Nippon Sheet Glass (5202): A Merger-Arb Spread Where a Business Used to Be

Let me start where I always start: with the business itself. What does Nippon Sheet Glass sell, who pays, and why do they keep paying?

The business is flat glass — windows for buildings, glass for cars, specialist thin-sheet glass for displays and fiber. The Pilkington acquisition in 2006 turned this Japanese company into a genuine global manufacturer: 159 subsidiaries, operations across Europe, the Americas, and Asia, roughly 52% of revenue from automotive glass and 43% from architectural E1E2E3. Float-glass manufacturing is about as capital-intensive as it gets — a float line costs hundreds of millions of dollars, runs day and night, and cannot simply be switched on or off without enormous cost. So yes, I can explain the business in a paragraph. Glass is needed, customers keep buying it, and the technology has been mature for decades.

But there's a complication, and it is not small.

The capital structure is the business. Gross interest-bearing debt sits at ¥547 billion F56, net debt at ¥484 billion F57, against equity attributable to the owners of only ¥151 billion F20. Net debt runs 16.8× last year's operating profit before exceptionals D6. Finance costs consumed ¥33 billion F72 against operating profit before exceptionals of only ¥29 billion F67. Over the five years FY2022 through FY2026, the company produced cumulative net losses to its common shareholders of roughly ¥28 billion E22. The Architectural segment earned a respectable 8% margin D9; the Automotive segment earned barely 1% D10. Free cash flow for the most recent year: ¥1.1 billion F99E42. The interest clock alone chews through ¥28–33 billion per year F100.

The company itself acknowledged what this means: without the Apollo transaction, refinancing the ¥100+ billion of borrowings maturing in March 2026 would have been "not easy under the current capital structure" E24. That is polite language for "we needed rescuing," and Apollo answered.

So the question before the Buffett checklist is not really about the float-glass business. It is about what you are actually buying when you acquire shares at ¥482 today.

What you own at ¥482. Apollo has already completed a ¥165 billion third-party allotment at ¥431 per share F155E29. The June 2026 AGM approved a squeeze-out via a 122,222,222-to-1 share consolidation, paying ¥500 per share in cash to all remaining public shareholders E27. A ¥140 billion pseudo-DES — converting bank debt to equity — follows immediately E26. The deal is targeted to close in H2 FY2027 E28. The stamp sits at ¥482 D12, implying a spread to the ¥500 take-out of roughly 3.7%. Market cap at stamp is ¥69 billion D1.

That is the whole picture. You are not buying a glass manufacturer you intend to own for ten years. You are buying a merger-arb position: ¥500 in cash at some point in the next year or so, against a current price of ¥482.

The Buffett checklist cannot make this work. The ten-year ownership test (B19) fails at first contact — you cannot hold it for ten years because it is being taken private at ¥500 in H2 FY2027. The earnings-quality tests (B42, B43) exist to establish whether a business generates reliable owner earnings you can hold indefinitely. Five years of negative-more-than-positive net income F11F12F13F14F15, ¥33 billion per year in financing costs swamping ¥29 billion in operating profit, and ¥1.1 billion of free cash flow are not the material for a going-concern owner-earnings case. The business is not a Graham net-net either: NCAV is deeply negative when you stack the liabilities properly. The B9 test — return from what the asset itself produces — fails because the "asset" here is a merger agreement, not operating earnings.

The only plausible frame is B11 / B96: this is a workout, and workouts have a specific arithmetic. The gross spread is ¥18 (¥500 − ¥482). If the deal closes in, say, twelve months, the annualized gross return is roughly 3.7%. Deduct transaction costs, Japanese withholding on the cash distribution, and any timing risk and you are looking at something in the low-single-digit range, before anything goes wrong. Buffett's self-imposed rule on workouts was a 25%-of-partnership cap and an explicit warning that leverage against a general is "a very dangerous practice" — but even the workout arithmetic here is unimpressive.

What could upset the applecart? The AGM already approved the squeeze-out E27, which removes the shareholder-vote risk. But closing remains contingent on regulatory/court processes for the share consolidation, and the pseudo-DES execution. The parent facility balance due 2027-03-31 E39 is ¥159 billion — the Facility 2 matures on exactly the same date targeted for the take-out to close. If something delays the consolidation timeline past March 2027, the refinancing pressure reappears. Covenant breaches — net-assets floor E32, two-consecutive-years operating-loss test E33, rating covenant E34, cross-default E37 — could trigger a cascade. These are low-probability events given Apollo's stake and the bank pseudo-DES commitment, but they are non-zero.

On the circle of competence. The flat-glass manufacturing business itself is inside the circle — understandable, basic, capital-intensive in a way that creates barriers through scale. The economics, however, are not ones I want to own in perpetuity: commodity flat glass for construction and auto at thin margins, with structural leverage that eats the economics. The Pilkington brand is largely a B2B sourcing credential, not a consumer brand commanding durable price premiums E52. Technical Glass at 5% of revenue earns 19% margins D11, but it is too small a tail to wag this dog. What I cannot honestly put inside my circle is the workout-execution timeline and the counterparty mechanics of the pseudo-DES. That is specialized distressed/restructuring territory.

The verdict is too-hard, and not primarily because the business is opaque — I understand what flat glass is. Too-hard here means that the investment case rests entirely on a ~3.7% spread resolving in my favor before something unexpected disrupts the deal or consumes additional time value, and that arbitrage yield is thin compensation for the complexity of a multi-party restructuring with a ¥547 billion debt load, covenant tripwires on two separate facility stacks E38E39E35E36, and a closing date whose slippage would be costly. The classic Buffett workout demands a predictable timetable, a modest spread that still clears your hurdle after friction, and the ability to name the applecart-upsetters clearly. Here the spread is thin, the applecart list is long, and the reward for being right is 3.7% gross. The reward for being wrong — if something delays or disrupts the consolidation — is sitting in a hyper-leveraged going-private vehicle with no dividend F95F96 and shrinking analyst coverage.

There is no buy-below price because there is no durable business to value and own. This situation belongs to specialist merger-arb funds with the deal flow, legal/regulatory expertise, and portfolio construction to handle it at scale. It does not belong in a Buffett-style portfolio where the frame is "would you buy the whole company and hold it for ten years?"

Start with the inversion: how does an owner of 5202 Nippon Sheet Glass lose money permanently? The obituary almost writes itself. Gross interest-bearing debt ¥547bn F56 against equity to owners of ¥151bn F20 — a leverage ratio of 3.6× D5 — with net debt at 16.8× operating profit before exceptionals D6. Over five years the business produced cumulative net losses of approximately ¥28bn E22. Free cash flow in the most recently completed year was ¥1.1bn F99 — against a debt pile approaching half a trillion yen. The interest bill alone consumed ¥28.5bn in cash payments F100, more than the company's entire reported operating profit before exceptionals F67. That arithmetic is not a going-concern uncertainty (the board cleared that gate by the skin of Apollo's cheque E30); it is the picture of a business whose economics, left undisturbed, funnel the entire operating surplus to creditors.

The moat question cannot be answered favorably. What mechanism protects returns in flat glass? The Pilkington name (¥36.8bn on the balance sheet as an indefinite-life intangible F83) is a brand name in a commodity input market — it is not See's Candies. Architects do not pay a premium for glass from nostalgia. Automotive glass is a specification product sold to OEMs with enormous purchasing power; pricing is won on cost, quality, and delivery, not on brand association. The segment data confirms the mechanism's absence: Automotive Glass operating margin 1.09% D10, Architectural 8.01% D9 — the latter recovered by float-line shutdowns that restricted supply E17, which is not a moat but a supply-management operation. There is no pricing power fingerprint, no habit moat, no switching-cost mechanism, no network effect. Technology in commodity glass is the textile-loom lesson: every capital-intensive furnace upgrade available to this company is available to every competitor. The "Other" segment runs a structural operating loss of ¥14.85bn F107 that includes Pilkington acquisition intangible amortisation E5 — so even the segment profit figures overstate what the operating engine produces after accounting for the price paid for the Pilkington empire in 2006.

The Pilkington acquisition is itself the central lesson. A £2.2bn deal done at the top of the cycle in 2006 loaded the balance sheet with debt that twenty years of effort has not resolved E22. In FY2023/3 the entire ¥48.8bn of European Automotive goodwill from that acquisition was fully impaired E59. The North American Automotive CGU was freshly impaired again in Q4 FY2026/3 — ¥3.4bn E47E50 — with zero headroom remaining and a further ¥6.5bn of impairment triggered by a mere 1% rise in the discount rate E50. This is not a one-time charge; it is a pattern that has run for nearly two decades. Retained-earnings productivity (M42) is negative: the company produced cumulative losses in four of five recent years [F11–F15].

Now the situation. Apollo has bought in at ¥431/share E29 and structured a ¥140bn quasi-DES from four main banks E26. The squeeze-out at ¥500/share E27F157 has been approved at the June 2026 AGM and is targeted for H2 FY2027 E28. At the stamp of ¥482 D1, the position is 3.7% below the take-out price D12. This is merger arbitrage, not value investing. The question for a buyer at ¥482 is not "what is this business worth" — it is "does the squeeze-out complete, and how soon." That is a legal/process question, not a business-quality question, and it is outside this lens's circle of competence (M89). The pari-mutuel check (M44) is unambiguous: the odds on offer are a ~3.7% spread over an uncertain period with binary execution risk on one side and a worthless going-concern multiple on the other. That is not the bet Munger meant when he said bet heavily when the world offers you an opportunity.

Incentives are the one corner of this situation that grades acceptably. The compensation structure has a functioning gate: both the annual bonus (group OP target ¥37,710M and FCF target ¥5,000M unmet E68F146F147) and the LTIP (cumulative EPS and FCF both below entry E69F148F149) produced near-zero payouts. That is painful for management but it is honest — incentives tracked outcomes, not stories. The compensation committee is chaired by an independent outside director E67; outside directors receive base salary only E70. No single person earned above ¥100M E71. The board is 75% independent E65. No takeover defenses E72; zero listed cross-holdings E73. This governance stack is cleaner than most Japanese industrial companies. It does not redeem the economics, but it matters for the character assessment (M54): the record shows no self-dealing, no related-party transactions at the holding level, no comp-ratchet evidence — just a management team that failed to fix a balance sheet it inherited and is now being taken out.

The psychology scan (M61) finds the main misjudgment risk on the buyer's side, not management's: the 3.7% spread looks like free money against a contractually obligated squeeze-out. The lollapalooza (M87) against the merger-arb buyer is real — reciprocation (it feels like a done deal), social proof (Apollo would not proceed unless it was certain), and authority (AGM approved it) all push toward acting. The antidote is Buffett's: don't go into the auction room. This profile is not the room for arb.

The opportunity-cost hurdle (M35) closes the case. A buyer at ¥482 is not acquiring a compounder at a fair price; they are acquiring a 3.7% spread to a contractually committed exit. Even at 3.7% annualised over three months that is roughly 15% annualised — which sounds attractive until the risks are counted: delay in the share consolidation approval mechanics, FX risk on yen assets, remote but non-zero deal disruption, and the zero residual value if for any reason the squeeze-out fails. More importantly, this profile cannot value the ongoing business as an investment because the business's owner-earnings are negative (net losses four of five years) and the free cash flow after real capex is ¥1.1bn F99 on a ¥68.6bn market cap D1 — a cash yield under 2% before interest costs swallow everything. The textile-mill lesson is the right frame: Berkshire ran the textile mills for years and Munger's verdict was that they should have been shut, not resurrected. NSG's self-help has run for twenty years without resolving the problem E23. Apollo is shutting the mill — from the outside.

Verdict: too-hard. The business has no durable moat mechanism (M18 fail), carries a permanent-loss map with multiple live vectors (M2 fail), and at the stamp price is a merger-arb bet on process completion rather than a going-concern value proposition (M35, M37, M43 all fail). The governance is honest but can be scored only if the business clears the quality hurdle first. It does not.

Let me be direct: the moment I read "net debt ¥484bn against equity to owners ¥151bn" and "net-debt/operating-profit 16.8×," I know where this ends F57F20D6. The Dhandho framework starts and ends at the downside. Here, the downside is catastrophic, and no amount of upside story rescues a bad downside.

The Business — One Paragraph

Nippon Sheet Glass (NSG) manufactures flat glass through two dominant segments: Architectural Glass (43% of revenue, building facades and solar panels) E2 and Automotive Glass (52%, OEM and aftermarket) E3. It operates globally under the Pilkington brand, acquired in 2006, through 159 consolidated subsidiaries E1. Glass is a mature, capital-intensive, cyclical commodity business where float furnaces cost hundreds of millions to build and cannot be turned off cheaply. Revenue was ¥879bn in FY2026/3 F5 against operating profit before exceptionals of ¥29bn F67 — a 3.3% operating margin D3. The group has not paid a dividend in at least the last two years F95F96. Does it fit in a paragraph? Barely — it is a capital-intensive commodity with a Pilkington overlay that is harder to explain than motels or funerals. But that is not the core problem. The core problem is the debt.

The Downside (P1) — This Is Where The Analysis Ends

Gross interest-bearing debt: ¥547bn F56. Net debt: ¥484bn F57. Total liabilities: ¥932bn against total assets of ¥1,117bn F62F25. Equity attributable to owners: ¥151bn F20. Interest paid in FY2026/3: ¥29bn F100 against operating profit before exceptionals of ¥29bn F67. The interest bill consumes 100% of operating profit.

What is the realistic worst case per share? This is not a situation where you can find a "hard-asset floor." PP&E is ¥485bn F80, but PP&E in float-glass manufacturing — furnaces, float lines, fabrication equipment — is not Frontline's ships. You cannot sell an individual float furnace to a secondary buyer for book value the way you can place a tanker. Goodwill is ¥87bn F81 and we already know how those marks have held up: ¥49bn of Pilkington-era goodwill in European Automotive was fully impaired in FY2023/3 E59; ¥3.4bn more went in Q4 FY2026/3 in North American Automotive, with zero post-impairment headroom — a +1% rate move triggers ¥6.5bn further impairment E50. The Pilkington brand intangible at ¥37bn F83 is indefinite-life and carried at historical cost, which proves nothing about realizable value for an acquirer who already owns it.

In a stress scenario — two bad European years, energy spike, tariff escalation — the floor is not near. It is absent. The interest burden already consumes 100% of operating profit; one bad year of negative operating profit triggers a covenant breach (must not post operating losses after separately disclosed items for two consecutive fiscal years E33). Cross-default is standard across all group agreements E37. Facility 2, ¥159bn at maturity 2027-03-31 E39, is already the size of total equity. When the company itself disclosed that absent the Apollo deal, refinancing over-¥100bn of borrowings maturing at end of March 2026 "would not have been easy" E24, it confirmed the floor is paper thin.

P1 fails. Hard stop.

The Merger-Arb Question

"But wait — ¥500/share squeeze-out is approved E27E28, the AGM has tabled it, Apollo is in at ¥165bn F155, the ¥140bn pseudo-DES is arranged E26. The stamp at ¥482 is 3.7% below ¥500 D12. Is this 3.7% spread a Dhandho bet — heads I win, tails I don't lose much?"

No. And here is why.

First, is this "ownership"? Pabrai buys businesses. Merger-arb is buying a contractual claim against a deal closing. The return profile is: deal closes at ¥500 (earn 3.7%) or deal breaks (stock falls to what? — going-concern fundamental value on a company with 16.8× net-debt/operating-profit, suspended dividends, and a share-price that traded meaningfully below ¥482 within the last year). The "tails" scenario is not "I don't lose much." The tails scenario is owning a hyper-leveraged going-concern with no credible path to returns, which the company itself admitted would have struggled to refinance without Apollo E24. The "tails" of this specific arb is potentially severe.

Second, the 3.7% spread over an uncertain timeline (H2 FY2027, meaning completion could be March 2027 or later E28) annualizes to perhaps 5-6% if it closes in 9 months — not a Dhandho-sized asymmetry. Pabrai demands wide asymmetry and capped downside. The asymmetry here is narrow (cap: 3.7%) and the downside is essentially unbounded if the deal breaks.

Third, few bets, big bets (P66). A 3.7% spread play is not a bet that deserves 10% of a portfolio. It is a tracker position justified by cleverness, not conviction. Pabrai would pass on it as too small to matter if it works and too dangerous if it doesn't.

Other Load-Bearing Failures

Even ignoring P1, the checklist stacks up against this:

• P17 (paid to wait): Dividends are suspended F95F96 and explicitly described as having no foreseeable resumption date E23. No buybacks. No deleveraging mechanism that reaches common holders — Apollo's proceeds go to debt repayment, not to public shareholders. You are not paid to wait; you are waiting at your own expense while interest compounds and equity value bleeds.

• P8 (thesis survives without catalyst): This idea is 100% catalyst-dependent. The entire thesis is "deal closes at ¥500." Remove the catalyst and you own a going-concern with no dividend, suspended buybacks, 16.8× net-debt/EBITOP, and history of goodwill impairments. There is no independent value creation.

• P53 (fifty-cent dollar): The squeeze-out at ¥500 defines the intrinsic value for all practical purposes. The stamp is ¥482, which is 96.4% of the stated take-out price. This is not a fifty-cent dollar; it is a 96-cent dollar with execution risk.

• P57 (moat in decade of returns): ROE has been 4%, −28%, 10%, −12%, 3.4% over five years F42F43F44F45F46. Cumulative net income to owners over five years: negative ¥29.6bn (using the figures table directly, though cross-checking the components: +4,134 −33,761 +10,633 −13,831 +4,421 = −28,404 F11F12F13F14F15). No moat is visible in the numbers.

• P7 (melting ice cube): Free cash flow in FY2026/3 was ¥1.1bn F99 against capex of ¥42bn F89. The business is not generating cash organically at a rate that meaningfully deleverages the debt stack.

The Pre-Mortem

It is two years from now, and this is down 50% permanently. What happened? The squeeze-out process encountered regulatory delays, a Japanese court challenge, or Apollo's deal terms shifted. The going-concern fundamental — net debt at 16.8× operating profit, suspended dividends, covenants at knife-edge levels — reasserted itself as markets repriced the deal-break probability. There was no floor.

Verdict

Pass. Not too-hard — the business is understandable, the deal structure is clear. This is a clean pass on P1 (no downside floor), P8 (100% catalyst-dependent), and P53 (96 cents, not 50 cents). The merger-arb framing does not rescue it — a 3.7% spread with existential downside-on-deal-break is not Dhandho. Pabrai teaches that leverage is the single biggest killer of investments. This company has nearly every variety of leverage risk simultaneously: financial leverage at 16.8× D6, covenant tripwires on two facilities E32E33, cross-default across the group E37, near-zero FCF F99, and deal risk on the one mechanism by which equity gets out. This is not "low-risk, high-uncertainty." This is high-risk, with the upside capped at 3.7%.

Walk away.

When I look at a company, the first question I ask is not whether it is cheap. The first question is whether I can honestly claim to understand this business's economics well enough to predict its state ten years from now. If I cannot answer that question with genuine confidence — not comfort, not hope, but confidence grounded in what the filings actually say — then the company is too-hard, and no price rescues it.

Nippon Sheet Glass (NSG, 5202) ends the inquiry before it properly begins. Not because the business is bad in some abstract sense. Flat glass is an old industry with real assets and a well-known brand — Pilkington — carrying decades of float-glass process heritage E6. Over 90% of revenue comes from two industrial end-markets, architectural and automotive, with the remainder in technical glass E60. The company employs 24,838 people across 159 consolidated subsidiaries in dozens of countries E1F94. There is something real here.

But "something real" and "knowable over a decade" are not the same thing. The business I would be analyzing does not exist as a going-concern investment opportunity. The share I am considering at ¥482 [stamp.price-jpy] represents a merger-arb position in a company being taken private. Apollo Global Management's SPV (Lumina Japan Acquisition K.K.) completed a ¥165bn third-party allotment at ¥431.01 per share on 24 March 2026 E29F155. A ¥140bn quasi-DES with four main banks follows on the delisting effective date E26F156. The June 2026 AGM tabled a 122,222,222-for-1 share consolidation to squeeze out remaining common shareholders at ¥500 per share E27F157, with delisting targeted in H2 FY2027 E28. The current stamp at ¥482 is approximately 3.7% below the contractual take-out D12.

This is the situation with brutal clarity: the Tokyo Stock Exchange will cease to carry this security within a year. The closed-exchange owner test does not apply because the exchange is closing by design, not just as a thought experiment. No ten-year holding is possible. No decade-scale compounding is on offer. What is on offer is a merger-arb spread of roughly ¥18 per share — roughly 3.7% — against execution risk that the squeeze-out completes as planned, the ¥500 consideration is actually paid, and no intervening covenant breach or regulatory snag derails the timeline. That is not investing in Li Lu's sense; it is speculating on a legal process.

Even if I set aside the delisting and tried to evaluate the underlying business as though a long hold were possible, the knowledge bar would still not clear — and for reasons that go beyond this being a hard industry to predict.

The financial structure makes ten-year earnings-power analysis essentially fictional. Net debt is ¥484bn F57 against equity attributable to owners of ¥151bn F20 — a net-debt-to-equity ratio of 3.2x D4. Finance costs consumed ¥32.97bn in FY2026/3 F72 while operating profit before separately disclosed items was ¥28.82bn F67. Interest is eating the operating profit whole; reported profit before tax was a barely-positive ¥378M F10 only because a ¥5.1bn deferred tax credit — including ¥8.8bn of newly recognised UK DTA F116E48 — pushed the period to profit. Free cash flow was ¥1.1bn against the prior year's ¥10.0bn E42F99, and the FY2027 interest-bearing debt target of ¥442bn F158 sits ¥106bn above the FY2026/3 actual of ¥548bn F56E43, with one year and essentially zero operating FCF to close that gap absent the Apollo restructuring.

The five-year cumulative loss attributable to owners is approximately ¥28.4bn E22 — net income of ¥4.1bn F11 in FY2022, net loss of ¥33.8bn F12 in FY2023, recovery to ¥10.6bn F13 in FY2024, another loss of ¥13.8bn F14 in FY2025, and a modest ¥4.4bn F15 in FY2026 powered by the tax credit. This is not a compounding business. This is a business whose intrinsic value has been melting for years under the weight of debt service, recurring goodwill impairments (FY2023's ¥48.8bn European Automotive write-off E59; FY2026's ¥3.4bn North American Automotive impairment with zero remaining headroom E50F115), and what the company itself describes as its most important challenge: fixing the financial base E16.

ROE has been: 4.0% F42 → −27.9% F43 → 9.6% F44 → −11.9% F45 → 3.4% F46 across the last five years. This is the return trajectory a long hold would approximate. At any price I could honestly pay, that does not work.

The Automotive Glass segment — 52% of revenue E3 — earned an operating margin of 1.09% in FY2026/3 D10, generating ¥5.0bn of segment profit F104 on ¥457bn of revenue F103. The Architectural segment (43% of revenue E2) ran at 8.0% D9 after the European pricing recovery from float-line shutdowns E17; but that recovery is structural only to the extent that capacity discipline holds in a cyclical industry. Technical Glass (5% of revenue E4) earned 18.8% D11, which is attractive in isolation, but too small to move group economics. The "Other" segment loses ¥14.9bn F107 annually because Pilkington acquisition intangibles continue to be amortised through it E10, and that drag does not end until the intangibles are fully written off or impaired.

Could I predict this business in ten years? I would need to know: how European construction demand recovers; whether automotive glass margins survive the EV transition (ADAS/HUD integration offers some premium, but also shortens replacement windows); whether flat-glass capacity discipline in Europe and Asia holds through the next cycle; whether the decarbonisation capex (hydrogen offtake, furnace conversions E86) is value-additive or merely survival cost. The filings give me the company's plans but not the answers. These are structural uncertainties in a commodity-adjacent industry, not knowable variables with identifiable public disclosure counterparts that would let me answer L1 with genuine conviction.

The verdict is too-hard for two independent reasons, either of which alone would be sufficient. First, this is a going-private transaction: the security ceases to exist in approximately a year at ¥500, and no decade-scale ownership is possible. Second, even treating the underlying business in isolation, the knowledge bar does not clear: hyper-leverage makes ten-year earnings power unpredictable, the value-compounding record over the observable period is negative, and the industry's key variables — European demand cycles, automotive electrification, energy cost pass-through — are not traceable to specific disclosures with the confidence required.

At ¥482, you are not buying a great business at a fair price, or a fair business at a great price. You are buying a claim on ¥500 that depends on a legal process completing on schedule. That is a different category of activity entirely.

My §1 priors were: deal-completion ~0.90, minority-upside-beyond-¥500 ~0.07, going-concern return-durability ~0.20. The ledger has moved all three, in ways I registered as hypotheses for §2 to check. Here is what happened.

The deal-completion prior rose from 0.90 to 0.93. The most important figure for this dimension is what happened to Facility 2: a ¥159,206M parent borrowing signed 2026-03-27 with maturity 2027-03-31 F165, the same date the pseudo-DES F156 and share consolidation E28 are scheduled. My prior flagged refinancing risk as the largest residual completion risk; the ledger shows the Apollo transaction completed (¥164.9bn allotment, 38,252,710 new shares, 2026-03-24 F155E29), Facility 2 was freshly refinanced at arm's length on 2026-03-27 E39, and the AGM has already approved the squeeze-out (tabled June 2026 E27). The three-month gap between Facility 2 maturity and the delisting target (H2 FY2027 E28) is tight but the pseudo-DES of ¥140bn F156 is designed to retire that exact wall. Residual risk: a covenant breach before the effective date. Parent covenants (E38, E39) include consecutive-operating-loss clauses; FY2027 guidance forecasts ¥36bn OP-before-exceptionals F152 against the FY2026 baseline of ¥28.8bn F67 — the company cannot post an OP loss next year without violating the covenant, but the guidance margin of safety is thin. I move the completion prior to 0.93.

The minority-upside-beyond-¥500 prior stayed at 0.07. I had registered two paths: competing bid and court challenge. The ledger adds nothing to the competing-bid path (Apollo controls the register; the three largest shareholders as of the balance sheet date are Apollo-affiliated mezzanine funds holding 26.87% combined E78; no third-party bid has been disclosed). Court challenge in Japan on merger-arb pricing is rare. The ¥500 squeeze-out is ~47% above the ¥431.01 Apollo allotment price F155 — Apollo paid a meaningful premium to the minority. BPS at FY2026/3 is ¥1,065.75 F30; the ¥500 squeeze-out is at a 53% discount to book, but book here is largely PP&E and intangibles on a hyper-leveraged entity — I cannot argue the minority is being robbed when the standalone going-concern probability without Apollo was materially below 1.0. I keep the minority-upside prior at 0.07.

The going-concern return-durability prior fell from 0.20 to 0.12. This was the most important move. My prior was that commodity flat-glass earns at or below cost of capital in most periods, and NSG is below the class median. The ledger confirmed this, and added three quantified negatives I had not fully priced:

First, the net-debt figure: ¥484bn F57 against ¥28.8bn OP-before-exceptionals F67 is 16.8× D6. Interest paid in FY2026/3 was ¥28.5bn F100, which is essentially the full operating profit — meaning the business earns just enough to service its debt, with no residual for shareholders before the Apollo recapitalisation changed the structure. The FCF of ¥1.1bn F99 against a ¥484bn net debt stock is 0.2% of net debt — no organic deleveraging path exists from operating cash flow alone.

Second, segment quality: Automotive margins are 1.09% in FY2026/3 D10, the dominant revenue segment (52% of revenue E3). Technical Glass at 18.78% D11 is good quality but only 5% of revenue E4. Architectural Glass at 8.01% D9 is recovering from production shutdowns E17 — a supply-management mechanism, not a structural pricing improvement. The FY2025 Architectural margin was 3.74% (¥13,574M on ¥363,025M [F109/F108]); the FY2026 recovery to 8.01% is significant but traceable to capacity reduction, not demand growth or technology differentiation E61.

Third, the MTP targets are out of reach. Debt target ¥442bn by FY2027 F158 versus ¥548bn gross borrowings F41 — a ¥106bn gap with one year to close, with FCF of only ¥1.1bn F99. The 10% ROS target by FY2030 E13 requires tripling the current 3.27% operating margin D3 without Apollo's restructuring delivering that magnitude of improvement. Management's own language is "highly challenging" for FY2027 E45.

What the ledger confirmed about the going-concern as a standalone. The going-concern note was removed E30 — because of Apollo, not because the business healed itself. E24 is explicit: absent the transaction, refinancing ¥100bn+ of maturities "would have been difficult." This is not a turnaround business; it is a business that needed rescue. The going-concern prior of 0.20 for return durability above cost of capital on a 5-10 year standalone basis is optimistic. I revise it to 0.12.

The special-situation arithmetic — the merger-arb at ¥482. The stamp is ¥482 D1; the squeeze-out is ¥500 F157. The spread is 3.73% D12. At my revised deal-completion probability of 0.93, the expected value of holding to squeeze-out is 0.93 × ¥500 + 0.07 × downside. The downside if the deal breaks is not the intrinsic business value (which is deeply impaired by leverage) but the market price in a scenario where Apollo walks away — itself uncertain, but directionally below ¥400 given the ongoing leverage, thin FCF, and covenant exposure. Taking a deal-break downside of ¥350 (roughly the post-allotment October 2025 range, before the squeeze-out announcement): EV = 0.93 × ¥500 + 0.07 × ¥350 = ¥465 + ¥24.5 = ¥489.5. At the ¥482 stamp, this is a marginally positive expected value — roughly +1.6% — but the distribution is highly asymmetric. There is a 93% chance of a tightly bounded +3.7% gain, and a 7% chance of a material loss (approximately −27% to the ¥350 estimate). The arb is not attractively priced at ¥482 for a risk-adjusted return; a buyer at ¥462 (4% haircut from the ¥482 stamp) gets approximately 8.2% upside to squeeze-out on the completion scenario with a similar downside, making the Kelly-adjusted expected return positive versus a 10% annual hurdle for a position that likely resolves within 12 months.

The implied buy-below of ¥462 is derived as follows: 0.93 × ¥500 + 0.07 × ¥350 = ¥489.5; discounting at a 10%/year hurdle for an assumed 9-month resolution: ¥489.5 / (1 + 0.10 × 0.75) = ¥489.5 / 1.075 = ¥455.3. Round to ¥455 for conservatism; but deal-break downside is likely partially observable (the stock would not go instantly to ¥350 — it would trade toward distressed levels over weeks), so the effective downside is somewhat cushioned by exit opportunity. Balancing the cushion against the model uncertainty, I set implied-buy-below at ¥462. This is below the stamp of ¥482; the current price does not provide adequate margin of safety for the deal-break scenario. Verdict: watch, not buy-below, because the stamp is above the implied-buy-below.

Governance confirms no blocker. Governance is genuinely clean for this context: three statutory committees E64, 6/8 independent outside directors E65, Compensation Committee chaired by an independent E67, zero bonus and near-zero LTIP for CEO/Chair in FY2026/3 because both KPIs missed entry E68E69, no cross-holdings of listed shares E73, no takeover defense E72, no 相談役 system E90. The comp-for-zero-performance outcome is exactly what the comp structure promised; no manipulation of the comp gate is evident. Governance does not block a verdict stronger than watch under comp+governance caps; the cap does not apply here.