EBARA JITSUGYO Co., Ltd. (6328): A Good Business on a Rented Moat, at a Price That Leaves No Room

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	¥1,850	B92/B89/B93; B23/B25
Munger	watch	— (conditional ~¥1,800–2,000)	M18; M82; M44
Pabrai	watch	¥1,250	P53; P18/P56; P1
Li Lu	watch	¥1,650	L1; L18/L5; L48
Claude	watch	implied ¥1,600	C35; C40/C9; C33

All five lenses independently reached watch — the batch's FIRST unanimous non-decline after nineteen studies, and the first name to clear the quality bar every deep-value name failed. The shared finding: this is a genuinely good business — durable 15–17% ROE [F47–F51], net cash +¥13,646M D14, asset-light F66, owner-friendly (payout 35%→40% [F322/F323] + a ¥1bn buyback E72), with an activist on the register F352 — offered at 2.01× book D1 / 12.78× earnings D2, which is fair, not cheap, and so carries no margin of safety. The panel waits for a ~30% cheaper entry (buy-belows ¥1,250–1,850). This directly answers the batch's question — does quality at a fair price get a buy? — No: it gets a disciplined watch. The crux every lens circles: is the moat owned or borrowed — a privileged-but-terminable Ebara agency (contract auto-extends only one year past 2026-09-30 E57; ~67% of merchandise purchases from Ebara [F304/D23]) plus government-budget demand (66% of sales to 官公庁 [F294/D24]) — and is FY2025 a peak (Engineering segment profit +76.8% D25; management's own EJ2027 plan guides operating profit ¥6,300M→¥5,500M, FY2026→FY2027 [F411/F412])?

The business

EBARA JITSUGYO (荏原実業) sells water and environmental-infrastructure equipment in Japan, and it does three distinct things under one roof E7. A shopkeeper can follow it in a sentence.

The Maker segment (メーカー) designs its own gadgets — ozone-concentration meters, deodorizing units, infection-control products, aquaculture equipment, home storage batteries — and has partner factories build them; the company owns no plant of its own and is fabless E8E38. It owns the IP and the production know-how, and outsources the metal-bending. In FY2025 this segment turned over ¥7,506M of external revenue at ¥1,247M of profit — down 5% on higher personnel cost F253F259E48.

The Engineering segment (エンジニアリング) takes contracts directly from city and prefectural governments (官公庁) to design and build water-treatment plants, buys the pumps and materials from the Ebara Group, and subcontracts the digging to a ~70-firm partner network it calls the Ginsuikai E9E31. This is the biggest arm — ¥22,450M of revenue F255, ¥4,332M of profit F261 — and it is 97% government by revenue (¥21,861M of ¥22,450M to 官公庁 F274).

The Trading segment (商社) resells Ebara-Group pumps, boilers and energy-saving HVAC to private buyers, earning a distribution spread E10 — ¥11,254M of revenue, ¥1,883M of profit [F257/F263].

Two facts sit at the center of the whole enterprise, and they are the reason the study exists. First, about two-thirds of what the company sells goes to the government: 66.3% of sales and 67.3% of orders are to 官公庁 [F294/D24, F288/D31], a budget-cycle-driven demand base with hard December-to-March seasonality [E33/E34]. Second, about two-thirds of the merchandise it buys to resell comes from one supplier — the Ebara Group (㈱荏原製作所), at 67.0% of merchandise purchases and 19.3% of product/construction COGS [F304/D23, F300/D22]. EBARA JITSUGYO has been Ebara's distribution agent (販売代理店) since July 1950 — seventy-five years of unbroken commercial relationship E3 — and it even holds Ebara shares (¥4,036M) as its largest policy cross-holding, expressly to protect that relationship E102. But there is no capital tie the other way: the company owns 0.24% of Ebara Corp F313, Ebara owns none of it, and the filing states plainly that Ebara is not a related party and not a parent [E11/E124/E125].

This is the borrowed-moat nuance the whole panel wrestles with. Customers keep paying because water infrastructure is aging and disaster-resilience spending is rising E23, and because a municipal engineer specifying a new sewage plant reaches the standing, certified channel — EBARA JITSUGYO — the way he reaches a habit. But that channel rests on two things the company does not control: a government's budget, and a supplier's contract. The basic agency contract for fluid-machinery products (including engineering and incidental work) is dated Oct 1 2024 and auto-extends for only one year after 2026-09-30 E57; management lists non-renewal or substantial shrinkage of the Ebara relationship as a named risk E36. The durability is real — seventy-five years of it — but for a decade-long owner it is rented durability, not owned. The aging-water tailwind is genuine and structural E23, but management itself names the counter-vector: population decline E24. The end-market is not the problem; whose the moat is, and how normal FY2025's earnings are, is the whole debate.

The numbers

At the ¥2,354 stamp (post the 2026-01-01 2-for-1 split E140), EBARA JITSUGYO is a genuinely good business offered at a fair price — the mirror of every deep-value name on the record.

The record is durable, and this is the finding that separates the name from its predecessors. Recurring profit (経常利益) compounded ¥4,110M → ¥6,316M over five years [F6/F10], never once a loss year; return on equity ran 17.62 / 11.55 / 15.93 / 14.20 / 17.06% [F47–F51] (recomputing cleanly to 17.06% D5), averaging ~15.3% with the trough at ~11.6% — a real, cyclical, un-confected series. FY2025 was a record across the board: revenue ¥41,211M (+9.9%), operating profit ¥6,121M (+44.0%), recurring profit ¥6,316M (+42.2%), net income ¥4,384M (+38.8%) [F5/F87/F10/F220, E45]. And it is earned with almost no debt. Total borrowings are ¥1,524M F171 against ¥15,170M of cash F153 — net cash of +¥13,646M D14, ~24% of the ¥56.6bn net market cap D4, about ¥528 a share D15. Equity ratio 57.7% F46; interest coverage 130× E55. The business is asset-light to the point of comedy: investing cash flow of just −¥205M on a ¥41bn revenue base F66, capex ¥428M (of which ¥190M was research-lab land E61, i.e. growth, not maintenance) against depreciation ¥234M [F331/F233], and tangible fixed assets of only ¥3,269M F162. Owner earnings are close to reported net income. Strip the idle cash and the operating business earns a very high return on the tangible capital it truly uses; on the whole book the ROIC proxy is ~14.8% D21.

The valuation is fair, not cheap — the exact opposite of a margin-of-safety bargain. P/B = 2,354 ÷ split-adj BPS ¥1,172.56 = 2.01× D1; P/E = 2,354 ÷ split-adj EPS ¥184.24 = 12.78× D2; dividend yield ¥60.00 ÷ 2,354 = 2.55% D13; EV ≈ ¥43.0bn D17, EV/経常 ≈ 6.8× on the peak recurring line D19 (but ~9.8× on mid-cycle recurring). Graham would walk right past it — net current assets are roughly ¥569 a share (current assets ¥35,168M less all liabilities ¥20,443M [F161/F166]) against a ¥2,354 price, about four times covered; this is not a net-net, and the net cash, though real, is only ~22% of the price. At 2× book there is no asset floor doing the analyst's work; the whole case rests on the quality holding up.

The record year leans on one lumpy segment, and management guides the peak down. The +44% jump in operating profit came almost entirely from Engineering segment profit +76.8% (¥2,450M → ¥4,332M) [D25/F260/F261], which the MD&A attributes to a high opening order backlog converting on schedule — timing, not run-rate E47. Maker was down 5% and Trading up 8% [F259/F263]; strip the backlog conversion and there is no record. The incremental operating margin FY2024→25 was ~50% — far above the ~14.9% average D7 — which reads "earnings high," not "moat widening." Management's own EJ2027 plan then puts operating profit at ¥6,300M (FY2026) → ¥5,500M (FY2027) [F411/F412] — declining — with the plan-period operating margin guided to 12.2% E21, below the 14.9% just booked F89; FY2025's ¥6,121M already exceeds the FY2027 plan. And the headline yield is flattered: FY2025's ¥120 pre-split DPS includes a ¥20 80th-anniversary commemorative (¥10 interim + ¥10 year-end) [F324/E74], so the recurring post-split yield is ~2.1%, not 2.55%. Cash conversion was lumpy this year — operating cash flow ¥1,413M against net income ¥4,384M [F61/F220] — because trade receivables and contract assets rose ¥3,157M as the Engineering backlog converted F235 (contract assets ¥7,856M → ¥11,128M E113); a benign, disclosed percentage-of-completion timing feature — the sole Key Audit Matter E134 — not an accrual game, but a reason to normalize earnings down.

The five lenses

Buffett — watch

Let me tell you what this company actually does, because that is where every one of these decisions starts. Ebara Jitsugyo sells water and environmental equipment in Japan, and it does three things under one roof — it designs its own gadgets and has other people build them E8, it takes contracts straight from city and prefectural governments to design and build water-treatment plants E9, and it runs a little trading desk reselling Ebara equipment E10. Two facts sit at the center: about two-thirds of what it sells goes to the government [F294/D24], and about two-thirds of the merchandise it buys comes from one supplier, Ebara Corp [F304/D23]. A shopkeeper could follow that in a sentence, so the circle-of-competence gate opens — I understand how the cash gets in the door.

The record is the kind I like to look at. Recurring profit went from ¥4,110M to ¥6,316M over five years [F6/F10], never once losing money, and return on equity has run 17.6, 11.6, 15.9, 14.2, 17.1 percent [F47–F51] — good returns, and here is the part that matters, earned with almost no debt. The whole balance sheet carries ¥1,524M of borrowings F171 against ¥15,170M of cash F153 — roughly ¥13.6bn of net cash D14, about ¥528 a share D15, equity ratio 57.7% F46. It throws off cash without needing to eat much — depreciation ¥234M against a capital budget of ¥428M, and ¥190M of even that was buying land for a research lab, not staying in business [F233/F331/E61]. That is not a good business; that is a very good business hiding behind a fortress balance sheet.

Now, what would I pay, and what does the market ask? It asks ¥2,354 — two times book D1 and about thirteen times earnings D2. My old teacher Ben Graham would have walked right past this: net current assets are about ¥569 a share [F161−F166] against the price, four times covered, nowhere near a net-net. This is not a cigar butt. If I buy it, I am buying a good business at a fair price, and the whole case has to rest on the quality holding up — because at 2× book, there is no asset floor doing my work.

So I have to be honest about the quality, and here I get uneasy. A genuine franchise is something customers need, can't easily substitute, and where nobody sets your price [B23]. Two-thirds of this business is selling to government bodies whose budgets can be squeezed and whose contracts are competitively bid E33 — a customer relationship, not pricing power. And the moat I can actually point to isn't Ebara Jitsugyo's at all — it's the Ebara agency contract [E3/E57], which gives them the products to sell, and which auto-renews but expires September 2026 and only extends one year at a time thereafter E57. A moat you rent from someone else, on a one-year rollover, is not the enduring kind I want to own for a decade without a quote. It has held for seventy-five years, which counts for a great deal — but I cannot underwrite it as if the company controls it, because they don't. The record this year sharpens the worry: the whole jump in profit came from one segment — Engineering profit up 76.8% D25 on a backlog burning off E47 — and management's own plan guides operating profit down the next two years [F411/F412]. I'd value it on the cycle, not on the peak. And the founder, born 1939, is both Chairman and CEO at 86 E88 — a succession question I cannot wave away.

Here is where I come out. This is a genuinely good, understandable business, conservatively financed, run by people who tell the truth and treat owners fairly — a payout raised 35%→40% and a billion yen of buyback [E73/E72]. But it is offered at a fair price, not a cheap one — an owner-earnings yield around 7.7% on the net market cap D4, which is fine, not fat, and the durable edge underneath it is a rented agency contract and a government-budget customer base. When the quality is arguable and the price is merely fair, I don't swing — there are no called strikes. I'd want this at a price where the owner-earnings yield reaches double digits and I'm paying closer to 1.6× book — around ¥1,850 — which also buys me a margin of safety against the day the Ebara contract or the government budget disappoints. Until then, I watch.

What a student should take from this: when the business is good but the moat belongs to somebody else — a supplier's contract, a government's budget — you are renting the durability, not owning it, and you must price for the day the lease ends. A fair price for a genuinely good business is a watch, not a buy; the discount, not the quality alone, is what protects you when the thing you can't control finally moves against you.

Munger — watch

Invert first, as always. How does this business die? The obituary writes itself in one paragraph: Ebara Corporation — the source of two-thirds of EJ's merchandise-purchase cost [D23/F304] and the dominant equipment supplier — declines to renew its agency contract, which it can do on one year's notice after September 2026 E57. EJ owns 0.24% of Ebara Corp F313; Ebara owns none of EJ E11. The capital relationship is nil; the commercial relationship is everything. Kill path two: a Japanese public-works austerity cycle — two-thirds of sales are to public bodies D24, and the Engineering segment that drove FY2025's 76.8% surge is 97% public-sector [E121/F274]. Kill path three: an 86-year-old founder who is simultaneously Chairman and CEO E88, holding relationship capital an org chart cannot transfer.

Now the business, plainly stated. EJ is a privileged distribution agent for the Ebara Group [E7/E11]. It does not make pumps; it sells them, installs water infrastructure for municipalities, and runs a smaller trading book. The moat has two distinct layers, and I must not conflate them. Layer one, the Ebara relationship — seventy-five years of unbroken commercial entanglement E3, a conditioned reflex where the public engineer specifying a plant reaches EJ the way he reaches his dentist. This is close to a franchise, but it is a borrowed one: the brand belongs to Ebara Corp, not EJ. Layer two, the Maker segment's own IP — ozone meters, PFAS filtration, storage-battery flood-gates E60 — genuine EJ inventions, but small (¥1,247M of profit, down 5% F259).

The figure table is genuinely good: five-year ROE 17.6 / 11.6 / 15.9 / 14.2 / 17.1% [F47–F51], the FY2022 dip to 11.6% showing this is real and cyclical, not confected. Net cash ≈+¥13,646M against ¥1,524M of borrowings [D14/F171], interest coverage 130× E55, investing cash flow −¥205M on a ¥41bn base F66. No bonds, no lease debt, no Level 3 mysteries; the non-trivial items — the Ebara stake at ¥4,036M F316 and contract assets ¥11,128M E113 — are transparent, and the Deloitte KAM zeroes precisely on the over-time revenue estimates with an unqualified opinion [E134/E137]. Incentives are the right three metrics — consolidated OP, net income, ROE E97 — no EBITDA theatre, no options E63, founder skin in the game at 2.48% F362. The activist (Nippon Active Value Fund 7.58% F352) and the payout upgrade to 40% E73 plus the ¥1bn buyback E141 keep the board focused on per-share value.

But what inflated FY2025 was Engineering profit +76.8% D25 on a high opening backlog delivering on schedule E47 — timing, not a permanent step-change. FY2025 is a record made partly by construction timing, and management's own EJ2027 plan guides operating profit ¥6,300M → ¥5,500M, FY2026→FY2027 [F411/F412], with margin to 12.2% E21. Strip the ¥20 commemorative dividend F324 and the sustainable yield is ~2.1%, not 2.55% D13. The load-bearing question is whether the Ebara agency is a durable privileged position or a revocable license — seventy-five years of renewal is strong evidence E3; zero capital tie E11 is the perpetual risk. I cannot put a decade-horizon compounding thesis on a contract that renews annually after 2026. At P/E 12.78× D2 on peak-cycle EPS, the market is pricing continued elevated Engineering margins; normalize Engineering toward its FY2024 level and consolidated OP is nearer ¥4.5bn, an EV/OP around 9.5× on normalized earnings — more than fair, not a bargain.

The verdict is watch. The business quality is genuine, the price is not clearly wrong, but the founder transition is a known unknown, the Ebara renewal is a dated cliff, and FY2025 earnings are elevated by construction timing. At a price that discounts a normalization and removes the commemorative dividend F324 — something in the ¥1,800–2,000 range on normalized EPS of ~¥150 at 12–13× — this becomes a candidate. At ¥2,354 on peak-cycle figures (12.78× D2), patience is the rational posture. There is nothing more to add until the contract renews past 2026, a succession event is disclosed, and Engineering normalizes without a profit collapse.

What a student should take from this: the borrowed-moat problem is the central lesson. When a firm's returns depend on a revocable commercial contract with a more powerful principal — however long it has run — you are not buying a franchise, you are buying a temporary license dressed in franchise statistics. The ROE and the margin look like a Wrigley; the contractual structure is closer to a car dealership. Distinguish between the moat the figures describe and the moat the legal structure protects; here both are real, but only one is unambiguously durable.

Pabrai — watch

Start where Dhandho always starts: what do I lose if I am wrong? Heads I win; tails I don't lose much. Only after that do I let myself look at the upside. The balance sheet here is the kind I like to find in Japan — cash and deposits ¥15,170M F153 against total borrowings of ¥1,524M F171, no bonds, no lease debt, leaving net cash of +¥13,646M D14, about ¥528 a share D15; equity ratio 57.7% F46, interest coverage 130× E55. On top of the cash sit listed securities: the ¥4,036M Ebara stake F316 plus pure-investment names carrying a ¥3,800M unrealized gain E104 — call it ~¥8.2bn of marketable stock. Before I value the operating business at one yen, I have roughly ¥20bn of liquid assets against a ¥56.6bn net market cap D4 — ~36 cents of every purchase yen sitting in cash and quoted stock. This is the IKEA-never-took-debt profile: this company simply cannot be killed by a lender's mood. The whole survival family passes without a fight — two lost years would leave it still cash-rich, still paying dividends, still buying back stock.

So the "tails" branch is real and hard. But here is the discipline: an asset floor is not the same as a fifty-cent dollar. At ¥2,354 the stock trades at 2.01× book D1 and 12.78× earnings D2, EV/経常 6.8× D19. My hard floor — net cash plus haircut securities — is maybe ¥900–1,100 a share; book value is ¥1,172 F35. If the market re-rated to its asset backing, I'd lose roughly a third to a half. That is not "I don't lose much" — the downside is cushioned, not absent.

Now the business, in five plain sentences (P50). EBARA JITSUGYO builds water-treatment plants for Japanese governments (~66% of sales to 官公庁 D24), sells self-designed environmental gear it has contract-manufactured because it owns no factories E38, and trades Ebara-Group pumps to private buyers. It is the distribution agent for Ebara Corp E11, buying ~67% of its merchandise from that group D23. Customers stay because water infrastructure is aging and disaster-resilience spending rising E23. That clears the circle-of-competence gate on simplicity, and it is a slow-changing industry — a sewage plant in 2005 looks like a sewage plant today.

The load-bearing worry is P18 — cyclical trough vs. what I'm actually being handed, which is the opposite: a record year. FY2025 was the best in the company's history — 経常利益 ¥6,316M F10 versus ¥4,110M four years earlier F6 — driven almost entirely by Engineering profit +76.8% [D25/F261] on a high opening backlog burning off E47. I am not buying a distressed business in a distressed industry; there is no fire in this theater, no fearful seller. The ROE series dropping to 11.6% in FY2022 F48 and 経常 to ¥2,929M F7 proves the stream breathes; value the company on a mid-cycle ¥4.5–5.0bn and 13× trailing is really ~16–18× normalized. That is fair value for quality, not a bargain. What I am paid to wait (P17) is genuine — net cash compounding, a dividend (though ¥10 of it is an 80th-anniversary one-off F324, so the durable yield is ~2.1%), a payout rising 35%→40% [F322/F323], a fresh ¥1.0bn buyback E72, and a real activist on the register (Nippon Active Value Fund 7.58% F352). But the excess returns come from an incumbent Ebara agency E11 whose basic contract expires 2026-09-30 with only a one-year auto-extension E57; the two-sided concentration (67% of purchases from one group E36; 66% of sales to budget-constrained governments E33) is exactly what P12 flags.

Verdict: watch. The downside is cushioned by a hard net-cash-and-securities floor (net cash +¥13,646M D14), so this is not a pass — nothing here can permanently impair capital. But the discount is not wide enough for a big Dhandho bet, and I'm being asked to pay a full multiple on peak earnings. Few bets, big bets, infrequent bets: this one doesn't clear the bar at ¥2,354. It clears it in the ¥1,000s — I'd be a confident buyer near ¥1,250, where net cash and securities cover most of the price and I'm buying the water franchise nearly for free.

What a student should take from this: a fortress balance sheet answers "will I go to zero?" — it does not answer "am I buying a fifty-cent dollar?" Those are two different questions, and a hard asset floor at roughly two-fifths of the price is a reason not to pass, not a reason to buy. The second lesson is the trap of the record year: when the cheapness is measured against peak earnings driven by one segment burning off a high backlog, normalize before you pay, or you will mistake the top of a cycle for the floor.

Li Lu — watch

Let me begin where I always begin: not with the price, but with whether I can honestly say I understand this business's next ten years better than most of the people who own it. Everything downstream — the two-times-book, the thirteen-times-earnings, the fortress balance sheet — is idle until that question is answered, because no price rescues a business you cannot predict.

The business is unusually legible, which I respect. Ebara Jitsugyo sells and builds water and environmental infrastructure through three segments — a fabless Maker E8, an Engineering arm taking public-sector orders directly E9, and a Trading arm E10 — and it is a distribution agent for the Ebara Group E7. The economics are genuinely good and durable on the face of the record: ROE 17.6 / 11.6 / 15.9 / 14.2 / 17.1% over five years [F47–F51], recurring profit compounding ¥4,110M to ¥6,316M [F6/F10], equity ratio 57.7% F46, net cash ~¥13.6bn D14 against a ~¥56.6bn net market cap D4. Investing cash flow of −¥205M F66 tells you this earns its returns on almost no plant. The auditor has been Deloitte continuously since 1984, unqualified, no going-concern doubt, the single KAM the percentage-of-completion accounting [E94/E134/E137]; no related-party transactions E124, Ebara explicitly not a parent E125.

So can I predict the earnings ten years out (L1)? The decisive variables are two. First, the public-sector water-infrastructure spend — ~66% of sales, ~67% of orders to 官公庁 [D24/D31] — which management itself frames as two opposing vectors, aging-infrastructure and resilience pulling up, population decline pulling down [E24/E23]. Second, the Ebara agency — ~67% of merchandise purchases Ebara-sourced D23, under a basic contract that "auto-extends only one year after 2026-09-30" E57. I can map each variable to a disclosure — that part of L1 passes. What I cannot do is answer them with the confidence my standard demands. The demand is a government budget line the company neither controls nor prices; the Ebara contract is the hinge of the whole enterprise, and the filing gives me a renewal date, not a renewal guarantee. That forces the honest ledger (L4): two load-bearing inputs are assumptions, not verified facts, and at 2× book with a cash-heavy balance sheet the price does not pre-absorb a bad resolution of either (L5). This is not Timberland, where the feared exposure was under 5% of earnings and the stock was priced for disaster; here the feared exposures are the majority of revenue, and the stock is priced for a good, well-run compounder — fairly, not fearfully.

Now the compounding case, because it is why this is a watch and not a pass. Is intrinsic value growing or melting (L35)? Growing: book value per share compounded ~10.6% p.a. D30, the ROIC proxy sits near 14.8% D21 and higher stripped of idle cash, and management's capital allocation is the virtue I prize most — payout raised 35%→40% E73, a ¥1.0bn buyback E141, no empire-building goodwill, eighty years inside its circle. The four-part gate (L50): superior company economics yes; superior industry economics only partly (public works is stable but not a great business, and the Engineering swing that drove +76.8% D25 was backlog timing E47, not a moat step-change); good management yes, on the action trail; right price not yet. Quality wrinkles too: FY2027 operating profit is guided below FY2026 in the company's own plan [F411/F412], and operating cash flow was ¥1,413M against ¥4,384M net income [F61/F220] because contract assets swelled ¥3,157M F235 — real profits, lumpy cash. And the key-man fact I will not wave away: the founder, born 1939, is Chairman and CEO E88, with a President/COO born 1962 beneath him E89.

Verdict: watch. The knowledge bar bars a buy and would push toward too-hard were the business not so clean and the downside so bounded — which is what pulls it up to watch. A concentrated position is warranted only below a price that pre-absorbs a bad resolution of the unknowns — well under the 2.01× book D1 the market asks — I'd own it near ¥1,650, and only after the contract and succession unknowns narrow.

What a student should take from this: a clean, cheap, net-cash compounder with honest books is not automatically a buy — the knowledge bar comes first, and it is not cleared by a low multiple but by whether you can genuinely predict the business a decade out. When the two variables that decide the outcome (a government's budget and a supplier's renewable contract) are things the company does not control and you must assume rather than verify, the correct answer is a disciplined watch at a price that pre-absorbs a bad resolution — not a pass on a good business, and not a purchase you are pretending to underwrite.

Claude — watch

My figures-blind priors, registered first, held up almost verbatim: I put P(watch) at 0.35 as the single most probable landing, with the tension pre-committed — "good business, right direction of travel, but priced too close to fair value to demand today." The ledger confirmed every load-bearing prior and overturned none of the important ones. (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.)

What the ledger confirmed. Quality is real, not a peak illusion at the ROE line — 17.62 / 11.55 / 15.93 / 14.20 / 17.06% across FY2021–25 [F47–F51], no loss year, on a 57.7% equity ratio F46. Net cash is a genuine cushion, the opposite of a levered residual — +¥13,646M D14, ~24% of the ¥56.6bn net market cap D4. Asset-light is confirmed to the yen — 投資CF −¥205M F66, depreciation ¥234M F233, tangible fixed assets only ¥3,269M F162. The moat is borrowed, my H1 gate — ~67.0% of merchandise purchases Ebara-sourced [F304/D23], the basic agency contract auto-extending only one year past 2026-09-30 E57, ~66.3% of sales to 官公庁 [F294/D24]. And the record year is Engineering-peak-flattered — consolidated OP +44% E45 almost entirely on Engineering segment profit +76.8% (¥2,450M→¥4,332M) [F260/F261/D25] on "a high opening order backlog" E47, with management's own EJ2027 guide putting FY2027 OP (¥5,500M F412) below FY2026 (¥6,300M F411) — an explicit admission the peak recedes. The read on the moat is more precise than "borrowed": it is mostly rented (the selling right and the demand base both), but carries some owned elements — direct 官公庁 incumbency, the Ginsuikai network E31, design IP E8, a 75-year relationship E3 — owned enough to earn 15%+ ROE, not owned enough to underwrite the peak as permanent.

The deciding work is the owner arithmetic. I value the whole company as a private buyer, downside first. Bear case (C33), each assumption from observed history: the FY2022 trough printed net income ¥2,169M F12; un-do the Engineering peak and let margins revert to a normalized bear net income ≈¥2,400M, capitalize at 9% no-growth (¥26.7bn) plus full net cash ¥13.6bn D14 = ¥40.3bn ÷ 24.06M net-float shares ≈ ¥1,675/share — a −29% draw, cushioned, not eliminated, by real cash. Private-owner yield (C35), multiple-free — the hinge: normalized owner earnings ≈¥3,400M against market cap net of ~¥11bn deployable cash = 7.4%, below a ~9% hurdle for a borrowed-moat, budget-dependent small-cap. Only on FY2025 peak earnings (¥4,384M) does the yield reach 9.6% and clear. That is the whole verdict in one line: the owner yield clears only if you credit the peak you were just told recedes. A reverse-DCF confirms it — at ¥2,354 the price embeds near-peak-no-growth or mid-cycle-plus-2%-growth, fair not cheap, and harder to beat than a bargain would need.

Verdict: watch. It is not a pass: the downside is real (−29%) but survivable and cash-cushioned, the capital-return trajectory is executing (payout 35%→40% [F322/F323], a ¥1bn buyback running E141), and an activist sits on the register F352. It is not a buy here: the private-owner yield does not clear on normalized earnings, the moat is borrowed at a contract the firm does not control, and I would be paying a full price for a peak year management itself guides down. And it is not too-hard: the price question decides the verdict at ¥2,354 before the genuine unknowables (agency renewal, founder succession) become load-bearing — those are reasons for the discount I demand, which is precisely a watch. Implied buy-below ¥1,600 (bear-case owner yield clears at the hurdle; mid-cycle yields 12.4%), ~32% below the stamp, derived blind to the price.

What a student should take from this: a high ROE, a fortress balance sheet, and a rising dividend do not by themselves make a buy — the question is always at what price, and a good business bought at a fair price is a fine business and a mediocre investment. The discipline that produced the verdict was refusing to capitalize a peak: normalize earnings to mid-cycle, strip the commemorative dividend, and require the private-owner yield to clear the hurdle on that number — when it only clears on the year management is guiding you away from, the honest answer is "own it cheaper," not "own it."

Synthesis

Where the lenses agree

For the first time in the record, all five lenses reach the same verdict without a single decline — and it is the first name to clear the quality bar that every deep-value predecessor failed. The agreement is total on three facts, and they are the whole study.

First, the quality is genuine and durable. ROE ran 17.6 / 11.6 / 15.9 / 14.2 / 17.1% across five years with no loss year [F47–F51], recurring profit compounded ¥4,110M → ¥6,316M [F6/F10], and it is earned on a fortress balance sheet — net cash +¥13,646M D14, equity ratio 57.7% F46, asset-light to −¥205M of investing outflow F66. Every lens credits this. Buffett: "a very good business hiding behind a fortress balance sheet." Munger: "the figure table is genuinely good." Claude: "my priors under-, not over-, weighted the quality." This is not a levered residual or a melting-ice cube; it is a real, above-average business.

Second, the price is fair, not cheap — there is no margin of safety. At 2.01× book D1 and 12.78× earnings D2, the panel is unanimous that the discount every deep-value name offered is simply absent. Buffett: "not a cigar butt … a fair price, not a cheap one." Pabrai: "an asset floor is not the same as a fifty-cent dollar — the downside is cushioned, not absent." Li Lu: "priced for a good, well-run compounder — fairly, not fearfully." Claude: "fair, not cheap, and harder to beat than a bargain would need." The net cash is real but only ~22–24% of the price D14; NCAV is ~4× covered [F161−F166]; the operating business is not being handed to you at a discount.

Third, the record year is a peak, and management says so in writing. FY2025's +44% operating-profit jump E45 is almost entirely Engineering segment profit +76.8% [D25/F261] on a high opening backlog converting on schedule E47 — timing, not run-rate. Maker fell 5%, Trading rose 8% [F259/F263]. And management's own EJ2027 plan guides operating profit ¥6,300M → ¥5,500M, FY2026→FY2027 [F411/F412], margin to 12.2% E21. Every lens normalizes the peak down; every lens strips the ¥20 commemorative dividend F324 to a recurring ~2.1% yield. This is the panel's operating-line discipline catching a flattered headline — the same discipline that caught one-off-flattered "records" earlier in the batch.

Where the lenses diverge

The verdict is unanimous, but the buy-below spread is wide — ¥1,250 to ¥1,850, a ~48% range — and the spread is not noise. It is three different definitions of the discount required, each driven by a specific item.

Pabrai anchors to the net-cash floor (¥1,250, the lowest). His P1 asks what he loses if wrong, and his answer is a hard asset floor — net cash plus haircut securities, ~¥900–1,100/share — that makes this not a pass. But P53 (the fifty-cent dollar) then demands he buy near that floor, "where net cash and securities cover most of the price and I'm buying the water franchise nearly for free." His discipline is the strictest on price because his floor is the most concrete: he will pay up only when the balance sheet does almost all the work. That drives the deepest buy-below on the panel.

Buffett prices the franchise, not the floor (¥1,850, the highest of the numbered belows). His crux is B23/B25 — is there a nameable barrier the company owns? — and his answer is no: the moat is the rented Ebara agency E57 plus budget-bound demand E33. So he does not demand a net-cash floor; he demands an owner-earnings yield that reaches double digits and ~1.6× book, "which also buys me a margin of safety against the day the Ebara contract or the government budget disappoints." He is paying for a good business, so he pays more than Pabrai — but he insists on the discount that a rented moat, not a floor, requires.

Li Lu prices the knowledge bar (¥1,650, in between). His L1 gate is neither the floor nor the franchise price but predictability: he can map the two decisive variables (the government budget, the Ebara contract) to disclosures, but cannot answer them "with the confidence my standard demands." His buy-below is the price that "pre-absorbs a bad resolution" of unknowns he cannot verify — a knowledge-discount, landing between Pabrai's asset-floor discipline and Buffett's franchise price.

Claude (implied ¥1,600) and Munger (~¥1,800–2,000) bracket the middle by the same normalization route — the price at which the normalized (not peak) owner yield clears a 9% hurdle. Claude's C35 is the sharpest statement of the shared crux: "the owner yield clears only if you credit the peak you were just told recedes." Munger reaches the same number from EV/normalized-OP (9.5×). The gap across the whole panel, then, is not whether to wait but how deep the discount must be — and that depends entirely on whether you protect yourself with the asset floor (Pabrai), the franchise price (Buffett), the knowledge margin (Li Lu), or the normalized-yield hurdle (Claude/Munger). All five agree ¥2,354 — 12.78× peak earnings D2 — is a watch; all five would own it ~30% lower.

The red team, engaged

Because five lenses converged on watch (a unanimous non-decline), a fresh adversary (ledger only) argued the stronger PASS case — that ¥2,354 (2.01× book D1) is not a "wait for cheaper" but a "drop it from the list," because the quality on display is borrowed, peaked, and flattered, so no future price reliably converts this into a sound minority holding. The synthesis must meet its strongest points by name. Each sharpens the crux; none breaks watch down to pass, and here is why.

• "The moat is not owned — it is rented, and the lease has a date on it." The bear's central point, and it is true as stated. EJ is a fabless E38 distribution agent buying 67% of merchandise from Ebara [D23/F304] under a contract that auto-extends only one year past 2026-09-30 E57, with non-renewal a named risk E36. But this is precisely the crux the whole panel already priced — it is why the buy-belows sit 30–48% below the stamp. The rebuttal to pass is that a rented moat is a reason to demand a discount, not a reason to declare the business un-ownable at any price: seventy-five years of unbroken renewal E3, a two-sided anchor (EJ holds ¥4,036M of Ebara stock to protect the tie E102), and the 15–17% ROE it has actually produced [F47–F51] mean the relationship is durable enough to own — at a price. The bear proves the moat is rented; it does not prove the rent will be called.

• "FY2025 is a peak — management says so in writing." Fully conceded, and the panel conceded it first. EJ2027 guides OP ¥6,300M → ¥5,500M [F411/F412], and FY2025's ¥6,121M already beats the FY2027 plan. This is exactly why every lens normalizes down and none pays 12.78× the peak. But a peak-year print is a reason to value on the cycle, not to conclude the through-cycle business is worthless — and the through-cycle record here is genuinely good (no loss year, ROE trough at 11.6% F48), unlike the loss-ridden cyclicals the batch declined. The guide-down caps the price you pay; it does not zero the franchise.

• "The record is un-cashed backlog — operating cash flow fell to ¥1,413M as contract assets ballooned +¥3,157M." Conceded on the facts [F61/F235] — and it is the reason the panel trusts the record less than the headline. But the driver is benign and disclosed: percentage-of-completion timing on Engineering backlog converting E113, the sole KAM the auditor examined and passed unqualified [E134/E137], not accrual manipulation. Cumulative five-year cash conversion is ~80%, below 90% but not alarming. It argues for normalizing earnings down — which the panel did — not for a pass.

• "An 86-year-old founder is Chairman and CEO — an undated, unpriced key-man discount." Conceded, and named by every lens E88. Munger and Li Lu both flag it as the concrete near-term risk. But a capable President/COO is in place E89, the Nomination Committee monitors succession E86, and — decisively for pass vs watch — it is a reason for the discount the panel already demands, not a fact that makes the business un-ownable. Succession risk is priced by waiting, not by dropping.

• "There is no reachable discount even at the watch price — the cash is inert, the activist has not forced a re-rating." This is the bear's real challenge to watch itself, and it is the honest tension. The rebuttal: the net cash is a genuine downside cushion whether or not it is activated (~24% of cap D14), the capital-return trajectory is executing (payout 35%→40% E73, a ¥1bn buyback with ¥62M already bought by the filing date E141), and the register is contestable, not management-locked F352. The bear is right that a lower price does not fix a rented, guided-down franchise — but the panel never claimed it would; the buy-belows assume the bear case (normalized earnings, no re-rating) and simply demand to buy the good-through-cycle business at a price where the normalized yield clears. The bear's own falsifier — sustained OP ≥¥6bn at ≥13% margin beyond the backlog year, beating the guide-down — is the same observable the panel is waiting on.

The red team did not move any verdict off watch, but it earned its keep: it is why the synthesis states the crux as sharply as it does — is the moat owned or borrowed, and is FY2025 a peak? The honest resolution is that both facts are true (the moat is rented; the year is a peak), and both are reasons for the ~30% discount the panel demands rather than reasons to declare the business uninvestable. The net cash (+¥13,646M D14) and the genuine, no-loss-year, durable-ROE history (ROE floor 11.55% F48) keep it a watch, not a pass: a good business, correctly priced as fair, worth owning ~30% cheaper — not a value trap to discard.

Self-distance note. The Claude lens holds one of the five verdicts compared above (watch) and wrote this synthesis; it also built the dual-blind reconciled figure table and 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.

Prediction-vs-actual: VOID. This was an autonomous headless cycle; the human blind prediction is voided (void: no-human-prediction, never forged). No prediction-vs-actual scoring applies.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp. Three lenses issued explicit buy-belows — Buffett ¥1,850, Pabrai ¥1,250, Li Lu ¥1,650; Claude publishes an implied buy-below of ¥1,600; Munger's is a stated conditional range (~¥1,800–2,000), not a hard number.
• pass / watch / too-hard are recorded but unscored in any future review. All five verdicts here are watch; the buy-belows sit below the ¥2,354 stamp and are the price 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. (The ¥2,354 stamp already reflects the 2026-01-01 2-for-1 split E140; all buy-belows are stated on the post-split basis.)

Red team

A consensus red-team (five of five lenses converged on watch, a unanimous non-decline) was dispatched to argue PASS — that ¥2,354 / 2.01× book is not a "wait for cheaper" but a name to drop, because the FY2025 quality is borrowed (a terminable Ebara agency), peaked (by management's own guide-down), and flattered (a commemorative dividend + a backlog-timing profit spike), so no future price reliably makes it a sound minority holding. Its strongest points, verbatim-faithful, and the synthesis's engagement with each, are in "The red team, engaged" above. In brief, the adversary's five ranked points were: (1) a borrowed moat with a dated expiry — 67% of purchases from Ebara [D23/F304], contract auto-renewing only one year past 2026-09-30 E57; (2) management's own EJ2027 plan says FY2025 peaked — OP guided ¥6,300M → ¥5,500M [F411/F412]; (3) the record is un-cashed backlog — operating cash flow fell to ¥1,413M as contract assets rose +¥3,157M [F61/F235], the sole KAM E134; (4) an 86-year-old founder is both Chairman and CEO E88; (5) the dividend record is one-off-flattered by a ¥20 commemorative F324. The synthesis concedes every factual point — the moat is rented, the year is a peak, the cash conversion was lumpy, the founder is 86, the dividend was flattered — and meets the load-bearing claim (that these make the business un-ownable at any price) by the reachability-of-the-discount test: each fact is a reason for the ~30% discount the panel already demands, not a reason to declare a no-loss-year, net-cash, durable-ROE business uninvestable. The bear's own falsifier (sustained OP ≥¥6bn at ≥13% margin beyond the backlog year, beating the guide-down [F411/F412/E21]) is the same observable the panel is waiting on. A consensus that faced its strongest opponent and emerged intact — sharpened on the crux (owned vs borrowed moat; peak vs plateau), unmoved on the verdict.

What would change our minds

Pre-registered falsifiers, per lens issuing a buy-below or holding the watch. The verdicts converge on a single observable: does the recurring earning power hold on a NORMAL (non-peak) year at a full tax charge, and does the Ebara agency renew / does founder succession resolve — versus the price simply falling into the ¥1,250–1,850 range.

• Buffett (watch, buy-below ¥1,850). Toward buy: a price near ¥1,850 (owner-earnings yield ~10%, ~1.6× book) clears the margin-of-safety bar [B93]. Toward too-hard: if the basic Ebara agency contract E57 is not renewed, materially re-priced, or narrowed, OR the ¥13.6bn net cash D14 is deployed into a full-priced acquisition rather than returned or reinvested at the historical ~15% ROE — the quality case that justifies paying ~2× book collapses.
• Munger (watch, conditional ~¥1,800–2,000). The falsifier is durability: sustainable recurring profit at or above ¥5.0bn for two consecutive years WITHOUT the ¥20 commemorative DPS F324 and WITHOUT an exceptional Engineering backlog conversion (Engineering segment profit staying above ¥3.0bn even in a lean-backlog year), confirming the moat is the relationship and the operator rather than a one-cycle backlog flush E47.
• Pabrai (watch, buy-below ¥1,250). Broken by cyclicality proving the record a peak: two consecutive years of falling public-sector orders F288 OR Engineering segment profit reverting below ~¥2,450M F260. Second falsifier: the Ebara basic agency contract E57 not renewed or materially narrowed — the fabless/agent model loses its supply spine.
• Li Lu (watch, buy-below ¥1,650). The Ebara contract E57 not renewed on comparable terms, OR the ~66–67% public-sector mix [D24/D31] contracting as a bloc — either severs the two variables that decide the ten-year earnings power. Concretely: a renewed contract on materially worse economics, or consolidated 官公庁 orders F288 falling for two consecutive years, flips this from watch to too-hard.
• Claude (watch, implied buy-below ¥1,600). A price at/below ~¥1,600 upgrades toward buy [C44]. Conversely, a materially unfavourable renewal of the Ebara agency at its ~2026-09/2027 expiry E57 moves it toward pass/too-hard; and Engineering segment profit sustaining near the FY2025 ¥4,332M peak F261 through FY2027, contradicting management's own guide [F412<F411], would break the "own it cheaper" framing toward buy.

The single observable most lenses converge on is whether the recurring (ordinary) line compounds on a NORMAL, full-tax year — the difference between a genuine plateau and a one-off-flattered peak [F411/F412] — together with whether the Ebara agency renews on comparable terms E57 and founder succession resolves E88 — versus the price falling to the ¥1,250–1,850 range where each lens's discount is satisfied.

What this taught the checklists

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

• Buffett — add a "rented-moat / third-party-dependence" item to the Business-economics section. B23–B31 test whether a moat exists and endures, but none directly asks: is the durable advantage owned by the company or licensed from a counterparty (supplier agency, franchise grant, sole-distributor contract) on renewable terms? Here the entire above-market return leans on the Ebara agency E57, yet the existing items let it score "moat present" (returns are real D21) without forcing the question of whose moat it is. Also: B26/B60 should prompt an explicit ex-cash operating-ROIC re-strike when the balance sheet is cash-heavy — reported ROE understates the operating business once ¥13.6bn idle cash D14 is removed.
• Munger — M18a (agent-dependency flag): add a sub-item to M18 distinguishing "owned franchise" from "privileged agency" — where the moat depends on a commercial contract with a more powerful principal, require stating the contract terms, renewal cadence, and what the principal loses by terminating, because the ROE fingerprint is identical between a true franchise and a contractually dependent agent. M82a (named-succession-plan requirement): tighten M82 to require not just evidence of succession planning but a disclosed, dated transition timeline where the CEO is over age 75 — this company shows the machinery can exist (active committee E86, identified successor E89) without a disclosed timeline.
• Pabrai — add a "quality-at-fair-price is a watch, not a buy" branch to Verdict guidance: a hard asset floor (P1 pass) combined with a full multiple on peak earnings (P18/P56 fail) is a distinct shape — the floor makes it not-a-pass, the peak-multiple makes it not-a-buy; a one-line note that "a fortress balance sheet answers P1 but never substitutes for P53" would sharpen the boundary. And P56 should ask for a segment-level peak flag — here the record was one segment (+76.8% Engineering D25) masking flat Maker/Trading; normalizing the driving segment against its own history catches the peak faster than the consolidated line.
• Li Lu — L1/L34 interaction, a "renewable-contract moat" sub-test: where the barrier is contractual E57 rather than a statutory license (L34) or a switching-cost lock-in (L43), the item must locate the contract's term and renewal history and treat an imminent renewal as an unknown that lowers the L1 knowledge grade — not as a moat. And an L48 lumpy-working-capital caveat: for percentage-of-completion EPC businesses, single-year CFO/NI can diverge sharply on contract-asset swings F235 without signaling accrual games — compute the multi-year cumulative conversion and flag contract-asset growth vs revenue growth specifically.
• Claude — two checklist sharpenings, plus a library seed. (1) A new item, "borrowed-vs-owned moat, sized": force a discrete computation of supplier/franchise dependence × contract term to expiry (67%-of-purchases F304 × a dated agency expiry E57), rather than distributing the inference across three items. (2) Sharpen C40 (cycle-position) to require management's own forward guide as a cross-check — the FY2027<FY2026 OP guide [F412<F411] was the single cleanest confirmation FY2025 is a peak, and it lives in the plan table, not the cycle history. Library (Class-level) — a new corner, the mirror of the deep-value/levered corners: "durable-return, net-cash, asset-light distribution-agent / public-EPC-integrator with a borrowed moat, priced fair-to-full." First member EBARA JITSUGYO (6328), N=1. Class definition: 15%+ through-cycle ROE with no loss year, net cash ~20%+ of market cap, fabless/subcontracted, where the selling right is a third-party agency contract and the demand base is a government budget (both rented), with an activist on the register and a founder key-man, at ~2× book / low-teens earnings. Outcome on the class's key axis (does the private-owner yield clear the hurdle at a full price?): No — at 2.0× book / 12.8× peak earnings [D1/D2] the normalized owner yield was ~7.4% [C35] vs a ~9% hurdle, clearing only on the peak; verdict watch, implied buy-below ~32% below stamp. Where the deep-value/levered corners ask whether a discount closes, this corner asks whether a fair price is worth paying — the mirror image, and it needs its own base rate: for good-but-rented-moat compounders at a fair price, how often does the market later offer the ~30% discount at which the owner yield clears?

---

Corrections