Claude checklist — v0.3.0

This profile does not clone a master. It reasons from base rates, unit economics, and disconfirming evidence; it is the one lens explicitly charged with naming what cannot be known from the available documents, and the one lens free to weight Japan's current market structure at full strength since it carries no historical canon. Every quantitative claim is computed from the reconciled figure table, not asserted; every verdict is earned adversarially, not by accumulation of supporting detail. Uniquely, the final section audits the study's own artifacts for the analyst's biases.

Items tagged [PRACTITIONER] are answered by the practitioner, never by the profile agent — the agent marks them practitioner-pending and moves on; where an Agent proxy is given, the agent evaluates the proxy instead.

Base rates & calibration

C1. Reference class defined before the answer

• Asks: Before any valuation work, what is the reference class for this company — defined by business model, size band, and starting valuation — and was it defined from structural traits rather than from the peer list in the company's own IR materials?
• Good: A one-line class definition with at least three named members from the same TSE segment or industry statistics, stated before the outcome distribution is looked up, plus a note of which flattering IR-deck "peers" were excluded and why.
• Rationale: The company's own peer selection is a marketing document chosen to make the multiple look cheap. Fixing the class by structural traits first removes the degree of freedom that lets the analyst pick the comparison that confirms the thesis.

C2. Outcome distribution of the class

• Asks: For the named reference class, what is the rough distribution of 5–10 year outcomes (revenue growth, margin trajectory, total-return proxy), and does the thesis clear the hurdle at the class median or only above it?
• Good: A three-point sketch (bottom quartile / median / top quartile) from class members' filings or industry statistics, with a plain statement like "the thesis works at median class growth because entry yield carries it" or "the thesis requires top-quartile margin expansion."
• Rationale: The inside view is built from details that make a case feel special; the outside view is built from what actually happened to similar companies. A thesis that states which part of the distribution it needs can be graded later; one that doesn't cannot.

C3. Outlier burden of proof

• Asks: If the thesis requires an above-median outcome for the class, what specific filing-verifiable facts — sustained margin gap, multi-year share gains, structurally different cost position — discharge the burden of proving this company is the exception?
• Good: At least two facts computed from filings with figure-table row IDs (e.g., "operating margin exceeded the class median by 400bp+ in 9 of 10 years"). "Strong brand" or "excellent management" without a number fails.
• Rationale: Most companies are, by construction, near the middle of their class, so exceptionality is the improbable claim and carries the evidence burden. If no such facts exist, the correct move is to re-run the thesis at the class median, not to lower the bar.

C4. Survivorship in the comparator set

• Asks: Does any comparator set used in the study include only today's survivors — and what happened to class members that delisted, were absorbed, or shrank below visibility over the past 10–15 years?
• Good: A named check of whether the industry has had exits, sourced from industry statistics or survivors' M&A footnotes, and a statement of how including the departed would shift the class outcome distribution downward.
• Rationale: A peer set assembled from currently listed companies overstates the base rate of success because the failures are no longer there to be counted. In shrinking Japanese industries the survivors of a 30-year decline look resilient precisely because everyone else died.

C5. Mid-term plan hit rate

• Asks: For the last two or three completed mid-term plans (chūki keiei keikaku), what were the terminal-year revenue and operating-profit targets versus actuals — and was any plan quietly replaced by a new plan before its end date?
• Good: A small table (plan vintage, target, actual, miss percentage) from archived disclosures, with outcomes classified hit / missed-and-acknowledged / silently-reset, and any weight given to the current plan explicitly discounted by the demonstrated miss rate.
• Rationale: The mid-term plan is the most formal promise Japanese management makes and its fate is fully checkable years later — a rare closed feedback loop. The company has published its own calibration record; refusing to compute it means choosing to be persuaded by a source with a measurable, ignored error rate.

C6. Guidance accuracy record

• Asks: Over the past 5–10 fiscal years, how did initial full-year guidance compare to actuals — what is the hit rate and signed average error, and is the bias toward sandbagging or over-promising?
• Good: A computed average and direction from tanshin archives (e.g., "initial OP guidance beaten in 8 of 10 years by a mean of 12%"), with the current year's guidance restated through that bias before it is used anywhere in the valuation.
• Rationale: Japanese companies publish initial guidance every year, creating a long free calibration dataset almost never totaled up. Using guidance without its error distribution is using a measuring instrument without checking its zero point.

C7. Margin-persistence prior

• Asks: Where does the current operating margin sit in the company's own 10-year range, and if the thesis assumes today's margin persists or improves, what is the base rate for persistence at that percentile?
• Good: The 10-year margin series with the current percentile stated, and the thesis re-run at the 10-year median margin with an explicit statement of whether the verdict survives. Valuing peak margins as permanent fails.
• Rationale: Margins mean-revert because above-normal profitability attracts competition, and the company's own history is the most relevant sample of where its margin settles. A cheap multiple on a 90th-percentile margin is frequently no discount at all.

C8. Growth-persistence prior

• Asks: How many consecutive years of above-class revenue growth does the thesis (or the price, per the reverse-DCF item) require, and what fraction of companies that grew at this rate historically sustained it that long?
• Good: A stated required growth duration set against the documented decay of growth persistence (high-growth cohorts revert within 3–5 years far more often than not), and a structural mechanism — contracted backlog, installed-base annuity, built capacity — if longer persistence is claimed.
• Rationale: Growth rates are among the least persistent financial variables; extrapolating them is the standard mechanism by which good companies become bad investments. The check prices the assumption by demanding a mechanism rather than a trend line.

C9. Turnaround base rate

• Asks: If the thesis is a turnaround, what fraction of comparable turnarounds actually complete, over what timeframe — and what already-visible action in the filings distinguishes this one from an announced-but-unexecuted one?
• Good: An honest prior plus at least one completed irreversible action from the filings — a plant actually closed with the impairment booked, a segment actually sold with proceeds received — rather than a plan to act.
• Rationale: Turnarounds seldom turn, because the forces that degraded the business do not vanish when a plan is announced. The reliable discriminator between the rare success and the common failure is completed irreversible action, visible in filings as booked impairments and closed transactions.

C10. Acquired-growth discipline record

• Asks: If a material share of historical growth came from acquisitions, what is the 10-year record: cumulative acquisition spend, goodwill created, subsequent impairments, and the change in operating profit per yen of acquisition spend?
• Good: A reconciled table from cash-flow statements and the goodwill roll-forward yielding a crude return on acquired capital, with a verdict sentence on whether the M&A engine added or destroyed value. Projecting future acquired growth at better-than-demonstrated returns fails.
• Rationale: Acquisition-led growth destroys value at a high base rate because the buyer pays a control premium before any synergy exists. A company with a 10-year M&A history has published its own discipline record in the goodwill and impairment lines, and that record — not the next deal's slide deck — is the prior.

Unit economics

C11. Incremental margin computed

• Asks: Over rolling 3-year and 5-year windows, what is incremental operating margin (change in OP ÷ change in revenue), and how does it compare to the reported average margin?
• Good: Both computed numbers with figure-table row IDs and a one-line reading: incremental above average means scale is working; below it means growth is being bought. If revenue shrank, the decremental margin is computed instead and read for fixed-cost downside exposure.
• Rationale: The average margin describes the past; the incremental margin describes what the next yen of revenue is worth, which is the thing a growth thesis actually buys. The two routinely diverge, and computing rather than asserting this is the cheapest test of whether growth deserves a multiple.

C12. Return on incremental invested capital

• Asks: Over the past 5 years, what is ROIIC — change in after-tax operating profit divided by cumulative reinvested capital (capex above depreciation, working-capital build, acquisitions net of divestitures) — and does it clear a stated cost-of-capital hurdle?
• Good: The computation shown with figure-table inputs, a stated hurdle, and a plain verdict: reinvestment compounds value, treads water, or destroys it. If the number is unstable (one acquisition dominates), windows are varied and the instability reported rather than smoothed.
• Rationale: Growth creates value only when the return on the capital funding it exceeds that capital's cost; otherwise retained earnings convert into less than face value. Many optically cheap Japanese companies are cheap precisely because ROIIC is poor — this single computed number separates a compounder from a value trap better than any moat discussion.

C13. Maintenance versus growth capex

• Asks: Split reported capex into maintenance and growth — anchoring maintenance on depreciation adjusted for asset-price inflation and asset age, cross-checked against any management disclosure — and recompute owner earnings on the maintenance figure.
• Good: A stated maintenance-capex estimate with its method shown and owner earnings recomputed on it. Treating all capex as growth, or all as maintenance, without argument fails.
• Rationale: Reported free cash flow conflates the cost of standing still with the cost of getting bigger. The owner's true yield depends on this split, so it must be estimated, not skipped.

C14. Working-capital intensity of growth

• Asks: What is the change in net working capital per yen of revenue change over the last 5 years, and what is the cash conversion cycle (DSO + DIO − DPO) trend?
• Good: Both numbers from the figure table plus a consequence statement: at the assumed growth rate, working capital absorbs ¥X per year of operating cash flow. A lengthening cycle during flat revenue is flagged separately, since it means the same business now needs more cash.
• Rationale: Revenue growth is not cash growth when each sale must first be financed through inventory and receivables. The intensity ratio converts a growth assumption directly into a cash cost, and the trend detects quiet deterioration in bargaining power that the P&L hides.

C15. Unit-level economics reconstructed

• Asks: From segment data, facility/store counts, subscriber counts, or per-employee disclosures, reconstruct one unit's economics: revenue per unit, contribution per unit, cost to add a unit, implied payback period.
• Good: An explicit per-unit table with each input traced to a filing figure. If the filings do not disclose enough to reconstruct any unit, that is stated as a limit rather than papered over with consolidated averages.
• Rationale: Consolidated statements average the mature and the immature into a blur; the decision-relevant question is whether the marginal unit earns its capital back. When the reconstruction is impossible, knowing that is itself material — the growth story cannot be audited.

C16. Operating leverage decomposed

• Asks: Using years in which revenue moved in different directions, decompose the cost base into fixed and variable components, and compute the revenue decline at which operating profit reaches zero.
• Good: A fixed/variable split estimated from at least one up-year and one down-year, and the computed breakeven revenue as a percentage of current revenue (e.g., "OP hits zero at roughly −18% revenue"), carried into the downside valuation rather than left as trivia.
• Rationale: Operating leverage is symmetric but theses only ever cite its upside; the same fixed costs that expanded margins in recovery set the floor through which profits fall next downturn. A "conservative" scenario milder than an already-observed drawdown is not conservative.

C17. Inflation pass-through arithmetic

• Asks: During the most recent input-cost spike, what happened to gross margin quarter by quarter — how much cost increase was recovered through price, at what lag, and did percentage margin or only yen margin recover?
• Good: A short series showing cost-of-sales ratio through the spike and recovery with a computed pass-through statement (fraction recovered, lag in quarters), distinguishing yen-per-unit recovery from percentage-margin recovery.
• Rationale: Pricing power is endlessly asserted and rarely measured; the recent cost shock gave nearly every Japanese company a natural experiment whose results sit in its quarterly filings. Pass-through fraction and lag are the measurable form of the moat claim.

C18. Labor cost versus productivity

• Asks: From the securities report's headcount and average-salary disclosures, compute revenue, operating profit, and personnel cost per employee over 5 years. Is shunto-era wage growth outrunning productivity growth?
• Good: The three per-employee series with figure-table references and the comparison stated: productivity above wage growth means labor inflation is absorbed; the reverse means margin erosion is structural until proven otherwise. Staffing-mix shifts noted where disclosed.
• Rationale: Japan has moved from three decades of wage stagnation to sustained increases, converting labor from a fixed assumption into a live margin variable. The mandatory headcount and salary disclosures make this one of the few cost lines auditable per unit from public documents.

C19. Segment cross-subsidy map

• Asks: For each reported segment, what is operating profit against allocated segment assets — which segments earn above a cost-of-capital hurdle, which destroy value, and how much of the good segments' cash flow do the bad ones consume?
• Good: A per-segment return table from the segment footnote, a named hurdle, and a quantified subsidy statement, with corporate/elimination lines allocated or explicitly noted as unallocated.
• Rationale: A consolidated multiple prices the average of a portfolio, but value lives at the segment level: a great business stapled to a chronic loser is worth the great business minus the subsidy. The decomposition also identifies the highest-value capital-allocation decision available to management, against which their behavior can later be graded.

C20. Steady-state margin behind growth spend

• Asks: If the company claims margins are depressed by growth investment, estimate the steady-state operating margin with growth spending cut to maintenance level — and verify the claimed spend actually appears as identifiable line items (R&D, advertising, hiring ahead of revenue).
• Good: A bridge from reported to steady-state margin with each adjustment tied to a disclosed line and not double-counted. If no growth spend is separable from the filings, the reported margin is the steady state and the "investment mode" narrative is rejected.
• Rationale: "Margins are depressed because we are investing" is either a computable claim or an excuse, and filings usually disclose enough to test it. Running the arithmetic in both directions prevents paying twice — once for the growth and once for the margin recovery it supposedly suppresses.

C21. Accrual gap over a decade

• Asks: Sum operating cash flow and net income over the past 5–10 years — what is the cumulative gap, and which balance-sheet lines (receivables, inventory, capitalized development, provisions) account for it?
• Good: The two cumulative totals with the gap decomposed into its largest drivers, each with figure-table references. Cash conversion persistently below ~90% requires a named benign explanation or it flags earnings quality directly in the verdict.
• Rationale: Over one year income and cash legitimately diverge; over a decade they must reconcile, because accruals are only ever timing. A persistent cumulative gap is the most reliable filings-level symptom of aggressive accounting or deteriorating economics.

C22. Organic growth isolated

• Asks: Decompose reported revenue growth into volume, price, FX translation, and consolidation-scope changes. What is the organic constant-currency growth rate?
• Good: A bridge for the last 3 years (e.g., "reported +9%: FX +4%, acquired +3%, organic +2%") with each component sourced, and the thesis's growth assumption compared against the organic number, not the reported one.
• Rationale: A weak yen and a couple of acquisitions can manufacture years of headline growth from a stagnant underlying business, and the multiple silently capitalizes the headline. The components are separable from standard Japanese disclosures, which makes not separating them a choice to be deceived.

Balance-sheet reality

C23. Debt maturity ladder versus liquidity

• Asks: From the borrowings footnote, what does the year-by-year debt maturity schedule look like against unrestricted cash, committed lines, and normalized free cash flow — is there a repayment wall operations cannot cover?
• Good: A maturities-versus-sources table from the footnotes, the largest single-year tower identified, and a stress sentence on whether the company survives a year in which rollover is unavailable. Covenants and pledged assets noted where disclosed.
• Rationale: Solvency is not a ratio; it is a calendar — companies fail at specific maturity dates, not at average leverage levels. With Japanese rates now positive, the assumption that banks roll everything at negligible cost is a claim requiring inspection.

C24. Deployable versus trapped cash

• Asks: Of reported cash and deposits, how much is genuinely deployable to shareholders after subtracting operating cash needs, overseas-subsidiary repatriation friction, restricted deposits, JV cash, and seasonal working-capital swings?
• Good: A waterfall from gross cash to deployable cash with each haircut stated and justified. The deployable figure — not gross net cash — feeds every subsequent valuation item.
• Rationale: Net-cash screens count every yen at face value, but a cash pile that exists to run the business is working capital wearing a disguise. Japanese balance sheets, with decades of accumulated deposits, are where this distinction changes verdicts most often.

C25. Securities portfolio marked and taxed

• Asks: Mark the investment-securities portfolio to market from the footnote, compute the tax due on realization (~30%), and state after-tax realizable value as a percentage of market cap.
• Good: The computed after-tax figure with the largest holdings named, plus the 3-year trend in portfolio book value (shrinking = actual selling; static = a plan without behavior). Value counted only at after-tax realizable value, weighted by the demonstrated disposal trend.
• Rationale: Cross-held securities are Japan's most visible "hidden asset" and the most routinely overcounted: analysts add gross market value while the shares sit unsold, untaxed, and voting for management. After-tax value discounted by demonstrated willingness to sell is the honest number.

C26. Pension position as debt

• Asks: From the retirement-benefit footnote, what is the funded status, what discount-rate and expected-return assumptions produce it, and are those assumptions defensible against current JGB yields?
• Good: The net position computed and treated as debt in enterprise value, the discount rate sanity-checked against the current curve, unrecognized actuarial losses noted, and materiality relative to market cap stated even when the answer is "immaterial."
• Rationale: A defined-benefit deficit is borrowed money owed to employees, and its reported size is manufactured by assumptions the company chooses itself. Decades of near-zero JGB yields let Japanese sponsors carry optimistic discount rates, so the one-line audit against the current curve has real valuation consequences.

C27. Leases and off-balance-sheet claims

• Asks: For JGAAP filers, capitalize the operating-lease footnote's future minimum payments into debt; for all filers, sum purchase commitments and guarantees from the commitments footnotes. How much does enterprise value change?
• Good: A single adjusted-EV computation with each addition itemized and sourced, lease-adjusted leverage compared to unadjusted, and the lease adjustment applied to margin comparisons against IFRS peers where relevant.
• Rationale: JGAAP still permits operating leases off the balance sheet, so identical retailers under JGAAP and IFRS report structurally different debt and EBITDA — an accounting artifact that screens misread as a valuation difference. Fixed obligations are debt regardless of the standard that classifies them.

C28. Real assets at disclosed fair value

• Asks: What does the investment-and-rental-property footnote disclose as fair value versus carrying value, and does the facilities table suggest material unrealized land value — and is there any realization path?
• Good: The fair-value gap computed where disclosed, major properties identified, and a two-part statement: the size of the mark, and whether any mechanism (announced sale, sale-leaseback history, activist pressure) plausibly converts it to cash. Value with no path is reported but excluded from the buy-below computation.
• Rationale: Japanese filings uniquely require fair-value disclosure for rental property, letting an outsider approximate a real-estate mark from public data. But land a company will never sell has an owner yield of zero; the mark matters only multiplied by the probability of realization.

C29. Full dilution accounting

• Asks: Compute fully diluted shares — options with strikes, convertibles with conversion prices, restricted and performance shares, minus treasury — and check whether treasury stock is cancelled or held for reissuance.
• Good: A reconciled bridge from basic to fully diluted count with each instrument sourced, the 5-year net trend (buybacks net of issuance), and all per-share figures in the study restated on the diluted count. Treasury retained "for strategic purposes" flagged as potential re-dilution.
• Rationale: Every per-share claim is a fraction whose denominator is quietly contested, and buybacks that merely offset stock-comp issuance return nothing while consuming real cash. The bridge is mechanical and the error from skipping it compounds through every subsequent computation.

C30. Provisions and allowance integrity

• Asks: Compare the doubtful-accounts allowance against receivables aging, inventory against revenue growth (with write-down history), and warranty/loss provisions against subsequent utilization. Are balance-sheet estimates absorbing reality or deferring it?
• Good: Three small ratio series each with a one-line reading. Inventory growing structurally faster than revenue without write-downs, or an allowance ratio falling as receivables age, carries into the earnings-quality assessment with a yen estimate of the deferred cost.
• Rationale: Provisions are management's estimates of future bad news, and softening them is the quietest way to manufacture current profit. Trends in these ratios detect deferred losses one to three years before they surface in the income statement.

C31. Minority-interest leakage

• Asks: What share of consolidated net income and equity is attributable to non-controlling interests, which subsidiaries drive it, and do consolidated multiples silently price cash flows that legally belong to minorities?
• Good: The NCI share of profit and equity computed for 3 years, major partially-owned subsidiaries named, and the valuation restated on parent-attributable figures — including the look-through multiple where a crown-jewel subsidiary is partially owned.
• Rationale: Consolidation counts 100% of a subsidiary's profit even when the parent owns 60%, so consolidated multiples overstate what the shareholder buys. Japanese groups are dense with partially-owned subsidiaries, making the leakage routinely material and the attributable-basis restatement a one-division fix.

C32. Listed affiliates at market

• Asks: For equity-method affiliates and consolidated subsidiaries that are themselves listed, what is carrying value versus market value of the stake, after tax — and how has the company historically treated these stakes?
• Good: A stake-by-stake table (ticker, ownership, book, market, after-tax mark) with the adjusted net-asset figure, plus any decade-long record of stake sales or buy-ins. The mark included at after-tax value weighted by that demonstrated behavior.
• Rationale: A listed affiliate has a public price, making its book-to-market gap the rare hidden asset requiring no estimation — only a tax adjustment and a realism check. Pricing the stake by observed behavior rather than theoretical value is what separates asset analysis from asset tourism.

C33. Receivables and factoring off-ramp

• Asks: Do the footnotes disclose transferred or factored receivables, securitized assets, or discounted notes — and how much reported operating cash flow comes from moving assets off the books rather than from collections?
• Good: Disclosed transfer balances quantified against reported OCF with the 3-year trend; growing reliance on receivables sales to hold DSO flat is restated as the borrowing it economically is. "No such disclosures found" is acceptable when the search is documented.
• Rationale: Selling receivables converts future collections into today's operating cash flow while flattering exactly the metrics (OCF, DSO) used to verify earnings quality. It is economically secured borrowing booked as a sale, and the check costs minutes.

Japan structure & behavior

C34. Effective float and allegiant block

• Asks: From the shareholder register, sum the stakes of cross-holders, allied banks and insurers, business partners, founding family, the ESOP, and treasury shares. What percentage of votes is effectively pledged to management, and what is the true free float?
• Good: A named-holder table summing to an allegiant-block percentage, the residual effective float, and the consequence drawn explicitly: whether any outside proposal (activist campaign, tender, AGM vote) can mathematically succeed without management's consent. Compared against the TSE tradable-share ratio, which overstates contestability.
• Rationale: Governance change in Japan happens through votes, and the register determines before any campaign begins whether votes are winnable. A 45% allegiant block means any unlock thesis depends entirely on management's voluntary conversion — a different probability than a contestable register, computable from a public table.

C35. Policy cross-holdings: policy versus disposals

• Asks: How many policy cross-shareholdings (seisaku hoyū kabushiki) does the company hold, at what value and share of net assets — and does the stated reduction policy contain numbers and deadlines matched by an actual 5-year record of sales, with proceeds traced?
• Good: Count, value, and percentage computed; the policy classified numeric or platitude; the disposal record from the named-holdings table compared against the policy; and sale proceeds traced to their destination (returned, reinvested, or accumulated). Behavior graded above words.
• Rationale: Cross-holdings are capital earning near-zero returns while voting to protect the status quo, and disclosure rules make the gap between stated policy and actual sales measurable rather than speculative. Follow-through here predicts follow-through on the company's other governance commitments.

C36. Parent and controlling-shareholder structure priced

• Asks: Is there a controlling or dominant (>33%) shareholder? If so, what related-party transactions flow across that boundary on what terms, how did minorities fare in past group transactions (TOBs, asset transfers, shared-service fees), and what premiums did comparable buy-ins or sell-downs pay?
• Good: The stake and board seats documented, transaction terms quoted from the footnotes, past conduct assembled, and a stated scenario range — thin-premium squeeze-out, status quo forever, or genuine independence — with the thesis naming which scenario it pays for and why the register supports it.
• Rationale: A listed subsidiary's minority shareholders own an asset whose exit price is set by a counterparty with control and superior information, and past boundary crossings are the only data on future ones. The discount on such companies is often a rational fee for structural disadvantage, not a mispricing.

C37. Takeover-defense inventory

• Asks: Does the company maintain a poison pill or other defense (renewed at recent AGMs?), unequal voting arrangements, or golden-share-like provisions — and combined with the allegiant-block arithmetic, is a change of control possible at any price without board consent?
• Good: A defense-by-defense inventory from the governance report and AGM notices, and a single yes/no conclusion on contestability with its arithmetic. Recent pill abolition under proxy-advisor pressure noted as a directional signal.
• Rationale: The value of balance-sheet assets to a minority shareholder is conditional on someone, someday, being able to acquire control or credibly threaten to. If defenses plus the register make that probability effectively zero, the correct weight on unlock scenarios is also near zero, however cheap the assets.

C38. TSE capital-cost disclosure graded on numbers

• Asks: Has the company published its response to the TSE's capital-cost-and-stock-price-conscious-management request — and does it contain a specific cost-of-capital figure, an honest P/B diagnosis, and dated actions, or slogans?
• Good: The disclosure located (or its absence noted despite P/B<1 — itself a data point), its cost-of-capital number compared against an independently computed one, a pass/platitude grade with the load-bearing quote, and prior-year commitments checked against actual execution.
• Rationale: The TSE request created a natural experiment: every low-P/B company faced the same prompt at the same time, so response quality is directly comparable across firms. A company that answered a regulator's direct question about its own undervaluation with platitudes has revealed how seriously it takes the problem an investor is underwriting.

C39. Activist register check

• Asks: What do EDINET large-shareholding (5%) reports show — which activist or engagement funds have filed, when, with what stated purpose, and what do amendments show about accumulation or exit? What happened to prior campaigns here?
• Good: A filing timeline plus the outcome record of any past engagement — proposals made, management response, whether the activist exited at a gain or gave up. "No 5% filings exist" is a complete answer that recalibrates any unlock thesis toward voluntary management action only.
• Rationale: An unlock thesis needs an unlocking agent, and the 5% regime makes the presence or absence of one a public fact rather than a hope. Past campaign outcomes at the same company are the best available predictor of what the next campaign achieves against this specific register.

C40. Distributable capacity versus stated policy

• Asks: Compute realistic shareholder-return capacity — distributable reserves, deployable net cash (C24), normalized FCF — against the stated dividend/buyback policy and the actual 5-year payout. What fraction of capacity is being used?
• Good: Capacity stock, annual capacity flow, actual payout, and the utilization ratio computed, with the capacity-versus-policy gap stated as the size of the prize an unlock would deliver — or the finding that the generous-looking policy already exhausts true capacity.
• Rationale: In most markets payout is constrained by cash generation; in Japan it is frequently constrained by policy alone, making the capacity-policy gap the quantified upside of any governance-change thesis. The number converts "they could return so much more" from a mood into a yen figure with a funding source.

C41. Dividend policy design consequences

• Asks: What is the stated dividend policy type (payout ratio, DOE, stable/progressive), and what does its formula imply on this balance sheet — does a DOE or stable-dividend policy on an overcapitalized base mathematically lock yield below capacity regardless of earnings?
• Good: The policy quoted, its implied dividend computed under flat / −30% / +30% earnings scenarios, and the design consequence stated plainly, with any recent policy change noted with its direction.
• Rationale: A dividend policy is a formula with computable consequences that marketing language obscures — DOE sounds shareholder-friendly while quietly indexing payout to the very balance-sheet bloat that suppresses returns. Reading the formula's implications is analysis; reading its label is not.

C42. AGM signal response

• Asks: What do published AGM voting results show — are approval rates for top-management elections trending down, have shareholder proposals appeared, and did management respond to weak votes with substantive change or silence?
• Good: Director approval percentages across the past three AGMs, any sub-80% result or shareholder proposal noted, and the following year's governance or policy changes checked for a response.
• Rationale: AGM votes are the one recurring, quantified referendum minorities hold, and the trend is more informative than the level because Japanese approval rates start high. Whether management treats a deteriorating vote as information or noise is a direct observation of its accountability loop.

C43. Comply-or-explain quality

• Asks: Which Corporate Governance Code principles does the company not comply with, and are the explanations reasoned and company-specific or interchangeable boilerplate?
• Good: The non-complied principles listed, each explanation classified substantive (names specific circumstances and an alternative mechanism) or template (pasteable into any report), and the ratio weighed as a disclosure-sincerity signal.
• Rationale: The comply-or-explain design deliberately leaves a free-text channel, and free text is where sincerity becomes measurable — a reasoned explanation costs thought, a template costs nothing. How a company uses its cheapest disclosure obligation calibrates how to read its expensive ones.

C44. Minority communication in practice

• Asks: Are results-briefing materials and Q&A records published, and when hard questions appear in them, are the answers substantive or deflective?
• Good: An inventory of what exists with at least one hard question traced to its answer and judged substantive or evasive; a company publishing no Q&A record at all is scored on that absence.
• Rationale: Formal filings are compelled; briefing behavior is chosen, so it reveals how management regards outside owners when disclosure is voluntary. The treatment of hard questions is the observable proxy for how minorities will be treated when interests actually conflict.

C45. English disclosure effort

• Asks: What is the actual scope and lag of English disclosures — tanshin only, or also the securities report, governance report, and briefing materials — relative to the TSE requirement and to sector peers?
• Good: An inventory of which documents exist in English and the typical lag in days behind the Japanese originals, positioned against the Prime-market requirement and two named peers.
• Rationale: English disclosure is a cost borne solely for the benefit of outside capital the company does not control, making it a clean revealed-preference signal about openness to scrutiny. The lag matters as much as the scope: translation timed after the price has moved serves ceremony, not investors.

C46. MD&A candor about failures

• Asks: In the past three years of MD&A and president's messages, does management ever name a failure as its own — decision identified, cost quantified, remedy stated — or are all setbacks externalized in passive voice?
• Good: At least one located instance of owned failure with citation, or a documented finding that three years of setbacks were uniformly externalized, scored as a candor negative.
• Rationale: Every multi-year record contains failures, so their absence from the narrative measures the narrative's honesty, not the company's perfection. Management that cannot name its own errors publicly is also not running the internal feedback loops that error correction requires.

C47. Dilutive financing history

• Asks: Does the ten-year financing history include instruments that transferred value from existing minorities — discounted third-party allotments, moving-strike convertibles, repeated issuance below book — and under what circumstances were they used?
• Good: All equity and equity-linked financings listed with terms, discount to prevailing price, and allottees, each assessed for minority dilution; a clean record explicitly confirmed rather than assumed from silence.
• Rationale: How a company raised money when it needed it reveals whose interests bind when cash is tight, which no policy statement can. This history directly predicts what happens to per-share value in the next stress.

C48. Liquidity as position arithmetic

• Asks: What is average daily trading value over the past 6–12 months, and how many days would building and exiting a meaningful position take at 20% of volume? Does the tradable-share ratio sit near listing-maintenance thresholds?
• Good: The ADV figure, days-to-build and days-to-exit at a stated position size (exit also run at half average volume), and a listing-criteria note where relevant — a company near maintenance lines has a structural incentive to act.
• Rationale: A position that takes forty days of patient selling to exit has a different risk profile than its multiple suggests, and the discount may be rational compensation for illiquidity. Listing-maintenance pressure is a structural catalyst hiding in exchange arithmetic.

C49. Index-flow mechanics

• Asks: Where does the company sit relative to mechanical index thresholds — TOPIX tradable-cap reweighting, MSCI bands, Nikkei 225 membership — and is there identifiable forced buying or selling at current prices?
• Good: Float-adjusted market cap set against the published thresholds with the direction of mechanical flow stated; if no threshold is nearby, one sentence saying so suffices.
• Rationale: Passive flows are price-insensitive by construction, so a company crossing an index threshold experiences buying or selling unrelated to value — either a headwind to endure or a supply of mispriced shares. The thresholds are published rules, making this one of the few forecastable order flows in markets.

Governance & management behavior

C50. Capital allocation: revealed versus stated

• Asks: Quote the stated capital-allocation policy and compare it against the actual ten-year deployment of operating cash flow across capex, M&A, dividends, buybacks, and cash accumulation. Do revealed preferences match stated ones?
• Good: The policy quoted with source, a ten-year deployment-share table from figure-table rows, and divergence quantified (e.g., "policy says 'proactive shareholder returns'; 60% of FCF accumulated as cash over the decade").
• Rationale: Capital allocation is the one domain where management's actions are fully recorded in audited statements, making stated-versus-revealed comparison exact rather than impressionistic. The revealed preference is the base rate for what happens to the next decade of cash flows an owner is buying.

C51. Buyback execution record

• Asks: For each buyback program announced in the past five years: authorized versus actually repurchased, over what period, at what average price relative to announcement day — and were the shares cancelled or parked in treasury?
• Good: A program-by-program table from timely disclosures with the aggregate completion rate and the 5-year net change in shares outstanding after stock-comp issuance. Announcement-heavy, execution-light programs graded as signaling, not capital return.
• Rationale: A buyback announcement is free; execution costs money, which is why the two diverge, and Japanese companies have a documented habit of letting authorizations lapse partially used. Net share count is the only line that cannot be gamed — it either fell or it did not.

C52. Metric definition drift

• Asks: Have the definitions of headline self-reported metrics (adjusted margin, ROIC formula, KPI denominators) changed across the past five years — and if so, was the change disclosed and prior years restated?
• Good: Each headline metric's definition quoted from filings at two or three points in time and diffed; any change flagged with whether it was announced, whether comparatives were restated, and whether it happened to flatter the trend.
• Rationale: A metric whose definition the measured party can quietly change is not a measurement; it is a communication. Definition drift that consistently flatters is a direct read on whether management treats disclosure as information or marketing.

C53. Miss acknowledgment in the new plan

• Asks: When the current mid-term plan was announced, did management explicitly reconcile against the previous plan's targets — naming the misses and diagnosing them — or did the new plan begin from a clean slate?
• Good: A citation showing an explicit prior-plan post-mortem, or a documented finding that no reconciliation exists anywhere in the announcement materials.
• Rationale: Acknowledging a miss costs face and is therefore expensive signaling; only management that actually runs a feedback loop pays it. Clean-slate plans indicate targets are decoration for announcement day, not commitments anyone inside expects to be scored on.

C54. Initiative persistence trace

• Asks: Pick three named strategic initiatives from the president's message or mid-term plan of three years ago — does each still appear in current materials as completed, in-progress with metrics, or vanished without mention?
• Good: Three initiatives traced by name across three years of communications, each classified done / progressing / silently-dropped, with citations at both ends. Two or more silent drops scores as a commitment-durability failure.
• Rationale: Strategy is only real if it survives the fiscal year it was announced in. Tracing named initiatives forward is a mechanical test of whether communications describe an actual multi-year program or an annually refreshed set of fashionable words.

C55. Related-party transaction pattern

• Asks: What does the related-party note show across the past five years — recurring transactions with founder-family or director-linked entities, growing balances, or terms a third party would not accept?
• Good: A five-year trend of transaction types, counterparties, and amounts, each material or growing item assessed for arm's-length plausibility; "none disclosed" verified against the note itself, not assumed.
• Rationale: Related-party dealings are the channel through which value leaves minority shareholders without ever appearing as a loss. The pattern over years matters more than any single year, because extraction that works tends to grow.

C56. Compensation-performance linkage

• Asks: Does disclosed executive compensation actually vary with results — comparing total director compensation and its structure across the company's best and worst recent years — or is it flat regardless of outcomes?
• Good: Compensation figures for at least one good and one bad year with the variable share's actual movement computed, set against the disclosed comp policy's language.
• Rationale: Incentives predict behavior better than statements do, and Japanese filings disclose enough to test the linkage empirically. Compensation that ignores results tells you the board's real principal is not the shareholder.

C57. Officer turnover clustering

• Asks: Do departures of the CFO, accounting officers, statutory auditors (kansayaku), or audit-committee members cluster within a year of restatements, large guidance misses, or impairments?
• Good: A timeline of financial-oversight personnel changes overlaid on the event timeline, clusters flagged with the disclosed departure reasons quoted.
• Rationale: The people closest to the numbers exit first when the numbers are wrong, because they bear personal liability and have the earliest knowledge. Turnover clustering is publicly reconstructible and hard to launder, unlike departure narratives.

C58. Auditor timeline

• Asks: Has the audit firm changed in the past ten years — from whom to whom, with what stated reason, and did the change fall near any accounting dispute, restatement, or qualified language?
• Good: A ten-year timeline; for any change, the disclosed reason quoted, the direction assessed (comparable rotation versus downgrade to a smaller firm), and proximity to any accounting event within ±1 year explicitly checked.
• Rationale: The auditor is the only party inside the reporting process paid to disagree with management, so a change in that relationship is informative even when the stated reason is bland. A caliber downgrade coinciding with accounting stress is one of the most reliable public warning patterns that exists.

C59. Key audit matter evolution

• Asks: What do the Key Audit Matters identify as the hardest judgment areas, how have they changed over the available years, and do any map onto risks the thesis depends on?
• Good: The KAM list per year with arrivals and departures noted, and each current KAM mapped to a thesis claim or marked orthogonal.
• Rationale: KAMs are a professional skeptic's ranked list of where this specific company's numbers most depend on judgment, published for free. Ignoring them means re-deriving from scratch what the auditor already flagged, and probably missing some of it.

C60. Disclosure quality versus sector peers

• Asks: Compared with two or three same-sector peers, how granular is this company's disclosure — segment count relative to complexity, KPI depth, note detail, briefing materials — and where is it conspicuously thinner?
• Good: A concrete comparison on at least three named dimensions, with each below-peer gap assessed for whether it hides something the thesis needs.
• Rationale: Disclosure granularity is a choice, and the peer set establishes what is feasible, stripping away the "industry practice" excuse. Systematically thinner disclosure means management is exercising an option to be opaque, and opacity is rarely exercised to hide good news.

Disconfirmation

C61. Strongest short case

• Asks: Using only these filings and the figure table, construct the most persuasive short/avoid case a competent skeptic could build — what three facts would they lead with?
• Good: Three specific cited facts anchored to figure-table rows (e.g., "customer concentration rose from 22% to 41% in three years") — not generic risks copied from the risk-factors section — followed by an honest assessment of whether the long thesis answers each fact or merely talks past it.
• Rationale: Confirmation bias is strongest exactly when a thesis is attractive, and the only reliable countermeasure is arguing the other side with full effort against the company's own numbers. If the strongest constructible short case cannot be answered from the documents, that is signal, not noise.

C62. Disappeared disclosures

• Asks: Diff the latest annual securities report against the versions from one and three years prior — which KPIs, segment splits, operational metrics, or narrative sections quietly disappeared?
• Good: An explicit list of dropped items, each explained by a disclosed reason or flagged as a deterioration-hiding candidate. "Nothing disappeared" is valid only if the diff was actually performed.
• Rationale: Companies rarely announce that a metric turned unfavorable; they stop reporting it. The absence of previously volunteered information is one of the few disclosures management makes involuntarily, so it carries more signal per word than anything they chose to say.

C63. Vanished major-customer disclosure

• Asks: Track the major-counterparty disclosure (customers over 10% of revenue) across the past five annual reports — did any named customer drop below threshold or vanish, and does the narrative anywhere acknowledge it?
• Good: A year-by-year table of disclosed major customers with revenue shares, any disappearance matched against segment revenue changes and MD&A commentary. Silence about a lost 10%+ customer is recorded as a candor failure.
• Rationale: Losing a customer worth more than a tenth of revenue is among the most material single events a company can experience, and the threshold disclosure makes it verifiable even when management says nothing. The gap between the event and the narrative measures how much the narrative can be trusted elsewhere.

C64. Revision timing asymmetry

• Asks: When guidance was revised mid-year over the past five years, in which direction and which half did revisions cluster — and were downward revisions disclosed promptly or bundled into results announcements?
• Good: A list of all guidance revisions with date, direction, and magnitude, plus a pattern statement (e.g., "downward revisions always arrive with Q3 results, never standalone"). Late-clustered downward revisions scored as a disclosure-behavior negative.
• Rationale: The timing of bad news is a management choice, and choices repeat. A company that structurally delays downward revisions until they can be buried is telling you exactly how it will handle the next piece of bad news you are underwriting.

C65. Segment deterioration masked by consolidation

• Asks: Compute five-year revenue and margin trends for each reported segment — is any segment in decline whose deterioration is invisible at the consolidated line because another segment offsets it?
• Good: A per-segment trend series from figure-table rows, any declining segment named with its consolidated footprint quantified, and the thesis addressing whether the decline is structural or cyclical.
• Rationale: Consolidated figures are averages, and averages are where problems hide. A rotting segment inside a healthy consolidated print is deferred bad news, not absent bad news.

C66. Segment reclassification coincidence

• Asks: How many times in the past ten years did the company change its segment definitions, and did any change coincide with deterioration in a segment the new structure made untrackable?
• Good: A timeline of segment-structure changes with before/after mappings, noting whether restated comparatives were provided and whether the last pre-change trend of any absorbed segment was negative. Changes that dissolve deteriorating segments are flagged.
• Rationale: Segment redefinition is legitimate when the business changes, but it also resets every trend line an outsider can draw. When the reset arrives precisely as a trend turns ugly, the burden of proof shifts to management, and the checklist enforces that shift.

C67. Customer-side corroboration

• Asks: For the company's stated demand drivers, does at least one public customer or end-market source (a listed customer's filings, industry shipment statistics, government statistics) corroborate or contradict the story?
• Good: At least one independent public data point checked and cited (e.g., "largest disclosed customer's own capex guidance is down 20%, contradicting the 'expanding orders' narrative"). A documented failed search is acceptable; not looking is not.
• Rationale: Everything inside a company's own filings shares one author with one incentive. A demand claim that survives contact with a second, differently-incentivized public source is categorically stronger evidence than the same claim repeated across ten of the company's own documents.

C68. Supplier and input-side corroboration

• Asks: Do the company's claims about input costs, procurement, or capacity match publicly observable supplier-side data — listed suppliers' filings, commodity price series, industry capacity statistics?
• Good: The one or two most margin-relevant input claims identified and each checked against an independent public series; contradictions carried into the epistemics items as load-bearing unknowns.
• Rationale: Cost attributions in MD&A are the easiest place to launder execution failure as external misfortune, because few readers check the input side. Public commodity and supplier data make many attributions cheaply falsifiable, so skipping the check is choosing not to know.

C69. Internal contradiction hunt

• Asks: Search the current-year filings for at least one internal contradiction — MD&A tone versus segment figures, tanshin outlook versus securities-report risk factors, IR deck versus audited notes.
• Good: One concrete contradiction cited on both sides and either resolved or carried as an open flag — or a documented statement that the cross-reading was performed across named document pairs and none was found.
• Rationale: A company's disclosure set is written by different hands under different liability regimes, so the audited documents constrain what the promotional documents can honestly say. Contradictions between them reveal which claims are aspiration rather than fact.

C70. Chronic one-offs

• Asks: Count the special items (tokubetsu son'eki) reported over the past ten years — how many years were "one-time" losses booked, and what does average profit look like with them treated as recurring?
• Good: A ten-year count and sum of special gains and losses from figure-table rows, average profit recomputed treating any item type appearing in three or more years as operating in nature, and the resulting hit to headline profit quality stated.
• Rationale: An item that occurs most years is not extraordinary regardless of its label; the label is a classification choice made by the party being measured. Recasting chronic one-offs as recurring is the simplest possible test of whether reported profitability is the real profitability.

C71. Risk-factor diff

• Asks: Diff the risk-factor section across the past three annual reports — which risks were added, deleted, or reworded, and does any change align with something the thesis relies on?
• Good: A short added/deleted/reworded list with years, each change interpreted, and boilerplate churn distinguished from substantive change.
• Rationale: Risk factors are written by people with legal exposure for omissions, which makes additions a rare channel of involuntary candor about what management actually fears. Deletions are equally informative and even less advertised.

Valuation & downside

C72. Bear case valued first

• Asks: Before any base case is written, construct the downside valuation: margins revert to the 10-year low, the operating-leverage breakeven (C16) applies to a revenue decline already observed in the company's own history, no unlock arrives, trapped assets stay trapped. What is per-share value and the implied loss from here?
• Good: A completed downside valuation with each assumption tied to an already-observed historical fact, the implied loss percentage, and the price at which the downside case alone produces roughly zero loss — written and dated before the base case, with the ordering visible in the study artifacts.
• Rationale: The first case an analyst builds anchors every subsequent one, so building the optimistic case first guarantees the "conservative" case is an optimistic case with haircuts. A downside constructed from the company's own observed history cannot be accused of pessimism, and its zero-loss price is the natural raw input for the buy-below figure.

C73. Reverse-DCF on the current price

• Asks: At a stated discount rate, what combination of revenue growth and steady-state margin does the current price imply over the next 10 years — and is that implied path harder or easier to beat than the reference-class base rate?
• Good: The implied assumptions solved for and written as a falsifiable sentence ("the price embeds 6% growth for a decade at peak-decile margins"), followed by an explicit verdict against the base-rate items (C2, C7, C8, C5).
• Rationale: A forward DCF launders the analyst's optimism through assumptions; a reverse DCF extracts the market's assumptions and puts the burden of proof on whoever disagrees with the price. It converts "undervalued" from an adjective into a specific disagreement with a specific implied forecast.

C74. Private-owner yield, multiple-free

• Asks: If this stock had no quoted price, would you own it at the implied private-owner yield — normalized owner earnings (after the C13 maintenance-capex and C21 accrual corrections) divided by market cap, plus deployable net cash (C24) and after-tax marked securities (C25) — against a JGB-plus-spread hurdle?
• Good: One equation with every input traceable to a prior item's computation, a stated required return with one line of justification, and a plain yes/no: an ownership-minded buyer takes the deal or doesn't. No terminal multiples, no peer comps, no "re-rating" term anywhere in the arithmetic. When normalization is impossible (the C13 reconstruction failed, capex absent), report the raw net-income yield with an explicit flag and do not issue a buy-below on this lens alone.
• Rationale: Removing the price removes the strongest source of motivated reasoning — the hope that someone else pays more later — leaving the business as an income-producing asset, which is the thing actually being bought. If value only appears when a multiple is assumed, the thesis is a bet on sentiment wearing valuation clothes.

C75. Sum-of-parts without wishful summing

• Asks: For a multi-segment group: value each segment at what an actual industry acquirer pays, listed stakes at after-tax market value, subtract unallocated corporate costs capitalized as a permanent drag, and apply a holding-company discount unless a specific separation catalyst is named. What is the honest SOTP?
• Good: A parts table where corporate overhead appears as a negative part, listed stakes carry the tax haircut, the conglomerate discount is applied at a stated rate, and any claim to close the discount names its mechanism. An SOTP that sums gross optimistic parts fails.
• Rationale: Sum-of-parts is the most abuse-prone valuation format because every part invites its own optimistic multiple while the costs of the assembly are quietly dropped. The conglomerate discount is substantially a rational price for those costs plus the improbability of separation.

C76. Normalization ledger for owner yield

• Asks: List every adjustment made to reported earnings to reach "normalized" owner earnings — with the yen size and direction of each — and show the yield's sensitivity to reversing the two largest adjustments.
• Good: A signed adjustment ledger, the resulting normalized figure and yield, and the sensitivity statement. Adjustments that all point the same favorable direction are called out as a red flag on the analyst's own arithmetic.
• Rationale: "Normalized" is where motivated reasoning hides in quantitative clothing — each adjustment defensible while the ensemble quietly rebuilds the bull case. Sensitivity to the largest adjustments reveals whether the yield is a property of the business or of the normalization.

C77. Earnings power versus asset value

• Asks: Compare earnings-power value (normalized owner earnings capitalized at the required return, no growth) against reproduction value of the operating assets (adjusted book excluding surplus cash and securities). Which is larger, and what does the gap assert?
• Good: Both numbers computed from prior items' inputs and the gap interpreted: EPV below asset value is a management/structure problem with the gap as the prize for fixing it; EPV above asset value is a franchise claim that must be defended with competitive facts from the filings, not assumed.
• Rationale: This reconciliation is the cleanest quantitative test of whether "moat" talk is warranted — a genuine franchise shows up as earnings power exceeding what the assets alone justify. It also correctly reframes the abundant Japanese case of good assets earning poorly as a governance option rather than a compounding machine.

C78. Asset floor with realization test

• Asks: Compute a haircut liquidation value — receivables ~80%, inventory ~50%, securities after tax, property at fair value less friction, minus all liabilities including off-balance-sheet items — and answer separately: does any mechanism exist by which this floor could actually be reached?
• Good: The haircut NAV per share compared to price, and the second half answered honestly: named forces that could compel realization (contestable register, activist filings, listing-maintenance pressure, MBO economics) or the admission that the floor is theoretical because control is entrenched.
• Rationale: Margin-of-safety logic depends on assets being ultimately accessible to their owners; in Japan the arithmetic half of that test passes far more often than the enforcement half. The floor still has value as downside insurance even when unrealizable — but insurance and upside must not be conflated.

C79. Cycle-position honesty

• Asks: Where are current earnings relative to the company's own cycle — current margin against the 10-year average, revenue against trend — and what does the multiple look like restated on mid-cycle earnings? Is the stock cheap, or are earnings high?
• Good: The mid-cycle earnings figure computed, both multiples shown side by side, and the classification stated. Any "this time is structural" override supported with a filing-verifiable structural break, not asserted by convenience.
• Rationale: The most reliable way to overpay for a cyclical is to capitalize peak earnings at an average multiple — the low P/E at the top is the market correctly forecasting the cycle. Requiring evidence for the structural-change escape hatch keeps it from being free.

C80. FX dependence of normalized earnings

• Asks: From the FX-sensitivity disclosures and geographic segments, what fraction of operating profit is attributable to yen weakness, and what do normalized earnings look like at a stated long-run exchange rate rather than spot?
• Good: The sensitivity figure quoted or derived, earnings recomputed at the long-run rate with the rate choice justified in one line, and the private-owner yield (C74) restated on the FX-normalized figure. A thesis that survives only at current spot says so explicitly.
• Rationale: A large share of recent Japanese profit growth is the arithmetic of a historically weak yen — a macro position the analyst holds whether or not they meant to. Separating the currency from the business is required before any claim about earnings durability means anything.

C81. Paid to wait

• Asks: If the unlock or re-rating takes five to ten years or never arrives, what is the realized return from cash yield alone — current dividend plus demonstrated (not announced) buyback pace against the current price? What is the bear-case IRR including this yield?
• Good: The demonstrated total-yield figure from actual 5-year payout behavior, bear-case IRRs over 5- and 10-year horizons with no re-rating, and the plain conclusion: the position pays an acceptable return during an indefinite wait, or it depends on an event with an unknowable date.
• Rationale: Value traps do not usually destroy capital; they destroy time, converting a correct valuation call into a poor annualized return through denominator years. Requiring the wait itself to be paid makes the thesis robust to its own catalyst never arriving — which is the base-rate outcome for catalysts.

C82. Per-share value after all leakage

• Asks: Restate the final per-share value on the fully diluted count (C29), net of minority-interest leakage (C31) and the annualized stock-compensation run rate. How much does the headline value shrink with all leakage applied at once?
• Good: A single bridge — headline value → full dilution → NCI attribution → capitalized stock-comp drag — with the total shrinkage stated and carried into the buy-below figure. Shrinkage above ~10% triggers rechecking earlier items for a flattering denominator.
• Rationale: Dilution, minority leakage, and stock compensation are each individually small enough to wave off and jointly large enough to change verdicts, which is exactly why they must be applied together in one visible bridge. The figure at the end of this bridge is the only number the owner of one share can spend.

Epistemics

C83. Epistemic-limits inventory

• Asks: List the load-bearing unknowns — the specific questions the verdict depends on that no available filing or figure-table reconciliation answers — ranked by how much of the verdict rests on each, and each tagged: resolvable now from public data (source named), resolvable only by time, or structurally unknowable.
• Good: A ranked list of at most five company-specific unknowns (generic "macro conditions" fails), where no unknown tagged resolvable-now remains unresolved in the final output, and the count of unknowable ones is carried directly into the verdict. An empty list is suspicious; a twenty-item list means the work above was not finished.
• Rationale: Every verdict rests on things not known, and the danger is not the unknowns but the unexamined ones; leaving a cheaply resolvable unknown unresolved is negligence, while agonizing over an unknowable one is waste. This is the item that makes this lens's differentiator — calibrated uncertainty — checkable rather than rhetorical.

C84. Pre-registered falsifiers

• Asks: What specific, observable events or figures within the next one to two reporting cycles would flip this verdict — stated with thresholds before those reports arrive?
• Good: At least three falsifiers of the form "if X appears in report Y at level Z, the verdict flips," each naming the document it will appear in. Vague triggers ("if fundamentals deteriorate") fail.
• Rationale: A verdict that no future observation could reverse is a belief, not an analysis. Pre-registering falsifiers before the data arrives is the only defense against post-hoc rationalization, because it commits the analyst while commitment is still cheap.

C85. Single-source dependency inventory

• Asks: Which thesis-critical claims rest on a single filing statement with no independent cross-check inside or outside the company's disclosure set?
• Good: An inventory of single-sourced claims among the load-bearing ones, each either cross-checked during the study or explicitly flagged, with the count of surviving single-source dependencies stated next to the verdict.
• Rationale: A claim's reliability is bounded by its weakest sourcing, and a thesis is bounded by its weakest load-bearing claim. Knowing exactly which claims would collapse if one document proved wrong is the difference between a structure and a house of cards that merely looks like one.

C86. Attribution honesty for the core signal

• Asks: For the single most thesis-supporting trend in the figure table, what are at least two alternative explanations (price versus volume versus mix versus FX versus accounting change), and which can the filings actually distinguish?
• Good: The core trend named, two or more rival causal stories listed, and for each the discriminating disclosure cited — or an admission that the filings cannot distinguish them, demoting the trend from "evidence of quality" to "consistent with quality."
• Rationale: A favorable trend is compatible with many causes, only some of which persist. The thesis depends not on the trend existing but on the favorable cause being the true one, and that is a separate claim requiring separate evidence.

C87. Locating the actual bet

• Asks: What would an informed bull and an informed bear both concede, and on which specific proposition do they irreducibly split — what is the actual bet?
• Good: A written statement of the common ground and the split point, with the split point tied to figure-table evidence and to the falsifier list (C84).
• Rationale: Most of any analysis covers ground where no reasonable person disagrees, and it is easy to mistake thoroughness there for edge. The verdict is only ever a position on the split point, so failing to name it means not knowing what one is betting on.

C88. Conviction-evidence proportionality

• Asks: Is the confidence in the verdict proportional to the evidence assembled — does the count of resolved load-bearing items exceed the count of unresolved ones for any verdict stronger than watch?
• Good: An explicit tally: N load-bearing items resolved with cited evidence versus M unresolved or unknowable. A buy-below verdict with M ≥ N fails and must be downgraded or the imbalance explicitly defended.
• Rationale: Confidence is a claim about evidence, and claims about evidence can be audited by counting. The mechanical check removes the most seductive failure mode in analysis: conviction that grew from time spent rather than facts established.

C89. Verdict-flipping item

• Asks: Which single checklist item, if its answer flipped, would change the verdict — and if no single item would, where did the verdict actually come from?
• Good: At least one named item with the counterfactual stated, or an honest admission that the verdict rests on something outside the checklist — itself recorded as a checklist-revision note.
• Rationale: If the verdict is insensitive to every individual answer, the items were decoration around a conclusion formed elsewhere. Identifying the hinge item both validates that the checklist did the deciding and tells the future reader exactly what to monitor.

C90. Surprise ledger

• Asks: What did the study reveal that genuinely surprised the analyst relative to expectations at the source-gate stage — and if nothing did, why not?
• Good: At least two recorded surprises with the prior expectation and the correcting evidence, or an explicit admission that priors were too weak to be surprised — never a silent absence.
• Rationale: An investigation that only confirms what was already believed either had perfect priors or was run in confirmation mode, and perfect priors are not the way to bet. The surprise ledger is a cheap running detector of whether the study is actually updating on evidence.

C91. Precision honesty

• Asks: Are numbers in the thesis stated with more precision than the evidence supports — buy-below prices to the yen, forecasts to a decimal — and does stated precision match the width of what is actually known?
• Good: The buy-below figure and forward estimates expressed at a rounding consistent with their derivation's error bar, with at least one sensitivity acknowledged. Four-significant-figure outputs from one-significant-figure inputs fail.
• Rationale: False precision is a confidence claim smuggled in through formatting, and it survives because no one audits decimal places. Matching stated precision to evidential precision keeps the artifact honest about how much it knows.

C92. Knowledge half-life

• Asks: How fast does this analysis go stale — which figure-table rows and thesis claims expire first, and what is the review-by date after which the verdict should not be trusted without refresh?
• Good: A staleness ranking and one explicit review-by date attached to the verdict. Commodity-linked or contract-renewal-dependent theses get short dates.
• Rationale: An analysis is a photograph of a moving object, and different parts of the object move at different speeds. Stating the decay rate at write-time prevents the silent error of acting on a stale verdict later while believing it fresh.

Study self-check (process bias)

These items audit the study's own artifacts — figures.md, predictions.md, the profile output — for the analyst's biases. They are unique to this lens.

C93. Narrative-first versus evidence-first audit

• Asks: Does the verdict sentence cite specific item IDs whose answers precede it, or does the item-by-item read as justification assembled after a conclusion was already formed?
• Good: The verdict traceable to named items answered above it; item answers containing verdict-flavored language ("consistent with our positive view") before the verdict section are counted as contamination, and zero or near-zero contamination passes.
• Rationale: The order in which conclusion and evidence were produced determines which one shaped the other, and the artifact itself preserves fingerprints of that order. Checking the artifact is the only version of this test that does not rely on the analyst's self-report.

C94. Stamp-price anchoring check

• Asks: Was the buy-below price derived from a valuation logic that never references the stamp price — and does the derivation show the value estimate before any comparison with the current price?
• Good: The buy-below derivation reproducible from figure-table inputs alone with the stamp price nowhere in the chain; if the final figure lands within ~15% of the stamp price, an explicit note addresses whether that proximity is coincidence or anchor gravity.
• Rationale: The current price is the single most available number in the study and therefore the strongest anchor, and anchored valuations converge toward whatever the market already says — destroying the point of independent valuation. The defense is structural: derive value blind, compare second.

C95. Trough-anchored framing check

• Asks: Do any growth rates, CAGRs, or trend claims in the study anchor on a trough or peak base year, and do the conclusions survive recomputation from two alternative base years?
• Good: Every multi-year growth figure cited in the thesis recomputed from at least one earlier and one later base year; any conclusion that flips or materially weakens under re-basing has the item answer flagged and the robust version stated in the revision note.
• Rationale: Base-year choice is the quietest way an honest table can still tell a slanted story, and the analyst who built the table is the least likely to notice. Re-basing is cheap, mechanical, and catches the error class where it actually lives — in the study's own artifact.

C96. Citation-period histogram

• Asks: Count this profile output's evidential citations by fiscal period — what share draws on the latest year versus the full ten-year record, and is any recency concentration justified by the claim types?
• Good: A rough histogram of this profile output's citations per period with the latest-year share stated; concentration above roughly half either justified (e.g., "the case is about the post-restructuring entity") or corrected by extending trend claims to full-record support.
• Rationale: The latest quarter is the most vivid and most available evidence, and vividness is not weight — one good year proves almost nothing about a durable characteristic. Counting citations by period makes recency weighting visible as a number instead of a vibe.

C97. Halo metric audit

• Asks: Is one spectacular metric doing a disproportionate share of the persuasion — measured by how much thesis attention it receives relative to its rank in the load-bearing item list?
• Good: The most-cited metric identified and its prose share compared against its load-bearing rank; if a metric outside the top load-bearing items dominates the prose, the halo is named and the thesis rebalanced or the imbalance defended.
• Rationale: One extraordinary number casts a glow that quietly upgrades every neighboring judgment, which is how a great ROIC buys a pass on a bad governance record. Auditing attention against declared importance catches the transfer, because the two should match in an evidence-driven document.

C98. Ugly-name test

• Asks: Restate the thesis using only figure-table row IDs and numbers — no company name, no sector romance, no story vocabulary. Does the same verdict still follow?
• Good: A short numbers-only paragraph followed by an honest assessment of what verdict a stranger would assign to it — and reconciliation if that differs from the actual verdict.
• Rationale: Stories attach to names and change how identical numbers feel; stripping the name is the cheapest way to measure how much of the verdict is carried by narrative rather than evidence. If the anonymous version reads pass and the named version reads buy-below, the difference is the story premium being paid.

C99. Archetype gravity check

• Asks: Which familiar investment archetype does this thesis most resemble ("neglected hidden gem", "cash-box awaiting catalyst", "quiet compounder", "turnaround inflection") — and what specific evidence shows this company deviating from that archetype's script?
• Good: The archetype named honestly and at least two concrete points where the company's record contradicts or complicates it, cited from the figure table; if no deviation can be found, a flag that the thesis may be template-completion rather than observation.
• Rationale: Archetypes are compression schemes that fill in unobserved details from the template, so a company that pattern-matches a beloved story gets credited with the whole story's attributes on partial evidence. Naming the template and hunting for mismatches forces the observed company back into focus.

C100. Hedged-answer share

• Asks: What fraction of this checklist's items were answered with hedges ("partially", "mixed", "unclear") — and is each hedge backed by a stated reason it could not be resolved, or is it pattern-matched filler?
• Good: A hedge count over total items, every hedged answer carrying either a resolvable-versus-unknowable tag or a citation showing the attempt to resolve; a hedge share above roughly a third triggers a note that the study may have run in coverage mode rather than investigation mode.
• Rationale: Hedged answers are how a checklist gets completed without being executed — each one looks diligent while deciding nothing. Counting them, and demanding a reason per hedge, converts the checklist from a form to fill into questions that must actually be fought with.

C101. Provenance leak check

• Asks: Does the item-by-item text contain any factual claim not traceable to a figure-table row or an archived source — including background "knowledge" about the company, sector lore, or numbers remembered rather than cited?
• Good: A pass over every factual assertion in the profile output; each traces to a row ID or source citation; any orphan claim removed, sourced, or flagged for figure-table addition. Zero surviving orphans.
• Rationale: The study's integrity model assumes every claim has checkable provenance, and the easiest breach is not fabrication but unexamined prior knowledge flowing in as if it were evidence. A mechanical orphan-claim sweep enforces the boundary the isolation design promises.

C102. Disconfirmation effort ledger

• Asks: Compare the effort spent seeking disconfirming versus confirming evidence in this study — sources consulted, searches run, figure-table rows computed for each side. Is the ratio defensible?
• Good: A rough two-column ledger of bull-supporting versus bear-supporting investigative actions; gross asymmetry toward confirmation either corrected with targeted disconfirmation work or explicitly acknowledged next to the verdict as a known tilt.
• Rationale: Search direction determines what gets found, so a study that mostly searched for support will mostly find support regardless of the company's quality. The ledger measures the input bias directly, which is more honest than assessing the output and hoping it balanced itself.

Verdict guidance

Epistemic honesty is this profile's differentiator, so too-hard is not a failure state — it is the correct output whenever a load-bearing unknown from C83 cannot be resolved from the documents, and the verdict must name the blocking item (e.g., "too-hard: C83 — sustainability of the top customer relationship is unknowable from filings"). Base rates discipline the story throughout: most companies are average (C1–C4), most margins and growth rates revert (C7–C8), most forecasts are optimistic (C5–C6), so a thesis surviving the base-rate section and the strongest short case (C61) has earned something rather than merely accumulated detail. Any verdict stronger than watch must also clear the proportionality tally (C88) and the study self-check section (C93–C102) — a verdict that fails its own process audit is downgraded regardless of how the company looks. Watch is issued when the unknowns are tagged resolvable (C83) — it comes with the specific work that would upgrade it and the falsifiers (C84) that would kill it.

Pass means the short case (C61) wins, growth destroys value (C12), or the register makes every unlock scenario unreachable (C34, C37) at a price that needs one. Buy-below-¥X is set from the downside case, not the base case: X is derived blind to the stamp price (C94) from the bear-case valuation (C72), such that the private-owner yield (C74) still clears the return hurdle on the fully-leaked per-share figure (C82) even if the skeptic's three facts largely play out, the wait is indefinite but paid (C81), and no C34–C49 unlock ever arrives. If the base case is needed to justify the price, the price is too high.

Archive-depth rule. When the archived sources are shallower than an item requires (fewer years, missing competitor/customer filings, absent disclosures), the item is answered data-insufficient with the missing source named — never estimated from general knowledge without an explicit [general-knowledge: not-from-archive] flag. Load-bearing items stuck at data-insufficient push the verdict toward too-hard or watch, not toward a guess.

A too-hard verdict names the specific item ID(s) that triggered it. Item-level answers use pass / fail / data-insufficient / practitioner-pending; items after a too-hard early exit are marked not-evaluated, never silently skipped.