Komaihaltec Inc. (5915): A Portfolio Worth More Than the Company, and No One Selling It

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	¥1,500	B92/B89 Sanborn discount that is neither net-net nor net-cash; B83/B84/B39 no Harry Bottle, no dated catalyst, no controlling holder; B43/B26/B86 1% ROE, no moat, payout > earnings
Munger	pass	null	M18 no nameable moat mechanism; M40 wind "turd" growing inside the bridge "raisin"; M77/M94 the cross-holdings-vs-cap frame is a contrast-anchor trick, napkin math goes the wrong way
Pabrai	watch	¥1,650	P53 ¥2,222 is ~67% of the ~¥3,300 floor, not the ≤50% he demands; P17/P8 reachability soft — needs the reduction to fire; P10/P12 wind + Shimizu are contained scars
Li Lu	watch	¥1,550	L35/L36 value melting not compounding (ROE never clears ~5%); L26/L24 cross-holdings a hoard with policy but no schedule; L50 four-gate fails 2 of 4
Claude	watch	implied ¥1,950	C48/C58 ~85%-of-cap after-tax cross-holdings but nil FY2026 disposals; C33/C39/C45 NAV ¥3,474/sh > tape yet un-closeable; C84/C102 the bet is reachability; C64/C42 payout erodes the anchor

Four of five lenses land on watch; Munger dissents to pass. No lens would buy at the ¥2,222 stamp — every watch verdict names a buy-below below the tape.

The business

Komaihaltec has fabricated steel since 1883 E1, reaching its present form in the 2010 merger with Haltec E2. It does two things a shopkeeper can understand and one thing it cannot yet price. First, it builds steel bridges — about 40% of sales — carrying a job in-house from cost estimate through design, fabrication, and field erection E6. Second, it fabricates the structural-steel skeletons of high-rises and public buildings — about 60% — and owns two of the plants certified by the Japanese government to the top S-grade that permits building super-high-rise frames without restriction E8E9. Those two, plus a small Osaka land-rental income and a printing subsidiary, are the company you can value E11E12.

The third thing is not a business yet. Komaihaltec has built an offshore-wind-turbine-tower plant at Futtsu, completed March 2026 under a NEDO Green Innovation grant E13. The filing is blunt: there are no tower sales, and development costs are running ahead of revenue E16. The Infra-Environment segment lost ¥599M this year F273 on ¥190M of revenue F259, with an explicit risk factor that if the market does not appear as assumed the advance costs may not be recovered and the plant may be impaired E34. Round-1 offshore-wind projects are being re-bid and operator schedules are slipping E15. Management's steel businesses are inside the circle; the wind venture is a cost with an option attached, unknowable a decade out — every lens treated it as such.

Who pays: the great general contractors and the state. Shimizu is 25.9% of sales, Obayashi 9.4% E44 — concentration that cuts both ways, since construction majors put work out to competitive bid. The company left the TSE Prime market for Standard in June 2025 E3, has no controlling shareholder and no parent E54, and is run by the founding Komai family — Emi Komai became president in April 2026, holding 81,914 shares E64, with Hiroshi Komai on the board E65. It is very thinly traded.

The numbers

The stock is ¥2,222. On 4,582,060 shares outstanding after treasury F233, that is a market capitalization of about ¥10.18 billion — against a book value per share of ¥7,635.78 F221, a P/B of 0.29×. And here is the fact that opens the file (Buffett's "baseball bat over the head," Munger's "third-grade math," Pabrai's "number that made me open it," Li Lu's "fact that made me open the file"): the parent's listed cross-holdings alone are carried at ¥11,679M across 29 issues F119 — ~33% of net assets D5 — more than the entire market capitalization. Behind them sit ¥6,097M of cash F47, ¥8,585M of land at decades-old cost F61, and a ¥52,820M order backlog F326, the bridge half of it growing 16.9% to ¥28,498M F328 on national-resilience demand while the structural-steel half falls 17.6% to ¥24,262M F330 and steel order intake collapses 40.2% F322.

Now walk the money through the operating business, because this is where the discount earns its keep. The five segments together earn ¥2,746M D4 — bridge ¥2,151M F269, steel ¥873M F271, real-estate ¥328M F275, less the wind loss −¥599M F273 and other −¥7M F277. Then a corporate/全社 overhead of −¥2,276M F283 falls on the whole like a stone, collapsing consolidated operating income to just ¥470M — a 1.37% margin D2F131. Net income to owners was ¥335M F15, ROE 1.00% F40. And of that ¥335M, ¥288M is dividends received on the cross-holdings F135 — 86% of net income is the securities portfolio's own yield, not the steel business. The five-year ROE series — 4.74 / 1.12 / 2.02 / 3.96 / 1.00% F36F37F38F39F40 — never once clears a hurdle a long-term owner would accept.

The balance sheet is not distressed, and that matters both ways. Total assets ¥66,539M F41 = total liabilities ¥31,552M F43 + net assets ¥34,987M F45; equity ratio 52.58% F35; net debt only ¥6,502M D1 against ¥12,599M of interest-bearing borrowings; operating cash flow positive ¥4,721M F176; the EY ShinNihon opinion is clean, with no going-concern doubt note E88E89. The company can survive an indefinite wait — which is exactly why the discount can persist.

Two disciplines keep the "cheaper than its portfolio" headline honest. First, the discount is not a Graham bargain. It is not net-cash: cash ¥6,097M F47 plus the securities ¥11,679M F119 is ¥17,776M, less than the ¥31,552M of total liabilities F43. And the ¥7,112M securities valuation reserve sits behind a ¥4,222M deferred-tax liability F107F89 — realizing the portfolio triggers roughly ¥3bn of tax, so the after-tax value is nearer ¥8.6bn than ¥11.7bn. Second, the payout is eroding the very book the discount is measured against. The ¥70 dividend F384 is 96.8% of FY2026 earnings F389 (102.9% on the parent F375); shareholders' equity ex-AOCI fell ¥27,642M → ¥27,431M F106F105 and retained earnings fell ¥13,920M → ¥13,882M F102F101 in a profitable year, because ¥324M of dividends F387 plus a ¥184M buyback F197 exceeded the ¥335M earned. And the FY2027 guide is worse: operating income ¥170M (−63.9%) F418, net income ¥150M (−55.3%) F420, EPS ¥32.74 F421, the dividend held at ¥70 F422 — a 213.8% payout. The one line that grew, investment securities ¥7,519M → ¥12,065M F64F63, grew entirely on the +¥3,112M mark-to-market F168, not on cash coming in.

The five lenses

Buffett — watch, buy below ¥1,500

Let me tell you what this company does before I say a word about the stock. It builds things out of steel — bridges, which the state and the expressway companies pay for, and the steel skeletons of tall buildings, which general contractors like Shimizu pay for E6E44. It owns two of the government's top S-grade plants E9. That part is inside my circle. Bolted on is a third thing that is not a business yet — an offshore-wind-tower plant that, in the filing's own plain words, records no sales while the costs run ahead E16. I cannot forecast that ten years out, so I value the two steel businesses and treat the wind venture as a cost with an option attached.

Now the reason anyone looks: the parent's listed cross-holdings carry at ¥11,679M F119 — more than the whole ¥10.18bn market cap — and management has written down a policy to reduce them E84. This is the Sanborn Map pattern (B92): a marketable portfolio worth more than the share price, with the operating business seemingly thrown in for less than nothing. But I have been burned by exactly this shape, so let me do the arithmetic honestly. Subtract all the prior claims (B63): cash ¥6,097M plus securities ¥11,679M is ¥17,776M — less than the ¥31,552M of total liabilities F43. So this is not a net-cash company (B89), and not a Graham net-net either — NCAV is about ¥921 a share against a ¥2,222 price. The discount is real, but it lives in the securities and land above the current-asset line, not in a classic margin of safety.

And the operating business barely earns. The five segments make ¥2,746M D4, but a −¥2,276M corporate overhead F283 and the wind loss leave ¥470M of operating income — a 1.4% margin D2, 1% ROE F40 — and they pay out 96.8% of earnings F389, guided above 100% next year F421F422. That is a business funding its dividend from the balance sheet. So which is it — a Sanborn bargain or a value trap? What keeps it a watch and not a pass is threefold: the securities are marked to market and the discount to them is enormous; management has stated a reduction policy E84; and the asset value is not melting — net assets grew this year F45 and cash flow is positive F176. But I have no catalyst with a timetable. There is no controlling shareholder to force conversion (top holder 5.0% E50); the founder family sits in the chairs E64; nothing in the record shows the securities actually being sold in size and the proceeds returned. Without a Harry Bottle turning assets into owners' cash (B83), a cheap portfolio can stay cheap for a decade. Price, don't time (B7): I want the whole thing at a wider discount — around ¥1,500 — before I let a family's goodwill and a wind venture's cash burn ride on my patience.

What a student should take from this: "cheaper than its portfolio" is only half a thesis; the other half is who turns the assets into cash for you, and when. When the operating business earns almost nothing and pays out more than it makes, the asset value must do all the work — so demand a wider margin of safety and a real conversion mechanism, not a policy sentence. Verdict: watch (buy below ¥1,500).

Munger — pass

Invert, always. The question is not how a steel fabricator at 0.29× book makes me money — it is how owning it becomes a permanent mistake. And the filing hands me the answer.

The kill paths, and the ugly part is they are already in motion. The wind venture is a capital incinerator with no floor — a plant completed March 2026 with zero revenue E13E16, a segment loss growing from −¥459M to −¥599M F274F273, explicit impairment risk E34. Structural steel is in secular contraction — demand below 4 million tonnes three years running E26, backlog down 17.6% F330, order intake down 40.2% F322. Earnings are thin and guided to collapse — operating income ¥470M F131 guided down 64% to ¥170M F418, with a ¥70 dividend paid above 100% of earnings F422. Customer concentration with no switching-cost moat — Shimizu is 25.9% of sales E44, and construction is competitive bid. Where is the moat? There isn't a nameable one (M18): the S-grade certification creates eligibility, not pricing power E9; five-year ROE ran 1.0–4.7% F36F40. A business that earns 1–4% on equity does not compound your capital; it occupies it.

Now the eyes-water fact — cross-holdings ¥11,679M F119 exceeding the ¥10.18bn cap. This is the contrast-anchor trick (M77). You are not buying the cross-holdings for free; they are illiquid relationship glue — Obayashi, Mitsubishi, Sumitomo, SMFG E85, the very customers and bank a fabricator cannot antagonize — held under a reduction policy E84 against every incentive to go slowly, with SMBC on the board and register E69. The napkin math the other way (M94): ¥150M guided net income, ¥70 dividend, ¥10.18bn cap — a 1.5% earnings yield, payout above 100%. That is not a no-brainer. The only attractive arithmetic is the cross-holdings gap, and it is not cheaply capturable.

The lollapalooza runs in reverse (M86): pre-revenue capex sinker + secular steel decline + construction-accounting cost exposure + customer concentration + a payout that exceeds earnings + ¥12.6bn of debt F73F75F85F87 + cross-holdings that cannot be quickly monetized. Each survivable alone; together a negative confluence with no margin for error. This is not too-hard — I understand it completely. It is simply not good enough, at any price I can recommend, to beat a Japanese index fund or any of dozens of better-capitalized franchises. Pass.

What a student should take from this: the cross-holdings exceeding market cap looks like a mathematical gift but isn't — you cannot extract the arbitrage while the operating business burns cash on a speculative venture and pays dividends it cannot earn. The absence of a nameable moat mechanism means statistical cheapness produces quotational gains when sentiment improves, not permanent wealth. When every checklist item reaches for the word "risk" and finds it, the verdict belongs in the pass pile. Verdict: pass.

Pabrai — watch, buy below ¥1,650

Let me tell you what I see, then why I'm not buying it today even though every bone in my Dhandho body wants to. Komaihaltec welds steel bridges for the government and the steel skeletons of skyscrapers for the big contractors E6E8, and it has done exactly this since 1883 E1 — one of the slowest-changing businesses on earth (P55). The number that made me open the file: the listed cross-holdings are ¥11,679M F119 — more than the entire ¥10.18bn market cap — and, unusually for Japan, they come with a written reduction policy E84. This is not the trap where a fifty-cent dollar sits behind a controller who will never let you touch it.

Downside first, before one word of upside (P1). Marked securities ¥11,679M + cash ¥6,097M − all interest-bearing debt ¥12,599M = ¥5,177M of net financial position, plus land at stale cost ¥8,585M F61 — I anchor a conservative floor near ¥3,000–¥3,300 a share, giving zero credit to ¥23,585M of receivables or the S-grade franchise. The floor sits above today's price, and the balance sheet is the good kind of Japanese one — no covenant tripwire disclosed, ¥12bn of committed lines, going-concern clean (P20) E88.

So why only watch? Because I count every rupee of the earnings, and they are a rounding error on top of the portfolio: ¥470M operating income F131, 1% ROE F40, a ¥70 dividend at 96.8% of earnings F389 funded off the balance sheet, guided down 64% next year F418. The thesis cannot lean on earnings; it leans entirely on the securities getting sold and returned. That makes reachability of the floor the whole game — and the reduction policy is stated but not yet demonstrated at scale in the numbers I can see, against dispersed ownership with no activist (P17 soft) E50. Two contained scars: the wind plant is a Horsehead-shaped bet, but small enough not to sink the ship (segment assets only ¥2,706M, subsidy-funded) F291; Shimizu at 25.9% E44 is above my line, but the asset floor survives its loss. Containment is what keeps this off the pass pile.

The arithmetic is a fifty-cent dollar on assets and a fair-to-rich price on earnings. Buffett-style, value is its own catalyst — but I want to pay half of the conservative floor, not half of book, because the earnings won't carry me and the return-of-capital pace is unproven. Half of that ~¥3,300 floor is my buy-below, ~¥1,650. The stamp is ¥2,222 — cheap, but not my fifty cents (P53). I wait.

What a student should take from this: an asset floor above the price is necessary but never sufficient — you must ask whether the floor is reachable and whether the operating business erodes it while you wait. When ROE is ~1% and the payout exceeds earnings, book value is not a reachable "dollar," so set your fifty-cent test against the marked-liquid asset floor, not against book. Cheap-on-assets plus thin-on-earnings plus dispersed-no-catalyst ownership is a watch, not a buy. Verdict: watch (buy below ¥1,650).

Li Lu — watch, buy below ¥1,550

I begin at the boundary of what I can honestly claim to know. Komaihaltec is a 142-year-old Osaka steel fabricator E1 — bridges (40%) and structural steel (60%) E6E8, two coveted S-grade plants E9, and now a pre-revenue bet on offshore-wind towers E13. At ¥2,222 it trades at 0.29× the ¥7,635.78 book F221 — a dollar for twenty-nine cents — and the parent's listed cross-shareholdings alone, ¥11,679M across 29 issues F119, exceed the entire market cap. This is the shape of my Korean hidden-asset walkthrough (L27): you are handed the securities portfolio, and the operating company, the ¥8,585M land F61, and the ¥52,820M backlog come on top. That is a real margin of safety in the asset column, and I will not pretend otherwise.

But my discipline is to ask whether intrinsic value is compounding or quietly melting (L35) — and here the answer is unflattering. Five segments earn ¥2,746M D4; a −¥2,276M corporate overhead F283 cuts consolidated operating income to ¥470M, a 1.37% margin, 1% ROE F40. The five-year ROE series never clears a hurdle a long-term owner would accept F36F40. On my closed-exchange test (L15), a part-owner at ¥2,222 collects a ¥70 dividend — a 3.15% yield — from a business earning 1% on its own equity, whose per-share earnings fell from ¥289 to ¥72 across the cycle F222F226. That is not compounding. And ¥288M of the ¥335M net income is dividends received on the cross-holdings F135 — the securities pile, not the steel business, carries the earnings — while the 96.8% payout F389, guided above 100% F421, means the retained capital sits idle in listed equity at deposit-like returns (L21, L24). The company states the right words about reducing the holdings E84, but the reduction is not yet visible.

Two things hold this at watch rather than too-hard. First, the accounting and the people appear trustworthy (L46): clean EY opinion, no going-concern note E88E89, related-party transactions none E87, modest compensation E82, a family holding real stock it did not merely grant itself E64. This is not a governance trap. Second, the asset backing is real and bounds the downside (L19), so I say watch, not too-hard. The four-part long-hold gate (L50) fails cleanly on business economics — a 1%-ROE compounder is no long hold at any price — so this can only ever be an asset-realization idea, and realization in Japan requires the cross-holdings to actually be sold and the cash to actually reach me. My buy-below of ¥1,550 is worst-case-anchored: securities haircut ~30% for tax and liquidity, plus cash, plus land near book, minus all debt, minus a reserve for the wind write-off and overhead drag, with the operating "franchise" and the wind bet folded to near-zero — a ~2× margin to that floor.

What a student should take from this: a book stuffed with listed securities worth more than the whole market cap is a genuine margin of safety on the downside — but it is not a franchise, and a 1%-ROE business wrapped around a portfolio does not compound; it waits. Separate the two questions cleanly — is capital safe here? (yes) and is capital compounding here? (no) — and when the second answer is no, the price must be low enough to be paid for the assets alone, and you must wait, on the record, for the catalyst. Verdict: watch (buy below ¥1,550).

Claude — watch, implied buy below ¥1,950

I registered this figures-blind as an at-to-below-median, sub-book, low-ROE cyclical fabricator whose whole verdict turned on one measurement: is the listed cross-holding portfolio large relative to market cap and actually being sold to minorities? The ledger answers the two halves in opposite directions, and that split is the study.

The discount is real and, by asset value, deep — my prior understated it. P/B is 0.29× on BPS ¥7,635.78 F221, with no minority leakage — NCI is nil F113. Marking the ¥11,679M cross-holdings F119 to after-tax realizable value — the ¥7,112M securities reserve F107 is already net of ~30% deferred tax, so realization triggers ~¥3.05bn of tax — the portfolio is worth ~¥8.6bn after tax, ~85% of the entire market cap. Add ¥8,585M of land at cost F61, a growing ¥28,498M bridge backlog F328, and ¥6,097M cash F47, and a haircut liquidation NAV I compute at ~¥3,474/share sits above the market cap. Strip the non-operating assets and the market pays only ~¥3.05bn for an operating business earning ~¥330M — priced nearly free. For a whole-company buyer the asset floor is reachable; my priors were too pessimistic on depth.

But the franchise earns almost nothing, and the payout is eroding the book that anchors the discount. ROE 1.0% F40; the five segments' ¥2,746M D4 is more than halved by the −¥2,276M corporate drag F283 plus the wind loss, to ¥470M operating income F131. Net income ¥335M is 86% dividends received on the portfolio F135, not operating profit. The ¥70 dividend is 96.8% of FY2026 earnings F389 and a 214% payout on the FY2027 guide F421F422; retained earnings already fell F102F101. The book that makes P/B 0.29 look safe is being spent down.

And the unlock has no forcing agent and — decisively — no behavior. The disposal trend is the tell: the ¥1,635M gain on securities sales was FY2025 F147; in FY2026 there were no disposals, and the investment-securities line rose F63F64, all of it mark-up F168 plus additions — the portfolio grew. Policy without behavior. The register is dispersed with no controlling shareholder E54, no activist appears E50, the Prime→Standard demotion removed the governance pressure E3, and the top-10 is threaded with reciprocal steel-and-bank cross-holders E85 — mutual glue, not a contestable float. Both a bull and a bear concede the assets; they split on one proposition: does that value ever reach a minority owner? This is the textbook shape of a value trap that is also genuinely cheap — the two are not contradictory. My jury of selves diverged one-directionally — the whole-owner self reaches buy-below, the realistic-minority and ugly-name selves do not — which caps the verdict at watch and widens the priced threshold down to ¥1,950. The two unknowns that decide it — will disposals fire, will the Futtsu plant impair — are resolvable by the next two yūhō, which is exactly why this is watch, not too-hard.

What a student should take from this: a deep asset discount and a value trap are not opposites — they routinely coexist, and the discount's depth tells you nothing about its reachability. Two numbers separate the cases: the disposal trend (is the hidden asset actually being sold — here, zero this year while the policy talks) and the payout's effect on book (a >100% payout spends the very asset that anchors the P/B). Cross-holdings that are 85% of market cap are worth exactly as much as the probability they are ever monetized to you. Verdict: watch (implied buy below ¥1,950).

Synthesis

Where the five lenses agree

The facts are not in dispute, and all five lenses read them the same way. Komaihaltec is a solvent, asset-rich, low-return steel fabricator trading at a deep discount to a portfolio it is not selling. The discount is specific and unanimous: listed cross-holdings of ¥11,679M F119 — ~33% of net assets D5 — that exceed the ¥10.18bn market cap, a P/B of 0.29× F221, ¥8,585M of land at ancient cost F61 and a ¥52,820M backlog F326 on top, against net debt of only ¥6,502M D1 on a 52.6% equity ratio F35. And the reasons the discount persists are equally unanimous: a 1.0% ROE F40 on a 1.37% operating margin D2, a −¥2,276M corporate overhead F283 that halves segment profit, 86% of net income coming from dividends received on the portfolio rather than operations F135, a dividend paid above earnings (96.8%, guided to 214%) that erodes the book F389F421, a pre-revenue wind venture burning cash with explicit impairment risk F273E34, and no forcing agent — no controlling holder, no activist, a founder family in the chairs, Prime-market pressure removed E50E54E3. The near-unanimous read: the asset floor is real, but its realization is a governance question the filings do not answer. Four lenses go watch (Buffett ¥1,500, Li Lu ¥1,550, Pabrai ¥1,650, Claude implied ¥1,950); Munger passes. Critically, every watch buy-below sits below the ¥2,222 stamp — no lens would act at today's price.

Where the lenses diverge

The divergence is subtle, and it is not "cheap enough to buy" — nobody buys at ¥2,222. It is watch versus pass: does a deeply-discounted-on-assets, solvent, time-resolvable name with no current catalyst earn a watchlist slot at a low buy-below, or the discard pile?

Buffett frames it as a Sanborn discount missing its Harry Bottle: "a marketable portfolio worth more than the share price … but Sanborn had a resolution mechanism; this does not (B83/B84). Without someone turning assets into owners' cash on a clock, a cheap portfolio can stay cheap for a decade" — so he watches, at ¥1,500. Pabrai reaches the same wall through Dhandho arithmetic: "the floor sits above today's price … but reachability of the floor is the whole game, and the reduction policy is stated, not demonstrated (P17) … ¥2,222 is ~67% of the ~¥3,300 floor, not my fifty cents (P53)" — watch, at ¥1,650. Li Lu gates on the compounding question: "is capital safe here? yes. Is capital compounding here? No — ROE never clears ~5%, per-share earnings fell ¥289→¥72 (L35/L36). This can only ever be an asset-realization idea" — watch, at ¥1,550. Claude prices the escape hatch: "both a bull and a bear concede the assets; they split on whether that value ever reaches a minority owner (C102) … a value trap that is also genuinely cheap" — watch, implied ¥1,950, with the jury capping the verdict.

Munger dissents to pass, and his dissent is the sharpest reasoning in the file: "This is not too-hard — I understand it completely (M101). It is simply not good enough, at any price I can recommend, to beat alternatives. The 'cross-holdings exceed market cap' frame is a contrast-anchor trick (M77) — you are not buying them for free; they are illiquid relationship glue held against every incentive to go slowly. A 1%-ROE business F40 (M18) with a negative lollapalooza of secular decline + wind burn + above-earnings payout + debt + concentration (M86) does not compound your capital; it occupies it." The whole of the disagreement is whether Munger's "occupies your capital" is decisive at any price (pass) or whether a low-enough entry against a marked, liquid, tax-adjusted floor buys a time-resolvable option worth monitoring (watch). Note the watchers themselves diverge on how to anchor the floor: Buffett and Li Lu are most conservative (~¥1,500–1,550, assets net of tax and debt, wind at zero); Pabrai takes half the conservative floor (¥1,650); Claude, valuing as a whole-owner and discounting for un-closeability, lands highest (¥1,950). But all four agree the stamp is too high.

The red team's challenge (and our answer)

Because four of five lenses agreed on watch, the consensus was put to a dedicated adversary (red-team.md) that argued the opposite verdict — pass — building the strongest value-trap case from the ledger alone. Its thesis: watch is the wrong verdict; this is a permanent value trap, and watching it is itself the mistake, because the years of monitoring are the cost.

The red team's four strongest points. (1) The NAV melts while you watch. In a profitable FY2026, shareholders' equity ex-AOCI fell ¥211M (¥27,642M → ¥27,431M F106F105) and retained earnings fell ¥38M F102F101, because cash returned (¥324M dividend F387 + ¥184M buyback F197) exceeded ¥335M earned F166. FY2027 guides a 213.8% payout (¥70 F422 ÷ ¥32.74 EPS F421 on ¥150M net income F420). "A large discount to a shrinking book is a falling knife, not a floor." (2) No forcing agent, structurally. Top holder 5.0%, top-10 just 32.6% E50E51, no controlling shareholder or parent E54, the Komai family in the presidency and multiple board seats setting their own pay E64E65E76, no activist, and the June-2025 Prime→Standard demotion E3 stripped away the one external governance lever. "Who closes this discount, and when? The honest answer from the ledger is nobody." (3) Zero disposal execution. The gain on securities sales was ¥1,635M in FY2025 but nil in FY2026 F147, while the securities book grew ¥4,546M F63F64 — entirely mark-to-market F168 — and the count held at 29 issues F121. "A reduction policy with zero disposals is management telling you, in the numbers, that the relationship-glue is staying put." (4) The 'free' operating company nets ~nothing. ROE 1.00% F40; the corporate drag F283 plus the wind loss F273 consume 86% of the ¥3,352M profitable-segment profit; 86% of net income is merely dividends received F135; the real enterprise earns ~¥47M on ¥34.4bn of revenue F123. The red team's verdict: pass — remove it from the watchlist unless traced disposals actually fire.

Our answer — conceding what is right. The red team's central facts are real, and the striking thing is that the watch consensus already priced most of them in. Every watch verdict independently (a) set a buy-below well below the stamp because of the no-catalyst, eroding-book, no-disposal facts the red team marshals, and (b) pre-registered a falsifier that is, in substance, "this converts to a pass if the disposals do not fire." Buffett names the missing Harry Bottle; Pabrai names the unproven reduction pace; Li Lu names the melting book; Claude computes the nil-disposal trend as the verdict hinge (C48/C58). The red team is not refuting the watchers so much as voicing Munger's dissent, sharpened — and Munger already carried that dissent to a pass inside the consensus. So three of the four red-team points are conceded and already load-bearing in the watch verdicts themselves.

The one genuinely additive blow is point (1), the NAV-melt arithmetic, because it attacks the anchor the watch rests on. Here is the honest answer, and it is a matter of rate: the erosion is real but slow relative to the resolution clock. Equity ex-AOCI fell ¥211M F106F105 — about 0.6% of ¥34,987M net assets F45 — while the securities mark added ¥3,112M F168, so total book actually rose this year. The two decisive catalysts (will disposals fire; will Futtsu impair) resolve within roughly 24 months on the next two yūhō — faster than a ~0.6%/year hard-book bleed matters — and every watcher's buy-below already sits far enough below the floor to be paid for that wait. Decisively, the red team's own separator — two consecutive years of traced, realized disposals F147F121 — is the falsifier the watchers pre-registered. The disagreement is therefore genuinely one of weight, not fact: the red team (and Munger) judge a 1%-ROE, un-catalyzed trap unworthy of a watchlist slot at any price; the four watchers judge that a marked, liquid, tax-adjusted floor at a low-enough entry buys a time-resolvable option worth holding on the list. It is a close call — and it resolves on the record by 2028, which is why both the watch thresholds and the red team's pass ship together below.

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 consensus red team that challenged the four watch verdicts 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.
• Four watch verdicts carry buy-below thresholds — Buffett ¥1,500, Li Lu ¥1,550, Pabrai ¥1,650, and Claude's implied ¥1,950 — all below the ¥2,222 stamp; they are recorded as the prices at which each lens would re-engage, and are the thresholds a future review reads against (see "What would change our minds"). Munger's pass carries no price. The red team's argued pass is recorded but, like the watches, unscored.

Red team

A consensus red-team ran because four of five lenses agreed on watch; the adversary argued pass (the value-trap bear case) from the ledger alone. Its strongest points, verbatim-faithful and cited, with the synthesis's engagement:

• "The NAV melts while you watch — a >100% payout spends the very book the discount is measured against." FY2026 equity ex-AOCI fell ¥211M in a profit year F106F105, retained earnings −¥38M F102F101; FY2027 payout 213.8% (¥70 F422 ÷ ¥32.74 F421 on ¥150M F420).
  • Engagement (conceded, then bounded by rate): real, and this is the red team's one additive blow, because it attacks the anchor. But the ~¥211M hard-book bleed is ~0.6% of net assets F45 while the securities mark added ¥3,112M F168, so total book rose this year; the two decisive catalysts resolve within ~24 months, faster than a sub-1%/year erosion matters, and every watch buy-below already sits far below the floor to be paid for the wait (Buffett B65, Claude C64/C42).
• "No forcing agent, structurally — who closes this discount, and when?" Top holder 5.0%, top-10 32.6% E50E51, no controller/parent E54, Komai-family presidency and board E64E65, activist absent, Prime→Standard demotion removed the lever E3.
  • Engagement (conceded — this is the watch, not a refutation of it): every watch verdict is built on this fact. It is precisely why Buffett found "no Harry Bottle" (B83/B84), Pabrai judged reachability "soft" (P17), and Claude located the bet on reachability and capped the verdict at watch (C102). The point does not move the consensus off watch; it is the consensus's central reservation.
• "Zero disposal execution on the reduction policy." FY2026 securities-sale gain nil vs ¥1,635M in FY2025 F147; the book grew on mark-to-market F63F64F168; 29 issues unchanged F121 — policy E84 without action.
  • Engagement (conceded — and it is the pre-registered falsifier): Claude's forensic read makes the nil-disposal trend the explicit verdict hinge (C48/C58/C104), and the red team's own separator — two consecutive years of traced disposals — is verbatim the observable the watchers pre-registered to convert watch → buy or watch → pass. This is agreement on the decisive test, not a challenge to it.
• "The 'free' operating company nets ~nothing and burns cash." ROE 1.00% F40; corporate drag −¥2,276M F283 + wind −¥599M F273 eat 86% of profitable-segment profit; 86% of net income is dividends received F135; operating income guided −63.9% to ¥170M F418.
  • Engagement (accepted — this is Munger's pass): the consensus does not dispute a word of it; it is exactly why Munger declined ("occupies your capital," M18/M86) and why all four watchers priced the operations at essentially zero and rested the case on the assets. The point sharpens the dissent; it does not overturn the four-lens judgment that a low-enough price buys a time-resolvable asset option.

The red team's concessions are equally on the record: the discount is real and large F221F119; the balance sheet is not distressed (52.58% equity F35, net debt ¥6,502M and falling D1, +¥4,721M operating cash flow F176); going concern is clean E88; and a genuine S-grade franchise exists underneath E8E9. Its own summary of why none of that rescues watch — "every concession describes why the discount persists — solvent, un-pressured, un-catalyzed — rather than why it would close" — is the sharpest one-line statement of the bear case, and the synthesis adopts it as the standing risk. The split is weight, not fact: the red team and Munger judge the trap unworthy of a watchlist slot at any price; the four watchers judge a marked, liquid, tax-adjusted floor at a low-enough entry worth monitoring. It resolves on the record by 2028; both thresholds ship.

What would change our minds

Four of the five lenses issued a watch with a falsifier:; Munger's pass carries none by construction. Future review notes score against these, not hindsight:

• Buffett (watch → the discount is real; buy below ¥1,500). Converts to pass if, within ~3 years, the ¥11,679M listed-securities book F119 is not materially reduced with proceeds returned to shareholders — or if consolidated operating income does not climb back above ~¥1,000M and stay there. Any sign the wind plant forces an impairment E34, or the family/board blocks a securities sell-down, confirms the trap.
• Pabrai (watch; buy below ¥1,650). Converts to pass if, by the FY2028/3 yūhō, the cross-holding book F119 has NOT fallen materially despite the reduction policy E84 AND the wind loss F273 has widened rather than narrowed. Converts to buy-below ¥1,650 if the price reaches roughly half the ~¥3,300 conservative floor while the reduction is visibly underway (a disclosed sale, a raised buyback, or a wind loss that stops widening).
• Li Lu (watch; buy below ¥1,550). Converts to pass — permanently if two consecutive years show corporate overhead unchanged at roughly −¥2.2bn or worse F283 while the wind segment still bleeds F273 and cross-holdings have not fallen materially below ¥11,679M F119. Converts toward buy only if the reverse appears: cross-holdings monetized on a stated schedule with proceeds returned or redeployed at a documented return, AND consolidated ROE climbing back above ~4% (FY2025 was 3.96% F39) on operations rather than securities gains.
• Claude (watch; implied buy below ¥1,950; review-by 2026-11-30). Upgrade on ≥20% cumulative cross-holding book reduction across the FY2027–FY2028 yūhō with proceeds traced to buybacks or special dividends F119F147; or on a ≥5% shareholder filing an EDINET 大量保有報告書 with a stated engagement purpose E50. Downgrade toward too-hard/pass on a Futtsu wind impairment E34 booked while the portfolio stays static.

The single observable that most cleanly separates "value trap" from "watchable," on which the red team, Claude, and Pabrai converge: two consecutive fiscal years of traced, realized cross-holding disposals — a non-nil 投資有価証券売却益 in both FY2027 and FY2028 (reversing the FY2026 "―" F147), with the listed-issue count falling below 29 F121 and the securities book shrinking net of mark-to-market F63. Execution, not policy. Absent it, the monitoring cost is real and the floor is a mirage.

What this taught the checklists

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

• Buffett — a B92/B104 deferred-tax-on-unrealized-gains line. When the hidden asset is an appreciated securities book, the ¥4,222M deferred-tax liability F89 converts a "securities exceed the whole market cap" headline into a materially smaller net discount; a standard deduction prompt stops a careless reader from double-counting the pre-tax mark. Rationale: the raw ¥11,679M F119 overstates the harvestable value by ~¥3bn.
• Munger — two rationale-only notes. (1) M40 (raisins/turds) should test loss-as-percent-of-operating-profit, not just capital employed — the wind segment is 4% of assets F291 but its −¥599M loss F273 is 22% of the ¥2,746M pre-corporate segment profit D4 — and exceeds consolidated operating income of ¥470M F131 outright — which the capital-employed threshold misses. (2) M94 should carry a backlog-to-revenue coverage ratio for construction-accounting names (here ¥52,820M / ¥34,414M ≈ 1.5× F326F123) as the natural napkin variable. (3) Consider a Japan-specific M6 note listing fabricator/heavy-industry collapse histories (Hitachi Zosen, IHI infra, domestic shipbuilding) to check against.
• Pabrai — a P53 refinement for asset-floor-above-price, earnings-thin names. When ROE is below the cost of capital over a decade, set the fifty-cent test against the marked-liquid asset floor (securities + cash − all debt + hard assets at conservative marks), not against book or full IV — an unreachable book value is not a dollar. Rationale: this study needed two anchors (half the floor ¥1,650 vs half of book), and only the marked-liquid floor is a legitimate "dollar" at 1% ROE F40.
• Li Lu — two notes. (1) A melting-book discriminator cross-referencing L27 and L35: when the hidden-asset test passes but the compounding test fails (assets > market cap AND value not compounding), the pair should cap the verdict at watch and force an explicit realization-catalyst test (L29), so a future agent is not seduced by the asset discount into a pass — or into a buy. (2) An L24 "hoard-in-listed-equity" flag distinct from a cash hoard: a ¥11,679M listed-equity cross-holding earning a dividend yield F119 reads as "invested" but is functionally the same idle hoard unless a dated reduction schedule exists.
• Claude — four, each with rationale. (1) Promote the disposal-trend test to C48's headline — the trend (nil FY2026 sales, rising book F147F63) was far more decisive than the after-tax mark; for cross-holdings-≥-market-cap names, behavior is the whole verdict. (2) Name reciprocal/mutual cross-holdings as a distinct un-closeable class in C45 — the company holds its own customers and banks who hold it back E85, a tighter lock than one-way policy holdings, warranting a steeper discount. (3) Add an archive-depth fallback to C7/C8 — both calibration items silently fail on a single-year archive; they should emit a resolvable-now source-gate flag routed to the practitioner rather than a bare data-insufficient. (4) Add a minimum-archive precondition to the multi-year-diff items (C25/C31/C32/C75/C77) so the hedged-answer share (C95) is not inflated by a source-gate scope decision the analyst cannot fix.

---

Corrections