SONY 37 min read

Sony Group Corporation (SONY) — Investment Tree v1

Stage 7 final essay. Bilingual companion: tree_v1_zh.md. Date: 2026-08-10 · Anchor price: SONY ADR $23.10 (2026-08-08 close, Tier B) · 6758.T ≈¥3,542 · cross-rate ¥153/USD (derived) · Market cap ≈¥20.8T / ≈$136B · Forward P/E ~17.3× on ¥206 guided FY2027 EPS · Dividend yield ~1.0% forward Archetype: premium-franchise composite priced on a consumer-durables blend — a diversified issuer whose peer-group label is set by its smallest, fastest-shrinking segment

SOURCE QUALITY — read this before the numbers. The FY2026 spine of this tree is Tier A: the 20-F filed 2026-06-18 and the 6-Ks through 2026-07-17, all in evidence_2026-07-18.jsonl. Three things that matter are not:

1. The Q1 FY2027 print (2026-07-30/31) — OI +40% YoY, guidance raised to ¥1.72T OI / ¥1.21T net income, ~¥80B tariff refund, 125M PlayStation MAU. All Tier B, from press aggregation of a 6-K that postdates the evidence cutoff and has not been cross-checked against the primary filing.
2. The 2026-07-28 M7.1 Kumamoto earthquake — Tier B, and its financial impact is quantified by nobody, including Sony, which called it "under assessment."
3. The ADR mechanics — depositary bank is contested (WebSearch says Citibank N.A.; this repo's watchlist presumes BNY Mellon; neither checked against the F-6) and the 1:1 ratio is TBC, corroborated only by observed ADR-vs-TSE price parity. Do not run share-count or cancellation-fee math off the USD figures below.

Peer multiples were sourced by WebSearch on 2026-08-10 and are Tier B. Every EV/EBIT band in the composite is †Tier C — analyst-derived, not quoted.

0. Company Fundamentals — what Sony is and how it earns

Figures FY2026 (year ended 2026-03-31) unless noted. Reported in JPY; USD translations at ¥150.7/USD for P&L, ¥153/USD for market data. Sony is a 20-F filer, not a 10-K filer. Primary listing is 6758.T on the Tokyo Stock Exchange; the analysis ticker is the ADR.

What it is & how it earns. Sony is five largely independent franchises inside one holding company, and the distance between how they earn is the whole story. FY2026 consolidated sales were ¥12,479.6B (~$82.8B, +3.7%) producing ¥1,447.5B of operating income (+13.4%, an 11.6% margin — a record). Segment operating income sums to ¥1,531.1B before a −¥83.6B corporate and non-core drag:

All rows [SONY-2026-06-18-20F-002 through -008], Tier A.

Read that table twice. The segment supplying Sony's market classification — consumer electronics — is 10.4% of the profit pool and shrinking at −16.9% a year. The two segments with genuinely premium standalone comp sets, I&SS and Music, are 52.5%, both at FY2026 records, and are the two fastest-growing.

Cash-flow anatomy. FY2026 continuing-operations free cash flow ≈¥1,182B (~$7.8B) — operating cash flow ¥1,966B less ¥784B of investing outflow. Group capex ¥804.9B (6.4% of sales), of which ¥265.3B (33%) went to I&SS, and ¥246.7B of that specifically to sensor-capacity expansion. The 3-year capex budget was raised ¥1.7T → ¥1.8T, I&SS the primary recipient. [SONY-2026-06-18-20F-004, -013]

Balance sheet & capital allocation. Two consecutive ¥500B buybacks: the FY2026 programme fully executed; the FY2027 programme (authorized to 2027-05-10, up to 230M shares = 3.89% of ex-treasury float) was 25.5% deployed in its first ~7 weeks — 37.1M shares at an average ¥3,439, i.e. roughly double a linear pace and below the current trading band. Dividend ¥25 → guided ¥35/share against a ~40% total-payout target. [SONY-2026-07-06-6K-001, -20F-013]

⚠️ What the balance sheet does not say here. Cash, gross debt, net debt and the maturity schedule were never extracted into the evidence corpus (primer.md gap #2). This tree therefore contains no 10-year ROIC series and its enterprise-value-to-equity bridge assumes net-debt-neutral. Both are stated as gaps, not filled with estimates. This is R2 request #1.

What drives it. Three things, in order: the mix drift (the premium block compounding at +12.8% against a legacy block shrinking at −8.0% — a 20.8-point annual spread); the buyback shrinking the share count ~4%/year; and one undated corporate event, the TSMC joint venture, that would put the first standalone number on a piece of Sony. Key risks: a single seismically-exposed fab, AI-driven memory-cost inflation into console hardware, and a demonstrated ¥192B-in-one-year record of writing off acquisitions.

I. One-sentence verdict

Sony's ~17.3× blended multiple is a default, not a judgment — no diversified peer exists at its scope, so the vendor classification falls back on ET&S, a segment that is 10.4% of profit and shrinking 16.9% a year — but the market has already overridden that label by 5.2 multiple points without naming what the override is for, so the honest segment-weighted composite lands at +6% at the midpoint, not +32%, and with the Kumamoto deduction that the same disaggregation forces into existence still unquantified by anyone including Sony, this is a 1–2% starter with a 3% hard cap and a dated re-test at CP1 (~2026-11) — not a conviction position, and not yet a position at all.

II. Company snapshot

Sony Group Corporation is a Japan-domiciled diversified entertainment-and-technology holding company (6758.T, Tokyo; ADR: SONY). It earns from five franchises: PlayStation gaming and network services; recorded music, publishing and anime IP; CMOS image sensors sold as a component to mobile and automotive OEMs; consumer electronics; and film/TV production. FY2026 continuing-operations net income was ¥1,030.9B — though the total consolidated result was a −¥326.9B net loss because the October 2025 partial spin-off of Sony Financial Group deconsolidated ~¥19.6T of assets and liabilities in a one-time accounting effect. [SONY-2026-06-18-20F-009]

CEO Hiroki Totoki has been in post since April 2025 (~16 months), CFO Lin Tao likewise, with Wendy Becker as board chair from June 2025. That management has three visible actions on the board already: the SFGI spin-off, the cancellation of the Sony Honda Mobility EV launch (¥44.9B), and the Bungie write-down (¥120.1B). The direction of travel is narrowing.

The stock is down ~16.4% year-to-date against record FY2026 results, a beat-and-raise Q1 FY2027, and a 21-of-22 Buy sell-side distribution. Neither published bull nor published bear explains a decline of that size. That unexplained gap is where this tree starts.

III. The six facts that drive everything

1. The classifying segment is a tenth of the profit pool and falling fastest. ET&S: sales ¥2,260.5B (−6.2%), OI ¥158.6B (−16.9%), 7.0% margin, Displays −20.3%. Management's stated response is "disciplined inventory management and fixed cost reduction" — the vocabulary of harvest, not reinvestment. ✅A
2. The two un-comped premium blocks are 52.5% of segment OI and both hit records. I&SS +19.6% sales / +36.8% OI at 16.6% margin; Music +15.1% / +25.1% at 21.1% margin — the highest in the company, and 19.4% after stripping the one quantified one-off (a ¥34.7B Peanuts remeasurement gain). ✅A
3. No segment sum-of-the-parts has been published by any of the 22 covering analysts — despite every required input sitting in the 20-F. Even the maximally bullish target (¥4,686.8) is expressed as a higher blended 22.7×, which mechanically requires a 7.0%-margin Displays business to re-rate alongside the sensor franchise. ✅B (absence over a non-exhaustive search)
4. The market has already overridden its own classification by 43%. Sony trades at 17.3× against a JP Consumer Durables median of 12.1× — independently corroborated this pass at Canon 11.51× and Panasonic 9.10×. The label is being applied; the multiple is not. ⚠️B
5. The moat and the concentration are the same physical asset. I&SS's process/yield lead is embodied in specific plant at Kumamoto/Koshi City — the site named in the TSMC MOU, and the site an M7.1 earthquake struck on 2026-07-28, three days before Sony reported record I&SS results. Staged restart began 2026-08-04 targeting pre-quake output by mid-August. Financial impact: "under assessment." Quantified by nobody. ⊗
6. Sony destroyed ~¥192B of acquired capital in a single fiscal year — Bungie ¥120.1B (≈40% of a ~$3.7B 2022 purchase, written off inside four years), Sony Honda Mobility ¥44.9B on a discontinued EV launch, Pixomondo ¥27.1B. That is 13.3% of consolidated operating income. ✅A

IV. The H-0 thesis

H-0 (one sentence): Sony's single blended multiple is a template artifact, not a judgment — it anchors on ET&S (10.4% of segment OI, shrinking 16.9%/yr) while I&SS and Music (52.5%, both at records) go un-comped; segment-weighted, and net of the fab-concentration deduction the same disaggregation forces into existence, Sony re-rates net upward.

Mispricing taxonomy: Structural blindness × Peer group (per mispricing.md). Every published valuation — the lone named bear's and the 21-of-22 Buy consensus alike — is one multiple × one consolidated EPS. Bernstein: ¥3,400 = 17.3× × ¥197 (cut). Today: ¥3,542 ≈ 17.3× × ¥206. Japan-side average: ¥4,686.8 = 22.7× × ¥206. The framework constrains even the people who disagree with the price. That is the signature of structural blindness rather than of considered disagreement.

Secondary mechanism: Cognitive bias × Option value. Sony is implicitly short a fab-continuity option on Kumamoto that carries no price in any published model. This is not a risk appendix — it is the reason H-0's resolution is directionally contestable rather than automatically bullish. A correct SoTP re-prices the premium and the discount in the same motion.

Where this build revises the thesis — read this before the tree

Three things changed when the analysis was actually run, and they all cut the same way.

(a) "Blindness" is too strong. The right word is default. There is no diversified peer at Sony's scope — consensus.md searched and found none. When no whole-company comp set exists, the vendor classification falls back on the largest historically-identifying segment. That is a default, not a failure of attention. Separately, peers.md found that I&SS's own comp set is thin and partly unobservable: Samsung System LSI (Sony's ~20.2%-share competitor) has no separate listing, STMicroelectronics and OmniVision forward P/Es were not retrievable, and onsemi's 42.7× sits on trough earnings. Part of the reason nobody has published an I&SS carve-out may simply be that there is little clean to carve it against.

(b) The market has already paid for part of what H-0 says it hasn't. The orphan premium is +5.2 multiple points (+43%) over the assigned peer group, and it decomposes almost entirely into an implicit, unstructured partial re-rate of the gaming and sensor blocks. The market has already done, badly and without writing it down, a meaningful fraction of the work H-0 says it has not done at all. What is genuinely missing is the architecture, not the credit. H-0's mechanism survives; H-0's magnitude shrinks considerably.

(c) One of the two premium comp sets does not deliver a premium. UMG trades at 14.81× forward P/E, WMG at ~15× — a rights-annuity median of ~14.9×, which is below Sony's own 17.3× blend. Carving Music out at a UMG/WMG P/E would lower Sony's sum, not raise it. The premium appears only on EV/EBITDA, and only if UMG (29.9×) rather than WMG (14.0×) is the anchor — a 2.1× spread on a two-name comp set. Music's premium is in the margin, not automatically in the multiple, and a real carve-out has to argue for that lens rather than assume it.

5 falsification conditions (per h0_thesis.md, status as of 2026-08-10):

• FF1 — a published SoTP carves I&SS and Music at standalone comps and still lands ≤¥3,500. Not fired — but peers.md makes it likelier than the scaffold assumed.
• FF2 — I&SS OI growth negative or margin <10% for 2 consecutive quarters, and/or Music margin <15%. Not fired; both carry buffer.
• FF3 — ET&S returns to >5% sales growth at >10% margin for a full year. Not fired, and not close.
• FF4 — the Kumamoto/concentration deduction exceeds the I&SS + Music premium. Untested. Resolves at CP1 (~2026-11).
• FF5 — the JV is abandoned or non-capitalized and the 6th mid-term plan carries no structural action. Untested. Resolves at CP3 (~2027-05).
• FF6 — the blend re-rates to ≥22× with no valuation object appearing. Not fired. Would vindicate the bulls, not this thesis.

IV-b. FX decomposition — mandatory for a non-US domicile

An ADR return is never one number. It is two, multiplied:

ADR return  ≈  (local-share return in JPY)  ×  (JPY return vs USD)

Decomposing the move from the last Tier-A price anchor (6758.T ¥3,438 on 2026-07-16, [SONY-2026-07-17-6K-001], at ~¥155/USD) to the current Tier-B anchor (ADR $23.10 on 2026-08-08, 6758.T ≈¥3,542, ~¥153/USD):

Two-sided FX risk statement — and which direction is adverse. Sony reports in JPY, earns the majority of revenue outside Japan, and is bought by this holder in USD. The two directions are not symmetric in their effect on the ADR holder:

• Yen weakness flatters Sony's reported JPY P&L — Sony's own disclosure puts the combined G&NS/ET&S/I&SS sales impact at ~¥33.0B per 1-yen move [SONY-2026-06-18-20F-014] — while simultaneously eroding the USD value of an unchanged 6758.T price. That is the adverse direction for a SONY ADR holder. The business gets better; the holder gets paid less.
• Yen strength does the reverse: a translation tailwind to the ADR against a headwind to the underlying earnings. That is what the table above happens to capture, and it should not be mistaken for the base case.

Neither direction is a mispricing mechanism — mispricing.md rules FX out explicitly, because it is named by both camps and flows through the numerator. It is a measurement obligation: every future update_*.md on this ticker must decompose the ADR move into local return × FX translation rather than reporting one undifferentiated percentage. A single blended figure invites the wrong diagnosis.

Limitation, stated: the more interesting decomposition — the −16.4% YTD move — cannot be run, because neither the January 2026 6758.T close nor the January 2026 cross-rate is in the evidence corpus. The 23-day window above is what is actually grounded. Closing the YTD decomposition is an R2 request.

V. Tree — five branches, nineteen leaves

H-0: SONY's blended 17.3x is a template artifact; segment-weighted
     and net of a fab-concentration deduction, SONY re-rates net upward
     ⚠️ 部分支持 — h0 51%  [⚠️ LOW EVIDENCE COVERAGE: 35.6% of tier weight excluded]
│
├── L1A — The semiconductor credit (I&SS)          ⚠️ partial
│   │     Relative Valuation + Porter's Five Forces
│   ├── 1.1 No I&SS carve-out published, 22 analysts   ✅B supports          [3]
│   ├── 1.2 I&SS margin >=10% through CP1/CP2          ⚠️B partial           [3]
│   ├── 1.3 "Cautious view" -> sub-10% FY2027 growth?  ⊗  not measurable     [1]
│   └── 1.4 ~43.4% CMOS share stable vs China low-end  ⚠️C partial           [3]
│
├── L1B — The owned-rights annuity (Music)         ✅ supported
│   │     Relative Valuation + ROIC
│   ├── 1.1 Music margin >=15% (19.4% clean)           ✅A supports          [3]
│   ├── 1.2 No UMG/WMG carve-out published             ✅B supports          [3]
│   ├── 1.3 Anime VM&P +33.1% durable?                 ⚠️C partial           [1]
│   └── 1.4 Catalog M&A ROIC-accretive?                ⊗  no denominator     [1]
│
├── L1C — The installed-base digital annuity       ⚠️ partial
│   │     Platform Economics + Relative Valuation (bounded)
│   ├── 1.1 Network Svcs +13.9% outpaces DS&AC +5.4%   ✅A supports          [1]
│   ├── 1.2 Hardware run-down degrades the annuity     ⛔B leans AGAINST     [3]
│   └── 1.3 C3's bounded OI allocation narrows         ⚠️C partial           [1]
│
├── L1D — The legacy tail (ET&S + hardware + slate) ⚠️ partial
│   │     Reverse DCF + Historical Analogue
│   ├── 1.1 ET&S clears FF3 (>5% at >10% margin)       ✅A supports (FF3 NO) [3]
│   ├── 1.2 ~Y80B tariff refund lands as guided        ⚠️B partial           [1]
│   ├── 1.3 Displays -20.3% is a managed run-down      ⚠️A partial           [1]
│   └── 1.4 Console-hardware OI pinned below zero      ⊗  never disclosed    [1]
│
└── L1E — Net deduction & conversion mechanism     ⊗ UNMEASURED
      │     Real Options + SoTP subtraction + Historical Analogue + ROIC
      ├── 1.1 Kumamoto deduction < premium (FF4)       ⊗  CRUX — no figure   [5]
      ├── 1.2 TSMC JV -> definitive, capitalized       ⊗  CRUX — undated     [5]
      ├── 1.3 6th mid-term plan structural signal      ⊗  plan not written   [3]
      └── 1.4 C6 drag stabilizes; buyback compresses   ⚠️A partial           [3]

Total: 5 ✅ / 7 ⚠️ / 1 ⛔ / 0 ✗ / 6 ⊗  across 19 leaves
[N] = declared tier weight (crux 5 · load-bearing 3 · supporting 1)
H-0 verdict: 部分支持 (PARTIALLY SUPPORTED), h0 = 14.65/29 = 51%

The shape of this tree is the finding. Read the branches left to right: L1A, L1B and L1D all support H-0's architecture claim. The classification is inherited, the premium blocks are un-comped, the anchor block is a tenth of the profit pool and nowhere near clearing the bar that would make it a defensible peer group. That half of the thesis is evidenced.

Then read L1E. Both crux leaves are ⊗. The deduction that must be netted against the premium has no figure from any party including Sony, and the event that would create the missing valuation column is undated by both signatories to the MOU. leaves.md computes it precisely: 35.6% of total declared tier weight is excluded from h0 as 证据不足, and both crux leaves are among the excluded. The h0 of 51% is an honest score over the evidenced half and says nothing about the half that decides direction.

No leaf in this tree carries 强力支持. That is deliberate. The strongest evidence available is one fiscal year of Tier-A segment disclosure plus an absence-of-coverage finding over a non-exhaustive search, and verdict-calibrator demotes 强力支持 to 支持 on one-Tier-A-source evidence. A tree whose crux is unmeasured should not contain a single maximal verdict.

VI. Key findings

Finding 1 — The composite is net upward by ~6%, not ~32%. This is the number that matters.

peers.md builds the column mispricing.md says nobody has built. FY2026 segment operating income at comp-set EV/EBIT bands, less a 5–10% single-site fab-concentration deduction (the TSMC-Taiwan analogue that honest disaggregation forces into existence):

The band straddles the current price. H-0's direction claim survives at the midpoint and the high end and is negative at the low end — before any Kumamoto quantification lands. That is a materially weaker claim than the published bull's ¥4,686.8 (+32%), and it is weaker for two checkable reasons named in Section IV-b: the Music comp set trades below Sony's own blend on forward P/E, and the semiconductor comp set has no usable listed anchor.

The bull case's arithmetic is confirmed incoherent — in H-0's favour. A blended 22.7× requires Canon-comparable ET&S and Disney-comparable Pictures to re-rate alongside the sensor business. The composite prices those blocks at 6–9× and 9–13×, which is where their comp sets actually trade. Disaggregation reaches a lower destination than the bull's blend, by a better route.

⚠️ The load-bearing caveat on this entire table: Sony's net debt is absent from the evidence corpus, so the enterprise-value-to-equity bridge assumes net-debt-neutral. If Sony carries material net debt post-SFGI, every figure above is too high. This is a data gap, not an analytical one.

Finding 2 — Music is the strongest block and the most poorly served by its own comp set

Music is 29.2% of segment OI at a 21.1% margin (19.4% clean), the highest in the company, with all four disclosed sub-lines growing — streaming +8.1%, live/merch +21.0%, publishing +10.6%, anime VM&P +33.1%. It is a whole reported segment: comping it against UMG/WMG requires zero allocation assumption. Its absence from published coverage is the least excusable fact in the entire universe of 22 analysts.

And then peers.md returns UMG at 14.81× and WMG at ~15× forward P/E — below Sony's own 17.3×. The margin is premium; the multiple is not. This is the honest weak point in the premium half of H-0, and it is why RF3 (FF1 — a published SoTP landing ≤¥3,500) is rated likelier than the scaffold assumed.

A second problem sits underneath: leaves.md L1B 1.4 is ⊗. Music's growth is bought with catalog M&A, and Sony discloses no segment invested capital, so the accretion question a rights-annuity analyst underwrites first cannot be run at all.

Finding 3 — The legacy run-down is not costing the annuity what the taxonomy feared

taxonomy.md flagged a real threat: console hardware is partly a customer-acquisition line for the digital annuity, so "run down the legacy tail" would not be costless. The evidence leans against it in its near-term form (⛔B). Through two consecutive years of −12.1% hardware decline, Network Services grew +13.9%, digital add-on content grew +5.4%, PlayStation MAU hit a record 125M, and Crunchyroll — a subscription leg with no console dependency at all — crossed 21M paid subscribers.

The honest qualification: two years is short relative to a console generation, and late-cycle installed bases are at peak engagement because the hardware wave already landed. This bounds the near-term coupling. The generation-transition test is unrun.

Finding 4 — Sony's own disclosure policy is the binding constraint on Sony's own re-rating

This is the finding that most cuts against H-0, and it should not be smoothed over.

Sony's largest profit block is the installed-base digital annuity — G&NS digital software and network services plus Pictures Media Networks, ¥3,656.7B, 29.3% of consolidated sales. There is no ¥3,656.7B line item anywhere. It is assembled from three sub-lines living in two reported segments. And Sony discloses no sub-segment operating income anywhere — not in the 20-F, not in the kessan tanshin.

So the market cannot build a valuation column for Sony's largest profit block even if every analyst wanted to. The conglomerate discount H-0 wants closed is partly self-inflicted and entirely within management's power to fix at zero cost — and management has not. H-0's mechanism is strong at I&SS and Music, where every input is published. It is weak here, where the input does not exist. Trigger T5 — Sony discloses sub-segment OI — is the highest-leverage, lowest-cost action available to this management team and it does not appear on anyone's list of catalysts.

Finding 5 — The acquisition record is one bad year from a fatal flag

FY2026 destroyed ~¥192B of acquired capital: Bungie ¥120.1B (≈40% of a ~$3.7B 2022 purchase, written off inside four years), Sony Honda Mobility ¥44.9B on a discontinued EV launch, Pixomondo ¥27.1B. That is 13.3% of consolidated operating income in one year.

Against it: the organic record is genuinely good. The 5th mid-term plan is beating its own targets on both growth (18% average annual OI growth vs 10%+) and margin (11.1% cumulative vs 10%+); the buyback is executed at roughly double a linear pace below the trading band; capex goes to the highest-return franchise. A company beating its own multi-year plan on growth and margin is not serially destroying capital. The correct characterisation is excellent organic allocator, poor acquirer — which caps durability_test.md Q3 at 3/5 without firing K.3.1.

But the call is close, and it is documented as close. One year cannot distinguish "new CEO clearing the prior regime's decks" from "first year of a pattern." RF5 is the trip-wire: another >¥100B non-core write-down in FY2027 or FY2028 re-cuts Q3 to 1/5, fires a fatal flag, and caps the position regardless of aggregate score.

Finding 6 — The market is not being irrational, and saying so is part of the analysis

Market-implied scenario weights, reverse-engineered from $23.10: 10% Bull / 60% Base / 30% Bear. Mine: 20/60/20. The Base case is not in dispute at all — the entire disagreement is 10 percentage points of tail weight, and after adjusting for a 6–8% required return the real gap is closer to 5–7pp.

The market's 30% Bear is more internally consistent than the published sell-side. A −16.4% YTD decline against record results and a 21-of-22 Buy distribution is exactly what a 30% Bear weight looks like. mispricing.md names the likely reason: price may already be moving on the fab-continuity option while the models have not caught up. If so, the market's 30% is not error, and my 20% is the number that needs defending. I hold it because the staged restart began within seven days of an M7.1 event and targets pre-quake output within ~three weeks. But this is the one call in the tree where I am explicitly taking the other side of the tape, and it should be re-tested at CP1 rather than defended.

VII. Three valuation scenarios

(Full derivation in scenarios.md. Probability prior 20/60/20 per K.3.3's second default — "current pricing is approximately fair and the asymmetric mispricing depends on a specific catalyst not playing out." Three candidate deviations were considered and explicitly rejected; see scenarios.md for the reasoning on each.)

Probability-weighted 12-month expected return: +3.9%. Asymmetry 1.50 : 1 favourable.

Two asymmetry numbers, and the gap between them is the whole sizing argument. On the structural composite alone the setup is 2.29 : 1. Once the unquantified Kumamoto tail widens the Bear band — a range adjustment, deliberately not a probability adjustment — it degrades to 1.50 : 1. The entire difference between "attractive" and "merely acceptable" is one number that does not exist yet. A 2.29:1 setup with a dated resolver supports a real position; a 1.50:1 setup with an undated one supports a starter and a calendar entry.

Note what the Bear case is not. It is not "Sony is a bad company." It is FF4 exactly as h0_thesis.md names it: the missing-column diagnosis is right, and building the column produces a smaller number than the blend. peers.md already shows the mechanism half-realized in the Music comps. That is why the Bear weight is 20% and not 15%.

VIII. Triggers and red flags

(Full node-linked table with 应对 playbooks in triggers_redflags.md. Every trigger below resolves against a node ID; no orphans.)

Triggers (H-0 confirms):

• T2 (~2026-11, CP1) — Kumamoto P&L impact quantified. L1E 1.1 [CRUX]. The gating event of this entire file. Fires bullish under ¥50B; neutral ¥50–120B; becomes RF1 above ~¥150B. Do not add to the position ahead of this print.
• T1 (undated) — TSMC JV converts to a definitive, separately-capitalized agreement. L1E 1.2 [CRUX]. The only event that mechanically creates the missing column: a disclosed equity split and TSMC capital contribution imply a market-validated value for the Kumamoto sensor asset. MOU signed 2026-05-08; conversion base rate 6–18 months; no committed date from either party.
• T3 (~2027-05, CP3) — 6th mid-term plan carries an I&SS structural signal. L1E 1.3. The dated venue; SFGI is direct precedent.
• T4 (undated) — any analyst publishes a segment SoTP. L1A 1.1, L1B 1.2. Requires zero new disclosure.
• T5 (anytime) — Sony discloses sub-segment operating income. L1C 1.3, L1D 1.4. Highest-leverage, zero-cost action available to management; narrows two of the composite's three widest bands at once.
• T6 (~2027-02, CP2) — first fully post-recovery quarter. L1A 1.2.

Red flags (position gets cut):

• RF1 (~2026-11) — Kumamoto impact >¥150B, or a structural design-win loss at a top-3 mobile OEM. FF4 fires. Exit the starter within 30 days. Score it as a falsification of H-0 as stated — do not quietly re-label it a partial support.
• RF3 (anytime) — a published SoTP carves I&SS and Music and still lands ≤¥3,500. FF1 fires — the strongest falsifier. It proves the column was omitted because it does not change the answer. peers.md makes this likelier than the scaffold assumed.
• RF4 (~2027-05) — no structural action at CP3 and the JV abandoned or non-capitalized. FF5 fires — thesis expiry. Retire the thesis; do not roll it forward.
• RF5 (FY2027/28) — another >¥100B non-core write-down. The K.3.1 fatal-flag trip-wire. Cut to ≤1% and re-run durability before any re-entry.
• RF6 (standing) — another JMA M6.5+ event in Kumamoto/Kyushu. Base rate two in ten years. Suspend all scaling for one quarter; do not average down into an unquantified outage.
• RF8 (anytime) — a re-rate to ≥22× with no valuation object appearing. FF6 fires — "right price, wrong reason." Take the gain, close the position, and record the thesis as unresolved. A tree that lets a favourable price launder an untested mechanism has stopped being a tree.
• RF9 (standing) — currency. See Section IV-b. Yen weakness is the adverse direction for the ADR holder.

Action mapping: 1 FF fires → cut to ≤1% and open a thesis-update cycle. 2 FFs → exit to zero; the tree is rebuilt, not updated. FF4 or FF5 alone is sufficient to exit — both are crux-node failures, and the h0 rollup already excludes both crux leaves, so there is no evidenced remainder to fall back on.

IX. Long-term holdability verdict

Per durability_test.md: aggregate 18/25 — Medium durability (17–21 band). 0 fatal flags. exceptional_positive_overrides: no.

Fatal-flag check — 0 fired, and two of the three are close calls that are documented as close. Q3 requires 1/5 (chronic ROIC < WACC with serial destruction); a company beating its own multi-year plan on both growth and margin is not that, so 3/5 stands — but RF5 is the trip-wire and another >¥100B write-down fires it. Q4 requires a threat >50% probability with high impact; seismic clears probability and fails impact, cloud clears impact and fails probability. Balance-sheet survivability is inferred from cash generation (¥1,182B FCF funding a ¥500B buyback internally), not read from a balance sheet, because the balance sheet is not in the corpus. No flag fires on the available evidence; the gap is recorded rather than assumed away.

Escalation check: 18 ≥ 17, so no long-term-unsuitability warning is escalated. But 18 sits one point above the Medium floor, not comfortably inside it.

Position-sizing recommendation

Durability 17–21 with 0 fatal flags maps to the K.3 selective hold, 1–3% tier. Layered against the tree's own state:

• Today (pre-CP1): 0% — WATCH. Both crux leaves are ⊗; 35.6% of tier weight is unmeasured; asymmetry is 1.50:1; and the largest deduction in the composite has no figure from any party. A 1–2% starter is defensible for a holder who wants to be positioned before CP1 — the composite midpoint is +6%, the buyback runs regardless, and the downside band is bounded at roughly −20%. It is not defensible to size beyond that ahead of a print that could quantify the whole thesis against you.
• Post-CP1 with the Kumamoto impact under ¥50B (T2 fires): the largest unquantified deduction collapses, asymmetry moves toward 2.29:1 — scale to 2–3%.
• Post-T1 (JV definitive agreement, undated): the missing column is created rather than argued — scale to the 3% cap within 30 days.
• Hard cap 3% absent a durability upgrade to ≥22/25. That upgrade requires the sub-segment-OI disclosure (T5), a clean FY2027 with no >¥50B non-core write-down, and a balance-sheet extraction. Do not exceed 3% on price action alone.
• Auto-cut rules: RF1 or RF3 fires → exit within 30 days. RF5 fires → cut to ≤1% and re-run the durability test before re-entry. RF8 fires → take the gain, close, and record the thesis unresolved.

Do not add ahead of CP1 (~2026-11). That is the single operational instruction this tree produces.

Correlated-exposure paragraph (per K.3.4)

Load-bearing macro factor: the Japan sovereign/FX complex — not the AI-capex cycle. This is worth stating plainly because SONY looks like an AI name and is not one in the way that matters for portfolio construction.

Cycle exposure: ai-capex-low, and marginally negative near-term. Sony does not sell into hyperscaler capex. Its first-order AI-cycle linkage runs the wrong way: AI-datacenter demand inflates DRAM and NAND, which compresses PS5 hardware margin — this is the entirety of Bernstein's quantified bear case, corroborated by Sony's own two PS5 price hikes in eight months and a near-total memory-card order suspension. The physical-AI upside (robotics and automotive sensors via the TSMC JV) targets ~2029 and is financially immaterial today. SONY is therefore a partial hedge against, not an addition to, the AI-capex correlated bucket — the one genuinely useful diversification property in this file. Against the K.4 operational rule (correlated bucket >15% of portfolio → drop each cap by 2pp), SONY does not count toward that bucket.

Sovereign exposure: sovereign-mixed. Global revenue (Japan is only ~10.6% of sales†) against a domestic cost, listing and regulatory base. The channel that matters is not revenue — it is the discount rate: this thesis's payoff mechanism is multiple expansion, which is the single mechanism most sensitive to a rising domestic risk-free anchor. A JGB repricing is a headwind to H-0 specifically, in a way it would not be to a cash-flow-growth thesis.

Other StockNews tickers sharing this load-bearing factor: TM, HMC, SFTBY, AJNMY, TOTDY, RESONAC — all Japan-domiciled, all carrying the same JPY translation drag against a USD holder, all exposed to the same JGB/discount-rate channel. If the holder's combined Japan-domiciled sleeve already exceeds ~12% of portfolio, drop SONY's hard cap by 1pp at each tier (3% → 2%). Note that AJNMY and RESONAC additionally sit in the AI-capex bucket, so a portfolio holding SONY + AJNMY + RESONAC is doubly concentrated on Japan while being split on the AI cycle — which is a real diversification benefit inside the Japan sleeve and should be scored as one.

The correlated-exposure line that matters most, and it is not macro: SONY's thesis and AJNMY's thesis are the same thesis. Both are structural-blindness-on-segment-reporting stories about a Japanese conglomerate whose highest-margin franchise is buried inside a reported segment and never comped standalone. Both resolve on a disclosure event rather than a business event. If the general proposition "Japanese issuers will start disclosing their good segments" is wrong, both positions are wrong together, for the same reason, at the same time — regardless of what image sensors or ABF film do. That is a correlation no cycle overlay captures and it should cap the combined SONY + AJNMY position at 5%, below the sum of their individual caps.

XII. Investment Scorecard (per MANUAL_en.md Part K.6 + K.3.5)

15-question scorecard — Format B (analytical-tree Q-list)

Format B is used, not Format A. SONY's archetype is a segment-attribution / sum-of-the-parts story — structurally the RESONAC and AMZN shape — and Format B's Q-list (what it does · why now · bull · bear · valuation · revenue · profits · FCF · debt · competitors · sell-triggers · falsification · 2036-relevance · moat · ROIC>WACC) maps directly onto it. Formats are not mixed.

Verdict tally: 9 ✅ · 5 ⚠️ · 0 ✗ · 1 ⊗ — Q1✅A, Q2✅B, Q3✅B, Q4⚠️B, Q5⚠️B, Q6✅A, Q7✅A, Q8✅A, Q9⊗, Q10⚠️B, Q11✅A, Q12✅A, Q13✅B, Q14⚠️B, Q15⚠️C.

K.3.5 Weighted-score derivation

Format B tier mapping per MANUAL §K.3.5: Critical (5×) Q1, Q9, Q14 · Load-bearing (3×) Q4, Q5, Q11, Q12 · Important (2×) Q3, Q6, Q7, Q15 · Confirming (1×) Q2, Q8, Q10, Q13. Verdict values: ✅ = 1.0, ⚠️ = 0.5, ⛔ = 0.25, ✗ = 0.0. ⊗ rows are excluded from both the numerator and the theoretical-max denominator.

xii_score = 27.0 / 34 = 79%.

Band: 65–85% → moderate buy with sizing discipline.

On the excluded Critical row — this is not a rounding detail. Q9 (debt) is a 5×-weight Critical question and it is ⊗, which drops the theoretical max from 39 to 34. Per K.3.5 the exclusion is mandatory: an honest "the evidence does not exist" must not be scored adjacent to a falsified row. But note the direction of the effect — excluding an unknown Critical row makes the score look better, not worse, because the remaining rows are stronger than average. A reader should treat 79% as "79% of what could be measured," and the thing that could not be measured is whether this company carries too much debt. That is not a small caveat and it is R2 request #1 for a reason.

xii_score 79% vs h0 51% — a 28-point gap, and the gap is the thesis. xii_score measures structural quality: what the 15-question scorecard says about the business, the balance sheet, the moat and the valuation. h0 measures thesis confidence: how well the specific H-0 statement is evidenced. Sony scores 79% because it is a genuinely good set of businesses — records in its two best segments, disciplined capital return, a moat with an 11.1%-of-sales maintenance budget. It scores 51% because the specific claim that the market will re-price those businesses has both of its crux tests unmeasured. The 28-point gap names precisely what catalyst-firing could close: T1 (JV converts), T2 (Kumamoto quantified small), T5 (sub-segment OI disclosed). Nothing in that gap requires Sony to become a better company. It requires Sony, or an analyst, to publish a number.

Scorecard summary

Final verdict: HOLD-WITH-SIZING — 1–2% starter, 3% hard cap, do not add before CP1 (~2026-11)

Not a Buy, and the reason is specific rather than temperamental: a thesis with both crux leaves unmeasured is not a conviction thesis, however good the underlying company is. Not an Avoid either — the composite is net upward at the midpoint, the durability is Medium with zero fatal flags, and a ¥500B buyback executed below the trading band pays the holder to wait.

For the owner specifically:

• Today: 0% held. WATCH is the correct state. A 1–2% starter is defensible for a holder who wants to be positioned before CP1; anything larger is not.
• ⚠️ Do not add ahead of CP1 (~2026-11). That print quantifies the largest unknown in the file and could resolve it against the thesis.
• ✅ Scale to 2–3% if T2 fires with a Kumamoto impact under ¥50B — asymmetry moves from 1.50:1 toward 2.29:1.
• ✅ Scale to the 3% cap within 30 days if T1 fires (JV definitive agreement) — the missing column is created rather than argued.
• 🔻 Exit within 30 days on RF1 or RF3. Cut to ≤1% on RF5.
• 🎯 Hard cap 3% absent a durability upgrade to ≥22/25. Combined SONY + AJNMY cap: 5% — same thesis, same resolution mechanism, same failure mode.
• 📅 Re-test at CP1 (~2026-11), then CP2 (~2027-02), then CP3 (~2027-05).

The 2-minute pitch:

"Sony gets valued as a Japanese consumer-electronics company. That business is 10% of its profit and shrinking 17% a year. The two segments that actually earn the money — image sensors at 16.6% margin with the #1 global share, and music at 21.1% margin with perpetual copyrights — are 52% of profit, both at records, and no analyst out of 22 has ever valued either one standalone. So I built the column nobody built. And here is the honest answer: it comes out about 6% above the current price, not 32%. Two reasons. The market already pays a 43% premium over the peer group it supposedly assigns Sony to, so a chunk of the credit is already in the price. And the music comps — Universal at 14.8×, Warner at 15× — actually trade below Sony's own 17.3×, so one of the two premium blocks does not deliver a premium multiple at all. Meanwhile an M7.1 earthquake hit Sony's single sensor fab on July 28th, three days before it reported record sensor results, and nobody, including Sony, has put a number on it. Both crux tests of this thesis are unmeasured. So: 1–2% starter, 3% hard cap, and do not add a share before the November print that quantifies the earthquake. Scale to 3% if that number comes in under ¥50B or if the TSMC joint venture converts to a definitive agreement. Sell if a published sum-of-the-parts lands at or below ¥3,500 — that would prove the column was omitted because it doesn't change the answer."

Risk types most relevant (per MANUAL_en.md Part K.4):

• Currency risk (mandatory for an ADR — see §IV-b) — yen weakness is the adverse direction for the ADR holder: it flatters Sony's reported JPY P&L (~¥33.0B combined sales impact per 1-yen move) while eroding the USD value of an unchanged 6758.T price.
• Execution risk — Kumamoto fab recovery; TSMC JV MOU→definitive conversion; and a demonstrated ¥192B-in-one-year acquisition write-off record.
• Cyclical risk — AI-driven DRAM/NAND inflation into console hardware; the smartphone sensor mix cycle management has itself guided down.
• Competition risk — Chinese CIS vendors eroding the low end; four separate competitive fronts with dominance on only one.
• Valuation risk — 17.3× is already +43% over the assigned peer group; the composite band straddles spot rather than sitting above it.
• Balance-sheet risk — not that leverage is known to be high, but that it is not known at all. Q9 is ⊗.
• Regulatory risk — US tariff regime (two-sided, with an unresolved N.D. Cal. class action on the refund); Japan FSA/TSE disclosure regime.
• Correlated-factor risk — Japan sovereign/FX sleeve shared with TM, HMC, SFTBY, AJNMY, TOTDY, RESONAC; and the deeper correlation with AJNMY, which is the same disclosure-resolution thesis wearing a different segment.

"When NOT to buy" anti-pattern check (per MANUAL_en.md Part K.5):

• ✅ NOT buying on a single news headline — the thesis is a segment-arithmetic argument built from a 20-F, not from the earthquake or the earnings beat.
• ✅ NOT buying on momentum — the stock is down 16.4% YTD. If anything the risk here is the opposite error (catching a decline nobody has explained), which is precisely why Finding 6 takes the market's 30% Bear weight seriously rather than dismissing it.
• ✅ NOT a low-nominal-price illusion — valuation is anchored on segment OI multiples and a per-share composite, not on the ADR's $23 handle.
• ⚠️ "It has AI attached to it" — flagged, and this is the one real anti-pattern hit. The published bull case (24/7 Wall St, 2026-05-27) is explicitly an NVDA-style physical-AI reframe built on a non-binding MOU targeting production in ~2029. This tree does not use that argument. Its I&SS credit rests on FY2026 disclosed economics — +19.6% sales, +36.8% OI, 16.6% margin, ¥238.5B R&D — not on robotics optionality, and its cycle band is ai-capex-low with a negative near-term sign. Stated so the reader can check that the AI narrative is not doing hidden work.
• ⚠️ Brand familiarity bias — flagged. "Sony" is one of the most recognizable brands on earth, and K.5 names brand recognition (Nokia, GE, Sears) as an anti-signal. Mitigation: every load-bearing number in this tree traces to a 20-F evidence ID, and the tree's own conclusion is that the famous consumer-electronics business is the weakest thing in the company and should be valued at 6–9× EV/EBIT.
• ❌ Not "the next Nvidia" pattern-matching — the archetype is explicitly value-with-overhang, not a growth analogue.

Net: 2 anti-pattern flags (AI narrative adjacency, brand familiarity), both mitigated in the text above rather than dismissed. Neither is disqualifying; both are reasons the position is a starter rather than a stake.

XIII. What's NOT in this tree (deferred / documented but not built)

• R2 data requests, in priority order. (1) Extract Sony's balance sheet — cash, gross debt, net debt, maturity schedule — from the FY2026 20-F; without it Q9 is ⊗, no ROIC series exists, and peers.md's EV→equity bridge is an assumption. (2) Confirm whether Sony discloses sub-segment operating income anywhere in the 20-F or the kessan tanshin; the answer re-shapes H-0's mechanism claim either way. (3) Resolve the ADR depositary conflict against the actual F-6 — WebSearch says Citibank N.A., this repo's watchlist presumes BNY Mellon, and the 1:1 ratio is corroborated only by observed price parity.
• Stage-0 extraction pass on the Q1 FY2027 6-K (2026-07-30/31) — the OI +40%, the raised ¥1.72T guide, the ~¥80B tariff refund and the 125M MAU are all Tier B and load-bearing to the Base case's numerator.
• The YTD FX decomposition — the −16.4% move cannot be decomposed because neither the January 2026 6758.T close nor the January 2026 cross-rate is in the corpus. Only the 23-day window in §IV-b is grounded.
• premortem-steelman.md — the Bear case in scenarios.md covers FF4 and FF5; a full premortem is deferred to the post-CP1 update.
• Exact 20-F sub-line tables for G&NS and Pictures — would close taxonomy.md's ¥128.6B unallocated residual (1.03% of consolidated sales) and ET&S's ¥782.9B unitemized residual.
• I&SS top-customer concentration (primer.md gap #5) — may not be publicly obtainable, which is itself worth confirming rather than assuming.
• update_{YYYY-MM-DD}.md — first scheduled at CP1 (~2026-11). Mandatory FX decomposition applies (non-US domicile).

Last updated 2026-08-10. Source quality: Tier-A FY2026 20-F + 6-K spine; Tier-B post-cutoff events and peer multiples; Tier-C derived EV/EBIT bands and cross-rate. K.3.6 evidence-strength suffixes applied to every leaf and every Section XII row. lint_leaves clean. Next refresh: CP1 (Q2/H1 FY2027 results, ~2026-11).

"Stories lie, structure doesn't. And when the structure has a hole in it, you say so and you size accordingly." — 90s.PM.Investing