to underwrite. Statistical cheapness plus a live, in-motion kill path is the classic value trap — name the one variable that decides it, admit you can't judge it, and walk. Verdict: too-hard.

Pabrai — pass

Let me teach you what Yorozu looks like through the Dhandho lens — and why it fails the fundamental test before we even get to the price. The business is simple: suspension and chassis parts, 5,405 employees, three geographic segments, ¥176,330M revenue F81E3F5. The product is fine — suspension parts are needed in EVs too E5. But here is what stops me cold, and I want you to sit with this number: Nissan-group sales were 63.8% of consolidated revenue E11D5. That is not a customer. That is a co-dependency, and Nissan is itself a struggling automaker. When your biggest customer needs a bailout discussion, you do not own a suspension-parts maker — you own an indirect Nissan derivative (P12, a hard stop).

Now the downside math, because Dhandho demands it. Headline P/B is 0.33× F45, but a book equity cushion is not an asset floor — only marked-down liquid and hard assets protect you. Stress it: haircut receivables 20% (46.1% concentrated in major customers E16), inventory to 50¢, and recognize that PP&E carries ¥194,705M of accumulated depreciation and has already been impaired ¥11,800M then ¥9,145M in exactly the Nissan-dependent Americas and Asia segments F361F327F324E88. Land is only ¥4,648M F148. Stressed liquidation value falls well short of ¥74,205M total liabilities F184; the company is net-debt D1 (P1 fails). The earnings history is damning — two big loss years, a one-year recovery, and management guiding the very next year down 47% F13F14F15F407 (P61 fails). Americas, the largest segment, earned ¥152M of operating profit — essentially zero margin F286. Could I clone this? The auto-parts makers that recovered did so when their primary OEM recovered — so I am not cloning a proven recovery, I am making a Nissan bet without holding Nissan. The verdict is pass. When tails includes "Nissan accelerates restructuring and we take year three of impairment," the game is not Dhandho — it is speculation on a turnaround at a struggling automaker I would never buy directly. Most companies studied should produce no action; Yorozu is this study's example of the system working.

What a student should take from this: cheap-on-book is not the same as having a downside floor — do the stressed liquidation math, and two consecutive impairment cycles tell you where the stress will land. When the business itself guides down 44–47% F406F407 the year after a recovery, do not confuse a single-year inflection with a restored earning-power trend. Verdict: pass.

Li Lu — too-hard

Let me begin where I always begin, with the honest question: could I, after this work, claim to predict this company's earnings power ten years out better than most of the people who own it? I cannot. And so, whatever the price, this is too-hard for me. Yorozu makes suspension and chassis parts, a real business with a global footprint of 19 subsidiaries E83E6, priced at a third of owners' equity of ¥54,154M F207 — a deep and genuine statistical discount. But look at the decade of value creation, because that is the whole game (L35): net income summed across five years is a cumulative loss of about ¥13bn F11–F15, with two consecutive heavy-impairment years F327F324, and book value per share round-trips — 2,375 → 2,636 → 2,724 → 2,140 → 2,629 F41–F45. It does not compound. I cannot tell melting from a pause, and that distinction is the entire question.

Then the fact that decides everything (L1, L11): 63.8% of revenue to the Nissan group, a struggling automaker F329D5E11. My Timberland test asks whether the feared exposure is small enough that a total loss costs under 5% of earnings — here the feared exposure is the company. I cannot predict Nissan's model volumes, its EV strategy, its survival, or whether it keeps sourcing from Yorozu, and neither can anyone from these filings. Layer on an EV transition the filings themselves call uncertain E33 and an auto cycle cut by tariffs and China E38, and a company that turns a corner then immediately guides earnings down by nearly half F406F407 is telling you the corner is not turned. Could the book discount rescue it as an asset play (L47)? I decompose it honestly: owners' equity includes ¥12,506M of volatile OCI (¥8,699M of it FX translation F195F194F198), and the tangible plant behind the 0.33× P/B has been impaired twice in three years — book value I have specific evidence is not reliably worth its carrying amount. On management I am fair: real buybacks, cross-holding cuts, majority-independent committees, no ¥100M earner, unqualified auditor, no going-concern flag E47E59E67E79E80E91 — a founder-controlled company E63 with an anti-takeover plan E64 but no evidence of minority milking. Yet good capital allocation on a business I cannot predict does not clear the bar; it only tells me the melting, if it is melting, is being partly returned to me along the way.

What a student should take from this: the knowledge bar gates everything, and no discount repairs a business whose next ten years you cannot honestly forecast. A P/B of 0.33× is a fact, not a thesis — "cheap on assets" and "melting ice cube" can be the same object, and you must tell which before you buy. Capital-allocation virtue is necessary but never sufficient; predictability comes first. Verdict: too-hard.

Claude — too-hard, implied buy below ¥520

I registered, figures-blind, that Yorozu is a customer-captive cyclical Tier-1 chassis supplier whose fate is a leveraged derivative of one publicly-distressed anchor customer, and I put the mass at-to-below the class median. The forensic read did not rescue it. The single figure that decides everything moved the wrong way: Nissan-group share is 63.8% and essentially flat versus ~64.0% the prior year F329D5F328 — the apparent drop in the formal note from three Nissan entities to two is a threshold artifact (Nissan Japan slipping below 10% F338F339), not diversification. The twice-impaired book is not the floor I need it to be (C39): a fire-sale haircut on specialized single-customer press lines and plant lands near ¥156/share (0.18× the stamp); the reported ¥54,154M self-capital F207 is reachable only if the plants sell near carrying value, which for assets management wrote down ¥11,800M then ¥9,145M F327F324 they will not. And the ¥64bn book is itself FX-inflated — ¥8,699M of translation adjustment plus ¥4,257M of securities marks F195F194 reverse if the yen strengthens, plausibly in the same risk-off a Nissan restructuring would bring. Worse, the floor is unreachable: founder family in control E63, an anti-takeover plan live to the 2027 AGM at a 20% trigger E64, and an allegiant register (Shido Holdings, JFE, Mizuho, Yokohama Bank, Suzuki, Nissan Trading, plus 17.8% treasury) that no outside proposal can outvote E50D6. The margin-of-safety arithmetic passes; the enforcement half fails.

The recovery is cyclical and the company says so: FY2026 operating margin 2.26% sits at the top of the five-year ordinary-margin band, and management guides FY2027 down 47% on net income D2F407. Downside-first (C33): in a real Nissan-driven down-year, after interest on ¥43.3bn of debt at a rising 4.78% short rate, after a foreign tax bill that gets paid even in a group-loss year (¥2,963M tax on −¥11,343M pretax in FY2025 F245F241), owner earnings go to roughly breakeven or negative — the two loss years are the precedent. So the downside rests on the asset floor, which I just showed is soft and unreachable. Why too-hard and not pass? Pass would assert the short wins at any price; the implied buy-below arithmetic shows a price (~¥520) at which the downside owner-yield could clear a hurdle if the Nissan path were knowable. It is not. The load-bearing unknown (C98) — Nissan's multi-year build volumes, platform awards, and restructuring severity — swings owner earnings from ~¥2bn to negative, lives entirely in Nissan's filings, and is disclosed nowhere in Yorozu's corpus. I publish the implied buy-below (¥520) per C44 so the record is falsifiable, but I do not issue a buy-below verdict.

What a student should take from this: a book discount is only a margin of safety when the assets are both real and reachable — twice-impaired single-customer plant, marked up by translation FX that can reverse, under founder control and a poison pill, is neither. "Too-hard" is a real answer, not a dodge: name the unknowable variable and decline; cheapness cannot buy your way out of an unknowable that large. Verdict: too-hard (implied buy below ¥520).

Synthesis

Where the five lenses agree

The facts are not in dispute, and all five lenses read them the same way. Yorozu trades at a deep discount to book — 0.33× owners' equity, ¥858 against ¥2,629.49 BPS F45 — and cash & deposits alone (¥29,767M F126) exceed the entire ¥17.67bn market cap. But the business behind that book is ~64% dependent on the Nissan group D5E11, a struggling automaker; the book has been twice-impaired — ¥11,800M then ¥9,145M, ¥20.9bn in two years F327F324 — and is only moderately levered but is not a fortress: net debt ¥9,338M, equity ratio 39.2% D1F35. The FY2026 recovery to +¥2,075M F15 is real but the company itself guides FY2027 down 44–47% F406F407. Every lens converged on the same read of the decisive variable: Nissan's multi-year trajectory is unknowable from Yorozu's own filings, and it is the variable that governs the outcome. Four lenses go too-hard; Pabrai passes. There is no bull among the five.

Where the lenses diverge

The divergence is unusually thin. It is not too-hard-versus-buy; it is too-hard versus pass — a distinction about whether you cannot judge it or can judge it and decline.

Buffett puts it as an estimation failure: "I cannot make a defensible ten-year earnings estimate for Yorozu without first making one for Nissan, and I cannot make one for Nissan" (B2). Munger arrives from inversion — the kill path is "already moving" — and lands on the same unjudgeable variable: "the decisive variable (M90) is Nissan's ten-year volume trajectory, and I cannot judge it from this ledger." Li Lu gates on the knowledge bar itself: "could I claim to predict this company's earnings power ten years out better than most of the people who own it? I cannot" (L1). Claude names the variable explicitly and prices the escape hatch shut: the unknown (C98) "swings owner earnings from ~¥2bn to negative … and is disclosed nowhere in Yorozu's corpus."

Pabrai does not disagree with any of that — he simply reaches a decision rather than a suspension. For him the 63.8% concentration D5 is not "unknowable," it is disqualifying: "at 63.8% Nissan-group dependence, this business has no independent earning power — its results are Nissan's production decisions" (P12), and the stressed floor fails (P1) and the guide-down kills convergence (P61). He can judge it well enough to say no. The four too-hard lenses decline to even score it. On the ledger both routes end at "no action" — but the epistemic posture differs, and it is the whole of the disagreement.

The red team's challenge (and our answer)

Because four of five lenses agreed, the consensus was put to a dedicated adversary (red-team.md), and its case is strong enough to state plainly and by name.

The red team's four strongest points. (1) You are not paying for the business; you are paying below the cash line. "Cash & deposits alone are ¥29,767M F126 — 1.68× the entire market cap. The market is capitalizing the equity below the gross cash line." Add listed securities and it is 2.15× market cap; net debt is only ¥9,338M D1 on a company throwing off ¥8,034M of operating cash flow a year F61. "You are buying ¥1 of clean, mostly-tangible book for 33 sen," against a near-goodwill-free book (intangibles ¥209M F152). (2) The impairments are behind you. "The two loss years were impairment years, not operating-collapse years" — operating income was positive every year (FY2024 ¥4,459M, FY2025 ¥298M, FY2026 ¥3,980M F122F123F217), the losses were non-cash, and impairment collapsed ¥11,800M → ¥9,145M → ¥105M F327F324F326. "The bad news the consensus is afraid of has already been charged through the P&L. You are not waiting for the write-down; you are buying after it." (3) Suspension/chassis is drivetrain-agnostic — "indispensable even in EVs," lightweighting a first-order tailwind, already spec'd into Toyota bZ4X and Nissan Leaf E5E34E82 — so the EV transition is a tailwind here, not the powertrain-supplier's death sentence. (4) An accretive buyback retiring "¥2,629 of book for ¥858," 14.8% authorized E47, funded by a ring-fenced cross-holding-liquidation program E59. The red team's verdict: WATCH, buy-below ¥660, on the argument that "'too-hard' confuses uncertainty in the business with uncertainty in the investment" — the downside is bounded by an asset floor the price already sits below, so the unknowable Nissan variable "only ranges you across degrees of win."

Our answer — conceding what is right. The red team's central fact is real and is the strongest bull point anywhere on this record: cash & deposits genuinely exceed the market cap F126, the book genuinely is near-goodwill-free, the operating losses genuinely were non-cash impairments, and the buyback genuinely is accretive at a third of book. We do not wave these away. Where the consensus holds — and where the red team's own honest self-audit concedes the ground — is on realizability and the softness of the floor, and here the too-hard lenses engage the red team's points directly:

• The cash sits inside Nissan-dependent operating plants, not in a distributable box. The red team itself concedes "~46% of PP&E and a big share of segment assets sit in Nissan-Americas plants that are single-customer-configured" — US PP&E ¥6,033M, Mexico ¥8,080M F353F354 — and that "the cash sits partly in overseas/JV entities (China JVs are 51%-owned E8; NCI is ¥9,406M F201) — it is not all freely distributable to the parent." This is Claude's C39/C47 and Pabrai's P1: a Nissan volume collapse drains the working-capital and cash-generation the floor depends on, and the plants that back the book impair toward scrap, not carrying value — which the FY2024–25 write-downs F327F324 prove management does the moment demand drops E88.
• The book is FX-inflated. ¥8,699M of translation adjustment plus ¥4,257M of securities marks sit inside the ¥64bn net assets F195F194F198 and reverse if the yen strengthens — plausibly in the same risk-off a Nissan restructuring brings. This is Claude's surprise-ledger point: a second, correlated way the floor fails exactly when the anchor's stress arrives.
• The catalyst is management's to grant. The floor is control-locked — founder Shido family plus an anti-takeover plan at a 20% trigger to the 2027 AGM plus an allegiant register E63E64E50 — so, as the red team itself admits under vulnerability #4, "the classic deep-value catalyst … is structurally blocked. The value can stay trapped at 0.33× indefinitely." Its rebuttal is that management is already self-liquidating the discount via the buyback — "but that is management's pace, not mine, and they can stop."
• This is a cigar-butt, not a compounder. The red team concedes ROE was −22.8% then 3.9% F39F40 and "the return is the re-rating of the discount, and if the discount never closes, you clip a 3.85% dividend on a low-ROE cyclical and wait."

Where it is genuinely close: the red team is right that "too-hard" carries a real risk of dressing a psychological claim (Nissan is scary) as an epistemic one (Nissan is unknowable). The honest reconciliation is that the four too-hard lenses do not rest on Nissan being scary — they rest on the asset floor being soft and unreachable enough that it does not neutralize the unknowable, which is precisely the test the red team says separates too-hard from buy-below. On that test the consensus and the adversary actually agree about the facts (cash > market cap; floor is plant-soft and control-locked) and split only on the weight: the red team judges the cash-above-market-cap line sufficient to demand a wider margin (¥660) rather than decline; the consensus judges the same softness sufficient to withhold a price entirely. Reasonable, and close — but the record ships both thresholds (below) so a future review can score which reading was right.

Self-distance note

The Claude lens holds one of the five verdicts compared above and also wrote this synthesis; the reconciled figure table all five lenses consumed, and the evidence ledger, were built by a (Claude-driven) dual-blind extraction (two independent passes per ledger, reconciled against page-delimited source text). The red team that challenged the consensus is likewise Claude-authored. Read the synthesis — and the by-name engagement above — with that concentration of authorship in mind: the adversary and the answerer share a model.

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 to this study.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp.
• pass / watch / too-hard are recorded but unscored in any future review.
• 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.
• Two priced thresholds are recorded for this study but were not issued as lens buy-below verdicts: the red team's falsifiable ¥660 watch-anchor and Claude's implied ¥520. Neither is a scored verdict; both are recorded as the priced thresholds a future review reads against (see "What would change our minds").

Red team

A consensus red-team ran because four of five lenses agreed on too-hard. The adversary's strongest points, verbatim-faithful and cited, with the synthesis's engagement:

• "Cash & deposits alone are ¥29,767M F126 — 1.68× the entire market cap. The market is capitalizing the equity below the gross cash line." Cash + listed securities ¥8,314M F154 = ¥38,081M ≈ 2.15× market cap; net debt only ¥9,338M D1 on ¥8,034M annual operating cash flow F61; "you are buying ¥1 of clean, mostly-tangible book for 33 sen," near-goodwill-free (intangibles ¥209M F152).
  • Engagement (conceded, then bounded): the fact is real and is the strongest bull point on the record. But the red team concedes the cash "sits partly in overseas/JV entities (China JVs are 51%-owned E8; NCI is ¥9,406M F201) — it is not all freely distributable to the parent," and ~46% of PP&E is Nissan-Americas plant F353F354. A Nissan volume collapse drains the very cash-generation the floor leans on (Pabrai P1, Claude C47).
• "The two loss years were impairment years, not operating-collapse years … The bad news the consensus is afraid of has already been charged through the P&L." Operating income positive every year F122F123F217; impairment ¥11,800M → ¥9,145M → ¥105M F327F324F326; the FY2025 net loss was ~¥9.1bn impairment plus a ¥2,234M FX loss, "non-cash and non-recurring."
  • Engagement (conceded, then bounded): true, and it defeats the naïve "another write-down is coming" reflex. But the impairments prove management marks single-customer plant down hard the instant demand drops E88 — so the same mechanism is the reason the floor is soft, not evidence it is safe. A third cycle is the bear case, not an impossibility (Munger M2/M42, Claude C39).
• "Suspension/chassis is drivetrain-agnostic — the EV transition is a tailwind here, not the powertrain-supplier's death sentence." Products "indispensable even in EVs," lightweighting "the center of business strategy," already on Toyota bZ4X and Nissan Leaf E5E34E82.
  • Engagement (accepted): the consensus agrees — every lens explicitly set the EV-displacement risk aside. This point is correct and is not why any lens declined; the decline is customer-trajectory, not terminal-value.
• "Buying back stock at ~0.33× book is not capital return, it is value transfer to continuing holders: every ¥858 spent retires ¥2,629 of book." 14.8% authorized E47, funded by a ring-fenced cross-holding-liquidation program E59; weighted share count already fell 24,388k → 22,861k F355F356.
  • Engagement (conceded, then bounded): genuinely accretive and insider-driven. But it is management's pace, not the minority's — the red team's own vulnerability #4 concedes the re-rating catalyst is "structurally blocked" by founder control + poison pill E63E64 and "they can stop" (Claude C57/C60). A per-share tailwind, not a discount-closing catalyst.

The red team's own verdict — "WATCH, with a buy-below of ¥660 … emphatically not 'too-hard'" — rests on the claim that "an asset floor the price already sits below is the definition of a knowable downside," and that too-hard "confuses uncertainty in the business with uncertainty in the investment." The consensus holds because, on the red team's own test (is the floor cheap-and-reachable enough that the anchor's fate stops mattering?), the floor is plant-soft, FX-inflated, and control-locked — so the unknowable is not neutralized. The disagreement is one of weight, not fact, and it is close; the record ships both priced thresholds so it stays falsifiable.

What would change our minds

No lens issued a buy-below or watch with a falsifier: line — the four too-hard verdicts carry no price falsifier by construction, and Pabrai's pass carries none. So there is nothing of that kind to pre-register from the lenses, and we say so plainly rather than manufacture one.

What the record does carry, as the priced thresholds a future review scores against:

• The red team's falsifiable buy-below: ¥660 (≈ 0.25× owners' equity F207, ~0.42× its stress-case recoverable estimate) — the adversary's genuine watch price, "a ~25% discount to even the punitively-haircut floor," paying for founder entrenchment and plant softness. A future review can score whether the stock reached it and what followed.
• Claude's implied buy-below: ¥520 (downside/normalized owner earnings ~¥1,100M — management's own guided trough F407 — at ~11% required yield ÷ 20,595,177 shares) — published for falsifiability, not issued as a verdict.

And the observable that would move the consensus off too-hard: the blocking unknown is Nissan's multi-year build-volume / platform-award / restructuring trajectory (Claude C98), which drives 63.8% of revenue D5 and is resolvable only by Nissan's own future disclosures and OEM sourcing decisions — not by any observation of Yorozu at a threshold. Concretely, the consensus would reconsider if Nissan's forward build-volume / platform-award visibility for Yorozu-sourced models improved to where a defensible ten-year estimate became possible, or if the discount closed via a realized catalyst (treasury cancellation reducing issued shares, a genuine re-rating toward ≥0.6× book, or a change in the control structure that made the floor reachable). Claude's pre-registered probabilities (in profiles/claude.md) put Nissan-group share staying ≥55% at 0.88 and the Nissan-trajectory-unknowable proposition at 0.85 through FY2027/3 — i.e. the consensus expects the blocker to persist, and has said so on the record.

What this taught the checklists

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

• Buffett — a single-customer / keiretsu-captive-supplier gate as a first-class item. The AmEx test (B38) rewards a diffuse franchise surviving a shock; its mirror — a supplier whose franchise is a single customer — is currently caught only at B2/B23 by inference. Proposal: a dedicated item asking whether any one customer exceeds roughly a quarter-to-two-fifths of sales and whether the position is that of a price-taking captive, making the most common Japanese auto-supplier failure mode a gate rather than an emergent finding. (Flag only; no grade change proposed.)
• Munger — a related guidance note: M40 (raisins/turds) and M42 (retained-earnings productivity) presume the bad segment is discardable and the impairment episodic; for a captive tier-1 supplier the value-destroying segment IS the largest (Americas) and the impairment is customer-demand-driven, not management folly. A one-line note that "concentration can make the turd un-excisable and the impairment exogenous" would sharpen the weighing. (No structural change.)
• Li Lu — L11's inverse. The "feared exposure quantified" item is framed entirely around the Timberland upside case (a feared exposure that turns out tiny). Its inverse — a feared exposure that turns out to be the company (here 63.8%) — is the more common Asian-supplier situation and is currently caught only implicitly. Proposal: an explicit clause that when the quantified feared exposure exceeds ~25–30% of revenue, the item flags a hard predictability constraint feeding L1.
• Claude — pre-fetch governance / discount-closure archive sources at the source gate. Items C61/C65/C67/C68/C71/C72 repeatedly hit data-insufficient because the corpus is the filing set, not the TSE capital-cost-response document, AGM voting results, or trading data — so for a company whose thesis is governance/discount-closure, the governance read is systematically thinner than the items demand. (Also: C98 should carry an explicit "unknowable-anchor-customer" sub-prompt; C39 should require both a fire-sale and a going-concern haircut, reported as a range — on Yorozu the two diverged 6× on the plant haircut alone.)
• Process observation (source-gate / commit-message fix). The stage-3 anchor commit subject leaked the stamp price to the figures-blind Claude §1 (outside-view) stage. The Claude lens excluded it and documented the exclusion (its buy-below chain is stamp-free, confirmed at C89), so no contamination entered the verdict — but the leak is a real mechanism. Flag for a source-gate / commit-message fix so the stamp price cannot reach a figures-blind stage via commit metadata.

---

Corrections