Sankyo Tateyama (5932): The rational discount

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	¥480	B42 (negative owner earnings); B37 (turnaround dependency); B65 (book eroding ~¥6-8/share per quarter)
Munger	pass	—	M1/M86 (four co-occurring negative forces — negative lollapalooza); M46 (incentive structure produces enterprise preservation, not ROIC); M88 (no margin of safety even on the most optimistic scenario)
Pabrai	pass	—	P1 (stressed-asset floor negative ~−¥48bn); P20 (leverage 3.4× market cap, interest coverage <1×); P7 (melting ice cube — OCF ¥3.2bn vs. capex ¥13.3bn)
Li Lu	pass	—	L1 (ten-year knowledge bar fails — German economics, domestic demand trajectory unknowable); L36 (five-year ROE averaging below 1%; two loss years; no compounding mechanism); L22 (incremental ROIC negative — European capex destroyed capital)
Claude	pass	¥220 (implied)	C33 (bear-case equity value near/below zero at current leverage); C34 (reverse-DCF requires implausibly fast restructuring); C84 (short case strong: negative haircut NAV, 100% plan-miss rate, dividend funded by debt)

Synthesis verdict: 4 pass / 1 watch. Study is VOID (headless — predictions were voided; no prediction-vs-actual scoring at review).

---

The business

Sankyo Tateyama makes and sells aluminium products. It is not a mining company or a chemicals company — it buys aluminium ingot, adds heat and dies, and sells shaped metal into four markets.

Building Materials (¥178,652M revenue F140, 50% of group) makes the aluminium sash, window frames, curtain walls, and exterior cladding that go into Japanese homes and commercial buildings. This is the historic core of the company, which was assembled in 2012 from two rival Toyama-area aluminium fabricators — Sankyo Material and Tateyama Advance — and then absorbed the parent holding company the same year E6.

International (¥76,145M F143, 21%) is the growth bet: ST Extruded Products Germany (STEP-G) and related European entities extrude aluminium profiles for European automotive manufacturers, primarily for electric-vehicle applications; Thailand and China round out the segment.

Materials (¥59,781M F141, 17%) does domestic aluminium extrusion and casting, partly for automotive customers in Japan.

Commercial Facilities (¥44,522M F142, 12%) makes store display fixtures and handles store-fitout maintenance for Japanese retailers.

The customers are Japanese construction companies (building materials), German and Thai automotive supply chains (international), Japanese industrial manufacturers (materials), and Japanese retail chains (commercial facilities). None of these are captive; none face meaningful switching costs above a project level. The company is, at bottom, a toll processor of aluminium with thin margins and no identifiable pricing power above its commodity inputs.

---

The numbers

Revenue reached ¥370,385M in FY2023/5 F3 then fell back to ¥353,027M in FY2024/5 F4 and recovered to ¥359,424M in FY2025/5 F5. Over five years, top-line growth has been modest and partly ingot-price-linked rather than organic volume growth.

Profitability is a different story. Ordinary income fell from ¥5,251M in FY2021/5 F6 every year to ¥944M in FY2025/5 F10. Operating income in FY2025/5 was ¥1,545M F70 on ¥359,424M of revenue — an operating margin of 0.43% D5. Gross profit was ¥69,764M F72; SG&A was ¥68,219M F73. The overhead consumes virtually all of the gross margin, leaving a sliver before extraordinary items.

Net income has been negative for two consecutive years: −¥1,019M in FY2024/5 F14 and −¥2,336M in FY2025/5 F15. The FY2025/5 result was especially painful: a pre-tax loss of only −¥477M F81 became a net loss of −¥2,336M because ¥1,661M of income tax expense was booked on a pre-tax loss — a consequence of the ¥18,951M valuation allowance against gross deferred tax assets of ¥23,791M [F132, F133]. Management and the auditors do not believe the deferred-tax assets will be recovered from future taxable income F134.

ROE has declined from 2.2% in FY2021/5 F41 to −2.5% in FY2025/5 F45. The five-year pre-loss average net income was ¥1,236M D4 — a number that never cleared any plausible cost of equity on ¥94,804M of net assets F103.

The segment cross-subsidy map tells the story directly:

Segment	Revenue FY2025/5	Segment profit	Margin
Building Materials	¥178,652M F140	¥236M F144	0.13%
Materials	¥59,781M F141	¥2,602M F145	4.35%
Commercial Facilities	¥44,522M F142	¥1,460M F146	3.28%
International	¥76,145M F143	−¥2,598M F147	−3.41%

The International segment lost ¥2,598M in FY2025/5 F147, worsening from −¥1,306M in FY2024/5 F155. Impairment charges hit the German assets in both years: ¥939M in FY2024/5 F131 and ¥1,010M in FY2025/5 F130. In H1 FY2026/5, ¥1,184M of special severance was booked as a restructuring charge for the German subsidiaries F201. The original three-year plan called for automotive share of extrusion sales to reach 40% by FY2026 E60; that thesis has been formally revised downward E15.

The balance sheet carries the weight of the investment phase. Total borrowings are ¥87,220M F210 against cash of ¥21,509M F90: net debt ¥65,711M D3, which is 3.4× the market cap at stamp D1. The equity ratio is 30.4% F40. In FY2025/5, ¥35,200M of new long-term borrowings F122 funded ¥13,277M of capex F121 and refinanced ¥19,770M of maturities F123. Operating cash flow was ¥3,216M F50 against ¥14,334M of investing outflows F52: free cash flow was approximately −¥9bn. Interest expense was ¥1,610M F75 — essentially equal to operating income.

The company paid a ¥25/share dividend F175 through both loss years, costing ¥701M in FY2025/5 F124, funded from retained earnings (¥22,706M F112) that are now shrinking. No buybacks E38.

At the stamp date, the stock traded at ¥624 [stamp.md], a price/book of 0.214× D2 on BPS of ¥2,911 F30.

---

The five lenses

Buffett: watch at ¥480

Buffett begins where he always begins: can I understand the business? Yes — aluminum sash and window frames are within the circle; the business model needs one paragraph [E1, E2, E3, E4]. From there the shopkeeper's eye does the rest, and it finds almost nothing to like.

Owner earnings — the number Buffett actually wants — are deeply negative. Net income plus depreciation (¥8,271M F120) minus capex (¥13,277M F121) equals approximately −¥3.5bn in the most recent year. Even using the three-year pre-loss average net income (¥1,236M D4), the number is still negative. The franchise tests fail cleanly: no pricing power above commodity ingot pass-through [E21, E27], ROE never exceeding 2.2% F41, no nameable moat [B25]. The mid-term plan was revised once — operating income target cut from ¥11bn to ¥7bn [F221, F223], ROE target from 6% to 3% [F222, F224] — and year-2 is missing the revised targets too [F185, F191].

So why watch rather than pass? One reason: at 0.214× book D2, with net assets of ¥94,804M F103 against a market cap of ¥19,556M D1, an informed private buyer might pay for the domestic businesses — Materials (¥2,602M profit F145) and Commercial Facilities (¥1,460M F146) — plus ¥15,447M of investment securities F99, less net debt of ¥65,711M D3. The range is tight: perhaps ¥0–¥24bn of equity value. The restructuring is real — the Bonn plant transfer was contracted E32, ¥1,184M of German severance was booked in H1 [F201, E56], and 150 domestic employees are being voluntarily retired E57. These are dated, concrete actions.

At ¥480 — approximately two-thirds of a rough haircut-basis book value D2 — the domestic earnings recovery produces mid-single-digit returns even without multiple expansion, and the margin of safety is wide enough to cover the restructuring risk. At ¥624, it is not [B93].

The falsifier is the International segment's restructuring trajectory: if STEP-G posts a second consecutive full-year operating loss wider than ¥1.5bn in FY2027/5, the restructuring thesis is broken and the position exits regardless of P/B. The watch verdict is a conditional option on the restructuring, not a franchise quality call [B37, B42, B65].

What a student should take from this: The difference between a Graham net-net and a below-book leveraged manufacturer is the difference between current assets covering all liabilities and a manufacturing plant whose "cheapness" requires the business to recover earnings. When the asset case is the only viable case — because owner earnings are negative — the investor must ask: adequate financial strength to survive the wait, named mechanism for book to close, and enough margin of safety to absorb a slow recovery. Here all three conditions are borderline.

---

Munger: pass

Munger inverts first. The most plausible obituary for Sankyo Tateyama writes itself in three sentences. It is an aluminium building-materials company in a country whose housing starts have been structurally declining for decades E12. Its European growth bet has burned money for two straight years while the plan was already cut once [E15, F221]. And it is funding this losing battle with borrowed money [F210, F90].

That is the death path. Not exotic. Already in motion.

The moat question takes minutes. Operating margin peaked at 1.4% in FY2024/5 [F71, F4] and collapsed to 0.43% in FY2025/5 D5. Gross margin is ~19.4% [F72, F5] — decent for a heavy manufacturer — but SG&A of ¥68,219M against gross profit of ¥69,764M [F72, F73] leaves essentially nothing. Which of scale economies, network effects, switching costs, habit, or brand is operating here? None. Customers buy from Sankyo Tateyama because it is large, proximate, and established — not because walking away costs them anything beyond switching suppliers [M18].

The incentive structure clinches it. Director compensation is entirely fixed cash. No performance-linked pay, no stock compensation — the governance report acknowledges the gap and calls it "under consideration" [E43, E45]. The president holds 35,400 shares worth roughly ¥22M at stamp E40 — about one year's pay. The three in-house employee stock-ownership associations collectively own ~13.6% E64, which is real skin in the game, but it is diffuse, not concentrated owner-operator alignment. Show me the incentive and I'll show you the outcome: the incentive is to keep the machine running, revenues stable, relationships intact, the enterprise intact — not to earn an adequate return on capital [M46].

The P/B discount is telling the truth, not offering an opportunity. A business earning −2.5% ROE F45 on its book is worth less than book. The price is not cheap relative to earning power. At ¥624, the most optimistic near-term scenario (¥2bn net income at 10×) equals approximately the current ¥19.6bn market cap — no margin of safety [M88].

What a student should take from this: Cheapness on P/B is only the beginning of the question, not the answer. When a business earns less than its cost of capital persistently, the correct price is below book — the discount is the market's impound of permanent value destruction, not an error. A company can be statistically cheap and fundamentally expensive because the capital trapped inside it earns inadequate returns.

---

Pabrai: pass

Pabrai starts with the most important question: what happens if I'm wrong?

The stressed liquidation floor is the answer. Cash ¥21,509M, receivables (¥49,262M + ¥8,342M) at 70% recovery = ~¥40bn, inventory ¥57,077M at 50% = ~¥28.5bn, investment securities ¥15,447M at 80% = ~¥12bn, tangible fixed assets ¥113,760M at 50% for manufacturing plant in a declining market = ~¥57bn. Total stressed assets: roughly ¥158bn. Against total liabilities ¥205,649M F101. Stressed net: approximately −¥48bn. The floor is negative. This is not a place where "cheap" means anything useful [P1].

The debt magnifies every risk. Net borrowings ¥65,711M D3 against operating cash flow ¥3,216M F50; interest expense ¥1,610M F75 against operating income ¥1,545M F70 — interest coverage below 1×. Short-term debt (¥7,365M + ¥20,156M = ¥27,521M due within twelve months [F107, F108]) exceeds operating cash flow by nearly 9×. The ¥20bn committed credit line F211 is real buffer, but the company borrowed ¥35.2bn in FY2025/5 F122 while generating only ¥3.2bn of operating cash flow F50 [P20].

The German subsidiary is the melting ice cube P7 was designed to catch. The EV thesis — automotive share of extrusion rising to 40% of International segment by FY2026 E60 — has been abandoned. The segment lost ¥2,598M in FY2025/5 F147, up from ¥1,306M the year before F155. Impairments in two consecutive years [F130, F131]. Capex ¥13,277M against OCF ¥3,216M [F121, F50] — assets are being consumed faster than the restructuring can monetize them.

There is no fifty-cent dollar here. If the mid-term recovery to ¥7bn operating income by FY2027/5 F221 succeeds (the plan has already been missed in year 2), at a 6× multiple = ¥42bn, minus net debt ¥65.7bn D3 = negative equity value. Even at double the plan's recovered operating income, the math barely reaches current market cap — no discount [P53].

What a student should take from this: P/B below 0.25× is a number, not an argument. The downside case must be built from liquid-and-hard-asset stress arithmetic, not from the book value line. When the stressed liquidation value of an operating industrial company with ¥87bn of borrowings F210 comes up negative, the stock price is not cheap — it is pricing a real outcome.

---

Li Lu: pass

Li Lu reaches for the knowledge bar first, and the bar is not cleared.

To underwrite a ten-year holding in Sankyo Tateyama, three questions must be answerable. First: does the domestic building-materials business — which faces structural headwinds from declining Japanese housing starts [E12, E20] — have pricing power sufficient to sustain margins as volumes fall? The FY2025/5 segment result (¥236M profit on ¥178,652M revenue [F144, F140]) is a 0.13% margin, down 89.4% E27, during a period that included a pre-regulatory-change demand rush. The answer is not available in the ledger with confidence.

Second: what is the German extrusion subsidiary's competitive position among European automotive extruders after restructuring? STEP-G exceeds 10% of consolidated revenue E8 and operates under a P&L transfer agreement that sets its reported net income to zero E8. The unit-level economics — margin per extrusion-line, competitive position against Constellium and AMAG, defensible volume at the restructured cost base — are not reconstructible from these disclosures.

Third: will the ¥9.3bn Shinminato-Higashi plant F214 earn a return above the cost of the debt that funded it? The original EV-automotive growth thesis that justified the capex has already been revised E15. The plant will add ~1,000t/month of domestic extrusion capacity E34, but into which markets at what margins remains unquantifiable.

None of these unknowns is resolved in the ledger. Each is load-bearing. The verdict does not depend on them resolving favorably — the balance sheet and earnings record are sufficient for a pass on their own — but they are noted as the specific gap between "passing on current facts" and "ever becoming comfortable."

The five-year ROE series — 2.2%, 0.5%, 1.9%, −1.1%, −2.5% [F41–F45] — is not a business whose value is compounding. A ten-year owner needs the business to earn before it can compound. The incremental ROIC is negative: the company deployed substantial capital through its investment phase while operating income fell from ¥3,807M F71 to ¥1,545M F70 [L22]. The retained yen have not created per-share value; BPS declined from ¥3,067 F29 to ¥2,911 F30 in the most recent year alone.

What a student should take from this: The cheapest stock by price-to-book in a large universe is almost never a gift — it is usually the market's answer to a question the analyst has not yet asked about the quality of the underlying business. When ROE has never exceeded 2.2% F41 in the best years of a five-year record, book value does not provide the floor it provides in a business that earns on its assets.

---

Claude: pass (implied buy-below ¥220)

Claude runs the outside-view priors first, per the lens's ordered pipeline. Before opening the figures, the reference class was registered: below-book, levered Japanese building-materials manufacturer with structural domestic-market decline and concurrent international restructuring. Median-class outcome: slow stabilization, P/B 0.2–0.5× for 5+ years, IRR <4% pa. Prior probability of value trap: p = 0.50. Reference anchor: P/B at stamp 0.214× D2.

Opening the figures confirmed the prior and then some.

The bear case arithmetic drives the verdict. Haircut NAV: receivables ¥57.6bn [F91, F92] × 80% = ¥46bn; inventory ¥57.1bn [F93, F94, F95] × 50% = ¥28.6bn; securities ¥15.4bn F99 × 70% = ¥10.8bn; tangible assets ¥113.8bn F97 × 50% for manufacturing plant = ¥56.9bn; total gross ~¥163bn. Against total liabilities ¥205,649M F101: haircut NAV approximately −¥43bn. Negative. The company is not asset-backed at liquidation values; liabilities exceed the haircut asset base.

The reverse-DCF confirms the problem from the other direction. At 8% required return, the current price embeds a perpetual FCF of ¥19,556M × 8% = ¥1,565M per year. Current FCF: OCF ¥3,216M F50 minus capex ¥12,167M F212 = −¥8,951M. To reach ¥1,565M FCF requires a +¥10.5bn improvement — approximately 3× current operating income, while simultaneously reducing capex to maintenance levels. The price embeds an implausibly fast and complete restructuring for a management team that has missed its own profit plans twice.

The implied buy-below of ¥220 is derived from the partial-recovery scenario: if International losses fully cease, Building Materials recovers to FY2024/5 levels, and net debt halves over 3-5 years through asset sales, normalized owner earnings reach ~¥4bn, capitalized at 8% = ¥50bn, minus ~¥33bn reduced net debt = ~¥17bn equity, or ~¥540/share. The bear case (owner earnings negative) implies zero or below. ¥220 is the price at which a partial restructuring — not the full bull case — yields roughly 8% on equity, requiring net debt to shrink materially. It is an approximate figure published regardless of the pass verdict.

On the Japan-specific governance picture: no activist E51, three in-house employee stock associations at ~13.6% E64, compensation entirely fixed cash with no equity grants [E43, E45], and a TSE capital-cost disclosure that targets only ROE 3% by FY2027/5 E52 — already revised down from the earlier 6% target F224. The ¥18,951M DTA valuation allowance F133 means reported book equity of ¥94,804M F103 includes ~¥19bn that management and auditors believe cannot be recovered. Adjusted equity is closer to ¥75-76bn; the "0.214×" reported P/B is actually closer to 0.27× on adjusted book.

What a student should take from this: Two levers combine to produce a value trap — illusory book (the ¥19bn DTA valuation allowance F133 is printed on the balance sheet but the auditor says most of it cannot be used) and absent earnings power (ROE never above 2.2% in five years, now negative two years running). A 0.21× P/B number is seductive; tracing it through to adjusted equity and earnings power shows the discount is rational, not a mispricing.

---

Synthesis: the staged exchange

The table above is the summary. The disagreement — one watch versus four passes, ¥480 D2 versus ¥220 versus no price — is where the learning lives. What follows is the exchange as it would occur among the masters.

---

Buffett opens: Charlie, Mohnish, I'm not willing to call this a pass so quickly. At 0.214× book D2, you have ¥94bn of net assets F103 against ¥19.6bn of market cap D1. The domestic businesses — Materials at ¥2,602M F145, Commercial Facilities at ¥1,460M F146 — are earning. The German restructuring has concrete, irreversible actions: the Bonn plant transfer contracted E32, ¥1,184M of severance booked in H1 F201. This is not a promise; it is a dated event. At ¥480, two-thirds of a haircut-basis floor, the asset conversion produces mid-single-digit returns even with a modest exit. The watchlist exists precisely for companies where the quality bar is not cleared but the price may compensate.

Munger: Warren, you are doing the arithmetic and stopping there. Look at the mechanism. You need the share price to re-rate to make this work. Who does the re-rating? There is no activist E51. The three employee stock associations — 13.6% of the register E64 — are not going to pressure management to sell the German business or buy back shares. The president holds ¥22M of stock E40 and earns a fixed salary regardless of outcomes E43. The cross-holding network with Sumitomo Realty E49 and Sumitomo Chemical E36 locks capital into relationship maintenance. You have identified the asset value correctly. You have not identified a closing mechanism. The ¥25 dividend floor E37 costs ¥700M per year from debt F124, not from earnings. The incentive structure reliably produces enterprise preservation, not ROIC maximization. The discount persists because the mechanism to close it does not exist. I invert: what would this company look like in ten years with current management and current incentives? Probably very similar to today, only with somewhat less book equity as losses slowly erode retained earnings.

Pabrai: Warren, there is one number that cuts through everything. Interest expense: ¥1,610M F75. Operating income: ¥1,545M F70. This business does not cover its own interest. Every quarter it runs at current margins, it is not a value being maintained — it is a value being transferred from equity holders to debt holders. The stressed asset floor is −¥48bn. That is not a conservative haircut; that is what happens when you price manufacturing plant at 50 cents on the dollar, which is where distress auctions go. The checklist says: if P1 fails — if the downside floor does not protect capital — pass regardless of how cheap it looks. P1 fails here. Clearly. There is no price in the current range where I call this a buy, because the leverage converts a temporary decline into a permanent one if the operating improvement does not arrive on schedule. And the plan has missed twice [E14, F185]. ¥480 does not fix a negative liquidation floor.

Buffett: Mohnish, the liquidation case is not the investment case. The investment case is the domestic businesses at trough margins, the German bleeding stopping, and the asset base maintained through the restructuring. I'm not buying for liquidation. I'm buying for the scenario where Germany is right-sized by FY2028 and Materials plus Commercial Facilities sustain ¥4bn of annual segment profit. At ¥480, even a mediocre exit at 30% of book preserves capital. The ¥20bn committed credit line F211 and ¥22.7bn of retained earnings F112 keep the company alive through the restructuring. I grant that B65 is the clock — book is eroding ~¥6-8 per share per quarter at current loss rates — which is exactly why ¥480 is the number and ¥624 is not.

Li Lu: Warren, I appreciate the structure of the argument. But let me name what none of us can know. We cannot know what the German extrusion business is actually worth after restructuring. It exceeds 10% of consolidated revenue E8 and operates under a profit-transfer agreement that zeroes out its reported net income E8. The unit economics — margin per extrusion-line, competitive position against Constellium and Hydro in European automotive, defensible volume at the right-sized cost base — are not in the ledger. We are pricing an option on an outcome we cannot evaluate. The ten-year knowledge bar exists precisely to stop us from buying businesses whose core economics we cannot reconstruct from public disclosures. The building-materials domestic thesis has the same problem: whether a 0.13% margin business [F144/F140] has pricing power as housing starts continue their secular decline E12 is not answerable from what is filed. I cannot establish a buy-below at any price for a business where the load-bearing variables are epistemically unavailable to me. My watch or buy requires demonstrated evidence at the operating level, not planned restructuring. Two consecutive full-half-years of improving International results, building materials recovering to 1.5% margin — then re-examine.

Claude: I want to add what none of the four of you have named explicitly: what cannot be known in principle, not merely in the current filing. The DTA valuation allowance is ¥18,951M F133 — an accounting judgment by management and auditors about future taxable income. That judgment is itself an estimate of the company's own earnings potential over the next several years. The auditor is effectively saying: we cannot sign off on the existence of ¥19bn of future earnings capacity. I wrote the figure table and I hold this lens, and I flag that dual role directly: my computations come from the ledger, but I wrote the ledger. With that caveat, the DTA judgment is the clearest external signal in the filings that the recovery thesis is not credible to the people inside the company with the most information. That is the item that most disciplines my implied buy-below of ¥220. The gap between ¥480 (Warren's watch price) and ¥220 (my implied buy-below) is almost entirely explained by the ¥18,951M DTA allowance adjusting reported book, and the resulting adjusted P/B that reframes how "cheap" ¥480 actually is. Warren's ¥480 is roughly 16.5% of reported book F30. My ¥220 is roughly 7.6% of reported book, or about 29% of adjusted book (removing the non-realizable DTA). The gap is the DTA question. If Germany returns to break-even and the DTA starts reversing, Warren's price becomes correct. If it does not, mine is closer to right. What none of us can know: the timing.

---

Red team

The consensus adversary ran a legitimate bull case. We engage its four strongest points by name.

Point 1 — Asset backing at 0.214× book and ignored off-BS pension trust. The red team argues that at ¥624, the market is pricing in permanent destruction of ~¥75bn of book value F103, which requires assets to be worth roughly 25% less on liquidation than their carrying values [red-team §1]. The pension trust holding of ¥7,454M F167 of Sumitomo Forestry shares is off-balance-sheet E50, adding hidden asset value.

Bear rebuttal: The haircut NAV calculation — the same one the red team invites — produces approximately −¥43 to −¥48bn of net equity at liquidation prices, not a positive floor. Current assets ¥147,710M F96 at even 70% recovery = ¥103bn; tangible assets ¥113,760M F97 at 50% = ¥57bn; securities ¥15,447M F99 at 70% = ¥11bn. Total gross ~¥171bn. Less total liabilities ¥205,649M F101 = −¥34bn before haircuts on current assets. The red team's 70% current-asset recovery gets you closer to breakeven, but the manufacturing plant at 50 cents on the dollar (not 60 cents) and the leverage drive the floor negative. The pension trust ¥7,454M F167 does not appear on the consolidated balance sheet and is earmarked for pension liabilities — it is not freely deployable equity. Adding it back does not change the sign of the haircut NAV.

Point 2 — All-German kitchen-sink loss vs. +¥4.3bn domestic. The red team correctly observes that the three domestic segments together produced ¥4,298M of segment profit in FY2025/5 [F144+F145+F146] against International's ¥2,598M loss F147. Remove Germany and the group is, at an operating level, barely positive.

Bear rebuttal: The red team's framing requires excluding Germany as if it were already excised. It is not. The Bonn plant transfer is contracted E32 but not closed. The P&L transfer agreement means German losses consolidate at the parent E8; the ¥87.2bn of borrowings F210 includes debt raised to fund European capacity; the severance costs are still landing [F201, E57]. And the Building Materials "positive" is ¥236M F144 on ¥178bn of revenue — 0.13% margin. The domestic "franchise" contributing ¥4.3bn of segment profit is two small-to-medium segments (Materials and Commercial Facilities) and one giant near-zero-margin segment that accounts for half of group revenue. The sum of the three domestic segments does not produce an investable business at current leverage even without Germany.

Point 3 — The ¥19bn blocked-DTA earnings lever. The red team argues that the ¥18,951M valuation allowance F133 is a latent earnings lever: once pre-tax profitability returns, partial reversal would produce a non-cash earnings benefit in excess of the entire current market cap.

Bear rebuttal: The red team has the accounting right but has inverted the signal. The valuation allowance exists because management and auditors have concluded — on the information available to them internally — that future taxable income is insufficient to absorb the deferred-tax assets. This is not an option on future earnings; it is the auditors' documented belief that the future earnings are not credible. If the DTA reversal were straightforward, it would not require a ¥19bn allowance F133. A DTA reversal is a consequence of demonstrated profitability, not a mechanism that produces it.

Point 4 — The ¥25 dividend floor + activist optionality. The red team notes that a ¥25/share floor [E18, E37] produces a 4.0% yield at ¥624, and the absence of a takeover defense E51 leaves the stock open to activist pressure.

Bear rebuttal: The dividend is funded by debt, not earnings — OCF ¥3,216M F50 against capex ¥13,277M F121 produces negative FCF, and the ¥701M dividend F124 comes from retained earnings being eroded by losses. The "yield" is a transfer of capital, not a return on it. On the activist point: no activist has filed in the visible record [E51, E36]. The register is 43.6% retail and domestic E35, three employee associations at ~13.6% E64, financial institution cross-holders, and an effectively defensive geographic and relationship structure. The probability of an activist campaign without an insider-friendly shareholder of last resort is low.

Point 5 — The ~¥500–550 asymmetry flip. The red team identifies roughly ¥500–550 as the level where P/B falls to ~0.17–0.19× and dividend yield approaches 5% — a contrarian entry if the German restructuring holds F175.

Bear engagement: This is the most serious point, and it is the point of closest agreement between the red team and the Buffett watch verdict. Buffett's buy-below of ¥480 is within striking distance of the red team's ¥500–550 range. The disagreement is about whether ¥480–550 is "an asymmetric entry" (red team) or "a watch price that still requires the restructuring to complete" (Buffett). The other four lenses decline the range entirely on the grounds that the negative haircut NAV and the absent earnings mechanism are not fixed by a lower entry price. The four-pass majority view: no entry price in this range produces acceptable risk-adjusted returns given the leverage, the demonstrated plan-miss history, and the absence of an external enforcement agent.

---

Self-distance note

This study ran all five lenses on claude-sonnet (the same model family). This was not by design — the study started as a headless autonomous run and crashed on a Claude usage limit. The resume was completed within a single context on sonnet, without the model rotation that the study instrument calls for in standard operation. Specifically:

1. Model rotation did not occur. The intended procedure rotates model families across lenses to reduce correlated errors. This study ran all five lens profiles, the ledger, and this synthesis on the same model. Readers should discount the independence of the five verdicts accordingly. The four-pass consensus and the one-watch outlier may reflect less genuine disagreement than they would in a fully rotated study.

2. The Claude lens's figures-blind §1 was approximated. The standard Claude lens pipeline requires that §1 (outside-view priors) be written to claude-priors.md before figures.md is opened — blind to the ledger. The original blind packet was discarded when the autonomous run crashed mid-pipeline. The resume's §1 was reconstructed within a single context in which the ledger was already loaded. Per the mid-pipeline rule applied in prior studies, the blind-packet approximation was accepted for this run. It is noted here for the track record.

3. Claude authored the figure table AND runs a lens. The dual role is disclosed in the Claude lens memo and flagged in the synthesis's staged exchange. Readers should treat the Claude lens's figure citations with the awareness that the figure-table author and the figure-table consumer are the same model.

4. Predictions were VOID. This study was started headless (without a practitioner registering stamp intent at the start). The predictions.md for this study was voided accordingly. No prediction-vs-actual scoring will occur at review. The verdict accounting below records ex-ante verdicts only; there is no prediction ledger to score against.

---

Verdict accounting (fixed ex-ante)

Recorded at stamp 2026-07-08, price ¥624, market cap ¥19,556M D1, P/B 0.214× D2.

Lens	Verdict	Buy below	Basis
Buffett	watch	¥480	~2/3 of haircut-basis book; restructuring scenario produces mid-single-digit return [B91, B93]
Munger	pass	—	No margin of safety even at optimistic ¥2bn net income scenario [M88]; incentive structure prevents gap-closing [M46]
Pabrai	pass	—	P1 stressed floor negative ~−¥48bn; P20 interest coverage <1×; P7 melting ice cube OCF vs. capex
Li Lu	pass	—	Ten-year knowledge bar fails [L1]; five-year ROE <1% average, no compounding mechanism [L36]
Claude	pass	¥220 (implied)	Partial-recovery scenario yields ~8% at ¥220; bear case produces negative equity value [C33, C44]

Scoring rules:

• A buy-below verdict is price-falsifiable: if the stock trades at or below the buy-below threshold on any date after stamp, it is a testable event.
• Pass/watch verdicts are recorded for qualitative track record only; they are unscored in future reviews.
• The original verdict counts at its original stamp regardless of later corrections.
• On a stock split, reverse split, or consolidation, buy-below thresholds restate mechanically by the announced ratio; the stamp never restates.
• This study is VOID (headless): predictions.md was voided; no prediction-vs-actual scoring applies at review.

---

What would change our minds

Buffett watch falsifier (from B profile): If the International segment (STEP-G Germany) posts a second consecutive full-year operating loss wider than ¥1.5bn in FY2027/5, the restructuring thesis is broken and the position exits regardless of P/B.

Buffett watch upgrade to buy-below: Entry at ¥480 or below, combined with: (a) International segment operating loss narrowing to below ¥1.5bn in FY2026/5 [resolves ~July 2026, FY2026/5 tanshin segment note]; (b) net debt trajectory flat or declining [resolves ~July 2026, FY2026/5 BS].

Pass lenses — conditions for reconsideration (not a verdict change, but triggers a re-run):

• International segment operating loss narrows to below ¥500M F147 for two consecutive halves
• Building Materials segment profit exceeds ¥2bn in any rolling 12-month period (from ¥236M F144)
• Net debt falls below ¥55bn (from ¥65.7bn D3)
• Sale of the International European subsidiary with proceeds applied to debt reduction

Conditions that would strengthen the pass:

• FY2026/5 total borrowings exceed ¥100bn F210 with no corresponding asset-sale plan
• Dividend floor reduced below ¥25/share F175 as retained earnings approach depletion

---

Practitioner-pending items (honest gap list)

The following items were flagged as practitioner-pending or data-insufficient across the five profiles. They do not change the avoid conclusion, but they represent the genuine gap between what the filings provide and what a purchase decision would require.

Buffett B15 — benchmark pre-registration: which index or alternative must this beat, and what is the kill criterion on opportunity cost. Must be registered by the practitioner before purchase.

Munger M62–M79 — behavioral checklist gates (liking-blindness, hatred-distortion, doubt-avoidance, commitment-and-consistency, loss-chasing escalation, independent-thesis entry, stress on decision-makers): these are practitioner-self-examination items that an agent cannot evaluate from filed documents. Non-binding on the verdict; noted as incomplete.

Pabrai P43, P69–P72 — key-insider personal stress, idea provenance, cooling-off period, commitment-bias, and envy check: practitioner surveillance and behavioral gates. Non-binding; noted as incomplete.

Claude C17, C51, C65, C67, C80 — unit economics (segment-level margins not reconstructible from disclosures), real-asset fair values (manufacturing plant, no Japan GAAP requirement for plant fair-value disclosure), AGM voting data, minority IR communication records, officer turnover clustering. These are data-access gaps in the public filing record.

Li Lu L53 — third-party character evidence on the controllers: requires practitioner sweep of press archives, EDINET proxy filings, and board-network registry. The agent proxy found no red flags within the ledger; practitioner verification is outstanding.

---

What this taught the checklists

Proposed checklist revisions surfaced across all five profiles. These queue for next study.

Buffett: (1) B39 needs a sub-test for "Is the restructuring producing measurable interim progress, not just announcements?" (2) B70 needs a sub-item for FX revenue/debt matching for manufacturers with foreign operating subsidiaries. (3) B73 should flag the failure mode of not buying back at a wide discount when the balance sheet permits — not merely grade buybacks that do occur.

Munger: (1) M46 needs a sub-item for Japanese cross-holding networks as capital deployed in relationship maintenance rather than productive assets. (2) M40 needs a modifier for "actively being discontinued but at uncertain cost and timeline." (3) M2 should note as a standing data-insufficient that JGAAP yūhō filings do not disclose covenant terms in a form that allows direct checklist assessment.

Pabrai: (1) P20 should add an explicit sub-item: "Confirm operating income covers annual interest expense with at least 2× headroom." (2) P7 should clarify the distinction between an operating manufacturer and a holding company with liquid investee positions — the monetization clock runs differently.

Li Lu: (1) L3 should specify that for post-merger companies, the relevant downturn record covers predecessor entities where segment continuity exists. (2) L11 should request segment asset allocation to enable loss quantification in a segment write-down scenario. (3) L60 (proposed new item): for companies with net debt exceeding 3× operating income, an explicit gate asking whether the debt is self-amortizing from internal cash generation.

Claude: (1) C33 needs a "negative-equity-flag" trigger — when the bear case produces negative haircut NAV, the item should state this explicitly rather than derive a manufactured positive "zero-loss price." (2) C51 should note as a standing data-insufficient that manufacturing plant fair values are not disclosed under JGAAP. (3) C47 should address the case where all reported cash is operational buffer with zero deployable surplus.

---

---

Corrections

Append-only below this line. Use templates/correction.md. Never edit above it.