FUJI SEAL INTERNATIONAL, INC. (7864): The Best Business in the Funnel, at Book — and the Year Was Borrowed

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	—	B26+B24; B93+B42; B72+B59/B54
Munger	watch	—	M46/M54/M83; M18/M34; M60
Pabrai	watch	¥1,600	P1; P53+P76; P17
Li Lu	watch	—	L18; L43; L46+L51
Claude	watch	implied ¥1,250	C19; C23/C44; C40

Five of five lenses reached watch — the fourth unanimous watch in the record, and the second in this batch. フジシールインターナショナル is pick #2 of the "cheap AND good" batch, and it was selected to invert pick #1's test. 東京鐵鋼 put a low multiple against a return whose durability was the open question; this name puts a durability far less in doubt against a price that is merely fair. Ordinary profit more than doubled in four years on rising revenue — ¥10,600M to ¥22,009M F6F10 against revenue of ¥170,321M to ¥217,752M F1F5 — free cash flow was positive in four of the five years D111D112D113D114D115, the equity ratio is 71.3% F45, net cash is ¥25,249M D89, and the whole company was offered at 1.04× book D6 and 8.13× reported earnings D8. On the raw arithmetic of business quality it is the best name twenty-four studies have surfaced. It still drew only watch, from all five. Pabrai carries the only explicit price, at roughly half the midpoint of his own value range; the Claude lens publishes an implied threshold of ¥1,250 with a band of ¥1,080–1,330 D1D118, and reports one jury divergence — C84, two of three selves calling the name a compelling avoid and one calling the long the winner — whose one-directional consequence capped its verdict at watch regardless of arithmetic. So the answer to the question this pick was built to ask is: quality does not clear the bar when the price is merely fair, and the binding constraint is neither price nor quality alone but their conjunction.

Two facts decided it, and both are about what the reported record is made of. The first is that the year is borrowed. FY2026 net income of ¥20,655M F15 carries a ¥4,984M gain on the liquidation of Fuji Seal Switzerland AG F1063E244 — an amount recycled out of the translation reserve as a ¥4,984M 組替調整額 F1280 and added straight back in the cash-flow statement F1134, creating no cash at all; comprehensive income tells the undramatic version of the same year, ¥18,776M to ¥20,715M, up 10.3% F19F20. On top of it sits an effective tax rate that fell from 31.73% to 15.09% D86D85 against a statutory 30.59% F1415, driven by the tax effect on investments in subsidiaries E262. Strip both and the year is roughly +10% on revenue up 2.5% D39; the headline 8.13× P/E D8 becomes 10.72× ex-gain D11 and 10.98× on the company's own guidance D15. Separately, ¥2,078M of the ¥3,686M rise in ordinary profit — about 56% — is a single foreign-exchange swing from a ¥779M loss to a ¥1,299M gain F1053F1046F9F10. The company itself excludes the gain when it restates its payout ratio, printing 20.9% headline beside 30.2% adjusted F703F704E138, an adjustment implying ¥6,356M excluded D136 — larger than the pre-tax gain itself. And it guides net income down 25.9% for FY2027 F390, a decline the filing never explains in a sentence anywhere in 122 pages E113.

The second is that the cash does not fully arrive, and the floor is scheduled to be spent. Cumulative five-year free cash flow is ¥30,784M D118 against cumulative net income of ¥56,117M F11F12F13F14F15 — 54.9%. The operating line is not the problem: operating cash flow converts at 1.05× operating profit D125 and 1.38× ex-gain net income D124. The investing line is, with capital spending at 2.02× depreciation D119 against 0.76× the prior year D120 and construction in progress running from ¥5,328M to ¥13,791M F948F949. Against that, management has published investment frames of ¥43.5bn to FY2027 and ¥56.5bn to FY2031 F274F275 — roughly ¥100bn against ¥30.8bn of five-year free cash flow D118 — and has written down its intention to raise the financial-leverage effect with a view to using borrowed capital in order to reach a price-to-book target of 1.5× E64F273. The ~¥25.2bn net-cash cushion the downside rests on D89, 15.0% of market value D93, is therefore not a permanent floor; it is a disclosed source of funds for a spending programme, with a stated intention to add debt on top.

The business

株式会社フジシールインターナショナル prints the plastic sleeve that is dropped over a bottle of yoghurt drink or a jug of detergent and shrunk tight with heat before the package leaves the filling line. That sleeve is destroyed with the package; the next bottle takes the next sleeve. The group makes and sells shrink-sleeve labels, pressure-sensitive ("tack") labels and flexible soft pouches, together with the labelling and packaging machinery sold alongside them E40, through 26 consolidated subsidiaries F194 and 5,670 employees F80 producing locally in Japan, the Americas, Europe and Asia. It began in October 1958 in Osaka as 株式会社藤尾製作所, making cap seals E9; shrink-label manufacture started at the Nabari plant in September 1981 E12; packaging machinery followed with the founding of 株式会社フジアステック in October 1990 E14; the Americas business came from the 2001 purchase of Owens-Illinois Labels, Inc. E18, Europe from the 2012 purchase of Pago Holding AG E30, a North Carolina plant was built in April 2023 for US shrink-label demand E37, and pharmaceutical packaging was bought the same year E38. The listed entity has been a pure holding company since October 2004 E24, with no parent above it E307 and no equity-method affiliates at all E227.

The economic idea is the pairing, and it is genuinely there. The label is a consumable destroyed on every unit the customer fills; the applicator that consumes it is qualified into the customer's own production line. The filing's revenue-recognition policy is the cleanest evidence for it: labels are recognised on arrival or shipment, but machinery revenue is recognised only when the customer accepts the machine after installation and trial operation inside the customer's plant E267 — and the auditor made that period attribution one of two Key Audit Matters precisely because the per-transaction machinery amounts are large relative to labels E293. There is ¥15,401M of machinery backlog allocated to remaining performance obligations, to be recognised over roughly one to three years F1482E269, and ¥8,221M of customer advances sitting as contract liabilities, up from ¥7,283M F1244F1243E268. Dedicated machinery manufacture-and-service subsidiaries exist in every region E44E46E48E49. Changing label supplier is not a purchasing decision; it is a line-requalification decision.

But the qualified half is the small half, and the filing will not tell you what it earns. Machinery is ¥23,170M against ¥194,581M of label sales F1821F1820 — 10.64% of external revenue D59 — against shrink labels at 61.68% D62 and soft pouches at 13.54% D63. The margin on the machinery line is nowhere separated, because the group declines to give production and order-backlog data by segment on the express ground that it 「単一事業である」 — that it is a single business E105 — while in the same document defining four geographic reportable segments reviewed by the board for resource allocation E270. That contradiction is the study's single most consequential disclosure fact: the moat's whole mechanism sits inside a product line whose economics the filing structurally does not print.

What the geography shows instead is a competent converter in four places. External revenue runs Japan ¥98,347M, the Americas ¥68,248M, Europe ¥33,178M and ASEAN ¥17,978M F1472F1473F1474F1475; segment operating margins run 9.58%, 9.67%, 6.78% and 8.86% D49D50D51D52. Japan is 45% of external revenue F1472F5 and, on the local-currency basis the company reports growth on, the only segment that went backwards this year, revenue −1.9% and segment profit −2.6% F314F315, with tack labels inside it down 11.9% E98; the Americas grew 4.6%, Europe 8.2% and ASEAN 0.8%, all in local currency F316F318F320. Customers are brand owners in food, beverages and household goods E85, and no single customer reaches 10% of consolidated revenue E274 — diffuse buyer power rather than one dominant buyer. The stated strategy is local production for local consumption E69, and the stated competitive strengths are a local production-sales-development structure, integrated technology and customer relationships E58 — inputs, not a measured advantage.

The company holds its customers' shares, which tells you which way the dependence runs. Fourteen listed cross-holdings carry ¥2,884M plus two unlisted at ¥159M F801F802F800; the issuers are named — Yakult, House Foods, Kobayashi Pharmaceutical, Ito En, with whom it co-develops label-to-label horizontal recycling E212E215E214. The company states it holds no shares for pure investment E209, describes an annual board verification of holding purpose E210, and explicitly declines to state any quantitative holding effect E216; ten issues were increased through business-partner shareholding associations and none reduced E211.

Control is concentrated in one person and disclosed only in pieces. 株式会社創包 (Sohou Co., Ltd.) is the largest shareholder at 7,803 thousand shares, 14.5% F639F640, and note 3 states that the shares in Sohou's name are beneficially held by the CEO, 岡﨑成子, who holds a further 247 thousand directly E126F668. Two public-interest foundations registered at the company's own head-office address hold 4,800 thousand shares, 8.9%, and 1,780 thousand, 3.3% F643F644F649F650; 藤尾弘子, carrying the founder's surname, holds 1,784 thousand, 3.3% F647F648; the COO holds 848 thousand F736; the top ten together are 58.1% F660. The filing never once adds these together E127. The same person chairs the board E144, chairs the Nominating Committee E150 and chairs the Remuneration Committee E152, and the COO is her child E172. The governance form is the most demanding of Japan's three — a 指名委員会等設置会社 with six directors, three of them outside, and an audit committee composed entirely of outside directors including a CPA E142E189 — and the Group Internal Audit Office is stated, in the filing's own words, to have no direct reporting line to the board E192.

Two risks are named by management itself. Raw-material market prices move with the world economy, the supply-demand balance and exchange rates, and the stated response is procurement and cost reduction — never price E87. And environmental regulation is already arriving: carbon taxes, plastic-packaging taxes and lifecycle-monitoring responsibility, aimed squarely at the plastic sleeve that is 61.68% of revenue E91D62, answered by an R&D programme in recyclable, thin-film, low-density, plant-derived and linerless products E116 costing ¥2,539M, 1.17% of revenue F404F5.

The numbers

The five-year record is a four-year climb, not a cycle — which is the opposite shape from pick #1. Revenue, 百万円: 170,321 → 184,035 → 196,624 → 212,345 → 217,752 F1F2F3F4F5. Ordinary profit: 10,600 → 8,426 → 14,732 → 18,323 → 22,009 F6F7F8F9F10. Profit attributable to owners: 6,117 → 6,869 → 10,277 → 12,199 → 20,655 F11F12F13F14F15. Return on equity: 5.8 / 6.0 / 8.1 / 8.8 / 13.5% F46F47F48F49F50, and 10.22% in the last year once the one-off is removed D70. Book value per share compounded ¥1,999.85 → ¥3,023.11 F31F35; earnings per share ¥111.70 → ¥387.43 F36F40, or ¥293.94 ex-gain D10; net assets ¥109,492M → ¥161,264M F21F25; total assets ¥164,646M → ¥226,082M F26F30; the equity ratio strengthened 66.5 → 71.3% F41F45. The one stress inside the window is FY2023: revenue rose 8.1% and ordinary profit fell 20.5% D36D43, the ordinary margin going 6.22% → 4.58% D25D26 before recovering to 7.49 / 8.63 / 10.11% D27D28D29. That is the signature of lagging pass-through — margin compresses on the way up and expands on the way down — not of pricing power.

At ¥3,150 the price is fair on almost every measure, and cheap on only one. Market capitalisation net of treasury is ¥168,033M D3 on 53,343,865 shares D1; that is 1.04× book D6, 8.13× reported earnings D8, 10.72× ex-one-off D11, 10.98× on the company's own FY2027 guidance D15, a 2.57% dividend yield rising to 2.76% on the guided ¥87 D12D13F136F705, and — after ¥25,249M of net cash D89 — an enterprise value of ¥142,784M D94 at 4.94× EBITDA D96. The stamp sits at 0.905× the fiscal year's high of ¥3,480 and 1.42× its low of ¥2,212 D18D19F188F193. Only the reported P/E and the EV/EBITDA make this look cheap, and both are artefacts: the first of a one-off gain and a 15.09% tax rate F1063D85, the second of a net-cash balance worth 15.0% of market value D93 sitting in front of capital spending running at twice depreciation D119.

The one-off, traced four ways. ¥4,984M appears as the 子会社清算益 line in the income statement F1063, as a ¥4,984M deduction in the cash-flow statement's non-cash reconciliation F1134, as a −4,984 組替調整額 recycled out of the foreign-currency translation account in comprehensive income F1280, and in the company's own restated payout ratio F704. It arose from the liquidation of Fuji Seal Switzerland AG E244E41 — the entity acquired in the 2012 Pago purchase E30, whose German sister's shares were written down ¥812M in the parent-only (単体) accounts the prior year F1726E287. It is, economically, last decade's yen weakness booked once through the P&L on the way out of the reserve. Comprehensive income for the year rose 10.3%, ¥18,776M → ¥20,715M F19F20, while reported net income rose 69.3% F309. Both describe the same twelve months.

The tax line does as much work as the gain. Total income taxes were ¥3,669M on pre-tax profit of ¥24,324M F1085F1079 — 15.09% D85F1416 — against ¥5,670M on ¥17,870M, 31.73%, the year before F1084F1078D86, and a printed statutory rate of 30.59% F1415. The reconciliation attributes the fall principally to the tax effect on investments in subsidiaries E262, which is the liquidation again. Every lens in this study normalises at the statutory rate for exactly that reason.

The operating line is where the business actually is, and it improved modestly. Operating profit went ¥18,844M → ¥20,463M, +8.6% F304F305D41, on gross profit of ¥44,989M → ¥48,337M F1038F1039 — a gross margin of 21.19% → 22.20% D20D21 — with cost of sales at 77.80% of revenue D48 and an operating margin of 9.40% D23. Two qualifications belong beside that. First, ¥482M of the ¥1,619M operating-profit improvement — about 30% — is the segment-to-consolidated unrealised-profit elimination swinging from −¥608M to −¥126M F1496F1533: inventory in transit, not trade. Second, the gross-margin series exists for two years only, because the five-year highlights table prints 経常利益 and not 営業利益 or 売上総利益 — a limit of this study's one-document archive, not of the company's disclosure, and the single unresolved fact that most constrains this study's verdict.

The balance sheet is a fortress and holds nothing hidden. Total assets ¥226,082M F977 against liabilities of ¥64,817M F1011 and net assets of ¥161,264M F1033, with no non-controlling interests at all. Interest-bearing debt is ¥10,552M F1603 against ¥35,801M of cash and deposits F905 — net cash of ¥25,249M D89, ¥473.33 a share D92 — at average rates of 2.5% on long-term borrowings F1601, no bonds outstanding E280, interest coverage of 73.5× F334, debt at 0.49 years of operating cash flow D171F332, and an R&I rating of A E109. The defined-benefit obligation of ¥2,986M is entirely unfunded with no plan assets at all F1375F1005E257 and is treated as debt throughout. Goodwill is ¥936M F959, six-tenths of one percent of equity, and sits entirely in Japan E277. Investment securities are ¥3,135M F965. There is nothing to re-mark in either direction: no land at 1960s cost with a market value disclosure, no listed stake worth three times its carrying amount. Whatever margin of safety exists here has to come from earnings.

The cash flow is where the reported earnings stop being owner earnings. Operating cash flow ran 14,021 → 8,269 → 19,930 → 21,339 → 21,553 F56F57F58F59F60; investing △7,804 → △11,014 → △10,568 → △12,459 → △12,483 F61F62F63F64F65. Free cash flow was positive in four years and negative ¥2,745M in FY2023 D111D112D113D114D115. Cumulatively: ¥85,112M of operating cash D116, ¥54,328M consumed investing D117, ¥30,784M of free cash flow D118 against ¥56,117M of reported profit F11F12F13F14F15 — 54.9 sen on the yen. Cash and cash equivalents went ¥24,105M → ¥35,344M over the window F71F75. Capital spending was ¥16,984M on the cash-flow basis F1182 and ¥17,614M on the 設備投資 basis F405F1557, against ¥8,423M of depreciation F1132 — 2.02× D119, after 0.76× the prior year on ¥6,648M against ¥8,750M F1181F1131D120. Construction in progress rose ¥5,328M → ¥13,791M F948F949 and net property, plant and equipment ¥65,097M → ¥75,129M F956F957. Two named projects carry it: the Tendō plant at a ¥12,500M budget running to 2026-06, and a new Thai plant at ¥7,338M running to 2027-08 F602F604E119. And one qualification the red team supplies: FY2026's reported free cash flow of ¥9,070M D115 includes ¥5,896M of time deposits withdrawn against ¥929M paid in F1180F1178, money moving between pockets; on a pure capex basis — operating cash flow less tangible and intangible purchases — the year produced about ¥4.3bn F60F1182F1186.

Working capital is a supplier's, not a franchise's. Working capital is ¥85,711M, 39.4% of revenue D102F5, up ¥8,856M in the year from ¥76,855M D103. Receivable days are 111.5, inventory days 62.0, payable days 54.0, and the cash conversion cycle 119.5 days D107D105D109D110. The company finances its blue-chip customers for nearly four months.

Capital return is real, disciplined on price, and small. Dividends paid ¥3,934M plus a ¥1,077M buyback returned ¥5,011M F1210F1204D126 — 24.3% of reported net income and 32.0% ex-one-off D127D128 — against a stated policy of a 30% consolidated payout in principle E137F380. The buyback bought 430,900 shares for ¥1,077M in FY2026 and 734,300 for ¥1,922M in FY2025 F691F692F689F690, an average near ¥2,574 against book of ¥3,023.11 F35: below book, as it should be. But nothing was cancelled E135; treasury stands at 11.33% of shares issued on the ESOP-inclusive scope D138, and 185,000 shares were reissued after year-end to an employee share-granting trust at ¥2,481 F1592F1593E302, a dilution of 0.31% F1595. Shares issued have been unchanged at 60,161,956 for five years F112F116F608, and there are no options, no rights plan and no potential dilutive shares at all E122E6.

Guidance, and the sentence that is missing. FY2027 is guided to revenue ¥228,600M (+5.0%), operating profit ¥22,200M (+8.5%), ordinary profit ¥22,400M (+1.8%) and net income ¥15,300M (−25.9%) F383F384F385F386F387F388F389F390. The step down between the operating line and the net line is arithmetic: guided net non-operating income collapses by roughly the size of this year's ¥1,299M exchange gain F1046, and the ¥4,984M liquidation gain does not repeat F1063. The filing states the numbers and never explains the decline in words E113.

And the market has been scoring this company for five years. The five-year total shareholder return index is 115.6 against TOPIX-with-dividends at 202.2 F178F183 — +15.6% against +102.2%. The filing's own printed price-earnings series runs 15.1 / 12.0 / … / 6.7 F51F52F55. The plan the company is asking to be judged on, FSG.30, targets revenue of ¥350bn by FY2031 F272 with double-digit ROE and a price-to-book ratio of 1.5× or more F273E61, and the operating-margin and ROE targets are printed as 「２桁％」 — as words, with no number E111. Against ¥217,752M of revenue today F5 and 6.34% four-year compound growth D40, ¥350bn requires roughly 10% a year.

The five lenses

Buffett — watch

Start with what gets sold. A brand owner fills a bottle, and before it leaves the line a printed plastic sleeve is dropped over it and shrunk tight. That sleeve is destroyed with the package. This company sells those sleeves — ¥134,314M of them last year, 61.7% of everything it sold F1454D62 — plus tack labels, pouches, and the machines that put the labels on, ¥23,170M of machinery against ¥194,581M of label F1821F1820, out of 26 subsidiaries F194 for food, beverage and household-goods brand owners E85, no one of whom is 10% of the total E274. I like the shape of that. It is a tollbooth: every unit the customer fills pays a toll, and the machine that collects it is bolted into the customer's own filling line and signed off only after installation and trial running E267, with ¥15,401M of order book still to be recognised F1482E269. A shopkeeper would understand every word of it. It sits inside the circle.

Now the harder question: is the tollbooth a franchise, or a toll you have to keep rebuilding? Three tests settle it, and all three come out the same way. First, pricing. In the year ended March 2023 revenue rose 8.1% and ordinary profit fell 20.5% D36D43; the margin went 6.22% to 4.58% D25D26. Input costs went up and the company could not pass them along — its own risk disclosure says that when raw materials spike it responds with cost reduction, not price E87. A franchise raises the price. Second, capital. Last year revenue rose ¥5,407M F5F4; to get it the company spent ¥16,984M on plant F1182 and put another ¥8,856M into working capital D102D103 — nearly five yen of investment for one yen of extra sales. Over the five fiscal years the archive covers it is ¥54,328M of investing outflow D117 against ¥47,431M of additional annual revenue between the first year of that window and the last F5F1, with construction in progress going ¥5,328M to ¥13,791M F948F949 and capex at twice depreciation D119, and the plan calls for ¥43.5bn more over three years and ¥56.5bn over the four after that F274F275 — about ¥100bn against the ¥85,112M of operating cash the company generated in the whole preceding five years D116. Third, what the capital earns. Strip the cash and the intangibles: operating profit after the statutory rate is ¥14,203M against ¥133,818M of net tangible capital D80F25F1603F959F961F905 — 10.6%, in the best of five years F46F47F48F49F50. Return on capital excluding cash is 10.44% D84. Charlie and I look for fifteen without leverage. This is ten, with none.

So it is an adequate business, not a wonderful one. Fine — we bought plenty of adequate businesses. Then the price has to be wonderful. Owner earnings: take operating profit taxed at 30.59% D80F1415, add the small net of interest and dividends received over interest paid F1043F1045F1052, add back goodwill amortisation F1136, then subtract what it costs to stay in the same place. Depreciation is ¥8,423M F1132. If maintenance capital equals depreciation, owner earnings are about ¥14.5bn; on the two-year average of actual spending, ¥11,816M F1181F1182, about ¥11.1bn. I cannot narrow it because the company publishes one combined ordinary-plus-strategic investment envelope and never splits it E62. It must be a guess, and I have given you both ends. Against a market value of ¥168.0bn D3 that is a 6.6% to 8.6% yield; against enterprise value of ¥142.8bn after net cash D94D89, 7.8% to 10.1%. Capitalise at ten to twelve times and add the cash back and the whole company is worth ¥136bn to ¥199bn D89D3 — call it ¥2,550 to ¥3,730 a share on 53,343,865 shares D1, with a middle around ¥3,140. Tangible book alone is ¥2,982 F1033F959F961D1.

Only now do I look at the quotation: ¥3,150 D3. That is within ten yen of the middle of my range, and there is no margin of safety in it at all. The reported price-earnings ratio of 8.13× D8 is what makes this look cheap, and it is an illusion — 10.72× without the Switzerland gain F1063D11, 10.98× on the company's own guidance for a 25.9% fall in net income F390D15. Nor is there a Graham floor: net current assets less all liabilities come to ¥1,383 a share F921F1011D1, so you are paying 2.3 times, and on a haircut liquidation — with the company itself telling you its used label plant has no resale market, valuing disposal assets at memorandum value 「売却可能性が見込めないため」 E248 and carrying the whole Yamagata works at nil while 63 people still work there E118F479 — I get about ¥1,487 F905F1011D1.

Two more things a partner should know. Of the ¥8,456M by which net income rose F15F14, only ¥1,619M came from operating profit F305F304; the rest was the Switzerland gain, a currency swing and a 15.09% tax rate against 30.59% F1416F1415. Credit where it is due — the company excluded that gain when it set the dividend and printed both 20.9% and 30.2% payouts side by side F703F704E138. That is honest. Less honest is a ¥1,205M litigation settlement disclosed in a single sentence with no counterparty, no subject, no court and no date anywhere in 122 pages F1074E249, alongside an extraordinary report filed on a materially-affecting event whose contents are simply not reproduced E308. I will not buy the whole of anything with an unlabelled box in it. And the five-year scoreboard says the rest: book per share compounded ¥1,999.85 to ¥3,023.11 F31F35 while total shareholder return was 115.6 against a market at 202.2 F178F183, with roughly two-thirds of earnings retained on the ex-one-off base D128 into capital earning ten percent D84. That is why it trades at 1.04 times book D6, and it is not a mistake the market is making.

Verdict: watch, with no buy-below written. The balance sheet is a fortress at 71.3% equity with net cash F45D89, the family owns about 15% of the float and is paid ¥109M F639F668F795, and buybacks were struck near ¥2,500 against ¥3,023 of book F691F692F35. All of that is right. But a fair business at a fair price clears nothing, and there is no called strike for standing there.

What a student should take from this: when a company earns ten percent on tangible capital D80F905 and retains two-thirds of its earnings on the ex-one-off base D128, the market will eventually price it near book no matter how clean the balance sheet or how honest the managers — and a low reported price-earnings ratio built on a one-off gain and a one-off tax rate F1063D85 is not cheapness, it is arithmetic waiting to be undone. The test that separates a toll bridge from a factory is not whether customers come back; it is whether you can raise the toll in a bad year and whether the booth has to be rebuilt to keep collecting it.

Munger — watch

Invert it first. How does this company die inside ten years? Not from the balance sheet — net cash of ¥25.2bn D89, interest coverage of 73.5× F334, no covenant or rating trigger anywhere. Not from a customer walking out — none reaches 10% of revenue E274, and the customer list running through the cross-shareholding table reads like a roll call of stable Japanese food and household names E212E215. The plausible kill path already in motion is regulatory, and management says so itself: carbon taxes and plastic-packaging taxes aimed exactly at the shrink-sleeve label that is 62% of revenue E91D62. The stated answer is R&D redirected toward recyclable, linerless, plant-based-ink labels E116. Whether that response holds the economics together is not something the ledger lets me check — one year of R&D spend, no trend, no competitor comparison F404.

That is the first honest disappointment. The moat this company is supposed to have — a consumable destroyed on every fill, an applicator qualified into the customer's own line — is a good story, and I cannot find its fingerprint in the figures I was given. No customer-retention data. No price-versus-volume decomposition. No competitor filing anywhere in this archive to show whether this company's unit costs or margins beat the next label maker's. Two years of gross margin rising 21.2% to 22.2% D20D21 is a data point, not a moat. I would rather say "I don't know" than dress up a plausible mechanism as a demonstrated one. That alone is not fatal. But it means the quality half of "great business at a fair price" is unproven, not proven.

What is provable, and where I spent the rest of my attention, is the incentive structure — because show me the incentive and I'll show you the outcome. The CEO chairs the board E144, chairs the Nominating Committee E150, and chairs the Remuneration Committee E152 — the very body that sets her own pay and grades her own targets E153E203. She is also the beneficial owner of the largest shareholder, at 14.5% E126F640, and she is the COO's parent E172. Layer on two foundations registered at the company's own head-office address, 8.9% and 3.3% F644F650E125, and a 3.3% holder carrying the founding surname F648, and you have a control bloc the filing itself never once sums E127. That is not proof of expropriation; I found none. It is the structure to worry about: three committees with a nominal outside-director majority, chaired end to end by one person who is also the largest owner and the heir-apparent's parent. Formal independence with a single point of override is not independence. And by the filing's own words the Group Internal Audit Office has no direct reporting line to the board E192 — so even the control function reports up through the executives it is meant to check.

Add two candor gaps a cleaner filer would not have. A ¥1,205M litigation settlement appears in the P&L with no counterparty, no subject matter, no jurisdiction, anywhere F1074E249. And FY2027 guidance shows net income falling 25.9% with not one sentence explaining why, in a document whose segment build-up above it is disclosed in granular detail F390E113. Neither is a lie. Both are silences exactly where the filing is capable of being precise everywhere else, which is what makes them noticeable.

None of this erases the good arithmetic. Revenue has compounded at 6.3% a year for four years D40. Return on capital excluding cash is 10.44% and ROE excluding the one-off 10.22%, rising D84D70. No dilution and no stock options E6E122; buybacks struck below book, at an average near ¥2,574 against ¥3,023.11 F691F692F689F690F35. This is a perfectly respectable, ungeared, growing industrial business. What the arithmetic will not do is ratify the plan: the FSG.30 target of ¥350bn by FY2031 F272 needs 9.96% a year from today's ¥217,752M F5, not the 6.34% delivered D40 — hold the delivered rate for five more years and revenue arrives at about ¥296bn, some ¥54bn short. That target is a wish wearing a spreadsheet after all, and the margin and ROE ambitions beside it are printed as 「２桁％」 with no number at all E111. It is just not, on this evidence, a demonstrated compounding machine with a moat I can see widening, run by people whose incentives I can fully trust.

Verdict: watch, and I issue no number. Price is not the problem here — 1.04× book and 10.7× ex-one-off earnings D6D11 is not an expensive stock. Trust is. No discount fixes a governance structure in which one person holds every veto. Sit on your hands. The re-check trigger is not a lower price; it is either the company earning back the benefit of the doubt on its own disclosures, or the moat finally showing up in numbers I can audit.

What a student should take from this: a board can carry a majority of outside directors and three committees and still have a single point of failure — check who chairs the committees, not just who sits on them, and always sum the pieces the filing declines to sum itself E127. And a moat claimed in the strategy section E58 needs a fingerprint in the figures — price realised, share held, retention shown, a competitor's worse numbers beside it — or it is a story, and stories are not moats however mechanically plausible they sound.

Pabrai — watch, buy below ¥1,600

Downside first, and I mean before anything else, because everything I am about to say depends on what happens to me if I am wrong.

So take the balance sheet apart the way a bank would in a bad year. Cash and deposits ¥35,801M F905, at par. Receivables ¥66,528M D106 — 111.5 days of sales D107, a long leash, but the customers are brand owners in food, beverages and household goods E85 and none is 10% of revenue E274, so eighty cents. Inventories ¥28,798M D104 — and here stop and think. A shrink-sleeve label is printed with somebody else's artwork. Work in process of ¥7,086M F913 and raw materials of ¥10,550M F915 are somebody else's brand, sitting on a roll. In a liquidation that is not inventory, it is confetti. Forty cents, and I am being kind. Land ¥9,331M F941 at par. Everything else in property, plant and equipment — the buildings, the presses, the ¥13,791M of construction in progress F949 — at thirty cents, and I take that number from the company's own filing, not from my imagination: when it wrote down the Fujitack East Yamagata plant it valued the assets held for disposal at memorandum value 「売却可能性が見込めないため」, because no possibility of sale is expected E248, and carries the plant at zero across every asset class while 63 people still work there E118. That is management telling you in the notes that there is no secondary market for a used label plant. Goodwill ¥936M F959 and deferred tax assets ¥3,287M F967 to zero. Add it up: about ¥138.1bn of stressed assets less every liability at face, ¥64,817M F1011, leaves roughly ¥73.3bn — about ¥1,374 a share on 53,343,865 shares D1. At the stamp that is a 56% hole D3.

And the softer, far likelier downside is the one worth teaching. This stock traded at 0.68 times book three years ago — book ¥2,201.40, earnings ¥125.43, at 12.0 times F32F37F52 — with substantially the same business. Put today's ¥3,023.11 of book F35 on 0.70 times and you are at ¥2,116, a 33% loss with nothing wrong at all. That is my answer to P1 and it is decisive: there is no hard floor here. Net cash of ¥25,249M D89 is real, ¥473 a share D92, but it is only 15.0% of the market capitalisation D93. Frontline was ships you could sell into a live charter market. This is presses you cannot give away.

Now the other side. Reported earnings of ¥387.43 F40 are a mirage — ¥4,984M is a liquidation gain on Fuji Seal Switzerland AG F1063, recycled out of the translation reserve as a △4,984 組替調整額 F1280, backed straight out of the cash-flow statement at △4,984 F1134. Do not take my word for it: the company itself excluded it from the dividend base and restated its payout from 20.9% to 30.2% F703F704. Work backwards from their own 30.2% and clean earnings are ¥268.21 a share D134 — 11.7 times at the stamp D3. Next year they guide net income down 25.9% to ¥15,300M F390F389 — 11.0 times D15. EV/EBITDA of 4.94 times D96 looks cheap only because a quarter of the capital base is cash D84. Eleven or twelve times clean earnings for a good business is a fair price, and be brutally honest about the difference. Normalise operating profit at last year's ¥18,844M rather than this year's ¥20,463M F304F305, tax it at the statutory 30.59% F1415, capitalise at ten to twelve times, add the net cash and the ¥3,135M of securities D89F965, and I get ¥2,984 to ¥3,474 a share D1. The price sits inside it, at 91–106% of value. My sell band is 90 to 100 cents on that dollar, ¥2,690 to ¥3,470 D1 — today's quotation is my exit price. When your buy price and your sell price are the same number, you are not looking at an investment, you are looking at a fully valued stock.

Why normalise down? Because this year flatters. Ordinary profit rose ¥3,686M but operating profit only ¥1,619M F10F9F305F304; the other ¥2,067M is a non-operating swing, a ¥1,299M exchange gain against last year's ¥779M loss F1046F1053. Fifty-six percent of the 20.1% ordinary-profit growth is currency D46. And the Japanese five-year table prints 経常利益 and never 営業利益 F6, so a lazy normaliser working the only long series available swallows that whole. Meanwhile Japan, 45% of external revenue F1472F5, shrank 1.9% with profit down 2.6% F314F315, and ASEAN shrank 0.6% D170. On the yen basis including inter-segment sales, two of four segments went backwards — ASEAN is up 0.8% once you strip the currency F320, which is the basis on which Japan is the only one that fell.

There is a great deal to like. The CEO beneficially owns Sohou's 14.5% E126 and 247 thousand shares besides F668 against ¥109M of total pay E207F795 — she cares about the share, not the salary. No stock options at all E122. Net cash, an A rating E109, interest cover of 73.5 times F334, and a business that will be here in twenty years E267. Survival is not the question. The question is price — and there is a second question the filing answers plainly and against me. Management has written down that it will heighten the financial-leverage effect with a view also to using borrowed capital in order to reach a price-to-book ratio of 1.5 times E64F273, against investment frames of ¥43.5bn to FY2027 and ¥56.5bn to FY2031 F274F275 and five-year cumulative free cash flow of ¥30,784M D118. Net cash grew ¥595M last year on ¥21.6bn of operating cash flow D91F60 because capex ran at 2.02 times depreciation D119, and the buyback authorisation is 93.2% spent with nothing left E134. So the cushion that makes the downside merely bad rather than catastrophic is scheduled to be spent and levered against. You are not paid to wait — 2.57% D12 — and the cash is not building while you do.

Verdict: watch, buy below ¥1,600. Few bets, big bets, infrequent bets; this is not one of them at the stamp D3. Put it on the list at ¥1,600 — half the midpoint of my ¥2,984–3,474 value range D1, 0.53 times book F35, about six times clean earnings D134, a price inside the ranges this stock actually printed in FY2023 and FY2024, ¥1,402 to ¥1,939 F190F185 and ¥1,455 to ¥2,150 F191F186. Most companies are a pass. That is the system working.

What a student should take from this: an equity cushion is not an asset floor. When a company has net cash, high book value and no distress, your downside is not set by liquidation arithmetic at all — it is set by the lowest multiple the security itself has carried on the same business, which you can recompute exactly from the five-year highlights table the filing already prints (¥125.43 × 12.0 ÷ ¥2,201.40 = 0.68× F37F52F32). And when management writes down that it intends to spend the cash and add leverage E64, the floor you were relying on is a wasting asset: score it at what they said they would do, not at today's balance sheet.

Li Lu — watch

A label is destroyed the moment it does its job. Every bottle that leaves a filling line in Osaka or Indiana carries away a sleeve that will never be sold again, and the machine that put it there — which this company also builds, sells and services E43 — stays behind, qualified into the customer's line, waiting for the next order. That is a lovely shape for a business, and it is why I read this filing twice. But the question is not whether the shape is lovely. It is whether I can say honestly what this company earns in 2036, and whether today's price already assumes the good answer.

Start where an owner starts: with the whole thing. At the stamp the market asks ¥168,033M for all of it D3. Behind that sits ¥161,264M of book equity F1033 — you are paying 1.04× book D6 — of which ¥25,249M is net cash D89, ¥66,528M trade receivables D106, ¥28,798M inventory D104, ¥75,129M plant F957. Goodwill is ¥936M F959, six-tenths of one percent of equity. This is a clean book. There is nothing to re-mark: no hotel carried at a decade-old cost, no listed stake worth three times its carrying value — the fourteen cross-holdings are already at fair value and total ¥2,884M F802. Whatever margin of safety exists here must come from earnings, not from a hidden balance sheet.

So what does the whole business earn? Unlevered, after full statutory tax at 30.59% F1415, ¥14,203M D80. Capitalise that at a ten-percent owner's hurdle and add the net cash D89 and you get ¥167.3bn — which is, to within half a percent, the asking price D3. That is the single most useful sentence I can write about this company. The market is offering you a ten-percent unlevered return with zero credit for growth. It is not offering you fifty cents on the dollar.

Now the growth, and the care it deserves. Five years took revenue from ¥170,321M to ¥217,752M, 6.34% a year F1F5D40. Underneath, the picture divides: Japan is 45% of external revenue F1472F5 and shrank 1.9% with segment profit down 2.6% F314F315; Europe grew 8.2% in local currency, the Americas 4.6%, ASEAN 0.8% F318F316F320 — in local currency Japan is the only one that fell. And the headline profit jump wants decomposing before it is admired. Ordinary profit rose ¥3,686M F10F9, of which ¥2,078M is simply the swing from a ¥779M exchange loss to a ¥1,299M exchange gain F1053F1046 — 56% of the year's ordinary-profit growth is currency. Above that, net income "rose 69.3%" F309 on a ¥4,984M gain from liquidating a Swiss subsidiary F1063, money recycled out of translation reserve as a △4,984 組替調整額 F1280 and backed out of the cash-flow statement F1134 — no cash at all. To management's credit they excluded it from the dividend base and said so E138. Their own restated payout implies they treat ¥14,299M as the true year D135 — lower even than my naive ex-gain figure of ¥15,671M D9. Believe the company: this was a ¥14.3bn year, an 11.75× multiple, not the 8.13× the reported number produces D8D3.

Against that, the returns — and here is where the moat gets measured rather than described. ROE excluding the one-off is 10.22% D70; ROIC excluding cash is 10.44% D84. Segment operating margins run 9.58% in Japan, 9.67% in the Americas, 6.78% in Europe, 8.86% in ASEAN D49D50D51D52. Look at those four together, because they answer the moat question better than any narrative. If the applicator standing in the customer's plant were a true switching cost — the kind that lets you price without asking — you would see it here. Instead you see the margins of a competent converter in four regions, none approaching monopoly economics. The lock-in is real: the auditor confirms revenue is recognised only on customer acceptance after installation and trial run E293. But that lock-in is worth about nine points of operating margin D23, not thirty. This industry has not crossed the line where the winner takes it all, and nothing in the filing suggests it is approaching it.

Management I like. The record is sixty-eight years of adaptation executed and paid for out of cash flow — cap seals 1958 E9, tack labels, shrink labels 1981 E12, machinery 1990 E14, Owens-Illinois 2001 E18, Vietnam and Indonesia, Thailand bought out, North Carolina 2023 E37, pharmaceutical packaging 2023 E38 — and they close as well as open: Switzerland liquidated E41, the Yamagata plant written to nil across every asset class while still employing sixty-three people E118. Share count has not moved in five years F112F116, and the buyback ran near ¥2,574 a share against book of ¥3,023.11 F690F692F689F691F35 — bought below book, as it should be.

Then the structure, which no one aggregates for you. Sohou holds 14.5% F640 and is beneficially the CEO's E126; the Fuji Seal Foundation holds 8.9% and Innovation of FUJI 3.3% F644F650, both registered at the company's own head-office address E125; 藤尾弘子, carrying the founder's name, holds 3.3% F648. That is about thirty percent of the float, and the filing never once adds it up E127. The same person chairs the board, the nominating committee and the remuneration committee E144E150E152 and is the parent of the COO E172. I do not read this as a warning — related-party transactions are immaterial at both levels E278, Deloitte has been continuously engaged for nineteen years E193 and signed an unqualified opinion on this year's accounts E291, and an owner-operator whose ¥109M of pay F795 sits beside roughly ¥25bn of shares F639D3 is precisely the alignment I want. I read it as a fact that must be known and is not stated.

What I cannot know is where this ends, and there are two things. First: the board monitors sustainability as a management priority E75 while naming plastic-packaging taxes and lifecycle-monitoring responsibility as live regulation E91 against a business 61.7% of whose revenue is one plastic sleeve D62 — and the environmentally-conscious-product ratio the board watches is never printed. That is the decisive ten-year variable, and the company's own scorecard against it is withheld. Second: ¥1,205M was paid to settle a suit brought against the company and the filing states nothing further — no counterparty, no subject, no court, no date F1074E249 — while listing an extraordinary report filed 2025-11-18 on events materially affecting results whose content is not reproduced E308. I cannot bound a loss whose nature is undisclosed, and I will not price one.

Verdict: watch, with no buy-below. On my own construction — ¥14,203M of unlevered earnings D80 at a ten-percent hurdle with three percent terminal growth, plus net cash D89 — intrinsic value is about ¥228bn, or roughly ¥4,277 a share D1. That is 1.36× the ¥168.0bn price D3, not the 2× this lens requires. Fifty cents on that dollar is near ¥2,100 D1, and I give you the number so you know where price stops being the obstacle. I do not issue it as a threshold, because price is not the only obstacle: opacity caps this at watch whatever the quote E249E308. Fair is not my price, and unfinished work is not my bet.

What a student should take from this: when a company tells you it has strong customer relationships and integrated technology E58, do not argue with the words — go to the segment note and read the margin. Four regions earning 6.78% to 9.67% D51D50 is the measured size of that moat, and it is a different answer from the one the strategy page gives. And when a headline profit leaps, decompose it before you admire it: here a currency swing and a liquidation gain account for nearly all of a 69.3% rise F309F1046F1063, and the company's own dividend arithmetic D135 quietly tells you which number it actually believes.

Claude — watch, implied buy-below ¥1,250

I registered my priors figures-blind and the ledger has now graded them. Let me score before I argue, because the scoring is the more useful half. (Self-distance: I hold this verdict, built the reconciled figure table all five lenses consumed, and wrote the synthesis below — read all three with that concentration of authorship in mind.)

The load-bearing claim was Claim P — that more than half of the multi-year profit improvement is FX translation, input-cost timing and non-recurring items rather than durable operating progress, registered at P ≈ 0.65. It is substantially true, and I was right for two of my three reasons and wrong about the one I weighted most. At the net line it is true past argument: the ¥4,984M gain F1063 plus the tax rate it dragged from 31.73% to 15.09% D86D85 account for essentially the whole 69.3% net-income jump F309. At the ordinary line it is true by exactly the margin the claim specifies — of the ¥3,686M rise, ¥2,078M is a single FX swing F1053F1046F9F10, 56%. And the input-cost-timing shape is what I predicted: the ordinary margin ran 6.22 → 4.58 → 7.49 → 8.63 → 10.11% D25D26D27D28D29, compressing through the spike and expanding on the down-leg, which is the signature of lagging pass-through, not of pricing power. Where I was wrong is the mechanism I leaned on hardest. FX translation was not the driver of the reporting year at all — the yen strengthened 1.4% against the dollar, ¥151.69 to ¥149.61 F310F311, and once the segment local-currency rates are weighted the net FX contribution to FY2026's +2.5% revenue growth D39 is approximately zero F314F316F318F320. The growth was organic. I let a structural fact — a majority-overseas earnings base — stand in for a rate move that did not happen. That is the single cleanest calibration lesson of this study.

Two other priors deserve their marks, and three misses will not be papered over. My best-calibrated call was the cash-conversion prediction, registered at P ≈ 0.70 that cumulative five-year free cash flow would come in below 60% of cumulative net income, band 30–80%: it printed 54.9%, ¥30,784M against ¥56,117M D118F11F12F13F14F15, near the centre of my band — and I also said this would be the number that decided the owner arithmetic, and it was. I predicted the guided net-income fall as the one-off's shadow without a figure, and it arrived at −25.9% F390. Against that: I predicted Japan's operating margin would sit below the group's; it sits above, 9.58% against 9.40% D49D23, and Europe at 6.78% D51 is the weak segment. I sized Europe at 20–35% of revenue; it is 15.2%, ¥33,178M of ¥217,752M F1474F5. And I predicted a P/E of 10–17× on flattered earnings and 13–22× normalised; the actual is 8.13× and 10.72× D8D11, with EV/EBITDA at 4.94× D96. I systematically over-predicted the rating, because I priced the class and ignored how much a net-cash balance worth 15.0% of market capitalisation D93 compresses enterprise value — and how thoroughly this stock has already been marked down. Five-year total shareholder return is 115.6 against TOPIX-with-dividends at 202.2 F178F183. The market has been scoring this company for five years and has not been asleep.

So what is it worth. I will not use the multiple. Normalised owner earnings start at operating profit of ¥20,463M F305, subtract ~¥700M of charges that are chronic rather than extraordinary — reorganisation costs and losses on the retirement and sale of fixed assets appear in both available years, ¥549M and ¥313M this year against ¥247M and ¥295M last F1072F1069F1071F1068 — tax at statutory F1415, add back depreciation of ¥8,423M F1132, subtract a maintenance capex of roughly ¥10.5bn (depreciation grossed up because the depreciable base is heavily written down on two-to-fifteen-year machinery lives F1220F1221), then subtract the ~¥2.6bn a year that standing still costs at working capital of 39.4% of revenue and a 119.5-day cash cycle D102D110. That lands near ¥10.1bn, at the midpoint of three independent selves who ran the same equation. Against ¥168,033M of net market capitalisation D3 less deployable net cash — not the ¥25,249M headline D89 but the ¥11,375M the waterfall leaves after ¥35,801M of cash F905 gives up ¥10,552M of debt F1603, ¥2,986M of wholly unfunded pension F1375 and ¥10,888M of operating float at 5% of revenue F5 — the private-owner yield is about 6.0–6.5%. My hurdle is 9%: the company's own long-term borrowing costs 2.5% F1601, and an equity buyer of an uncontestable converter that has committed ¥100bn of investment over seven years F274F275 against ¥17bn a year of demonstrated operating cash flow D116 needs a wide spread over that. No whole-company buyer takes this at the stamp D3. All three of my selves said no independently.

The honest framing is cycle position. Current ordinary margin is the highest of the five available years D29; at the five-year mean of 7.41% the shares trade at roughly 15× mid-cycle earnings, not 8.13× reported D8D3. The stock is not cheap — the earnings are high. Every one of the four segments earns between 7.11% and 9.23% pre-tax on its own segment assets — profit of 9,637, 6,602, 2,628 and 1,722 F1528F1529F1530F1531 over assets of 104,356, 71,584, 33,406 and 24,219 F1534F1535F1536F1537 — below any honest pre-tax hurdle; nothing subsidises anything, because nothing clears. And the register makes the gap permanent rather than a prize: the six named holders sum to 17,262千株 F639F643F649F647F668F736 against 53,645,700 fully-voting shares F675, 32.2% of votes in an allegiant bloc — 32.8% counting the 357,700 shares of the employee ESOP trust F681, and the COO's block is dated 2026-06-23 while the other five are 2026-03-31. That is fractionally short of the one-third of outstanding votes that would block a special resolution outright, and comfortably enough to block one on any realistic turnout — so no outside proposal, tender or campaign can succeed without management's consent. Earnings-power value at 9% on ~¥10.1bn is about ¥112bn against net assets of ¥161,264M F1033: EPV sits below asset value, and there is no mechanism to close it.

What stops me short of pass is that the business is not deteriorating and the price is not absurd. Revenue has never fallen in the window F1F5. Incremental operating margin from FY2025 to FY2026 was 29.9% against a 9.40% average D23, and stays above average on every computable window and every re-basing. Europe and ASEAN inflected in local currency F318F320. Guidance has been sandbagged where I can check it — FY2025 operating profit came in at 122.4% of the initial forecast F785. The balance sheet is unarmed F45D89. This is a competent, capital-hungry, low-return manufacturer whose owners cannot fire anyone, priced at what that deserves. And what stops me short of anything stronger is a specific hole. The one test that would settle whether FY2026's margin is a plateau or a peak — what percentage gross margin did through the FY2023 input-cost spike — cannot be run, because the five-year highlights table prints 経常利益 and not 売上総利益 or 営業利益 F6F10, and the archive is one filing deep. All three of my selves ranked that unknown first, unprompted. It is resolvable now, from EDINET, and it is unresolved. A verdict is not allowed to be more confident than that.

Verdict: watch, implied buy-below ¥1,250, band ¥1,080–1,330. The threshold is derived blind to the stamp from a downside built only from observed facts: revenue flat at ¥217,752M because it has never declined F5F1, the ordinary margin reverting to the observed five-year low of 4.58% D26, chronic charges recast as recurring, less the ¥244M of net financial income the cash credit already counts F1043F1045F1052, taxed at the statutory 30.59% F1415, capitalised at 10% with no growth and no unlock, plus half of deployable net cash — ¥11,375M, not the ¥25,249M headline D89, being cash F905 less debt F1603, less the unfunded pension F1375 and less a 5%-of-revenue operating float F5 — halved because the ¥100bn plan pre-commits it F274F275 and a ~32% bloc F675 means nothing forces distribution. My own chain gives ¥1,281; three independent selves gave ¥1,080, ¥1,220 and ¥1,329; published at the median of the four D1. Two independent caps hold the verdict at watch and would hold it there at any price on this run: C98's resolvable-now unknown left unresolved, and C84's jury divergence — two selves calling this a compelling avoid, one calling the long the winner at the stamp on ~¥15.7bn of clean earnings and 1.04× book D6. A verdict may not be more confident than its most cautious diverging self and may not borrow confidence from its most optimistic one.

What a student should take from this: four things. One: a structural fact is not a rate move — I predicted FX flattery from the geography of the earnings base, and geography establishes the exposure while only the period's average rate produces the effect; the yen strengthened F310F311 and the flattery arrived through a tax line and a liquidation gain instead. Two: operating-cash conversion and free-cash conversion are different questions and only one of them is about accounting — this company converts 152% of net income into operating cash D116 and 54.9% into free cash flow D118; the gap is not earnings quality, it is reinvestment intensity, and it is the whole of the owner's problem. Three: read the adjustment line — ¥482M of the ¥1,619M operating-profit improvement, 30% of it, came from the inter-segment unrealised-profit elimination swinging from −¥608M to −¥126M F1496F1533; averages hide problems, eliminations hide improvements. Four: report the misses as prominently as the hits. My headline claim scored as a hit and two of its three legs were wrong, and being right for the wrong reason is a calibration failure the pre-registration exists to catch.

Synthesis

Where the five lenses agree

All five reached watch, and beneath that unanimity the panel agrees on five things, none of them soft.

First, this is a good business and everyone says so. The pairing is real — a consumable destroyed on every fill, applied by a machine the same company sells, services and books revenue on only at customer acceptance inside the plant E267E293, with ¥15,401M of backlog and ¥8,221M of customer advances behind it F1482F1244. Survival is not the question: 71.3% equity F45, net cash of ¥25,249M D89, interest cover of 73.5× F334, debt at 0.49 years of cash flow D171, and revenue that has never fallen in the window F1F5. Second, the reported year is not the run rate. The ¥4,984M liquidation gain F1063 and the 15.09% tax rate D85 are not repeatable, and management's own restated payout ratio says so in the company's own hand F703F704D136. Third, the reported earnings badly overstate owner cash — ¥56,117M of five-year profit F11F12F13F14F15 converting to ¥30,784M of free cash D118, with the operating line converting at 1.05× D125 and the investing line consuming the difference D117. Fourth, the moat cannot be measured from this filing — machinery is 10.64% of external revenue D59 and its margin is nowhere separated, because the group calls itself a single business when declining product detail E105 while reporting four geographic segments E270. Fifth, the price is fair rather than cheap, and every named threshold sits far below the stamp D3.

Where they diverge — a staged exchange

The unanimity is on the verdict, not on the reasoning; and the reasoning is what the checklists are for. Five disagreements are worth staging in the masters' own voices, each traced to the items that drive it.

---

On what kind of business this is: Buffett and Li Lu agree it is a competent converter and get there by different arithmetic.

BUFFETT (B26, returns on net tangible capital; B24, pricing power with capital-light growth). I do not need a valuation to answer this. Strip the cash and the intangibles and operating profit after the statutory rate is ¥14,203M against ¥133,818M of net tangible capital D80F25F1603F959F961F905 — 10.6%, and that is the best of five years F46F47F48F49F50. Then look at what the growth costs. Revenue rose ¥5,407M last year F5F4; getting it took ¥16,984M of plant F1182 and ¥8,856M of working capital D102D103 — ¥4.78 of investment per ¥1 of incremental revenue. A franchise grows on other people's money. This one buys revenue with capital, and the price of the revenue is nearly five times the revenue. That pair settles the question. Everything downstream — the multiple I will pay, the growth premium at nil, the tolerance for retaining two-thirds of earnings on the ex-one-off base D128 — follows from it.

LI LU (L18, a dollar at fifty cents; L15, the closed-exchange owner test). We agree, and my arithmetic never touches the plant. Take the same ¥14,203M of unlevered after-tax earnings D80, capitalise at a ten-percent owner's hurdle with three percent terminal growth, add the ¥25,249M of net cash D89, and conservative owner value is about ¥228bn — 1.36× the ¥168.0bn asking price D3, not the 2× I require. And here is the sentence that made me put the file down: capitalise the same NOPAT at ten percent with zero growth and add the cash, and you get ¥167.3bn — the asking price, to within half a percent D3. The market is offering a ten-percent unlevered return with no credit for growth. That is not a mispricing; it is a correct price for exactly what Warren just described.

BUFFETT. And note where our two roads would separate, because that is the useful part. Mine is a judgement about physical plant and incremental capital that a man can make from the fixed-asset note and the cash-flow statement. Yours is a discount-rate identity that needs only one earnings number and a hurdle. They converge here because the business is ordinary. On a company with a genuinely capital-light growth engine they would not converge, and the checklists should be able to tell those cases apart.

LI LU. They would also separate if the segment note said something different. It does not. Four segment operating margins of 9.58, 9.67, 6.78 and 8.86 percent D49D50D51D52 are the moat measured rather than described — the applicator lock-in is worth about nine points of operating margin D23, not thirty. I sized it instead of arguing about it, and that is the whole of my L43.

---

On price: Pabrai alone names a number, and gets there by attacking the floor.

PABRAI (P1, maximum permanent loss; P5, hidden assets — inverted; P53, the fifty-cent dollar; P76, the pre-set sell band). Four of you decline to name a price. I will, and the reason I can is that I did the downside first. In this hunting ground you learn to expect the Japanese balance sheet to hide value. This one hides nothing in either direction, and it tells you so itself. When it wrote down the Yamagata plant it valued the disposal assets at memorandum value 「売却可能性が見込めないため」 — because no sale is expected E248 — and carries the works at nil while sixty-three people still work there E118. That is management telling you there is no secondary market for a used label plant. So mark it honestly: cash at par F905, receivables at eighty D106, inventories at forty because a shrink sleeve is printed with somebody else's artwork D104, plant at thirty on the filing's own precedent F957, goodwill and deferred tax to zero F959F967, less every liability at face F1011. That is roughly ¥1,374 a share D1 — a 56% hole from the stamp D3.

PABRAI, continued. But the stress case is not the binding one, and this is the part worth teaching. The likelier downside is a de-rate, and this security has already shown you its own floor: three years ago it carried 0.68 times book — ¥125.43 of earnings at 12.0 times against ¥2,201.40 of book F37F52F32 — on substantially the same business. Put 0.70 times on today's ¥3,023.11 F35 and you are at ¥2,116, a 33% loss with nothing wrong at all. Now the value. Normalise on last year's ¥18,844M of operating profit rather than this year's ¥20,463M F304F305, tax at 30.59% F1415, capitalise at ten to twelve times, add net cash and securities D89F965: ¥2,984 to ¥3,474 a share D1. The stamp is 91 to 106 percent of that D3 — and it sits inside my own pre-set sell band of ¥2,690 to ¥3,470 D1. Buy price and sell price cannot be the same number. So: watch, and I want it at half the midpoint of value, which is ¥1,600 — a price inside the ranges this stock actually printed in FY2023 and FY2024, ¥1,402 to ¥1,939 F190F185 and ¥1,455 to ¥2,150 F191F186.

BUFFETT. I reach nearly the same arithmetic and decline to write it down. My private-owner range is ¥2,550 to ¥3,730 with a middle near ¥3,140 D1, my haircut liquidation about ¥1,487 F905F1011D1, and one-third off the midpoint lands near ¥2,095 D1 — which is the number I actually watch. The reason I do not publish it is B43 and B13: five years of record where I want ten F1F5, and a ¥1,205M charge with no counterparty, subject, jurisdiction or date F1074E249 beside an extraordinary report on a materially-affecting event whose contents are not reproduced E308. A price I would name would imply I had finished work I have not finished.

PABRAI. And I will concede the sharpest thing against my own number, because P17 is where it lives. The cushion my floor rests on is scheduled to be spent. Management has written down that it will heighten the financial-leverage effect with a view also to using borrowed capital to reach a price-to-book target of 1.5 times E64F273, against investment frames of ¥43.5bn and ¥56.5bn F274F275 and five-year free cash flow of ¥30,784M D118. Net cash grew ¥595M last year on ¥21.6bn of operating cash flow D91F60 because capex ran at twice depreciation D119, and the buyback authority is spent E134. That is a disclosed, dated, falsifiable threat to the only part of the downside that currently holds, and it is why my threshold is half of value rather than a comfortable discount.

---

On character: Munger's block is not arithmetic at all.

MUNGER (M46, incentives; M54, related-party and the unaggregated bloc; M83, whether the CEO can be overruled). Let me say the uncomfortable half first, because it is the honest half: the arithmetic here is perfectly respectable and I still will not pay. Revenue compounding 6.3% a year D40, return on capital ex-cash of 10.44% D84, ROE ex-one-off of 10.22% and rising D70, no dilution, no options E122E6, buybacks struck below book F692F691F35. Price is not the problem — 1.04× book and 10.7× ex-one-off earnings D6D11 is not an expensive stock. Trust is. The CEO chairs the board E144, chairs the Nominating Committee E150, chairs the Remuneration Committee E152 — the very body that sets her own pay and grades her own targets E153E203 — beneficially owns the largest shareholder at 14.5% E126F640, and is the COO's parent E172. Two foundations at the company's own head-office address hold 8.9% and 3.3% F644F650E125 and a founding-surname holder 3.3% F648. Every one of those facts is disclosed. The filing never once adds them up E127. And by its own words the Group Internal Audit Office has no direct reporting line to the board E192 — the control function reports up through the executives it exists to check.

MUNGER, continued. Show me the incentive and I will show you the outcome; here the incentive is graded by the person being graded. Add the two silences — ¥1,205M settled with no counterparty, no subject, no jurisdiction F1074E249, and a 25.9% guided fall in net income with not one explanatory sentence F390E113 — in a document capable of granular precision everywhere else. Neither is a lie. Both sit exactly where the filing could have been precise. No discount fixes a structure in which one person holds every veto, so I name no number at all: the re-check trigger is behaviour, not price.

LI LU. I read the same register and reach a softer conclusion, and the disagreement is worth stating rather than smoothing. Related-party transactions are immaterial at both levels E278, Deloitte has been continuously engaged for nineteen years E193 and signed an unqualified opinion on this year's accounts E291, and an owner-operator whose ¥109M of annual pay F795 sits beside roughly ¥25bn of stock F639D3 is the alignment I want rather than the conflict I fear. My block is not the character; it is the opacity — L46 and L51 — a settlement I cannot bound E249 and an extraordinary report I cannot read E308. Charlie's cap is permanent until behaviour changes. Mine lifts the day the company answers two questions.

MUNGER. Then we differ on what counts as a remedy, not on what we see. You would be satisfied by a filing. I want the committee chairs separated E150E152 and the internal audit office given a line to the board E192 — because a company that discloses everything and aggregates nothing is telling you which questions it would rather you did not assemble.

---

On self-grading: Claude's contribution is that it scored its own priors, and the misses are the useful part.

CLAUDE (C1–C6, the outside view; C23 and C44, cash conversion; C40, cycle position). My figures-blind section registered two propositions that decided most of what followed. P ≈ 0.65 that more than half the multi-year improvement was flattered rather than durable, and P ≈ 0.85 that the reporting year's net income specifically was one-off-flattered. Both resolved true. But the first resolved true partly for the wrong mechanism. I weighted FX translation of a majority-overseas base as the lead driver; in the reporting year the yen strengthened 1.4% against the dollar, ¥151.69 to ¥149.61 F310F311, and once the segment local-currency growth is weighted the net FX contribution to the +2.5% revenue growth D39 is approximately zero F314F316F318F320. The growth was organic. The flattery arrived instead through the two legs I ranked lower — the liquidation gain F1063 and the tax rate it dragged from 31.73% to 15.09% D86D85. A structural fact is not a rate move, and being right for the wrong reason is a calibration failure even when the headline scores as a hit.

CLAUDE, continued. Three misses I will not bury. Japan's operating margin sits above the group's, 9.58% against 9.40% D49D23, against my explicit prediction of the opposite, and Europe at 6.78% D51 is the weak segment — that inverted the story. I sized Europe at 20–35% of revenue; it is 15.2% F1474F5. And I predicted a multiple of 10–17× on flattered earnings against an actual 8.13× D8, because I priced the class and ignored both the net-cash compression D93 and how thoroughly this stock has already been marked down — 115.6 against 202.2 over five years F178F183. The one call that was well calibrated is the one that decided the study: P ≈ 0.70 that cumulative five-year free cash flow would print below 60% of cumulative net income, band 30–80%. It printed 54.9% D118F11F12F13F14F15, and I had registered in advance that this rather than the multiple would decide the owner arithmetic. It did.

BUFFETT. That is the same finding as my B48 and my B45, arrived at from the other end. I got there by looking at machinery lives and a capex ratio of 2.02 times depreciation D119; you got there by writing the ratio down before you saw it. The pre-registration is worth something precisely because the number could have come back at eighty percent and neither of us would have been able to pretend otherwise.

CLAUDE. And the honest reconciliation of the "cheap" reading is one line of arithmetic. The current ordinary margin is the highest of the five available years D29; on the five-year mean the shares trade at roughly 15× mid-cycle earnings rather than 8.13× reported D8D3. The stock is not cheap; the earnings are high. Nothing changed except which number went in the denominator.

---

On where the moat actually lives — the disagreement that is really a shared complaint about the filing.

MUNGER (M18, name the moat's mechanism; M34, its fingerprint in the figures). I can name the mechanism, which is more than I usually can: an applicator qualified into the customer's own line, with revenue booked only at acceptance after installation and trial run E267 and the auditor making that period attribution a Key Audit Matter E293. That is a real mechanism, not a story. But I cannot find its fingerprint. No retention rate. No price-versus-volume split. No competitor filing. Two years of gross margin, 21.19% and 22.20% D20D21. A moat I can name but cannot measure is one I must take partly on faith, and faith is not what the ledger is for.

LI LU. Sharper than that, and it is the study's cleanest finding. The machinery line — the plausible lock-in — is 10.64% of external revenue D59F1468, ¥23,170M against ¥194,581M of label F1821F1820, and its margin is nowhere separated. The filing declines production and order-backlog detail by segment on the express ground that the group 「単一事業である」 E105 while defining four geographic reportable segments in the same document E270. So the moat lives in a tail, and the tail's economics are structurally unprinted. What is printed is the geography, and I sized the moat with it rather than describing it: 6.78% to 9.67% across four segments D51D50 is a competent converter, not a monopoly.

CLAUDE. And the one hard forward number that exists points gently the wrong way: machinery remaining performance obligations fell ¥15,766M to ¥15,401M F1476F1482 in the year machinery revenue rose from ¥21,114M to ¥23,170M F1442F1468. One year of a book covering roughly one to three years E269 is noise, not evidence — but it is the only forward series the filing gives, and it is not rising.

MUNGER. Which is precisely the distinction worth teaching. The mechanism's existence is established. Its size, its source and its durability are not, and a single-segment filer with four geographic segments will never establish them. That is why my verdict is not a pass — the economics are judgeable — and why it is not a buy at any price this year: I am not paying for a mechanism whose size the company has structurally declined to print.

---

The bet, located

Strip the agreement away and one proposition is left, and the panel is unanimous that it is the only one that matters: does the capital now being spent — capex at 2.02× depreciation D119, construction in progress from ¥5,328M to ¥13,791M F948F949, ¥100bn committed over seven years F274F275 — earn the ~30% incremental operating margin the FY2025-to-FY2026 step implied, or does its depreciation land on a business that reverts toward its 7.41% five-year mean ordinary margin? Operationally that reduces to one question the archive cannot answer: whether percentage gross margin held through the FY2023 input-cost spike, unavailable because the five-year highlights table prints 経常利益 rather than 売上総利益 F6F10 and this study archived one filing. If it held, the franchise reading is earned, the through-cycle margin assumption rises, and every threshold in this thesis rises materially with it. If it did not, FY2026 is a peak and the five-year mean is the ceiling. Everything else — the borrowed year F1063D85, the ~55% cash conversion D118, the fortress balance sheet F45D89, the ~32% allegiant bloc F639F643F649F647F668F736F675, the unmeasurable moat E105D59, the fair-not-cheap price D6D11 — is agreed. Notably, management's own FY2027 guidance already embeds an incremental operating margin of 16.0%, being ¥1,737M of operating profit on ¥10,848M of revenue F385F305F383F5, against the 29.9% delivered in FY2026 D23. It resolves in the FY2027/3 to FY2029/3 operating-margin and free-cash-flow prints.

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.

Self-distance note. The Claude lens holds one of the five verdicts compared above (watch, implied buy-below ¥1,250) 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. Two 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 — including three misses reported as prominently as the hits; and its jury of selves produced one divergence, on C84, that capped its own verdict, which is the mechanism working rather than a claim that it always will.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp. One lens issued an explicit threshold — Pabrai at ¥1,600; Claude publishes an implied threshold of ¥1,250 with a published band of ¥1,080–1,330; Buffett, Munger and Li Lu issue no number, each for a stated reason (Buffett: a five-year archive where he wants ten, and an unlabelled ¥1,205M charge F1074; Munger: the block is character and disclosure, not price; Li Lu: opacity caps the verdict whatever the quote E249E308).
• pass / watch / too-hard are recorded but unscored in any future review. All five verdicts here are watch; both named prices sit far below the ¥3,150 stamp D3 and are the level at which each lens would revisit toward buy.
• 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. Note that there is no stock split inside the five-year window — the only split in the document is the 2-for-1 of 2017-01-01 F611F612, which predates the window, and 発行済株式総数 has been 60,161,956株 at every one of the five year-ends F112F116F608 — so every per-share figure in this thesis is on the ordinary, unadjusted basis. What does move the denominator is the 従業員持株ＥＳＯＰ信託口. Treasury is disclosed on three scopes at 2026-03-31 — 6,460,391株, 6,460,300株 and 6,818,091株 F636F683F1299 — and this thesis uses the trust-inclusive scope throughout, 53,343,865 shares net of treasury D1, because that is the denominator the filing's own BPS F35 and EPS F40 are struck on. Any future corporate action restates from that basis.
• Every FY2026 profitability figure in this thesis states its basis. Reported: net income ¥20,655M F15, EPS ¥387.43 F40, ROE 13.5% F50, P/E 8.13× D8, effective tax rate 15.09% D85, payout 20.9% F703. Ex-one-off: net income ¥15,671M D9, EPS ¥293.94 D10, ROE 10.22% D70, P/E 10.72× D11, and on the company's own restatement an adjusted net income of ¥14,299M D135 behind a 30.2% payout F704. Where a lens normalises at the statutory 30.59% rate F1415 rather than the reported rate, it says so.
• 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 — five of five lenses had converged on watch, a non-decline — to argue pass: that a watchlist slot spent on 7864 is a slot wasted. It is unusual in the record for having stated its own limits before its case, and those limits are worth recording first, in its own words.

Where the bear case does not win: price. "At 1.04× book D6, 15% of market value in net cash D93, 4.94× EV/EBITDA D96 and 10.98× on the company's own guidance D15, this is not a demanding valuation. I cannot honestly argue that the price embeds optimism. It embeds roughly what the bear case describes. Anyone claiming pass on the grounds of 'too expensive' is wrong." Where the bear case does not win: solvency. "71.3% equity F45, 0.49 years of debt-to-cash-flow D171, an A rating E109 and 73.5× interest cover F334 mean there is no path from here to permanent capital loss through the balance sheet, however the leverage rhetoric plays out." Its case therefore rests entirely on the two things a watch presupposes: that earning power is knowable, and that the record is falsifiable at the points that matter. Its three ranked arguments, engaged by name:

1. "The Borrowed Year." FY2026's 69.3% profit growth F309 is an accounting event, not an economic one. The ¥4,984M gain F1063 is recycled out of 為替換算調整勘定 as a △4,984 組替調整額 F1280 and adds nothing to comprehensive income, which grew 10.3%, ¥18,776M to ¥20,715M F19F20. The effective tax rate fell 31.73% to 15.09% D86D85 on a tax effect tied to investments in subsidiaries E262 worth roughly ¥2,885M against pre-tax profit of ¥24,324M F1079 — so the after-tax benefit is larger than the pre-tax gain the ledger flags. Strip the gain and tax the remainder at the printed statutory 30.59% F1415 and FY2026 net income is about ¥13,424M against ¥12,199M F14 — up roughly 10%, on revenue up 2.5% D39. "The multiple you are paying is 10.98× guidance D15 or ~12.5× normalised, not 8.13× D8." Fully conceded, and it is the panel's own first finding — Buffett's B46, Pabrai's P56, Li Lu's L39 and Claude's outside-view Claim P reached it independently, and no thesis in this study uses the reported multiple. The one qualification the red team itself supplies is real: a statutory-rate normalisation is punitive because this group's structural rate is below 30.59% even in a normal year, and FY2027 guidance of ¥15,300M F389 sits above the company's own adjusted FY2026 of ¥14,299M D135. Read that way, management is guiding underlying profit up, not down — which is why the guided −25.9% F390 is arithmetic rather than deterioration, and why watch rather than pass survives this point.

2. "The Cash That Did Not Arrive." Cumulative five-year free cash flow of ¥30,784M against ¥56,117M of cumulative net income D118F11F12F13F14F15 is 54.9%; the ¥168.0bn market capitalisation D3 is 27.3× the five-year average and enterprise value of ¥142,784M D94 is 23.2×. And FY2026's reported ¥9,070M D115 is worse than it looks: ¥5,896M of time deposits withdrawn against ¥929M paid in F1180F1178 is money moving between pockets, so on a capex basis — operating cash flow less tangible and intangible purchases — the year produced about ¥4.3bn F60F1182F1186, against roughly ¥14.4bn the prior year F59F1181F1185. Conceded, and it is the study's load-bearing arithmetic — Buffett's B48, Munger's M38, Li Lu's L48 and Claude's C23 all turn on it, and it is the one prediction the Claude lens registered in advance and got right. The honest counterweight, which the red team also writes: FY2026 is a deliberate capacity build, capex at 2.02× depreciation against 0.76× the year before D119D120, with two dated, budgeted, part-paid projects behind it — Tendō at ¥12,500M completing 2026-06 and Thailand at ¥7,338M completing 2027-08 F602F604E119 — and judging a converter mid-build on that year's free cash flow charges the cost while refusing the benefit. Operating cash flow converts at 1.05× operating profit D125. That is exactly why the capex line is the thing to watch rather than the thing already decided.

3. "The ¥1.2bn Silence." A ¥1,205M litigation settlement F1074 — 7.7% of ex-gain net income D9 — disclosed in one sentence with no counterparty, no subject, no jurisdiction and no date in 122 pages E249, charged at the parent too F1728 with a ¥379M deferred tax asset F1795, so the listed filer itself was the defendant; the risk section contemplates 訴訟 only as a hypothetical consequence of intellectual-property infringement E90 and names neither this litigation nor commercial litigation generally among its risks, the nearest being product claims E93 and changes in legal regulation E94; an 臨時報告書 filed 2025-11-18 under the material-effect trigger is listed and never reproduced E308; the Group Internal Audit Office has no direct board reporting line E192; and the family-and-foundation bloc is printed in six pieces and never summed F639F643F649F647F668F736E127. "The company certified to the regulator that something materially affected its accounts and did not tell its shareholders what." This is where the red team wins outright, and the synthesis adopts it: it is Buffett's B54 and B59, Munger's M60 and M10, Li Lu's L51 and Claude's C98(3), reached independently by four lenses, and it is why three of five decline to name any price at all. The defences the red team wrote against itself are honest and partial — Japanese practice routinely omits counterparty detail in a settled 訴訟関連損失 under a confidentiality term; the 参考情報 section is an index rather than a reproduction; the internal-audit office is deliberately placed under an audit committee that is 100% outside directors including a CPA E189, which is arguably stronger than reporting to a board the CEO chairs E144. Good defences, individually. Together they still describe a filer whose default under discretion is the legal minimum.

The red team's own verdict, in its words: "watch is defensible; it is not obviously wrong. What I believe is that the unanimity is doing work the evidence has not earned" — five lenses looking at a 69.3% profit print F309, a 1.04× book multiple D6 and a net-cash balance sheet D89, and each finding enough not to reject. Its narrower honest position: "there is a price for this asset, and no reliable way from this filing to know what it is." And its closing correction is adopted here on the record: "the correction that would change my mind is not a lower price — it is a filing that answers what the company was sued about and what its machinery business earns." Both are cheap for the company to supply, neither is guaranteed, and the second is structurally unlikely while the group reports as a single business across four geographic segments E105E270. Read watch on 7864 accordingly: watch the capex line, the gross-margin series and the two disclosures — explicitly NOT "cheap at 8.1× and 1.04× book," because on this filing's own arithmetic that framing is false D8D6D11.

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 two observables — the capex line and the margin level — which is precisely the red team's recommended emphasis.

• Buffett (watch, no buy-below). Once the Tendō and Thai plants are finished F602F604 and capital spending falls back to or below depreciation of ¥8,423M F1132, three consecutive years of return on net tangible capital above 15% together with free cash flow above 70% of reported net income would show the economics are better than the last five years said — and this lens wrong to treat the company as a converter rather than a franchise D80D118.
• Munger (watch, no buy-below). Toward pass: a future filing showing the CEO-chaired Nominating or Remuneration Committee approving a decision that visibly favours the founder bloc over minority shareholders E150E152E126, or the ¥1,205M settlement's particulars surfacing to reveal a recurring compliance or product-liability pattern rather than a closed one-off dispute F1074E249. Toward a priced verdict: the company voluntarily aggregating and disclosing the combined founder-linked stake E127, giving the Group Internal Audit Office a direct board-reporting line E192, naming the litigation counterparty, and the guided 25.9% net-income decline resolving to a stated cause F390E113 — and a competitor filing or customer-retention disclosure turning up that gives the machine-and-consumable switching-cost claim an actual fingerprint in the numbers E267.
• Pabrai (watch, buy below ¥1,600). Withdraw the threshold if either (a) a later yūhō prints consolidated 営業利益 below ¥17,000M, which removes the ¥18,844M normalisation the value is struck on F304, or (b) 借入金等明細表 合計 当期末残高 exceeds 現金及び預金 — net cash turning negative as the ¥43.5bn and ¥56.5bn investment frames F274F275 and the stated intention to raise financial leverage E64 are executed, which removes the ¥25,249M cushion the value is struck on D89F1603F905.
• Li Lu (watch, no buy-below). This watch is wrong if consolidated operating margin reaches and holds at or above 10% for two consecutive fiscal years while Japan segment revenue stops declining D23F314 — that would falsify the reading of this as an ordinary ~9%-margin converter and make the FSG.30 double-digit ambition E61F273 an economic claim rather than an aspiration. Conversely, machinery remaining-performance-obligation backlog falling two years running alongside shrinking machinery revenue F1482F1468 would falsify the applicator-lock-in premise and move the verdict to pass.
• Claude (watch, implied buy-below ¥1,250). Three, each with a threshold and a named resolving document. (a) Level: if the FY2027/3 yūhō 連結損益計算書 shows operating profit below ¥20,463M F305, FY2026 was a peak — verdict to pass. (b) Japan: if the FY2027/3 or FY2028/3 セグメント情報 shows Japan segment profit below ¥9,000M F1528, or a second consolidated 組織再編費用 or 減損損失 line of ¥300M or more appears F1072F1073, the domestic restructuring is wider than disclosed — verdict to pass. (c) Conversion: if the FY2029/3 yūhō shows cumulative five-year free cash flow still below 60% of cumulative net income while capex remains above 1.4× depreciation D118D119, the reinvestment is confirmed value-destroying — verdict to pass, permanently. Symmetric upgrade trigger: operating margin at or above 10.0% in two consecutive years in which 為替差益 is not a gain D23F1046, with operating profit at or above ¥25,000M F305. Review by 2027-07-31.

The single observable most lenses converge on is whether capital spending reverts toward depreciation once Tendō and Thailand complete, and whether free cash flow recovers with it D119F602F604D115 — together with whether the FY2023 gross-margin series, resolvable now from EDINET's 第65期 and 第66期 filings, shows percentage margin holding through the input-cost spike D20D21D26. The first resolves by time; the second could have resolved this week.

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, of which two are structural. B42 needs a maintenance-capex fallback for the "combined investment envelope" disclosure: the item's fallbacks assume maintenance capex and depreciation are either disclosed or absent, but this filer publishes a single seven-year frame explicitly combining ordinary and strategic spending E62F274F275 and never splits it. Propose a named third fallback — where a filer publishes a combined envelope, compute owner earnings at both the depreciation floor and the trailing multi-year mean capex, report the range, and never take the single point. No item asks what a filing declines to say: B54 covers footnotes that resist explanation, and nothing covers a charge that is quantified but entirely unnarrated — ¥1,205M with no counterparty, subject, jurisdiction or date F1074E249 beside an extraordinary report whose contents are not reproduced E308. Propose a new Earnings-quality item, undisclosed material events: enumerate every charge above ~5% of operating profit whose narrative does not identify counterparty and cause, and treat each as an unbounded liability in B14's permanent-loss inventory. Also: B45 and B52 assume a ten-year window a single-yūhō archive cannot supply and should route to data-insufficient with the shortfall named in years rather than letting a two-of-two-years reading masquerade as a decade test; B33 breaks when the five-year table prints 経常利益 but not 営業利益 F6F10 and should name the ordinary-profit proxy while deducting the translation-adjustment movement from the denominator, since currency inflates invested capital without owners contributing it F1027; and B26's inflation fingerprint needs a companion FCF-conversion test folded into B48 with a threshold — here the sharper number than tangible-assets-to-sales was cumulative free cash flow at 54.9% of cumulative net income D118.
• Munger — M20/M21/M28/M30 assume a consumer-facing brand and do not map onto a B2B industrial-consumable-plus-machinery model; a future version should flag them explicitly as consumer-brand-only and mark them not-applicable for pure B2B rather than leaving the agent to infer it each study. M31's consumer-psychology framework has the same gap: an industrial switching-cost moat — an installed machine and a requalification cost E267 — deserves its own item alongside M27 rather than being asked to fit the consumer-tendency catalogue. Both recur across Japanese industrial suppliers and both cost this lens real answer quality here.
• Pabrai — three, all failure-derived. P1 needs a de-rate floor clause. The item already says an equity cushion is not an asset floor, which did work here, but it only says what not to count and gives no positive floor for the common Japanese case of a healthy, net-cash, no-distress company where liquidation arithmetic is far below the price and therefore not binding. Add to Good: where the company is in net cash with no distress, recompute the lowest price-to-book the security itself carried inside the filed window — price = printed EPS × printed PER, divided by printed BPS, all from 主要な経営指標等の推移 — and treat that multiple on today's book as the working floor. Here ¥125.43 × 12.0 ÷ ¥2,201.40 = 0.68× F37F52F32 produced a −33% floor the liquidation test would have buried under a −56% number nobody believes. P20 should score stated future leverage, not only current leverage: this company passes P20 comfortably today and in the same document discloses an intention to add borrowed capital E64 alongside ¥100bn of frames F274F275 — credit the cushion at what management says will remain, not at today's balance. P56 must require operating profit to be separated from non-operating FX in a multi-currency group: the Japanese five-year table prints 経常利益 and not 営業利益 F6, so the only long series available to a normaliser already contains the FX result, and here ¥2,067M of a ¥3,686M ordinary-profit improvement F10F9 was a currency swing F1046F1053. That is a general Japan-filing trap, not a quirk.
• Li Lu — three, each grounded in a source the item it amends already cites. L43 needs a measurement instruction: the present wording asks for lock-in evidence and a winner-take-all judgement but never tells the agent to size the moat. Propose: where the filing reports segment or product-line margins, state the moat's measured size in margin points before assessing any narrative claim about it; a switching cost that coexists with single-digit operating margins is a real cost of a small size and must be described that way. This study's cleanest finding came from doing that unprompted D49D50D51D52. L2 should require the worst case to declare the length of its window — where the filed window is shorter than one full cycle, build the worst case from the worst year inside it and label it an incomplete floor, naming the missing filing years; here the five-year window contains neither 2008–09 nor 2020, and FY2023's 4.58% ordinary margin D26 is the only stress available. L46 should name the withheld-own-KPI pattern explicitly: an indicator the filing states its board monitors but does not quantify is a disclosure failure of the same class as a consolidated blob and should count toward the opacity cap — here the board's own environmental-product measure E75 and the 「２桁％」 targets printed without numbers E111 are the two most decision-relevant figures in the document and neither is printed.
• Claude — four checklist sharpenings, plus two library entries. (1) C28 should extend its Asks to the segment-to-consolidated adjustment line: ¥482M of this group's ¥1,619M operating-profit improvement — 30% of it — came from the unrealised-profit elimination swinging −¥608M to −¥126M F1496F1533, and nothing in the checklist looks at the reconciling line; it was found only by footing the segment note back to the consolidated statement. (2) C9 must be made robust to the Japanese five-year table: the item asks for a ten-year operating-margin range, and 主要な経営指標等の推移 routinely prints 経常利益 and not 営業利益 F6F10, so the item silently degrades to data-insufficient on the exact metric it names — permit the ordinary-margin proxy with its FX contamination stated. (3) C44 must require the hurdle and publish a band: the implied threshold moved from ¥1,150 at a 10% hurdle to ¥1,420 at 8% on identical inputs with full cash credit D1, and three jury selves used a different hurdle from the lens's own; a single number implies a precision the derivation does not have. (4) C1/C6 should register a prior on relative five-year total shareholder return: this company returned 115.6 against the index's 202.2 F178F183 and no prior was registered on it, because the outside view was thinking about business quality rather than about the market's own scorecard on that quality — the 株主総利回り series is an index, not a price, so it leaks no stamp and is fully compatible with a figures-blind stage, and it is the most direct evidence available on whether the mispricing a lens is hunting has already been adjudicated.
• Claude library (v0.1.0 → next). Class-level, N=1, no rate claimed: a mid-cap, globally-footprinted converter of decorative primary packaging — shrink sleeves, pressure-sensitive labels, pouches — with a minority captive applicator-machinery arm, majority-overseas revenue, founder-family controlled, TSE-listed. Datapoint: FUJI SEAL (7864) FY2026 — through-cycle post-tax ROIC roughly 6.5% on invested capital including cash and 8.2% excluding it D81D82, struck at the five-year mean ordinary margin of 7.41% D25D26D27D28D29 rather than the reporting year's peak, against a peak-year 10.44% ex-cash D84; cumulative five-year free cash flow 54.9% of cumulative net income D118; five-year total shareholder return 115.6 against 202.2 F178F183; machinery 10.64% of external revenue with its margin never separable D59E105. Second candidate class awaiting a member: a Japanese issuer whose reporting-year net income is flattered by a subsidiary-liquidation gain recycled out of the cumulative translation adjustment — the transferable mechanic being that the magnitude is derivable before any income-statement figure is seen, from the two payout ratios the company itself prints, ¥81.00 ÷ 30.2% implying adjusted net income and the difference from reported being the exclusion F136F704D134D135D136. Case-level, so it never anchors a future outside view: "A structural fact is not a rate move. An FX-flattery prior must be conditioned on the period's average-rate change, never on the geographic mix of earnings alone; where the packet withholds the rate series, register the prior explicitly as conditional on the rate having moved, and say so in the same sentence." Source: this study — the outside view weighted FX translation as the lead flattery mechanism, and the yen strengthened 1.4% F310F311 while the year's revenue growth turned out essentially all organic D39.

---

Corrections