KEYENCE CORPORATION (6861): Two Businesses on One Balance Sheet, and a Threefold Disagreement on Price

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	¥37,000	B71/B84; B93; B72
Munger	watch	¥58,000 (plus a disclosure gate)	M88; M54/M82; M42/M46
Pabrai	watch	¥19,100	P1/P4; P60; P59
Li Lu	too-hard	— (arithmetic shown, no threshold issued)	L1; L24/L21; L46
Claude	watch	implied ¥21,000, band ¥16,000–¥33,000	C13; C34/C35; C73/C74

Four watch and one too-hard — and the four named prices run from ¥19,100 to ¥58,000, a spread of 3.04× on one ledger, against a stamp of ¥78,970 F1430. That spread is itself the finding of this study. Every lens read the same figure table, agreed almost completely on the facts, and then disagreed about what to pay by a factor of three. Nobody argued about the business. The disagreement is entirely about what object a buy-below price is.

The load-bearing observation: there are two businesses here, and the famous one is not the larger. 82.50% of total assets is a cash-and-securities portfolio D15 — ¥3,028,413百万円 D13, ¥12,487.01 a share D16 — held against interest-bearing debt of exactly zero D18. Strip it out and the operating business earns 105.94% on the capital it actually uses D32, with an honest floor of 42.49% if you insist on calling every yen of cash operating D33. The blended figure the filing prints — return on equity of 13.53% D29 — is neither. It is a magnificent business bolted to a very large deposit account, and separating the two is the whole arithmetic of this study.

What that does to the price. At ¥78,970 F1430 the whole company trades at 5.5170× book D4 and 43.0207× reported earnings D5 on a 0.696% dividend yield D6. Net of the portfolio, the price of the operating business is ¥66,482.99 D25 against after-tax operating earnings of ¥1,720.14 a share D24 — an ex-cash multiple of 38.6497× D26. The portfolio is only 15.81% of what you pay D17; the other 84.19% is the operating business, and 38.65× D26 is the number the five lenses argued about.

The business

株式会社キーエンス develops and sells the small parts that let a factory see: sensors, measuring instruments, vision systems, laser markers, microscopes and code readers E45. It was founded in 1972 E12, started making sensors in 1973 E13, opened its first overseas subsidiary in the United States in 1985 E16, listed in 1987 E19 and entered China in 2001 E21. Today it is a TSE Prime company E19 headquartered in Osaka E5 with 39 consolidated subsidiaries F204, almost all wholly owned sales companies on a uniform pattern E28, and 12,784 employees F80 producing ¥1,169,289百万円 of revenue F5 — ¥91.47百万円 a head D57.

The stated mechanism is direct selling, and it is asserted rather than described. Management names 「当社の強みであるグローバル直販体制」 — its strength, the global direct-sales structure — and pairs it with the sustained creation of 「世界初」「業界初」 products E43E44. The overseas build-out of that same direct structure is the growth plan E48, with penetration described as still small and the room for growth large E47. That is the entire account. 直販 appears exactly twice in seventy-five pages E267. There is no sales-office count, no salesforce headcount, no customer count, no description of how direct contact feeds product planning.

The commercial shape is unusual and it is disclosed. The company ships from stock: 「即納体制を敷いているため、受注はほぼ売上高と均衡しており、受注残高に重要性はありません」 — immediate delivery, orders roughly equal to sales, no material order backlog E84. No customer reaches 10% of revenue in the year reported — both passages state the current year only E85E217. There are no contract assets and contract liabilities are immaterial E211, and no performance obligation extends beyond a year E212. Every sale is a fresh transaction on the disclosed record; whether that reflects repeat pull or transactional fragility, the filing does not say.

Manufacturing is ambiguous by disclosure. The word ファブレス — fabless — appears once in the entire document, undefined, inside a product-quality risk factor E62, while the parent files a full manufacturing-cost statement whose structure is 73.9% materials, 14.0% outsourced processing and 2.4% labour E224, and the group runs its own facilities with stated manufacturing headcount E90E23. The ledger records the juxtaposition and declines to resolve it E258; so does this thesis.

Disaggregation stops at geography. There is one reportable segment E213E33, and the product-and-service note is omitted outright because a single category exceeds 90% of revenue E214 — so no product-group revenue exists anywhere in the filing E262. Geography is the only cut offered E215: 国内 390,066百万円 F1038, 米国 225,751百万円 F1039, 中国 186,550百万円 F1040 and その他 366,920百万円 F1041, an overseas share of 66.64% D53, up 1.83pp in a year D54.

The culture disclosures are unusually concrete, and the governance disclosures unusually thin. The philosophy is to generate maximum added value with minimum capital and people E37; the company uses no job titles internally and works in open-plan offices E52; its fairness rules bar hiring relatives within the third degree of kinship and bar giving or receiving gifts and entertainment E53. It publishes no forecast, on the stated ground that it considers 「合理的な業績予想及び目標を算出することは困難」 — reasonable forecasts and targets difficult to calculate E40, sets no objective indicator for judging attainment of its goals E82, and watches three plain GAAP lines — sales, gross profit and operating profit E39. Against that: the register carries 株式会社ティ・ティ at 15.07% F372, the 公益財団法人キーエンス財団 at 4.57% F378 registered at the Company's own head-office address E257E5, and the founder 滝崎武光 personally at 3.15% F380 — a founder-linked block of 22.79% whose internal relationships the filing never explains, beside a related-party note reading 該当事項はありません E220. Foreign entities excluding individuals hold 50.36% F363 and the top ten hold 51.99% F392. An acquisition of CADENAS Technologies AG added eleven consolidated subsidiaries E180 with no consideration, no goodwill and no purchase-price allocation anywhere E256, sitting beside 「重要な契約等 該当事項はありません」 E86. Deloitte has audited continuously since 1987 F467E153, with unqualified opinions on the consolidated, parent-only and internal-control reports E242E253E251.

The numbers

Five years, consolidated, in 百万円. Revenue 755,174 → 922,422 → 967,288 → 1,059,145 → 1,169,289 F1F2F3F4F5; ordinary profit 431,240 → 512,830 → 519,295 → 561,010 → 635,756 F6F7F8F9F10; profit attributable to owners 303,360 → 362,963 → 369,642 → 398,656 → 445,185 F11F12F13F14F15. Revenue compounded at 11.55% D50, 1.5484× cumulatively D51, with no down year anywhere in the window D46D47D48D49. Earnings per share ran ¥1,250.83 → ¥1,835.63 F36F40; book value per share ¥8,962.26 → ¥14,313.86 F31F35; the equity ratio 93.5% → 94.6% F41F45, recomputed at 94.57% D62.

The profitability, and the honest limit on it. Gross margin is 83.02% D37, after 83.81% the year before D36; operating margin 50.95% D39, after 51.91% D38. Those are the only two years the filing prints either line — the statutory five-year highlights table carries 売上高, 経常利益, 当期純利益 and 包括利益 and omits both 売上総利益 and 営業利益. The only profit margin available for all five years is the ordinary margin: 57.10 / 55.60 / 53.69 / 52.97 / 54.37% D40D41D42D43D44, a band of 4.14pp D45 — down four years running, then up in the fifth.

And the fifth-year turn is not operating. Ordinary profit rose 74,745百万円 F294 while operating profit rose 45,983百万円 F292; of the 28,762百万円 residual, 27,772百万円 — 96.6% — is interest income and the currency swing: 受取利息 8,968 → 16,249百万円 F639F640, and 為替差益 16,270百万円 F643 against a prior-year 為替差損 of 4,221百万円 F648. Management says as much in its own profit bridge E72. Underneath the screen both printed margin lines fell — gross 83.81 → 83.02% D36D37 and operating 51.91 → 50.95% D38D39, which on a two-point series is one observable decline apiece and not a trend — both cost lines outgrew revenue (売上原価 171,444 → 198,552百万円 F631F632 and 販売費及び一般管理費 337,925 → 374,978百万円 F635F636 against revenue growth of 1.1040× D49), and the incremental operating margin was 41.75% (45,983 F292 ÷ 110,144 F289) against a 50.95% average D39.

The balance sheet has no other side. Total liabilities are 199,183百万円 F606 against total assets of 3,670,655百万円 F590; net assets 3,471,472百万円 F626; working capital 1,791,674百万円 with a current ratio of 10.54× D64. Interest-bearing debt is zero of any kind D18 — 社債明細表, 借入金等明細表 and 資産除去債務明細表 each read 該当事項はありません E222. There are no potential shares of any kind E9E93, and the issued count has been 243,207千株 in every one of the five years F101F105.

The portfolio, in four lines. 現金及び預金 596,976 F536 plus 有価証券 896,913 gives 1,493,889百万円 D11; adding 投資有価証券 1,514,304 gives 3,008,193百万円 D12; adding 長期性預金 20,220 gives 3,028,413百万円 D13. That is 82.50% of total assets D15 and ¥12,487.01 a share D16 — 87.24% of book value per share D16F35 — but only 15.81% of the market capitalisation at the stamp D17. It is short and safe: JGBs, corporate bonds and negotiable certificates of deposit E201, 37.46% maturing within a year D70, nothing beyond five years, no Level 3 in either year D71, and 25.44% of the fair-valued book at Level 1 D71. It sits 17,500百万円 below cost D72, which at this size is a rounding error. It yielded 0.561% in cash D35 — 受取利息 16,249百万円 F640 on an average portfolio of 2,896,216百万円, the prior-year portfolio being 2,764,019百万円 D14.

Cash conversion is near unity and the plant is trivial. Five-year cumulative operating cash flow is 1,802,222百万円 D83 against cumulative net income of 1,879,806百万円 D87 — 95.87% D88. Capex was 28,371百万円 F803F310, 1.6469× depreciation of 17,227百万円 D81F773 after 0.944× the prior year on capex of 14,342百万円 F802, the step named as a new logistics centre E89. Capital intensity is 2.43% of revenue D82; property, plant and equipment is 94,718百万円 F570. Free cash flow before securities activity was 402,309百万円 D78, after 395,180百万円 D79 — a 2.10% yield at the stamp D89. R&D was 32,839百万円 F854, 2.81% of revenue D55; SG&A ran 32.07% D56. The one working-capital line that is moving is receivables: days rose 110.4 → 118.5 D66, receivables growing 1.1851× against revenue's 1.1040× D49D66, with the doubtful-accounts allowance up from 1,426 to 2,135百万円 F545F546; payables fell 14,890 → 10,890百万円 F591F592, leaving 20.0 payable days D67 against 156.8 inventory days D65 and a cash-conversion cycle near 255 days financed entirely out of equity.

The two returns, side by side. Operating invested capital — equity less the portfolio, there being no debt to add — was 344,533百万円 at the start of the year and 443,059百万円 at the end D28D27. NOPAT excluding every yen of financial income was 417,178百万円 D23, on an effective tax rate of 29.98% D22. That is 105.94% on average operating invested capital D32, 42.49% on the conservative bracket that calls all cash operating D33, and 12.68% on the blended balance sheet D30. Return on equity, blended, ran 14.85 / 15.56 / 13.95 / 13.48 / 13.53% F46F47F48F49F50 — falling, across five years in which the operating business got better and revenue grew half again.

What was done with the money. Over five years the company earned 1,879,806百万円 D87 and declared 412,293百万円 of dividends D99 — a payout of 21.93% D100, with 78.07% retained. Dividends per share went ¥200 → ¥300 → ¥300 → ¥350 → ¥550 F121F122F123F124F125 and the payout ratio 15.99% → 29.96% D94D98, which is a real and recent improvement. Against that: there has never been a buy-back or a cancellation D93E101, though the articles expressly permit one by board resolution E129; the only treasury movement in the year is odd-lot mechanics, leaving 682,548 treasury shares, 0.28% of issued F427F420. Cumulative investing outflow was 1,130,412百万円 D84, of which 251,890百万円 in the reporting year alone was securities and deposit activity D80 against 28,371百万円 of capex F803. The whole stated policy for a three-trillion-yen portfolio is one sentence about investing in highly safe bonds E197, with no target, no definition of surplus, no timetable and no terminal use E261. At the parent (提出会社), 522,000百万円 was swept into 別途積立金 across two years — 250,000 F1244 then 272,000 F1294 — while distributable 繰越利益剰余金 rose only 54,576百万円, from 323,481 to 378,057百万円 F1229F1323.

And the company printed its own scorecard. Five-year shareholder total return reads 104.0 / 114.8 / 136.3 / 117.4 / 114.9 F161F162F163F164F165 against a dividend-inclusive TOPIX of 96.9 / 100.6 / 147.1 / 153.0 / 202.4 F166F167F168F169F170. The owner earned +14.9% while the index earned +102.4% F165F170, over five years in which revenue compounded 11.55% D50 and the ordinary-margin band was 4.14pp D45. The mechanism is on the same page: the printed price-earnings ratio fell 43.81 → 40.26 → 46.77 → 37.07 → 32.19× F51F52F53F54F55 while earnings per share rose 46.8% F36F40. The business worked. The shareholder did not get it.

Where the stamp sits. ¥78,970 F1430 is 1.1510× the 第57期 high of ¥68,610 D8F175 and 1.5864× its low of ¥49,780 D9F180. It is also above every 最高株価 the filing prints — the series is ¥76,210 / ¥62,210 / ¥75,040 / ¥77,400 / ¥68,610 F171F172F173F174F175, and 78,970 F1430 ÷ 77,400 F174 = 1.0203×. One stamp day's turnover was 86,601,214千円 on 1,095.9千株 F1434F1433, 0.452% of the net market capitalisation D10.

The five lenses

Buffett — watch, buy below ¥37,000

Let me tell you what this company does, because it is simpler than the price suggests. It makes the small parts that let a factory see. A plant manager somewhere has a line running and a problem — a part upside down, a weld short, a label that won't scan — and somebody from this company walks in, watches the line, and sells him a device — the filing will say only that these selling prices are 「比較的少額」, relatively small E263 — that saves him a shift of downtime. Then the plant standardises on it, and the next line gets one too. They sell these themselves, direct: management names the global direct-sales structure as its own strength E43, and no passage in the whole filing describes a distributor or an agent anywhere E267. They started in 1973 and are doing the same thing today E13. Twelve thousand seven hundred and eighty-four people F80 produced ¥1,169,289百万円 of sales F5 — ¥91.47百万円 a head D57.

Now the economics, and here is where you sit up. It costs them 198,552百万円 to make what they sell for 1,169,289百万円 F632F5 — a gross margin of 83.02% D37. After every selling and administrative expense the operating margin is 50.95% D39. Half of every yen through the door is profit. They carry no debt of any kind D18; 94.6% of the balance sheet is equity F45; earnings turn into cash almost one for one, 95.87% over five years D88. And what does it take to run? The whole operating business — every bench, every mould, every yen of inventory and receivable, net of what it owes — runs on 443,059百万円 of capital D27, and that capital earned 417,178百万円 after tax D23. It earns roughly its own weight every year: 105.94% D32. Be stubborn and call all the cash working capital and it still earns 42.49% D33. I can count on my fingers the businesses like that. Charlie would say it is not a business, it is a royalty.

So why isn't this the easiest yes I ever wrote? Because there are two businesses on this balance sheet and only one of them is any good. Cash and securities come to 3,028,413百万円 — 82.50% of total assets D13D15, ¥12,487.01 a share D16. It is all short paper, nothing beyond five years D70. Last year it paid 0.561% D35. Meanwhile the company retained 78.07% of five years' profits D100, bought back not one share D93, and put the money in that pile. Over those same five years the owner's total-return index came to 114.9 against a dividend-inclusive TOPIX at 202.4 F165F170 — printed in the company's own report. That is the Sanborn Map pattern in a modern suit: a fine little business with a growing hoard of securities stapled to it, and the owner getting a fifth of the earnings while the hoard compounds at half a percent. For every yen retained over the window, per-share earnings rose about ten sen — earnings per share up ¥584.80 F36F40 against roughly ¥6,051 a share retained D87D99D1 — just under the bar I set.

Now let me value the thing, before I look at any quotation. Two columns. The portfolio first: ¥12,487.01 a share D16, taken at face, sitting 17,500百万円 under cost D72, which at this size is a rounding error. Second, the operating business stripped of every yen of interest, equity-method and currency income: ¥1,720.14 a share after tax D24. Depreciation of 17,227百万円 F773 and the spending needed to hold volume are near enough the same number — last year's capex of 14,342百万円 ran below depreciation F802D81, and this year's 28,371百万円 was mostly the next logistics centre F803E89, which is growth, not upkeep. So owner earnings and reported earnings are close cousins here.

What is that ¥1,720 worth? On no growth at all, a business with those returns and no debt is worth fourteen times: ¥24,100. Crediting half the growth it has actually delivered — sales compounding 11.55% D50 on capital spending of 2.43% of sales D82 — twenty-five times, ¥43,000. Add the portfolio and the whole company is worth ¥36,500 to ¥55,500 a share. Now take the upper end of that range — ¥55,500, the growth-credited column, not the austere one — and charge the old margin-of-safety third against it: ¥37,000. That the discount off the generous appraisal lands within ¥500 of the austere appraisal itself is the point. Two roads, one answer.

The quotation is ¥78,970 F1430. Take the portfolio out and you are paying ¥66,482.99 D25 for a business that runs on about ¥1,827 of real operating capital a share D27D1 — 38.6497 times its earnings D26, and a goodwill premium over net tangible assets — book value per share of ¥14,313.86 F35 less the 29,181百万円 of intangibles F574 — of some 37.7 years of after-tax operating earnings D24. The figures are not hitting me over the head with a baseball bat. They are not even tapping me on the shoulder.

And there is one more thing, which is why this is a watch and not merely a price. When you buy the whole company you take on partners, and I cannot tell you who mine would be. A company called 株式会社ティ・ティ owns 15.07% F372; the filing does not say what it is, who owns it, or why it holds the stake E257. A foundation registered at the company's own front door owns 4.57% F378E257. The founder owns 3.15% F380, attends all twelve board meetings F439E113, and does not appear on the officer roster at all E266. Twenty-two point seven nine percent, unexplained, with not one word about succession E266 — beside 「関連当事者情報 該当事項はありません」 E220, and 「重要な契約等 該当事項はありません」 next to an acquisition that pulled eleven subsidiaries into the group with no price, no goodwill and no purchase-price allocation E256E180. The moat everyone talks about is described in two clauses E267. All twelve officers together own 600 shares F454; the president owns none E132.

None of that is fraud. The auditor has been there since 1987 and signed clean on all three reports F467E242E251E253. It is something else: a company that tells you the least the law allows and asks you to admire the results. I can admire the results. Verdict: watch, buy below ¥37,000. At that price the margin of safety pays for the darkness. At today's price I am the one paying.

What a student should take from this: when a business keeps a bank vault in its basement, insist on two columns and value them separately, because the blend lies to you in both directions — a headline 13.53% return on equity D29 understates a shop earning 105.94% on the capital it actually uses D32, and 43.02× earnings D5 overstates nothing at all once you see the vault is only 15.81% of the price D17. Then ask what the retained yen earned: here, 0.561% in short bonds D35 instead of the hundred percent available in the shop. A great business run as a savings account compounds like a savings account, and a five-year owner return of 114.9 against a market at 202.4 F165F170 is what that looks like on the scoreboard. Quality is not a substitute for disclosure; you may own a wonderful business without knowing your partners only if the price pays you for the ignorance.

Munger — watch, buy below ¥58,000, and a second gate

I want to invert before I praise, because a balance sheet this clean is exactly the kind that seduces you into skipping the step. Where does this business die? Not from debt — there isn't any D18. Not from a customer walking — nobody is 10% of sales E85. The two live paths are quieter. One: 22.79% of this company sits in a block the filing itself declines to explain — 株式会社ティ・ティ, an entity whose business, ownership and purpose the filing never describes, at 15.07% F372E96E257, a foundation registered at the company's own head-office address at 4.57% F378E97, the founder personally at 3.15% F380 — with the related-party note reading, in full, 該当事項はありません E220. Two: an acquisition that brought eleven subsidiaries into consolidation, disclosed in one sentence, with no price, no goodwill and no purchase-price allocation, sitting directly beneath a line saying there were no material contracts at all E180E256E86. Neither is proof of wrongdoing. Both are exactly the kind of undocumented door a filing this careful about everything else has no business leaving open.

Now the part that is genuinely hard to walk past, because it's real. Strip out the portfolio — 82.50% of the balance sheet D15 — and what is left earns between 42.49% and 105.94% on the capital actually working in it D33D32. No debt anywhere D18. Gross margin 83.02% D37. Revenue compounding at 11.55% D50, entirely self-funded. I went looking for the mechanism behind it and mostly found silence: direct sales, the stated source of the whole advantage, is invoked exactly twice in seventy-five pages and described no further E43E267. I can't name the moat's fingerprint the way my checklist wants. What I can say is that the fingerprint of something durable is all over the returns, even though the filing won't tell me what it is.

Here is where the checklist's own discipline earns its keep: a great business is not automatically a good purchase, because the price has to be fair too. This one is not fair right now. The stamp sits 15.10% above the fiscal year's own trading high D8, and the reported multiple of 43.0207× D5 sits near the top of the company's own printed five-year band — 43.81 / 40.26 / 46.77 / 37.07 / 32.19× F51F52F53F54F55 — the low end of which was struck at the very same fiscal year-end, four months before the stamp E8. That is not a company the market has overlooked. It is a company everybody can see is wonderful, priced accordingly: the pari-mutuel trap, where the good horse pays three-to-two.

Then there is what management did with the roughly 1,467,513百万円 it retained over five years D87D99D100. Almost none of it went into the operating business, which needed only another 98,526百万円 of invested capital to keep compounding at triple-digit returns D27D28. The rest went into a portfolio earning 0.561% D35. There has never been a buy-back, in a year of 445,185百万円 of net income F15D93, and no valuation trigger for one is stated anywhere. Retaining earnings you cannot deploy at anything close to your own return on capital, indefinitely, with no stated plan, is not conservatism. It is an allocation decision made silently, year after year, and it is the single largest thing wrong with an otherwise admirable business.

The compensation structure compounds the doubt rather than resolving it. The stated policy links 60% to 75% of the pay base to operating profit F485F486E163, on a base that is the prior year's employee salary multiplied by a coefficient capped at 3.0× F484E162 — yet every officer's reported remuneration this year is booked entirely as fixed, with a flat 「−」 on the performance-linked line for every category E167. That may be an honest artefact of a formula lagged one year through the employee base; the filing does not say so, and I am not going to invent the reconciliation. And the succession picture is the same shape again: the founder attends all twelve board meetings F439E113, is styled 取締役名誉会長 in the present tense E12, is the only named officer whose birth date is omitted, appears nowhere in the twelve-person roster the filing itself totals E116E144, and the words successor, succession and founding family appear zero times in the document E266.

Verdict: watch, buy below ¥58,000 — and I want it understood that price alone does not clear this name. ¥58,000 is not an appraisal. It is roughly the low end of the company's own printed five-year multiple band, about 32× on reported earnings per share of ¥1,835.63 F55F40, which is where this stock was valued at its own fiscal year-end four months before the stamp. Both gates have to open. The second is a subsequent filing that actually explains what 株式会社ティ・ティ is E257 and gives the CADENAS acquisition a disclosed consideration, goodwill figure or purchase-price allocation E256. I would revisit at the price. I would only buy after the disclosure.

What a student should take from this: a business can clear every quality test you own — no debt, extraordinary returns on capital, decades of discipline — and still not be buyable, because quality is not the only gate. When the disclosure around related parties, an acquisition, or a founder's own governance status goes quiet in a filing otherwise scrupulous about everything else, that silence is the finding, not a formality to wave through because the operating numbers are beautiful. And "a great business at a fair price" requires proving the second half of the sentence as rigorously as the first: a company priced above its own year's trading high D8 is not a bargain merely because it deserves to be expensive.

Pabrai — watch, buy below ¥19,100

Let me start where I always start, and where the answer here is uncomfortable: what do I lose if I am wrong?

You pay ¥78,970 F1430. Behind that share sits a portfolio of cash, deposits, JGBs, corporate bonds and negotiable CDs worth ¥12,487.01 a share D16 — 82.50% of total assets D15, nothing maturing beyond five years D70, nothing in Level 3 D71 — and, in front of it, interest-bearing debt of exactly zero D18. Not "low." Zero E222. Total liabilities are 199,183百万円 F606 against 3,670,655百万円 of assets F590, a 94.57% equity ratio D62 and a current ratio of 10.54× D64.

Now do the crayon arithmetic. Receivables 379,590百万円 F538 at eighty cents; inventory 85,273百万円 F542 at fifty; all property, plant and equipment 94,718百万円 F570 at fifty; add the JustSystems stake E259 carried at 4,517百万円 against 99,243百万円 of quoted value, a 94,726百万円 hidden asset F1357F939; subtract every liability F606. That much comes to ¥1,192 a share. Then add the ingredient the answer actually needs — the portfolio itself, 3,028,413百万円 at face D13 — and you get roughly ¥13,700. One caution on the JustSystems line, because it is not on the same basis as everything around it: the 94,726百万円 gap is a parent-only (提出会社) 関連会社株式 figure E231, while the consolidated sheet being marked here already carries that affiliate under the equity method inside 投資有価証券 — which holds 50,015百万円 of investments in unconsolidated subsidiaries and affiliates F838 — so dropping the full parent-basis gap into a consolidated net-asset figure double-counts whatever the equity method has booked above cost. Against ¥78,970 F1430 the crayon is about seventeen cents on the dollar, and the portfolio alone is 15.81% of the market capitalisation D17.

So here is the sentence a student needs to write down: eighty-four cents of every dollar you pay for KEYENCE buys earning power and not portfolio D17 — and on the crayon marks above, better than four-fifths of the price has no liquidation asset behind it at all F1430. The balance sheet is unbreakable — zero interest-bearing debt D18 against a 94.57% equity ratio D62 — and it protects almost none of your purchase price. That is not a contradiction — it is arithmetic. An unbreakable balance sheet on a company priced at 43.0207× earnings D5 and 5.5170× book D4 gives you survival, not a floor. Survival is the thing that saved me from Horsehead; a floor is the thing that would let me sleep. I have one and not the other.

Strip the portfolio out and the operating business is a marvel — ROIC 105.94% D32, 42.49% on the punitive bracket D33, gross margin 83.02% D37, operating margin 50.95% D39, the ordinary margin inside a 4.14pp band across five years D45 while revenue compounded at 11.55% D50, cash conversion 95.87% D88. Then I go looking for the mechanism and it is not there. 直販 appears exactly twice in seventy-five pages E267. ファブレス appears once, undefined, buried in a risk factor E62E258, while the parent files a full manufacturing-cost statement E224. There is no product-group revenue anywhere E262E214, no pricing commentary at all E263. I am being asked to pay 38.6497× D26 for an arbitrage spread the company names in one clause and evidences nowhere. When I cannot name and date the spread, I do not get to extrapolate it.

Now the part that decides it. Ask what happens if this merely becomes a very good company instead of the best one. Operating earnings after tax are ¥1,720.14 a share D24. Put a faded fifteen times on that — my judgment, not the archive's — add the portfolio at face D16, and conservative intrinsic value is about ¥38,300. From ¥78,970 F1430 that is minus fifty-one percent, and there is no asset to catch you on the way down. Meanwhile the upside case requires the multiple to survive. My scenario table says the same thing: multiple holds and revenue compounds, about ¥92,000 at a fifth of the weight; a de-rate to 25× on operating earnings plus the portfolio, about ¥55,500; a de-rate to 15×, about ¥38,300; a capex recession taking the margin down eight points on a 15× multiple, about ¥30,000. Expected value about ¥57,900 — below the stamp F1430. Downside a half, upside a fraction. That is the coin flip inverted.

And I am not paid to wait. Dividend yield 0.696% D6; no buy-back and no cancellation in a year of 445,185百万円 of net income D93F15, with the articles already granting the board that power E129; 21.93% of five years' earnings paid out D100 and the other seventy-eight percent into a bond book earning 0.561% D35. That is the whole reason headline return on equity drifted from 14.85% to 13.53% F46F50 while the operating business got better. Nobody is going to fix this on my timetable: twelve directors and auditors own 600 shares between them F454 against 334百万円 of annual board pay F493F496, with one director alone at 182百万円 F499 — 中田有, the Company's president from December 2019 until December 2025 E136. There is no owner in that room.

Where does the market's fear come in? It doesn't. The stamp sits 15.10% above the fiscal year's high D8 and 58.64% above its low D9. There is no uncertainty priced as risk here — there is quality priced as certainty. That is the trophy case, not the bargain bin.

Verdict: watch, buy below ¥19,100. That is half of my conservative value of about ¥38,300 — the fifty-cent dollar, and nothing else. At that price the portfolio D16 becomes roughly two-thirds of what you pay instead of a sixth D17, and the operating business trades near four times its after-tax earnings D24. I will say plainly what that means: it is a price this security has never traded at in the archived record, whose lowest printed low is ¥44,880 F177. The number is honest, and it is not near.

What a student should take from this: a fortress balance sheet is not a margin of safety — price is. Zero debt D18 and a three-trillion-yen portfolio D13 guarantee that this company survives anything; they guarantee nothing about what you get back, because at 43.02× earnings D5 the assets cover only a sixth of what you paid D17. And when you cannot find the mechanism of a moat in the filing — only the assertion of it E267 — you are extrapolating a number rather than underwriting a business, which is precisely how a great company becomes a bad investment.

Li Lu — too-hard

A student should study this company precisely because it is not cheap and not doubtful. My whole method rests on a single question — can I honestly claim to understand this business's next ten years better than almost anyone who owns it? — and the honest answer here is no. Not because the business is poor. Because the filing declines to describe the one thing that makes it good.

Start with the arithmetic, since it reframes everything. Eighty-two and a half percent of the balance sheet is a cash-and-securities portfolio D15 — 3,028,413百万円 D13, ¥12,487.01 a share D16 — against interest-bearing debt of exactly zero D18. Strip it out and the operating business earns 105.94% on its invested capital D32, with an honest floor of 42.49% D33. The reported 13.53% return on equity D29 is not the business's return; it is the blended return of a magnificent business bolted to a very large deposit account. Most owners of this stock, reading the 13.53% line F50, do not see it.

Now what I can actually verify. Five years of ordinary margin from 57.10% D40 to 54.37% D44, never below 52.97% D43 — a band of 4.14 points D45 across a period in which revenue grew 1.5484× D51. Earnings turn into cash at 95.87% D88. The balance sheet foots exactly at both dates D63. No goodwill amount anywhere in the document — のれん occurs only as two segment-note headings, with no figure attached E256, against intangible fixed assets carried as a single line, 6,338 → 29,181百万円 F573F574 — no borrowings, bond or asset-retirement schedule E222, no dilutive security of any kind E9, a share count unmoved in five years F105, an auditor in the chair since 1987 E153. This is a genuinely excellent enterprise and nothing in my conclusion disputes that.

But excellence is not the bar. The bar is prediction. To forecast this company's earnings power in 2036 I need three things: whether the direct-sales structure that produces an 83.02% gross margin D37 is defensible, what the product mix is and how it prices, and how much runway the overseas build-out really has. The filing answers none of the first two and asserts the third without a number E47. 直販 appears exactly twice in seventy-five pages E267. The product-group revenue split is declined outright because one category exceeds ninety percent of sales E214E262. Gross profit is named as one of three indicators management watches E39 and then never discussed — its level, its movement, its drivers appear nowhere E263. ファブレス appears once, undefined, inside a risk factor E62, and the partner factories it refers to are never counted, located or characterised E258, while the parent's own manufacturing cost runs 73.9% materials and 2.4% labour E224. So the physical product is largely bought in, and the only thing standing between this company and commoditisation is a sales organisation the document will not describe. I am asked to take a moat on faith. I am not permitted to.

The second failure decides it for a lens built on capital allocation. Over five years the company earned 1,879,806百万円 D87 and returned 412,293百万円 D99 — a payout of 21.93% D100. The other roughly 1.47tn went into government bonds, corporate bonds and negotiable certificates of deposit E201 yielding 0.561% D35. Watch what that does to the owner's return: return on equity falls from 14.85% to 13.53% F46F50 across the very years in which revenue grew half again and margins held. The rate of value creation is melting even as the absolute compounds, and the mechanism is arithmetic — the marginal yen goes to the deposit, not to the business earning forty to a hundred percent D33D32. There is no buyback and no cancellation in a 445,185百万円 year D93F15 despite standing articles authority E129. The entire stated policy for a three-trillion-yen portfolio is one sentence E197: no target, no definition of surplus, no timetable, no terminal use E261. 資本コスト, PBR and any return target appear zero times in the document E260, and the Company states plainly that it sets no objective indicator at all E82. A company whose own philosophy is to generate maximum added value with minimum capital and people E37 carries idle capital equal to 82.50% of its own balance sheet D15.

Then the three 該当事項はありません declarations sitting beside contradicting facts: no material contracts E86 against an acquisition adding eleven subsidiaries with no consideration, goodwill or allocation E256E180; no related-party transactions E220 against a founder-linked block of 22.79% including a foundation at the Company's own head office and an entity the filing never once describes E257; no critical accounting estimates E189E228 against a 2,430,622百万円 securities book E200. Add the absent subsequent-events note E264, the absent contingencies note E265, and a founder in all twelve board meetings F439E113 but no officer roster E266. I do not think this is dishonesty — every omission is statutorily permitted, the opinions are unqualified E242, the numbers reconcile D63. It is something else: a company that does not intend to be understood by outsiders. For my purposes the consequence is identical.

So, the price, because a student should always be made to name one — and then be told why I will not issue it as a threshold. Take after-tax operating earnings of ¥1,720.14 a share D24. Capitalise them at twenty times — a multiple I can defend from the filed margin band D45, the zero-debt sheet D18 and the cash conversion D88, and from nothing about the moat, because I have nothing about the moat. That is ¥34,403. Add the portfolio at ¥12,487.01 D16, its fair-value uplift of about ¥203 a share E200, and roughly ¥200 to ¥390 for the JustSystems stake E259 carried at 4,517百万円 against 99,243百万円 of quoted value on the parent-only 関連会社株式 line F1357E231 — a range rather than the whole gap, because the consolidated sheet already carries that affiliate under the equity method F838. Conservative owner value: about ¥47,400. A dollar at fifty cents puts the entry near ¥23,700 — below any price in the five-year record, whose lowest printed low is ¥44,880 F177. The stamp of ¥78,970 F1430 is 1.67× that conservative value, 38.6497× operating earnings after stripping every yen of cash D26, and 15.10% above the highest price printed for the fiscal year just reported D8. The cash does not rescue it: the portfolio is 15.81% of what you pay D17, and the market is paying the other 84.19% for something I cannot see.

Verdict: too-hard, on L1 first and on L24, L21 and L46 behind it. I publish the arithmetic and issue no threshold. That is not a dodge and it is not modesty. A buy-below is a promise that at that price I would act, and I would not — because ¥23,700 is a number derived from a twenty-times multiple I cannot defend from any disclosure about why the margin exists. The honest object is the sentence, not the number: two of the three variables that decide the next decade are absent from the public record, and no price makes an unknowable business knowable.

What a student should take from this: verifiable and knowable are not the same thing, and the numbers are not the knowledge. Everything in this company that can be checked from a statement passes — the margins D37D39, the conversion D88, the footing D63, the absence of debt D18 — and everything that requires understanding why the margins exist fails, because the filing declines to say E263E267. When a company will not describe the mechanism of its own advantage, the correct response is not to admire the outcome and infer the cause; it is to say you do not know, and let that lower the verdict no matter how fine the business or how patient your horizon.

Claude — watch, implied buy-below ¥21,000, band ¥16,000–¥33,000

I registered my priors figures-blind and then opened a ledger that made me look, in one place, badly wrong; in most places roughly right; and in one place wrong in a way I want on the record because it is the most useful thing this study produced.

What I got right, and one of them exactly. Revenue CAGR predicted 8–15% central ~11%, actual 11.55% D50. Overseas share 55–70% and rising, actual 66.64%, up 1.83pp D53D54. Blended ROE 10–22% central ~15%, actual 13.53% D29. Return on operating capital 35–80% central ~50%, actual bracketed 42.49–105.94% D33D32. P/B 3.5–7× central ~4.5×, actual 5.5170× D4. P/E 25–45× central ~33×, actual 43.0207× D5. Payout and yield inside band at 29.96% and 0.696% D98D6. Capex and R&D intensity both 2–5%, actual 2.43% and 2.81% D82D55. Cash conversion slightly under one, actual 95.87% D88. Zero buy-back at p=0.05, actual none, ever D93. And the pile: predicted at 8–20% of market capitalisation with a ~14% central, actual 15.81% D17 — with the registered consequence spelled out before the ledger opened, that this is not a net-cash value case and stripping the cash moves the multiple by roughly a tenth rather than by half. The ledger prints 38.6497× ex-cash against 43.0207× reported D26D5, a ratio of 0.899×. That is as close to exact as a pre-registration gets, and it is the reason no floor thesis appears anywhere in my file.

What I got wrong. The operating margin: I registered a band of 25–48% with a ~35% central and put P(≥50%) at 0.15. The truth is 50.95% D39 — above my whole band. That miss runs against the contamination I declared, since I feared a remembered anchor would push my band too high and instead I set it eight points too low; the blinding was not leaking upward. The pile against total assets came in at 82.50% D15 against a 55–75% band. The equity ratio at 94.57% D62 narrowly cleared a 88–94% band. Inventory days at 156.8 D65 exceeded a 60–130 band.

And the miss that matters most is one I would rather bury. I put 0.40 on this company underperforming its own index over five years, given a large and growing financial pile and an operating business compounding above the index. Both conditions held — revenue compounded 11.55% D50, the pile grew from 2,764,019 to 3,028,413百万円 D14D13 — and the shareholder still earned +14.9% against a dividend-inclusive TOPIX at +102.4% F165F170. It happened, by 87 points, because the multiple de-rated 43.81× → 32.19× F51F55 while earnings per share rose 46.8% F36F40. The retention identity I wrote out at the outside-view stage as arithmetic is not a theory here; it is a completed five-year experiment with a published result. I underweighted it, and I underweighted it in the direction of being too kind to a quality franchise.

And one registration could not be scored at all. My entire moat test was quantitative: does gross margin hold through a downcycle? — operationalised as a gross-margin series varying by under 400bp across the available multi-year window including the weakest year of customer capital investment. The five-year highlights table omits both 売上総利益 and 営業利益, so the archive prints gross margin for two years D36D37 and contains no down year at all D46D47D48D49. Both of my pre-registered too-hard triggers fired. I record MOAT-REAL as unresolved on this archive — not confirmed by two good years, not refuted. That is too-hard on the moat. It is not too-hard on the verdict, because the price fails the owner arithmetic at the highest margin ever printed.

The arithmetic, then. Ex-cash price ¥66,482.99 D25 on after-tax operating earnings of ¥1,720.14 D24 is 38.6497× D26. Subtract what growth actually costs — working capital of about 42.5 sen per yen of incremental revenue (receivables 320,312 → 379,590 F537F538 plus inventories 77,892 → 85,273 F541F542 less current liabilities 168,040 → 187,831百万円 F599F600, against incremental revenue of 110,144百万円 F289) and growth capex above depreciation F773F803 — and free owner earnings are roughly ¥1,481 a share. At a 7% owner hurdle the price embeds about 10.7% revenue growth for a decade at the 50.95% margin held flat D39, then 3% forever. Re-based off the second year of the window — 第54期 FY2023, 922,422百万円 F2 — the delivered CAGR to 1,169,289百万円 F5 is 8.23%; off the third, 967,288百万円 F3, it is 9.95%; the fourth gives 10.40% D49. My headline 11.55% D50 is flattered by a pre-boom base year. So the price asks for slightly more than the best four years just delivered, with no reversion at all — while the ledger says reversion has already started: the incremental operating margin is 41.75% (45,983 F292 ÷ 110,144 F289) against the 50.95% average D39, both printed margin lines fell — gross D36D37 and operating D38D39, one observable decline each — and personnel cost inside SG&A rose 10.39% (152,959 → 168,844百万円 F849F850) against revenue's 10.40% D49 on headcount up 4.27% F79F80 — no leverage in the largest cost line.

The price I will pay, and what moved it. My own bear case — margin cut to 42%, anchored on the observed incremental margin F292F289 and below every point in the archive; revenue held flat because the archive contains no decline D46; capex at depreciation F773; the portfolio at half of face D16 because it has never once been returned D93 — produces ¥26,438 a share, a zero-loss price near ¥26,000. Three jury selves re-ran the same item and returned ¥15,800, ¥17,400 and ¥21,300; three more, working from a ledger with every market row stripped out, returned buy-belows of ¥18,000, ¥21,500 and ¥33,000. Every one of the first three sat below my figure, because each substituted a real revenue decline where I had substituted flat — the softest reading available. Divergence in this lens may only subtract, so I publish ¥21,000, with an honest band of ¥16,000 to ¥33,000. I will not dress that up: it is 27% of the stamp F1430, it implies about 3.5× enterprise value to operating profit and about 4.9× ex-cash operating earnings D24, and this security has never traded there — its five-year low is ¥44,880 F177.

Why watch and not pass, when the threshold is that far away. A watch needing a 73% decline is a watch in name and a pass in effect on any horizon I can see, and I say so. I keep the label because the checklist's pass triggers do not fire: growth creates value here at a 32.7% return on incremental operating capital (ΔNOPAT — 営業利益 549,775 → 595,759百万円 F637F638, taxed at 29.98% D22 — on Δoperating invested capital of 98,526百万円 D27D28), the short case is an avoid rather than a short, and the register is contestable — 22.79% founder-linked F372F378F380 against 50.36% foreign F363 with no poison pill and no rights plan E93. And because C44 makes me name a price either way, so nothing is hidden by the choice.

Two things would change this materially and both are observable. A year in which revenue falls and the margin holds within 200bp would discharge, on the real test, the proposition this archive could not test. A first-ever buyback would tell me the 0.561% sink D35 is not permanent. Neither has ever happened.

This is a superb business. Zero debt D18, zero dilution E9, zero minority leakage, 94.57% equity D62, 95.87% five-year cash conversion D88, no special items in either printed year, 42.49–105.94% return on operating capital D33D32, and a cost base at which operating profit only reaches zero on roughly a 61% revenue decline D37F636. I would own it. Verdict: watch, implied buy-below ¥21,000. Not here, and not near here.

What a student should take from this: a moat can be entirely real and the shareholder can still lose. This company's five-year record is the cleanest demonstration you will find — the operating business did everything asked of it, an 11.55% revenue CAGR D50, a margin band of 4.14pp D45, cash conversion of 96% D88 — and the holder earned +14.9% while the index earned +102.4% F165F170. Two things ate it: the multiple started high and fell F51F55, and 78.07% of five years' earnings D100 were retained into paper yielding 0.561% D35 by a company that has never repurchased a share D93. A business that reinvests at 106% in its operations D32 but sends four-fifths of its profit into government bonds is not compounding at 106%; it is compounding at the blend, which is 13.53% D29 and falling. Second lesson, about method: when the company will not print the number that decides your question, say so and stop. Two good years is not a margin history, and the honest output is "unresolved," not a verdict dressed up from the years you happen to have.

Synthesis

Where the five lenses agree

First, the business is extraordinary and nobody disputes it. Ex-portfolio return on operating capital of 105.94% D32 with a conservative floor of 42.49% D33; gross margin 83.02% D37; operating margin 50.95% D39; zero interest-bearing debt of any kind D18; a 94.57% equity ratio D62; five-year cash conversion of 95.87% D88; no dilution, no options, no non-controlling interests, no potential shares E9E93. Every lens, including the one that declined, wrote a paragraph saying so.

Second, the balance sheet is two things and only one is a business. 82.50% of assets is a portfolio D15 of 3,028,413百万円 D13, ¥12,487.01 a share D16, yielding 0.561% D35 — and it is only 15.81% of the price D17. All five separated the columns before valuing anything, and all five concluded the portfolio does not rescue the multiple.

Third, the capital allocation has already been graded, by the company's own printout. 78.07% of five years' earnings retained D100, never a buy-back D93, one sentence of stated policy E197, no cost-of-capital, PBR or return-target language anywhere E260, no objective indicator set E82 — against shareholder total return of 114.9 versus a dividend-inclusive TOPIX of 202.4 F165F170. Four of five lenses named this as load-bearing.

Fourth, the disclosure has a consistent tilt, and all five name the same instances: 「関連当事者情報 該当事項はありません」 E220 against a 22.79% founder-linked block the filing never describes F372F378F380E257; 「重要な契約等 該当事項はありません」 E86 against an eleven-subsidiary acquisition with no consideration, goodwill or purchase-price allocation E180E256; no critical accounting estimates identified E189E228 against a 2,430,622百万円 securities book E200E259; and two notes absent altogether, subsequent events E264 and contingencies E265.

Fifth, the moat's mechanism is undocumented. 直販 twice in seventy-five pages E267; ファブレス once, undefined, in a risk factor E62E258; no product-group revenue E262E214; no pricing or gross-margin commentary E263. Five lenses looked; five found the same two clauses.

Where they diverge — a staged exchange

---

On the price: four thresholds between ¥19,100 and ¥58,000, read off one ledger against a ¥78,970 stamp F1430 — and the spread is the finding.

BUFFETT (B91, the private-owner appraisal; B93, the required discount from the conservative figure). My object is an appraisal, and then a discount from the bottom of it. The same ¥1,720.14 of after-tax operating earnings we are all handed D24 is worth fourteen times with no growth credited and twenty-five times crediting half the growth actually delivered — sales compounding 11.55% D50 on capital spending of 2.43% of sales D82. That is ¥24,100 to ¥43,000, plus the portfolio at face, ¥12,487.01 D16, for a whole-company value of ¥36,500 to ¥55,500. A third off the upper end of that range is ¥37,000 — within ¥500 of its own low end, which is why I call the answer two roads rather than one. Note what I did not do: I did not haircut the portfolio, and I did not go to a trough, because the archive has no trough to go to D46D47D48D49. The three columns are the appraisal; the third off is my protection against the appraisal being wrong.

PABRAI (P53, the fifty-cent dollar; P3, the probability-weighted tree; P1, the floor). *We start from the identical ¥1,720.14 D24 and I arrive at less than half your number, and the difference is entirely in the object. Yours is a discount from an appraisal; mine is half of a conservative intrinsic value, and I computed that value by fading the multiple rather than the earnings — fifteen times on the same operating earnings D24 plus the same portfolio at face D16 is about ¥38,300. Half is ¥19,150, so I write ¥19,100. And note where our numbers actually touch: my second scenario, a de-rate to twenty-five times, gives about ¥55,500 — your growth-credited appraisal is my second-best branch. The reason I halve rather than discount by a third is the floor, and the floor is the whole finding: mark this balance sheet the way a bank would — receivables 379,590百万円 at eighty F538, inventory 85,273百万円 at fifty F542, all property at fifty on 94,718百万円 F570, plus the JustSystems gap of 94,726百万円 F1357, plus the portfolio at face, 3,028,413百万円 D13, without which the same list nets to ¥1,192 — and you have about ¥13,700 a share against a ¥78,970 price F1430. One basis caveat I owe the reader: that JustSystems gap is a parent-only (提出会社) figure E231 and the consolidated sheet already carries the affiliate under the equity method F838, so my crayon takes a gap that is partly double-counted; Li Lu's ¥200 to ¥390 a share for the same stake E259 hedges exactly that, and mine does not. An unbreakable balance sheet at a premium multiple protects the enterprise, not the purchase price.*

MUNGER (M88, the margin of safety; M76, the pari-mutuel; M54, the character gate). And I am the outlier at ¥58,000, so let me be plain about what my number is, because a reader would otherwise assume it is a valuation and it is not. It is a historical-multiple gate, and it is only the first of two. The company prints its own five-year price-earnings band — 43.81 / 40.26 / 46.77 / 37.07 / 32.19× F51F52F53F54F55 — and the bottom of it, roughly thirty-two times reported earnings per share of ¥1,835.63 F40F55, is about ¥58,000. That is where this stock was valued at its own fiscal year-end, four months before the stamp. I decline to pretend to more precision than "back to the cheap end of what the market has actually paid." What makes my number the highest and my verdict no looser than yours is the second gate: I will not buy at ¥58,000 or at any price until a filing explains what 株式会社ティ・ティ is E257 and gives CADENAS a consideration, a goodwill figure or an allocation E256. Warren discounts an appraisal. Mohnish halves one. I am pricing a re-rating and gating it on a disclosure, and the two together are stricter than a number alone.

BUFFETT. Then the spread is not disagreement about the ledger at all. Four of us used the same ¥1,720.14 D24 and the same portfolio D16 and produced ¥19,100, ¥21,000, ¥37,000 and ¥58,000 because we applied four different operations to them: half of a faded value, a bear case with the portfolio marked down, a third off an appraisal, and the low end of a historical multiple with a disclosure gate attached. That is worth writing down as a finding rather than smoothing into an average. When a business's whole premium rests on a mechanism the filing will not describe E267, reasonable methods disagree by a factor of three — 58,000 ÷ 19,100 is 3.04× — and the honest object is the interval, not any point inside it.

---

On what settles it for me — the one-dollar test, and it is the decisive line in this study.

BUFFETT (B72, the retained-dollar test; B86, the payout decision). Everything else here is a matter of degree. This is not. A business earning 105.94% on the capital it uses D32 and 0.561% on the capital it keeps D35 is two businesses, and the second one is destroying what the first one makes. Look at what it took to keep the first one compounding: operating invested capital went from 344,533百万円 to 443,059百万円 D28D27 — ninety-eight and a half billion yen — in a year that earned 445,185百万円 F15. Everything else had somewhere to go and went into short paper. Retention is 78.07% D100; the marginal yen of retained earnings earns roughly nothing after tax; and the company printed the consequence itself, an owner index of 114.9 against a market at 202.4 F165F170. Per-share earnings rose about ten sen for every retained yen F36F40D87D99 — under my bar. The question is not whether this is a wonderful business. It is whether a wonderful business run as a savings account is a wonderful investment, and the five-year answer is printed on page three of its own report.

MUNGER. And I would add the one thing that makes it a watch rather than a pass: it is a choice, not a constraint, and choices reverse. Capital intensity is 2.43% of revenue D82 and capex is already 1.6469× depreciation D81 — there is no reinvestment runway that could absorb 445,185百万円 a year F15. The board has standing authority to repurchase E129 and has never used it D93. Meanwhile dividends per share went ¥200 to ¥550 F121F125 and the payout 15.99% to 29.96% D94D98 inside this very window. A company whose distribution policy is visibly in motion is not a company that has finished deciding.

PABRAI. Which is exactly why I will not be paid to wait at this price. Yield 0.696% D6, no repurchase D93, and nobody in the room with anything at stake — 600 shares across twelve officers F454 against 334百万円 of annual board pay F493F496. My discipline is not "will they eventually fix it"; it is "what do I earn if they never do." At ¥78,970 F1430 the answer is a 2.10% free-cash yield D89 and a multiple that has to survive a decade. At ¥19,100 the answer is that the portfolio alone is two-thirds of my cost D16 and the operating business is nearly free.

---

On why Li Lu alone declines — and it is a knowability claim, not a value claim.

LI LU (L1, the ten-year knowledge bar). Let me name exactly what I could not know, because "too hard" without the sentence is a posture. Three variables decide the outcome and the record answers one. (i) Is the direct-sales structure that produces an 83.02% gross margin D37 defensible? The entire description in seventy-five pages is two clauses: 「当社の強みであるグローバル直販体制を活かし」 and the pursuit of 「世界初」「業界初」 products E43E44 — 直販 occurs exactly twice E267, with no sales-office count, no salesforce headcount, no customer count. (ii) What is the product mix and how does it price? The segment note declines the split outright because one category exceeds ninety percent of revenue E214E262, and there is no pricing or gross-margin commentary anywhere E263. (iii) How much overseas runway is there? Asserted — penetration still small, room large E47 — with no number. And I will name the third silence, because it changes the shape of the first: ファブレス appears once, undefined, inside a risk factor E62E258, while the parent's manufacturing cost is 73.9% materials and 2.4% labour E224. If the physical product is largely bought in, then the whole moat is the sales layer, and the sales layer is two clauses. The decisive variable cannot be answered from the public record. That is not a low value; it is an absent one.

BUFFETT. And I looked at the same three gaps and did not call it too-hard, which is the disagreement worth staging. My reason is narrower than a claim to know: the returns themselves are filed, audited and enormous, and they are not a projection. 105.94% on operating capital D32, 42.49% on the hostile bracket D33, on a business that needs 2.43% of revenue in capital spending D82 and converts 95.87% of earnings to cash D88. I do not have to know why the margin exists to write down what it earned. I do have to be paid for not knowing, and that is what the third off my appraisal is for.

LI LU. Then we differ on what the shortfall costs, and my own item settles it against you. Priced at worst case, the unknowns are not covered. The portfolio is 15.81% of the price D17; 84.19% is paid for earnings power whose mechanism is undisclosed. I ran the arithmetic anyway, and a student should see it: twenty times ¥1,720.14 D24 is ¥34,403, plus the portfolio at ¥12,487.01 D16, plus about ¥203 of fair-value uplift E200 and ¥200 to ¥390 for the JustSystems stake E259, whose 94,726百万円 quoted-versus-carrying gap is a parent-only figure F1357 that the consolidated equity method already partly carries F838, is a conservative owner value near ¥47,400 — and fifty cents on that dollar is about ¥23,700, below any price in the five-year record F177. I publish that and issue no threshold. A buy-below is a promise that at that price I would act, and I would not, because the twenty-times multiple is defended by the margin band D45 and the balance sheet D18 and by nothing whatever about why the margin exists. Charlie's watch and my too-hard are the same reading of the facts with a different rule about what to do when the decisive disclosure does not exist.

MUNGER. And I hold the other rule openly. My test is whether the decisive variables are nameable, not whether they are currently known. They are nameable — an operating margin that holds through a down year; a filing that identifies its own 15.07% holder F372; a consideration figure for CADENAS E256 — and each resolves in a document with a date on it. A verdict of too-hard says the question cannot be posed. This one can. It just cannot be answered today, and the correct response to that is a demanding price plus a gate, not silence.

---

On the Claude lens's own scorecard: the registered consequence landed, the margin did not, and the moat test could not be run at all.

CLAUDE (C1–C6, the outside view; C109, the surprise ledger). Report the hits and the misses at the same volume, so both first. The hit that mattered was not a number, it was a consequence. Figures-blind, I put the pile at 8–20% of market capitalisation with a ~14% central and wrote the registered consequence in advance: this is not a net-cash value case; stripping the cash moves the multiple by roughly a tenth, not by half, and no floor thesis, no adjusted-NAV and no unlock arithmetic can carry a verdict here. The ledger printed 15.81% D17 and an ex-cash multiple of 38.6497× against 43.0207× reported D26D5 — 0.899×. Registering the consequence rather than only the number is what kept a floor thesis out of my file, and it is the single thing I would keep from this run. The miss that mattered is the operating margin: 50.95% D39 against a registered band of 25–48% with P(≥50%) = 0.15. The truth is above my whole band, in a direction that runs against the contamination I declared, so the blinding held even where the estimate failed.

CLAUDE, continued. And the miss I would rather bury. I put 0.40 on this company underperforming its own index over five years and it happened by 87 points — total shareholder return 114.9 against a dividend-inclusive TOPIX of 202.4 F165F170 — with both of my stated conditions holding: a large and growing pile, 2,764,019 to 3,028,413百万円 D14D13, and an operating business compounding above the index at 11.55% D50. I wrote the retention identity out as arithmetic at the outside-view stage and then assigned it 0.40. It is not a theory here; it is a completed experiment with a printed result, and the mechanism is printed beside it — the multiple fell 43.81× to 32.19× F51F55 while earnings per share rose 46.8% F36F40. I underweighted my own argument in the direction of being kind to a quality franchise, which is the specific bias this instrument exists to catch.

CLAUDE, on the test that could not be run. MOAT-REAL is not scoreable, and that is a finding about the archive, not a hedge. I converted the entire moat question into one quantitative test at the outside-view stage — gross margin varying by under 400bp across the available series including the weakest customer-capex year — and pre-registered p = 0.70. The five-year highlights table omits both 売上総利益 and 営業利益, so gross margin exists for two years D36D37; and the archive contains no down year, revenue having risen in every transition D46D47D48D49. Both of my too-hard triggers fired on the moat. I record it unresolved: not confirmed by two good years, not refuted. It does not drag the verdict, because the price fails the owner arithmetic at the highest margin ever printed D39 — but a reader should know that the proposition carrying the most weight in the bull case was, in this study, untestable.

PABRAI. And I reached the same wall from the other side without any pre-registration, which is the useful corroboration. I could not name or date the spread E267, so I would not extrapolate it, and I priced the consequence at half of a faded value. Li Lu could not name it either and declined to price it at all. Three lenses hit one absence and produced a threshold, a threshold and a refusal. The absence is doing more work in this study than any number in it.

---

On the jury: four of six items diverged, and every divergence subtracted.

CLAUDE (C33, C35, C84, C89, C98, C102; C108, one-directional consumption). Six items were run past three independent within-lens selves each. Four came back divergent, and every one of the four moved a published figure down or a claim weaker. On the bear case, three selves returned ¥15,800, ¥17,400 and ¥21,300 against my own ¥26,000 — all three below me, because each substituted a real revenue decline where I had held revenue flat, the archive containing none D46. The threshold moved from ¥26,000 to ¥21,000 and the band widened to ¥16,000–¥33,000. On the anchoring test, three selves given the ledger with every market row removed returned ¥18,000, ¥21,500 and ¥33,000 — an 83% spread driven entirely by whether growth is credited, with my price-aware figure inside their range at 33% of the stamp F1430, nowhere near the proximity band that would suggest anchor gravity. On the short case, gradings came back moderate, strong and moderate, and I adopted the more cautious; two selves supplied a computation I had not made and I took it — the ordinary-margin recovery is 96.6% interest income and the currency swing F639F640F643F648F292F294. On locating the bet, the framings diverged and I published the wider joint form rather than choosing. The two convergent items — the private-owner yield and the epistemic inventory — are recorded as "no divergence detected" and cited as support for nothing. Convergence among copies of one reasoner is not evidence.

MUNGER. Then say what the price-blind spread is evidence of, because the temptation is to call it noise in the instrument. It is not. Three copies of one reasoner, handed the identical rows with the market stripped out, valued this company across an 83% range because the rows do not determine the answer — the answer turns on whether a decade of growth is credited, and the filing publishes no forecast E40, sets no objective indicator E82 and prints an operating margin for two years D38D39. That is a statement about the company's disclosure, not about the jury. And it is the same statement Li Lu made in different words.

The bet, located

Strip the agreement away and one falsifiable proposition is left, in two legs that multiply. Leg one: is 50.95% D39 the steady state, or is 41.75% (45,983 F292 ÷ 110,144 F289) the truer forward signal? Leg two: does the 43.0207× multiple D5 hold at exit? — on which the archive contains a completed experiment whose answer was no, 43.81× to 32.19× F51F55 with the owner index at 114.9 against 202.4 F165F170. The Claude lens puts 0.30 on the bull side of the joint proposition, matching all three of its selves. The bull mechanism is on the record: a gross margin of 83.02% D37 against 83.81% the year before D36 — the only two years the filing prints, so a level and not a four-year persistence claim — while revenue compounded 1.5484× over four years D51, a 4.14pp ordinary-margin band D45, 42.49–105.94% on operating capital D33D32, and an overseas share at 66.64% and rising D53D54 on a build-out management says is still early E47. The bear mechanism is equally on the record: an incremental operating margin nine points below the average F292F289D39, both printed margin lines falling, gross D36D37 and operating D38D39, both cost lines outgrowing revenue F631F632F635F636D49, receivable days at 118.5 D66, and a retention machine that has never once returned a share D93D100. It resolves in the FY2027/3 yūhō, on a date, in a document anyone can read.

Prediction-vs-actual: VOID

This was an autonomous headless cycle. predictions.md carries void: no-human-prediction, and every prediction verdict is null by design — no practitioner was present at run time, and a blind call is never forged to fill the slot. No prediction-vs-actual scoring applies to this study, and it renders as void in docs/calibration.md, visibly distinct from a practitioner who declined. The five profile verdicts above still count in full for verdict accounting; only the human calibration half is skipped. The Claude lens's own figures-blind priors are a separate instrument and are scored, above and in its profile run — including the operating-margin miss, the 0.40 index-underperformance miss that happened by 87 points F165F170, and the MOAT-REAL registration that could not be scored at all because the disclosure it needed does not exist.

Self-distance note. The Claude lens holds one of the five verdicts compared above (watch, implied buy-below ¥21,000) and wrote this synthesis; it also built the reconciled figure table and the evidence ledger all five lenses consumed, and the red team ran on the same model family. That is an unusual concentration of authorship — the answerer, the ledger-builder, one of the five voters, and the adversary are the same system. Read the synthesis with that in mind. Three partial mitigations are on the record and should be weighed for what they are worth: the Claude lens ran figures-blind through its outside-view stage, so its priors were registered and pushed before any magnitude was visible and are scored against the ledger above — with two registered claims scored as failures and the largest one scored as unrunnable; its jury of selves produced divergences on four of six items, every one of which moved a published number down or a claim weaker; and the Munger run was executed on a different model, which is the only genuine cross-model check in this study.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp. Three lenses issued explicit thresholds — Pabrai ¥19,100, Buffett ¥37,000, Munger ¥58,000 — and the Claude lens publishes an implied threshold of ¥21,000 with a published band of ¥16,000–¥33,000. Munger's threshold is conjunctive: the price gate alone does not clear it; a disclosure gate on 株式会社ティ・ティ E257 and the CADENAS consideration E256 must also open. Future scoring must treat it as such.
• Li Lu issues no number. He computed a conservative owner value near ¥47,400 and a fifty-cent entry near ¥23,700 from ¥1,720.14 of operating earnings D24 and ¥12,487.01 of portfolio D16, and declined to publish either as a threshold, because too-hard is a statement that the decisive variable is absent from the record E267E262, not that the price is too high.
• pass / watch / too-hard are recorded but unscored in any future review. Four verdicts here are watch and one is too-hard; every named price sits far below the ¥78,970 stamp F1430, and the highest of them, ¥58,000, is roughly the low end of the company's own printed five-year multiple band F55F40.
• 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. There is no split inside the five-year window — 発行済株式総数 has been 243,207千株 at every one of the five year-ends F101F105, the only split on the capital-history table being the company's own two-for-one dated 2019-11-21, before the window E94. The denominator used throughout is 242,525,136 shares net of treasury D1, being 243,207,684 issued F338 less 682,548 treasury F427, which reproduces the filing's own per-share figures against a weighted denominator of 242,525,222 F1056.
• Every profitability figure in this thesis states which of the two businesses it belongs to. Blended: ROE 13.53% D29, ROIC 12.68% D30, P/E 43.0207× D5, P/B 5.5170× D4, dividend yield 0.696% D6. Operating business only: ROIC 105.94% D32 with the conservative bracket at 42.49% D33, after-tax operating EPS ¥1,720.14 D24, ex-cash price ¥66,482.99 D25, ex-cash P/E 38.6497× D26, EV/EBIT 27.0643× D90. Portfolio only: 3,028,413百万円 D13, 82.50% of assets D15, ¥12,487.01 a share D16, 15.81% of market capitalisation D17, yielding 0.561% D35. The blend is not a valid basis for either half, and no figure above mixes them silently.
• Two years is not a margin history. 売上総利益 and 営業利益 are printed for FY2025 and FY2026 only; the five-year highlights omit both. Every gross- and operating-margin claim in this thesis D36D37D38D39 rests on two observations, and the five-year proxy is the ordinary margin D40D41D42D43D44, which contains non-operating income by construction.
• This is a VOID study for prediction-vs-actual purposes only; the verdict accounting above is unaffected.

Red team

A consensus red team was dispatched — four of five lenses had converged on watch, a non-decline — to argue pass: that a watchlist slot spent on 6861 is a slot wasted. It concluded plainly that its own case does not defeat the consensus, and it did something better than that first: it reported two of its own best arguments turning against it.

Its three strongest surviving points, engaged by name.

1. "The financial-income screen." FY2026's margin recovery is 96.6% non-operating, and the operating business decelerated behind it. Ordinary profit grew 74,745百万円 F294 against operating profit's 45,983百万円 F292; of the 28,762百万円 residual, 27,772百万円 is interest income (16,249 F640 less 8,968 F639) plus the currency swing (16,270 F643 plus 4,221 F648) — 96.6%. In the same year the incremental operating margin was 41.75% (45,983 F292 ÷ 110,144 F289) against a 50.95% average D39; both printed margin lines fell, gross D36D37 and operating D38D39; and both cost lines outgrew revenue F631F632F635F636D49. Its corollary is sharper still: 24.2% of the year's interest and dividend income was accrued and not received — 16,368百万円 F775 against 12,406百万円 collected F793, a 3,962百万円 gap that was 611百万円 the year before (9,083 F774 less 8,472 F792). Fully conceded, and it is the point the synthesis adopts as binding. The red team's own rebuttal is honest and partial: operating profit still grew 8.4%, the margin fell 0.96pp off a level essentially no industrial company reaches D38D39, an incremental margin of 41.75% is above what most businesses earn on their average yen, interest income is cash on shareholders' assets rather than an accounting screen, and SG&A outgrowing revenue by 0.56pp on headcount up from 12,261 to 12,784 F79F80 and R&D at 2.81% of revenue D55 is what a deliberate direct-sales build looks like. And it convicted itself of its own charge: "the red-team here commits the exact sin it charges in point 3 — it builds a trend claim on a two-point operating series." What survives is the arithmetic; what does not is the inference that the operating business is deteriorating.

2. "The stamp is an all-window high, and the re-rating happened after the numbers were known." ¥78,970 F1430 is above every 最高株価 the filing prints — ¥76,210 / ¥62,210 / ¥75,040 / ¥77,400 / ¥68,610 F171F172F173F174F175 — the highest being ¥77,400 F174, giving 78,970 F1430 ÷ 77,400 F174 = 1.0203×. Worse: the filing's own 株価収益率 of 32.19× F55 on the 2026-03-20 price implies a fiscal-year-end price of 32.19 F55 × ¥1,835.63 F40 = ¥59,089, so the stamp is 33.65% above the price implied by the filing's own multiple, in roughly four months, against unchanged reported earnings. On the ex-cash basis the move is larger — about 27.09× at the year-end price against 38.6497× at the stamp D26D16D24. Conceded on the arithmetic, and it must be stated where the reader can see it. But the red team's own answer is the right one: this is a fact about the price, not about the business, and pass is supposed to be a judgement about the business or about a price no plausible future can fix. The 第57期 range alone ran ¥68,610 to ¥49,780 F175F180, a 1.378× spread inside twelve months. The point argues for patience, not for deletion — and it is precisely why Munger's threshold is the low end of the printed multiple band F55 rather than an appraisal.

3. "The retention machine has already returned its verdict." 78.07% of five years' earnings retained D100 into paper yielding 0.561% D35; no buy-back in any year, ever D93E101, with standing articles authority unused E129; and at the parent (提出会社) 522,000百万円 formally swept into 別途積立金 — 250,000 F1244 plus 272,000 F1294 — while distributable 繰越利益剰余金 rose only 54,576百万円, 323,481 to 378,057百万円 F1229F1323. Meanwhile ROIC including the portfolio is 12.68% D30 against 105.94% excluding it D32, and management chose to make the portfolio larger, 2,764,019 to 3,028,413百万円 D14D13. Conceded on every fact, and it is where four of the five lenses independently landed. The 別途積立金 sweep is the sharpest single disclosure in the red team's file and no lens computed it: a formal, deliberate quarantine of retained profit from distribution, on the parent's own statement of changes in equity.

Its two self-corrections, recorded before its case is scored, because both narrow the bear argument this study started with.

Correction 1 — the TSR shortfall cuts the other way. The red team's single most quotable number is the owner index of 114.9 against a dividend-inclusive TOPIX of 202.4 F165F170. On inspection it is substantially a de-rating: the printed price-earnings ratio fell 43.81× to 32.19× F51F55, −26.5%, while earnings per share rose 46.8%, ¥1,250.83 to ¥1,835.63 F36F40. In its own words: "A price that has been losing ground to its own business for four years is the classic reason to keep a name on a list, not to remove it. My single most quotable number argues for watch." Recorded unedited. It is the same fact Buffett uses for the one-dollar test D32D35 and the red team uses against itself — and both readings are correct, because retention drag and multiple compression are separable causes of the same shortfall.

Correction 2 — "will not distribute" is refuted inside the window. Dividends per share went ¥200 to ¥550 F121F125, a 2.75× rise, and the consolidated payout ratio 15.99% to 29.96% D94D98. "That is a company changing its behaviour... I would have had to suppress those rows to make the claim stand." Recorded unedited. The retention criticism survives on the buy-back and on the 別途積立金 sweep D93F1244F1294; it does not survive as a claim that nothing is moving.

The red team's own verdict, in its words: "The pass case can be argued forcefully, and it is honestly available on price and on disclosure. It does not defeat the consensus, and I say so plainly." Its reasoning: pass requires either that the business is not good enough to re-underwrite or that something disqualifies it, and neither survives contact with ex-portfolio returns bracketed at 42.49% to 105.94% D33D32, an 83.02% gross margin D37, 95.87% cash conversion D88, zero debt D18 and revenue compounding 11.55% D50. "Every point I made above is a price point or a monitoring point. Not one of them is a business-quality point, and I could not construct one that the record would carry."

And its surviving leg terminates somewhere other than where it started. "The one leg that genuinely holds is the disclosure leg — no operating-margin history, no product split, no PPA on an eleven-subsidiary acquisition, no critical estimates against a ¥3.03tn portfolio D13, no related parties against a 22.79% founder-linked block. But that argument terminates in too-hard, not pass. It says the instrument cannot see the thing clearly; it does not say the thing is not worth seeing. The fifth lens reached exactly that conclusion, and my strongest surviving argument reinforces it rather than the pass case. If the red-team moved any verdict, it would be to widen the too-hard minority — not to convert the consensus." Adopted on the record. Read the four watch verdicts accordingly: the consensus survives, its price discipline is unchanged, and the strongest adversarial argument available against it strengthens Li Lu's verdict, not the pass case. The red team also fixed the boundary condition its own point 4 leaves behind: a twenty-times ex-cash entry is ¥46,890 (20 × ¥1,720.14 D24 plus ¥12,487.01 D16), a 40.6% fall from the stamp F1430 — below the 第57期 low of ¥49,780 F180 but above the 第54期 low of ¥44,880 F177; a fifteen-times entry has never occurred in the archived record. Three of the four published thresholds sit below that line.

One point the red team raised and lost outright, recorded because it did the work honestly. Its receivables lead — days 110.4 to 118.5 D66, the allowance up 1,426 to 2,135百万円 F545F546, payables down 14,890 to 10,890百万円 F591F592 — is real but small: the allowance increase is 709百万円 against 1,169,289百万円 of revenue F5, and "I record explicitly that the inventory lead fails: inventory days fell, 165.8 to 156.8" D65. An adversary that prints its own failed sub-argument is worth more than one that wins.

What would change our minds

Pre-registered falsifiers, per lens, taken from each profile run's falsifier: line. Future review notes score against these, not hindsight. Four of the five converge on the same document — the FY2027/3 yūhō (第58期) — and on the same two observables: the operating margin and the disclosure.

• Buffett (watch, buy below ¥37,000). The watch is wrong — and the disclosure discount is one nobody else ever charges — if consolidated ordinary margin holds at or above 50% D40 and revenue grows in each of the next three fiscal years D49 while payout stays below 35% of net income D100, and the filing still names no product-group revenue E262 and still says nothing about who 株式会社ティ・ティ is E257, and the shares nonetheless compound ahead of dividend-inclusive TOPIX F170 over that window. That combination would say the market prices the opacity at zero and always has, and that the ¥37,000 level was a fantasy priced off an appraisal nobody else uses.
• Munger (watch, buy below ¥58,000, conjunctive). Upgrade requires both, not either: (1) price back near the low end of the company's own printed five-year multiple band, roughly ≤32× on current earnings per share, about ¥58,000 E8F55F40, against a stamp already 15.10% above the fiscal year's own trading high D8; and (2) a subsequent filing that actually explains what 株式会社ティ・ティ is and its relationship to the founder E96E97E257, and gives CADENAS Technologies AG a disclosed consideration, goodwill figure or purchase-price allocation E256. Absent either, the position stays capped at watch. It falls to pass if a later filing shows a related-party transaction with 株式会社ティ・ティ or the foundation that this filing did not flag, or shows interest-bearing debt or a covenant of any kind on a balance sheet that today has none D18.
• Pabrai (watch, buy below ¥19,100). A year-on-year revenue decline — which has not occurred anywhere in the archived window D46D47D48D49 — in which the operating margin D39 nonetheless holds inside the 4.14pp five-year ordinary-margin band D45 and ex-portfolio ROIC D32 stays above the 42.49% conservative bracket D33. That would prove the earning power trough-proof, make 38.6497× ex-cash D26 a price for demonstrated durability rather than an untested-peak bet, and force this decline to be re-underwritten on multiple alone.
• Li Lu (too-hard, no threshold). Any one of three overturns it. (1) A future yūhō discloses either a revenue split by product group or any sales-organisation metric — office count, salesforce headcount, customer count — and the disclosed structure shows the 83.02% gross margin D37 resting on something a competitor cannot buy from the same partner factories E267E262. (2) A filed year shows revenue down more than 15% with the ordinary margin still above 50%, which would turn the untested 4.14pp band D45 into evidence of durability rather than evidence of a good five years. (3) The Company states a capital policy for the 3,028,413百万円 portfolio D13 — a target level, a definition of surplus, a timetable — beyond the single sentence it now gives E197, and payout clears 60% D98 or a buy-back is resolved under the standing articles authority E129. Conversely the verdict is confirmed rather than overturned if the portfolio share of total assets rises above 85% D15 while payout stays under 35% D98.
• Claude (watch, implied buy-below ¥21,000). Five, each thresholded with a named resolving document. (1) FY2027/3 operating margin below 47.0% D39 → pass at any price above ¥40,000. (2) FY2027/3 incremental operating margin at or above 50% → the 41.75% reading F292F289 was an artefact and the threshold rises toward ¥27,000. (3) A repurchase of 100,000百万円 or more resolved and disclosed by FY2029/3 D93 → the capital-allocation discount halves and the threshold rises about ¥4,000 to ¥6,000. (4) Revenue declining year on year in FY2027/3 or FY2028/3 with the operating margin holding within 200bp D39 → MOAT-REAL discharged on the test this archive could not run D36D37, and the threshold rises materially. (5) A consolidated goodwill balance above 50,000百万円 or a CADENAS impairment in FY2027/3 E256 → the M&A prior confirmed early and the threshold falls. Review by 2027-06-30.

The single observable all five converge on is where the FY2027/3 operating margin settles against the 50.95% just printed D39 and the 41.75% incremental reading F292F289 — together with whether the filing closes any one of the three disclosure gaps E256E257E262. Both resolve in one document, on a date, and neither requires anyone's judgement.

What this taught the checklists

Queued for the next study (F2 revision proposals; see docs/process/evolution.md), attributed per lens. Checklist versions froze at this study's stamp commit; these take effect next time.

• Buffett — five. B99 must name the discount rate's source or say there isn't one: the item instructs valuation "discounted at the long-term government bond rate," and a single-company ledger may contain no market interest rate at all — this one does not, the only rate available being the company's own realised portfolio yield of 0.561% D35, which is not a discount rate. Propose: "where the ledger carries no market long-bond rate, state the hurdle as an explicit judgment with its number, and do not substitute a company-specific realised yield for it." B84 should widen from a blocker map into a control map: the problem here was not a blocker to value realisation but an undescribed register — a 15.07% holder named once and described nowhere F372E257, a foundation at the company's own address E257, a founder attending every board meeting F439 while absent from the roster E266. Propose: "list every holder above 5%, state the filing's own description of each, and record explicitly where the filing gives none; an undescribed holder above 10% is a fail regardless of whether anything is being blocked." B104 should widen its trigger beyond holdcos: the item is written for "holdcos, insurers, and conglomerates," and this was a single-segment operating company with 82.50% of assets in a securities portfolio D15; propose adding "…or any company where non-operating assets exceed roughly 25% of total assets." B42 should ask for the maintenance/growth capex split, not just the level: capex ran 1.6469× depreciation D81 with the excess named as a logistics centre E89, and the triangulation ladder handles undisclosed capex but not a disclosed capex whose split is undisclosed. B67 should ask for the sign of the float: the item assumes float is money held, and this company supplies it — 118.5 receivable days D66 against 20.0 payable days D67.
• Munger — three, all bright-line tripwires. M82 should add a first-order mechanical check: "is a birth date or age disclosed for every named officer, including any honorary or emeritus title?" The founder is the only named officer without one E266, and this study found the gap only by comparing the twelve officer bios on the roster by hand E116. M46's Good criterion needs an explicit cross-check: where a filing states a percentage-linked compensation design — here 60% to 75% of the base tied to operating profit F485F486E163 — but the reported table shows 0% performance-linked for every officer E167, that mismatch should be named as a documented contradiction requiring resolution before the item can pass, not folded into a general judgment call. It can be entirely consistent with honest accounting through a one-year-lagged formula E162 and still be a transparency failure worth its own tripwire. M53 should distinguish "hard to compute a return on acquired capital" from "the filing discloses literally zero consideration, goodwill or allocation for an acquisition that changed the consolidation scope by eleven subsidiaries" E180E256. The second is a brighter-line case and the item's present wording treats them the same.
• Pabrai — two, both failure-derived. New item P82, the excess-capital return mechanism at a great business: when a business earns far above its cost of capital on operating capital but retains most of its earnings in financial assets, is there a filing-visible mechanism — a stated target cash level, a definition of surplus, a repurchase authority actually used, a timetable — by which that capital reaches owners? Fail condition: retained capital compounding at a rate an owner could beat with a savings account, with no stated terminal use. Rationale: a 105.94% ROIC operating business D32 whose retained 78.07% D100 earns 0.561% D35, with the whole disclosed policy one sentence E197 and no terminal use stated E261 — P17 asks whether value accrues to owners and P46 asks about counter-cyclical allocation, and neither catches this. P1's Good text should warn that net cash is not automatically a floor either, when it is a small fraction of price: "where net cash and securities cover less than roughly a third of market capitalisation, treat the floor as the stress-test value alone and state that fraction explicitly." Here zero debt D18 and a 94.57% equity ratio D62 sit against a portfolio at 15.81% of price D17, and without the explicit fraction "no debt, huge cash" reads as P1 passing when it plainly fails.
• Li Lu — three. L21 needs a value test, not only a book test: the retained-earnings audit's stated metric passed here by arithmetic, because retained yen become book yen whether invested at 100% or at 0.561% D35; propose a second required computation, incremental return on retained capital, with an explicit fail when it falls below the company's own existing return on invested capital D32. L26 should split its two halves into separately gradeable answers: cross-holdings, parent-child structure and minority treatment pass cleanly here — policy holdings 814百万円 unlisted F502 plus 5,532百万円 listed F504 under a stated in-principle-none policy E170, no parent E237, one share class — the voting-rights table carries 普通株式 and nothing else F411F412F414F415 — while the TSE capital-efficiency response fails absolutely, 資本コスト, PBR and any return target occurring zero times E260E82. Structural subordination is disqualifying; reform silence is informative; a single verdict loses the distinction. The archive-depth rule needs a "verification-impossible" grade distinct from data-insufficient: eight items failed here for one structural reason — no competitor, customer, affiliate or court filings exist in a single-company archive — and eight identical data-insufficient answers overstate the number of independent failures while understating the single structural one.
• Claude — four checklist sharpenings, plus two library entries. (1) A new §3 item, "required-return discipline." C104 found that this verdict rests on the pair (required return, credited growth), which appears in the checklist only as a parenthetical inside C35; at a 7% hurdle with no growth credited the price fails by roughly two-thirds, and at 5% with a decade of 8–10% compounding it clears. The item should ask what required return is used, what it is anchored to, whether it is the same rate used in the previous study of a comparable-risk company, and what the verdict looks like at ±200bp — otherwise the lens's buy-belows are not comparable across studies and the calibration record cannot separate a judgement about a company from a drifting hurdle rate. (2) C33 must name the substitution when no decline exists. The item instructs the analyst to apply "a revenue decline already observed in the company's own history"; the archive has none D46D47D48D49, and three jury selves and the lens all silently substituted something different, producing a 68% spread in the bear value. Require the substitution to be stated with its magnitude and source, and the bear value reported as a range across at least two substitutions. (3) A new sub-test in C9/C13 — the incremental margin as the moat test when no downcycle is available. The pre-registered moat test needed a gross-margin series through a downcycle and the archive had neither; the incremental operating margin reached the same mix-erosion question from a single pair of years and produced the study's most informative adverse finding, 41.75% against 50.95% F292F289D39. C13 computes it as a scale diagnostic; it should be promoted explicitly as the substitute moat test. (4) An archive-adequacy pre-check in §1 (C6). Every pre-registered proposition should name the specific disclosure that will resolve it, and the figures-blind packet should confirm that disclosure exists in the archive — so a pre-commitment cannot be spent on a test that is unrunnable by construction, as MOAT-REAL was here.
• Claude library (v0.1.0 → next), staged for the approval contract. Class-level, N=1, no rate claimed: a Japanese franchise with a top-decile operating margin, zero interest-bearing debt, and a financial portfolio exceeding half of total assets, retaining most of its earnings into low-yield paper. KEYENCE (6861) is the first member and its outcome on the class's key axis is measurable: five-year total shareholder return of 114.9 against a dividend-inclusive TOPIX comparator of 202.4 F165F170, an 87-point shortfall over a period in which revenue compounded 11.55% D50 and the ordinary-margin band was 4.14pp D45 — decomposing into a multiple that fell 43.81× to 32.19× F51F55 against earnings per share that rose 46.8% F36F40. The lens further proposes an explicit amendment to the Workman (7564) candidate axis, and states the obstacle rather than paraphrasing around it. That axis reads, verbatim in profiles/claude/library.md v0.1.0, "High-equity Japanese specialty retailer, declining ROE while cash accumulates" — and KEYENCE is not a specialty retailer, so as written it cannot join. What KEYENCE does satisfy is the axis's mechanism: equity ratio 93.5% to 94.6% F41F45 against ROE 14.85% to 13.53% F46F50, equity accumulating while ROE declines. The proposal is therefore to relax the sector qualifier — from "specialty retailer" to any high-equity Japanese company — on the ground that the mechanism the axis tracks is a capital-allocation identity (retained yen entering the equity base faster than they earn) that has nothing to do with what the company sells; if the practitioner declines the relaxation, the axis stays at N=1 and KEYENCE joins only the new class above. Recorded either way as the lens's proposal, subject to the library approval contract, not as an accepted entry. Case-level, so it never anchors a future outside view: "Before pre-registering a quantitative test at the outside-view stage, verify from the packet that the series the test needs is actually printed in the archive." Source: this study — MOAT-REAL was registered on a multi-year gross-margin series that the five-year highlights table does not carry D36D37.
• A process note that belongs in the record. The study's extraction brief initially specified a 31 March fiscal year-end. KEYENCE's year runs 自 2025年3月21日 至 2026年3月20日, and one of the two blind figures passes caught the error against the source before the ledger froze; every balance-sheet, share-count and employee-count date in this study is 03-20. The lesson is the gate's, not the analyst's: a dual-blind extraction pass caught a date error that would have silently mislabelled 1,434 rows, and the cost of the second pass was repaid by that one catch. Separately, and consistent with the rule adopted after study #24: no ordinal or comparative claim about the record — "the first", "the highest yet", "the Nth" — enters a stamp, thesis or deck without being checked against docs/calibration.md and the repository first. This thesis drafted three such claims, not one, and the rule as written caught only the first. The checkable one — that twenty-five prior studies produced zero buy-below verdicts — was verified against docs/calibration.md before it was written. The other two were not checkable there at all: "the finest balance sheet I have seen in this hunting ground" and "the largest pile of idle capital I have examined" are comparisons across the whole studied record, and docs/calibration.md records verdicts and predictions — not balance-sheet quality, not cash-pile size. Both were restated before publication as claims about this company: zero interest-bearing debt D18 against a 94.57% equity ratio D62, and idle capital at 82.50% of assets D15. The lesson belongs to the rule, not to the lenses: it must fire on comparative superlatives in a profile's voice — "the finest", "the largest I have seen" — and not only on numbered ordinals, and it must first ask which file would settle this before asking whether that file was consulted.

---

Corrections