MANUAL — Stock & StockNews System Reference (English)

A plain-language reference for reading the StockNews investment trees and understanding the abbreviations and concepts used throughout. Designed for the system owner (Ming) — assumes no prior finance background but goes deep enough to read a 10-K with confidence.

For the Chinese version, see MANUAL_zh.md.

Part A — How to read a tree report (5-minute orientation)

Every ticker T analyzed by this system produces a folder reports/T/ containing several files. Read them in this order:

1. primer.md — Baseline facts about the company (what it does, who buys, financials by segment, governance). Read this first.
2. developments.md — Recent material events (last ~6 months). Reading this together with the primer gives you the current state.
3. evidence_{date}.jsonl — Every load-bearing fact this system uses, with source citation and tier rating (A/B/C). Spot-check by opening; read in detail only if questioning a claim.
4. h0_thesis.md — The single one-paragraph statement of what this system thinks the market is mispricing. The whole tree exists to test this statement.
5. tree_v1_en.md — The investment tree itself. Five branches (L1A, L1B, L1C, L1D, L1E), each with sub-leaves. Each leaf has a verdict (✅ supports / ⚠️ partial / ✗ falsified) and the evidence that drove the verdict.
6. scenarios.md + implied_prob.md — Three regimes (Bull / Base / Bear), each with its own peer group, multiple, narrative, and implied probability (so you can see what the current price is actually betting on).
7. triggers_redflags.md + dashboard.md — Forward events to watch (catalysts that would confirm/deny the thesis) + a one-page status board.
8. durability_test.md (NEW, for long-term hold tickers) — Six-question test of whether this is a 5-10 year hold candidate.

The system deliberately does NOT emit: price targets or pure buy/sell ratings.

The system DOES emit (added 2026-05-02 per Section XII Investment Scorecard, K.6): a Buy / Hold-with-sizing / Avoid band per ticker, a position-sizing range (e.g., "1-3%"), a 2-minute pitch, and the K.3.5 weighted-score that aggregates the 15-question scorecard. You still make the investment decision — the system gives you the structured reasoning plus a bottom-line decision artifact at the end of every tree, so the gap between "is this worth owning?" and "is this worth owning at this price?" is explicit rather than implicit.

Part B — Beginner stock concepts

What is a stock?

A share of ownership in a company. If a company has 1 billion shares outstanding and you own 1,000, you own 0.0001% of the company.

Share price

What one share costs to buy on the public market right now. Determined by supply and demand of buyers and sellers.

Market capitalization (market cap)

Share price × total shares outstanding. The total dollar value the market assigns to the company's equity.

• Example: Ford trades at $11.89 with ~4 billion shares = $47.4 billion market cap.
• Categories: Large cap (>$10B), Mid cap ($2-10B), Small cap ($300M-$2B), Micro cap (<$300M).

Dividend

Cash payment from a company to shareholders, usually quarterly. Companies that pay dividends are typically mature, cash-generating businesses.

• Ford pays $0.15 per share quarterly = $0.60 per share per year.

Dividend yield

Annual dividend ÷ share price. Tells you the cash return as a percentage.

• Ford: $0.60 ÷ $11.89 = 5.05% yield.
• Compare to: 10-year US Treasury bond yield (~4-4.5% in 2026). Ford's yield being above Treasury makes it a "yield substitute" for some investors.

Earnings per share (EPS)

A company's net income (profit) divided by total shares outstanding. The per-share profit.

• GAAP EPS = strict accounting standard (Generally Accepted Accounting Principles). Includes ALL items including one-time charges, write-downs, etc.
• Non-GAAP EPS / Adjusted EPS = company's preferred metric, usually excludes one-time items management considers non-recurring. Usually higher than GAAP. Read both.

P/E ratio (price-to-earnings)

Share price ÷ EPS. Tells you how many years of current earnings it would take to earn back your share price.

• Trailing P/E = uses last 12 months' actual EPS (backward-looking).
• Forward P/E = uses next 12 months' estimated EPS (forward-looking, what the market is paying for future earnings).
• Lower P/E = cheaper relative to earnings. Higher P/E = more expensive (or higher growth expected).
• Reference: S&P 500 forward P/E typically ~18-22x. Ford at 8.5x is well below market — pricing in decline. Nvidia at 25x is above market — pricing in growth.

Free cash flow (FCF)

The cash a company generates from operations after spending on capital expenditures (capex). The most important number for long-term investors because dividends, buybacks, and debt reduction all come from FCF.

• Formula: Operating Cash Flow − Capital Expenditures = Free Cash Flow.

Capital expenditures (capex)

Money spent on long-term assets (factories, equipment, R&D capitalized). Capex-heavy businesses (autos, telecom, oil) need to constantly reinvest to maintain their position. Capex-light businesses (software, consumer brands) can return more cash to shareholders.

Part C — Intermediate stock concepts

Return on Invested Capital (ROIC)

The most important measure of management quality. Net Operating Profit After Tax ÷ Invested Capital.

• Tells you what return management is earning on every dollar deployed in the business.
• Compare to WACC (cost of capital): if ROIC > WACC consistently, the business creates value. If ROIC < WACC, it destroys value.
• Reference: Industrial companies typically WACC 7-9%; tech 8-12%.
• Ford: ROIC averaged 2-4% in 2020-2024 — chronically below WACC, meaning capital deployed has destroyed shareholder value.

Return on Equity (ROE)

Net Income ÷ Shareholder Equity. Measures return on the equity capital. Less rigorous than ROIC because it ignores debt.

Return on Assets (ROA)

Net Income ÷ Total Assets. Measures asset productivity. Useful for comparing companies in capital-intensive industries.

Weighted Average Cost of Capital (WACC)

The blended cost of debt + equity that a company must clear to create value.

• Formula: (Equity / Total Capital) × Cost of Equity + (Debt / Total Capital) × Cost of Debt × (1 − Tax Rate)
• Conceptually: the hurdle rate. If a project earns above WACC, it adds shareholder value.

EBIT (Earnings Before Interest and Taxes)

Operating profit before financial expenses and taxes. Useful for comparing operating performance across companies with different debt structures.

EBITDA (EBIT + Depreciation + Amortization)

Operating profit before non-cash D&A charges. A rough proxy for cash generation. Critics say it's misleading because depreciation reflects real wearing-out of assets that must be replaced (capex). "EBITDA = bullshit earnings" — Charlie Munger.

Adjusted EBIT / Adjusted EBITDA

Company-defined operating profit excluding "one-time" or "special" items. Always read the reconciliation table — companies have wide latitude in what they exclude.

Operating margin

Operating Profit ÷ Revenue. Tells you how much profit the company makes per dollar of sales.

• Auto OEMs: typically 4-8%.
• Software companies: often 25-40%.
• Tech leaders (Nvidia): up to 60%+.

Gross margin

Revenue minus Cost of Goods Sold, divided by Revenue. The first profitability filter. High gross margin = pricing power or low input costs.

• Ford gross margin: ~10-12% (auto).
• Nvidia: ~71% (chip design with software moat).

Working capital

Current Assets − Current Liabilities. The cash tied up in day-to-day operations (inventory, receivables, payables). Negative working capital is great (customers pay you before you pay suppliers — Costco, Walmart).

Net debt

Total Debt − Cash. The "true" debt burden. Companies with negative net debt (more cash than debt) like Apple are net cash positive.

Leverage ratios

• Debt/EBITDA: how many years of EBITDA to pay back all debt.
• Net Debt/Equity: leverage relative to equity capital.
• Interest coverage: EBIT ÷ Interest Expense (how many times interest is covered).

Beta

Stock's volatility relative to the market. Beta of 1.0 = moves with market; >1.0 = more volatile; <1.0 = less volatile. Ford has historically traded with beta ~1.0-1.2 (lower than typical cyclical because of dividend support).

Dividend coverage ratio

Free Cash Flow ÷ Dividend payments. Tells you how many times the dividend is covered by cash generation. <1.5x is risky; >2.0x is healthy.

Part D — Valuation concepts

Multiple expansion / compression

When a stock's P/E ratio (or other multiple) goes up or down without earnings changing. Multiple expansion = market willing to pay more per dollar of earnings. Multiple compression = the opposite. A stock can rise via earnings growth, multiple expansion, or both.

Sum-of-Parts (SOTP) valuation

Valuing each business segment separately and adding them up. Useful for conglomerates or companies where one segment has materially different economics from another. Example: Ford Pro might be worth more on a services multiple than on the auto OEM multiple applied to the whole company.

Discounted Cash Flow (DCF)

Projecting future cash flows and discounting them back to present value using a discount rate (usually WACC). The "fundamental" valuation approach. Highly sensitive to assumptions (terminal growth rate, discount rate).

Reverse DCF

Instead of projecting cash flows to derive a price, take the current price and back out what cash flow growth rate the price implies. Tells you "what is the market actually pricing in?" Useful for stress-testing whether market expectations are realistic.

Implied probability

For a 3-scenario valuation (Bull / Base / Bear), back out the implied probability weights from the current price: if Bull = $X, Base = $Y, Bear = $Z, and price = $P, what probability weights produce $P as the expected value?

Mean reversion

Long-term tendency for above-average or below-average results to drift back toward average. Used by value investors as a basis for buying out-of-favor stocks. Doesn't always work — sometimes "things really are different."

Margin of safety

The discount between your estimated intrinsic value and the current price. Benjamin Graham's principle: "buy with a margin of safety to protect against the uncertainty of estimation." Typically 30-50% discount needed for high uncertainty.

Part E — Tree methodology vocabulary (StockNews-specific)

MECE (Mutually Exclusive, Collectively Exhaustive)

Every set of sibling branches in a tree must (a) not overlap and (b) together cover the parent question completely. McKinsey-origin discipline. Example failure: "macro factors" and "interest rates" as siblings (overlap, since rates are a macro factor).

Falsifiability (Karl Popper)

A claim is scientific only if you can specify what observation would prove it wrong. Every leaf in a StockNews tree must have a falsification condition. Without one, a claim is opinion, not hypothesis.

Mauboussin 2×2 (Moat Strength × Trajectory)

Michael Mauboussin's framework for assessing competitive moats. X-axis: moat strength (weak / strong). Y-axis: moat trajectory (eroding / widening). Strong-widening = best (Costco). Strong-eroding = warning (some legacy media). Weak-widening = improving (some startups). Weak-eroding = avoid.

Profit pool

The total economic profit available in an industry. Where does it sit (chip designers vs fabs vs OEMs)? Is it migrating? Profit-pool analysis is more powerful than market-share analysis.

S-curve

The lifecycle of a product, technology, or company: slow start, rapid scaling, plateau, decline. Used to identify where in the curve a company sits, and to forecast what's next.

Strategic Inflection Point (Andy Grove)

A moment when the fundamentals of a business shift by 10x in some dimension (technology, regulation, customer behavior). Past examples: digital photography for Kodak, mobile for Nokia. Identifying inflections early is the key to surviving them.

Mispricing mechanism (4 types)

This system uses 4 categories for WHY a stock might be mispriced:

1. Cognitive bias — market anchored on past patterns, representativeness heuristic, etc.
2. Structural blindness — information exists but is systematically uncomputed
3. Framework / category misalignment — wrong peer group or wrong identity
4. Time-horizon mismatch — short-duration capital pricing long-duration cash flows

Tier A / B / C evidence

• Tier A = primary source, issuer-stated, dated (10-K, 10-Q, 8-K, official press release, conference call transcript)
• Tier B = reputable secondary source (Reuters, Bloomberg, FT, WSJ, sell-side research)
• Tier C = derivative or generated (training-knowledge reconstructions, blog aggregators)

The system requires Tier A or B for any load-bearing claim before Stage 3 hypothesis testing.

Verdict calibration (5 tiers)

Every leaf gets one of these verdicts:

• 强力支持 (Strongly Supported) — multiple independent primary sources, no credible counter-evidence
• 支持 (Supported) — primary evidence exists but alternative interpretations remain
• 部分支持 (Partially Supported) — evidence directionally right but threshold not met, or evidence indirect
• 证据不足 (Insufficient Evidence) — testable but data doesn't exist yet
• 已证伪 (Falsified) — evidence affirmatively contradicts the hypothesis

A tree where every leaf is 强力支持 is almost certainly cooked.

H-0 thesis

The "null hypothesis" of the entire tree: the single one-sentence statement of what the system believes the market is mispricing. The whole tree exists to test this statement against evidence.

Three layers of every output

The system enforces these three layers running concurrently:

1. Analytical engine — classify, test hypotheses, falsify
2. Output template — every leaf has Question → Framework → Hypothesis → Evidence → Verdict → Falsification
3. Narrative wrapper — voice that translates structure into memorable language

Reading the index page (visual legend)

The StockNews index page packs each ticker's current state into one card. Here is what every visual element means:

Verdict tally — e.g. 13 ✅ · 11 ⚠️ · 4 ✗ · 7 ⊗. Counts of each verdict class across the tree's leaves. Color-coded:

• Green ✅ supported — multiple Tier A/B evidence consistently support the leaf
• Amber ⚠️ partial — direction correct but evidence inconclusive, or forward-looking and not yet testable
• Red ✗ falsified — current evidence conflicts with the leaf's hypothesis (already disproven)
• Grey ⊗ untestable — the question is real but no evidence is currently available; treated as "evidence not in"

A healthy tree has a mix; an all-green tree is almost certainly post-hoc rationalization rather than honest analysis.

H-0 score pill — e.g. 82%. The weighted aggregate confidence in the H-0 thesis (the one-sentence claim about what the market is mispricing). Color bands:

• Green ≥75% — H-0 strongly supported; thesis intact
• Amber 55-74% — H-0 partially supported; thesis on probation
• Red <55% — H-0 weakly supported or close to broken; reconsider holding

Probability bar — e.g. 25/50/25 rendered as a horizontal bar split into three colored segments:

• Green segment (left) — Bull probability — % chance the bull-case scenario plays out
• Blue segment (middle) — Base probability — % chance the base-case scenario holds
• Red segment (right) — Bear probability — % chance the bear-case scenario fires

Numbers are reverse-engineered from the current market price relative to the three scenario targets — they describe what the market is implicitly assuming, not what the analyst forecasts. If your own bull/base/bear distribution differs materially from the market's, that's the trade signal.

Price — the anchor price used at the time the tree was built or last updated. Not real-time; refresh it manually after each material event.

Left-edge color band on the card — same color logic as the H-0 pill (green = strong, amber = partial, red = weak). Lets you scan the card grid for thesis health at a glance without reading every label.

Updated date — the date the analytical state of the tree last changed (verdicts, scenarios, score, price). Sourced from the tree's INDEX_META.updated field, NOT from file modification time — a Chinese-translation cleanup or typo fix does not bump this date. Treat anything more than 90 days old with suspicion; the underlying facts have likely moved. The tree's INDEX_META.review_due field carries the next-review deadline (typically ~1 week for weekly-cadence tickers, ~3 months for quarterly-cadence holds).

Part F — SEC filing types explained

10-K (Annual Report)

The most important filing. Filed once per year (~60 days after fiscal year end). Contains:

• Business description (Item 1)
• Risk factors (Item 1A) — read this critically
• Properties (Item 2)
• Legal proceedings (Item 3)
• Management discussion & analysis (MD&A, Item 7) — management's view of financial trends
• Financial statements (Item 8) — audited income statement, balance sheet, cash flow statement
• Notes to financial statements — where the truth often hides

10-Q (Quarterly Report)

Filed ~45 days after each fiscal quarter end (3 per year; the 4th quarter is rolled into the 10-K). Less detailed than the 10-K but updates trends.

8-K (Current Report)

Filed within 4 business days of any "material event" — earnings release, executive change, M&A announcement, major contract, bankruptcy filing, etc. The fastest way to learn about new information.

DEF 14A (Proxy Statement)

Filed before annual shareholder meeting. Contains executive compensation, board composition, related-party transactions, governance disclosures. Underrated source — pay close attention.

S-1 (IPO Registration)

Filed when a company plans to go public. The most important document for evaluating a pre-IPO investment. Contains business description, risk factors, financial history, capital structure, and use of proceeds.

Form 4 (Insider Transactions)

Filed when corporate insiders (officers, directors, 10%+ shareholders) buy or sell shares. Public information on insider behavior.

13F (Institutional Holdings)

Filed quarterly by institutional investment managers >$100M. Shows what stocks they own. Useful for tracking what big investors are doing.

Part G — Common abbreviations cheat sheet

Financial metrics

Stock market terms

Regulatory bodies

StockNews system terms

Part H — Reading a StockNews tree report — example walkthrough

Suppose you open reports/F/tree_v1_en.md. Here's how to read it:

1. Read the H-0 statement first (single one-sentence thesis). This is the entire claim being tested.
2. Look at the Level-1 decomposition: 5-6 sibling branches, each addressing one aspect of H-0.
3. For each branch, look at the verdict tally at the bottom: how many leaves are ✅ (supports), ⚠️ (partial), ✗ (falsified)?
4. Read the leaves you find most contestable — the ones marked ⚠️ or ✗ are usually most informative because they show where the thesis is weakest.
5. Cross-reference with scenarios.md — the three regimes show what 2028-2030 looks like under each branch outcome.
6. Cross-reference with triggers_redflags.md — what forward events would resolve uncertainty?
7. For long-term hold tickers, also read durability_test.md — six-question test that goes beyond the 12-24 month price scenarios.
8. Don't skip the bilingual essay (tree_v1_zh.md) — sometimes the Chinese version captures nuance the English misses, or vice versa.

Part I — Mistakes to avoid (signal you're being sold a story)

When reading any analyst report (mine or anyone else's), watch for these red flags:

• All bull or all bear — real analysis acknowledges the other side. If every leaf supports the thesis, the analysis is cooked.
• No falsification conditions — if the analyst can't tell you what would prove them wrong, they're not being honest.
• Vague timeframes — "soon," "in the near term," "eventually" are weasel words. Demand specific dates or quarters.
• Risks listed but not weighted — a risk section that doesn't tie to the verdict is decorative. Each risk should change at least one leaf if it materializes.
• Hidden assumptions — every numerical claim ("the market is pricing in 20% growth") should unpack to explicit components (revenue base, margin, multiple).
• Story drift — if the analyst keeps moving the goalposts ("OK, that didn't happen, but here's why I'm still right"), the thesis is being protected from data.
• Authority arguments — "Buffett does this" or "Munger said that" are interesting but not sufficient. Demand the underlying mechanism.
• Conflict of interest unflagged — if the analyst has a position, financial relationship, or other bias, it should be disclosed upfront.

Part J — How to update this manual

Add new terms as they come up in tree reports. The cheat sheet (Part G) is meant to grow over time. When a new abbreviation or concept appears in a tree, add it here so the next reader doesn't have to re-derive its meaning.

If a term is ticker-specific (e.g., "Ford Pro" or "Stargate" for OpenAI), add it to the per-ticker glossary at reports/{TICKER}/glossary.md, NOT to this global manual.

Part K — Investment Decision Framework

This section synthesizes the investor mental models and pre-purchase discipline that complement the technical valuation concepts in Parts B-D. Adapted in part from a ChatGPT cheat sheet (2026-05-02 Nokia evaluation), filtered + augmented for use with the StockNews tree methodology.

K.1 — Foundational mental models

1. Stock price ≠ company quality. A great company can be a bad stock if the price is too high; a mediocre company can be a great stock if the price is low enough. Always ask: "At this price, what future success is already priced in?"

• Great company at 80x earnings = expensive (Nvidia 2024)
• Okay company at 8x earnings = maybe cheap (Ford 2026 — but maybe a value trap)
• Bad company at 5x earnings = still could be a trap

2. Expectations gap. Stock prices move based on the gap between what investors expected vs what actually happened. A company can report good results and the stock falls if expectations were higher; a company can report bad results and the stock rises if results were less bad than feared. The H-0 thesis in every StockNews tree is fundamentally about identifying expectations gaps.

3. Earnings ≠ cash. A company can show "profit" on paper (GAAP EPS) while generating little real cash (FCF). The most important question for long-term investors: does the business actually produce cash that can be returned to shareholders or reinvested at high IRR?

4. The 2-minute rule. Before buying any individual stock, you should be able to explain — in under 2 minutes — (a) what the company does, (b) what the valuation is, (c) what the main risk is, and (d) what would make you sell. If you can't, you don't understand it well enough.

K.2 — Stock types (very different risk profiles)

Why this matters: Different stock types need different evaluation frameworks. A "growth at any price" approach to a value stock will produce poor returns; a "yield substitution" approach to a growth stock misses the thesis entirely.

K.3 — Portfolio rules (position sizing discipline)

Beginner-appropriate baseline (US-resident retail investor):

• Broad ETF core: 60-90% — VOO (S&P 500), VTI (total US market), VXUS (international), QQQM (Nasdaq tech-heavy), SCHD/VIG (dividend-focused)
• Individual stocks: 10-40% — where you try to outperform; also where you create concentrated risk
• Single stock cap: 5-10% maximum unless you deeply understand it (and even then, monitor closely)

StockNews tree position-sizing convention:

• High-conviction long-term hold (durability score 22-25, no fatal flags): up to 5-7%
• Selective hold (durability score 17-21, no fatal flags): 1-3%
• Short-term opportunistic (durability score 12-16): 0-2%
• Avoid (durability score <12): 0%
• Any ticker with a fatal flag fired (see K.3.1 below): cap at "selective hold" tier (≤3%) regardless of aggregate score; 2+ fatal flags → 0%

K.3.1 — Fatal-flag override on the durability score

A high aggregate score can mask a categorical defect. A 5+5+1+5+5 = 21/25 reads as "high durability" but actually says "high-quality assets run by capital destroyers" — different bet entirely. The fatal-flag rule prevents this hiding.

A fatal flag fires when any single question scores 1/5 (已证伪) in:

• Q3 — 10-year capital allocation (chronic ROIC < WACC; serial value destruction)
• Q4 — disruption survival (a credible threat scores >50% probability with high impact)
• Balance-sheet survivability — informed by the parent tree's L1B/L1D credit verdicts: can the company survive a 2-3 year demand trough without dilution or distressed financing?

Override rule:

• 0 fatal flags → headline rating uses aggregate score directly
• 1 fatal flag → headline rating capped at "Medium" (max 21/25 effective)
• 2+ fatal flags → "Low / Unownable for long-term hold"

Worked example: Ford (F) scores 17/25 with Q3 = 1/5. Total already lands in Medium, but the fatal flag means the right framing is "Medium with a binding capital-allocation flaw," not "Medium durability — selective hold." The headline number is no longer independent confirmation of investability — it is a description that includes the flaw. Position recommendation must reflect this (1-3% is upper bound, not target).

Core principle: If a single stock's bad day can ruin your year, the position is too large. Position size should reflect both conviction AND diversification need — and conviction must be checked against fatal flags, not just the aggregate score.

K.3.2 — Durability cap convention (M2 — added 2026-05-03 per Codex review)

Per Codex's project-direction review (2026-05-02), the original "27/25 effectively capped at 25" framing used in some trees (notably AJNMY's AI-first re-framing) is removed. Going forward, the durability test produces TWO fields:

score: N/25                          # always within 0-25; never reports >25
exceptional_positive_overrides: yes/no    # noted separately when an asset has structurally above-framework strengths

Why: capping is useful because it prevents heroic quality claims from hiding valuation/cycle risk. But the system can separately note "above-framework strengths" — e.g., Ajinomoto's AI-sleeve runway is genuinely above what the 6-question test fully captures, so exceptional_positive_overrides: yes is appropriate but the score: 25/25 ceiling holds.

A score of 25/25 is the maximum; any tree previously claiming "27/25" should be re-stated as score: 25/25, exceptional_positive_overrides: yes with the additional strengths documented in the durability test prose.

K.3.3 — Default scenario probability prior (M4 — added 2026-05-03 per Codex review)

Per Codex's review (2026-05-02): scenario probability assignments (Bull/Base/Bear) are subjective and previously not derived from any explicit prior, which lets analyst enthusiasm quietly encode itself into the probability vector.

New rule: the default probability prior for any tree is 25/50/25 (Bull/Base/Bear) OR 20/60/20 for trees where the H-0 thesis is "current pricing is approximately fair and the asymmetric mispricing depends on a specific catalyst not playing out."

Deviations from default require explicit reasoning in scenarios.md. Acceptable reasons:

• Specific catalyst already partially de-risked (e.g., GOOGL post-DOJ-ruling: Bull weight raised from 25% to 30% because a key Bear-case branch was pre-empted)
• Stock has already absorbed a meaningful re-rating event (e.g., TOTDY post-May-2 +18%: Bear weight raised from 20% to 23% because momentum may unwind)
• Activist or governance catalyst has well-documented base-rate (e.g., AJNMY pre-May-7: Bull weight raised to 30% because Palliser is a top-25 holder with a track record)

Unacceptable reasons: "feels right," "this company is special," anchoring on management's outlook.

When applying the default, document it: scenarios.md should say "Defaulted to 25/50/25 prior; no specific evidence supports deviation."

K.3.4 — Correlated-exposure line item (M6 — added 2026-05-03 per Codex review)

Per Codex's review (2026-05-02): every position recommendation must include a correlated-exposure line item. Two positions with great individual asymmetry can still be a poor combined trade if they share a single load-bearing factor (e.g., AI semiconductor capex; rate-sensitive financials; energy prices).

New rule: the long-term holdability verdict section of every tree (Section XI of the standard tree template) must include a correlated-exposure paragraph that:

1. Identifies the load-bearing macro factor the position depends on (e.g., "this position depends on continued AI infrastructure capex through 2030")
2. States the cap-adjustment rule if the holder already has heavy exposure to that factor (e.g., "if existing NVDA + semis + QQQ exposure exceeds 30% of portfolio, drop the hard cap by 1-2 pp at each tier")
3. Names the specific other tickers in the StockNews library that share the same load-bearing factor (e.g., "AJNMY + TOTDY + NVDA all depend on AI capex; combined cap should not exceed N%")

Decision-journal entries (per dashboards/DECISION_JOURNAL_SCHEMA.md) must record the holder's actual existing correlated-exposure when the decision is made, in the correlated_exposure_acknowledged field. This is the data point that distinguishes a thoughtful concentrated-bet from accidental factor pyramiding.

K.3.5 — Section XII Investment Scorecard weighting (M3 — added 2026-05-03 per Codex review)

Per Codex's review (2026-05-02): Section XII Scorecard mixes binary ✅/⚠️/✗ judgments without explicit scoring weight. Codex's worked example: "Is a 13 ✅ / 2 ⚠️ / 0 ✗ score (AJNMY) meaningfully different from 5 ✅ / 9 ⚠️ / 0 ✗ (TOTDY)? Should there be category weights (e.g., Q12 fatal flags worth 10x Q9 insider alignment)?"

The pre-2026-05-03 convention treated all Section XII rows equally, which let two trees with very different load-bearing-question outcomes look superficially similar.

New rule (2026-05-03): Section XII rows are weighted into 4 tiers. The weighted score is computed as Σ(weight × verdict_value) where verdict values are ✅ = 1.0, ⚠️ = 0.5, ✗ = 0.0.

Weighted-score interpretation (theoretical max = 5×3 + 3×4 + 2×4 + 1×4 = 39 if 15-question scorecard):

• ≥85% of max = high-conviction buy signal
• 65-85% = moderate buy with sizing discipline
• 45-65% = wait or skip
• <45% = avoid

Two canonical Q-list tier mappings (adopted 2026-05-03 during full retroactive recalibration):

Both mappings yield max=39 and align with the K.3.5 spec language. Pick the mapping that matches the tree's actual Q-list structure; do not mix.

Worked example: AJNMY actual 15-row derivation (Format A)

(See reports/AJNMY/tree_v1_en.md Section XII for the source table; this is the K.3.5 weighted-score derivation.)

91% = high-conviction buy signal. The two ⚠️s (Q4 capital-allocation Forge / Q10 valuation no-cushion) are the visible tightening factors that justify the "5% pre-reform / 6-7% post-reform" cap and the "(a) wait for May 7 / (b) pre-position 1-2%" sizing dichotomy. K.3.5 score holds as long as Forge does not impair ¥30B+ AND the May 7 catalyst does not falsify the activist disclosure thesis.

K.3.1 fatal-flag override precedence (canonical example: Ford): K.3.5 weighted-score is a score, not the final verdict. K.3.1 sits above it. F (Ford) scored 86% raw — high-conviction band — but Q15 ✗ (10-year ROIC < WACC fatal flag from durability test Q3) caps the position type at "Value + Income hold, 1-3%" rather than "Compounder, 5-7%." This is the canonical example of why fatal-flag override binds even when the weighted score reads strong: the score reflects structural fundamentals; the fatal flag reflects a binding constraint on sizing that cannot be averaged away.

Status (2026-05-03): Full retroactive recalibration across all 9 trees shipped. Every tree_v1_en.md Section XII now contains an inline ### K.3.5 Weighted-score derivation subsection with per-Q tier assignment and weighted-score total. Final tally:

Going forward: weighted score is required (a) when building any new Section XII, (b) when a tree gets a tree_v2 update — derive against the actual Q-list structure of that tree using the appropriate Format A or Format B tier mapping.

K.3.6 — Evidence-strength suffix convention (M1 — added 2026-05-03 per Codex review)

Per Codex's review (2026-05-02): the system uses ✅/⚠️/✗ verdict markers heavily but does not require evidence-strength sub-grading. A ✅ from Tier-A primary-source evidence (10-K disclosure, audited filing) is structurally different from a ✅ from Tier-C training-data interpolation, but the system was treating them as identical.

New convention (going forward): every leaf-level verdict marker in tree_v1_en.md, leaves.md, durability_test.md, and Section XII tables MUST include an evidence-strength suffix:

• ✅A / ⚠️A / ✗A — Tier A evidence (primary-source filing, audited disclosure, dated official statement)
• ✅B / ⚠️B / ✗B — Tier B evidence (reputable secondary source: Reuters / Bloomberg / FT / WSJ / sell-side research)
• ✅C / ⚠️C / ✗C — Tier C evidence (training-data interpolation; community sources; activist deck claims; consultancy estimates without methodology)

Applies to:

• New trees being built going forward — required from Stage 3 onward
• Any tree getting a tree_v2 update
• The AJNMY tree where the load-bearing claims are at the highest evidence-quality stake (worked first-pass below)

Worked example: AJNMY L1D Leaf 4.1 (glass core does not displace ABF build-up dielectric)

Original: ✅ supported Suffix-graded: ✅A — primary-source-verified because Codex 2026-05-03 confirmed via Intel newsroom PDF (2023-09-18), IEEE ECTC 2025 Special Session 4 TSMC slide, Semiconductor Engineering 2025, and Micromachines 2025 academic paper. This is the strongest possible verdict in the AJNMY tree.

By contrast, AJNMY L1A Leaf 1.1 ("ABF holds approximately 95% global share") was originally ✅ supported but should be ✅B — sources are TrendForce + Palliser deck + Ajinomoto IR (B-tier or activist-deck self-interested). The 95% figure is widely repeated but the source chain partly recycles management/activist estimates rather than independent audit.

Migration plan: retroactive suffix-grading across ~80-120 leaves in 9 trees is multi-hour work. Deferred to a dedicated session. Going forward, all new leaves must include the suffix; trees getting tree_v2 updates apply suffixes during the update.

Why this matters: position-sizing discipline depends on knowing which load-bearing claims are primary-source-verified vs. consensus-asserted. A tree with mostly ✅A leaves is structurally more durable than a tree with mostly ✅C leaves, even if the headline verdict counts look identical.

K.4 — Risk taxonomy (10 categories)

Risk is not just "the stock goes down." It comes in distinct flavors that require different mitigation:

The StockNews tree's triggers_redflags.md file should map each red flag (RF1, RF2, ...) to one or more of these risk types.

Correlated-factor risk — operational rule (added 2026-05-19): the AI-capex-cycle overlay at dashboards/ai_capex_cycle_overlay.md tags each ticker with a cycle_exposure band (ai-capex-high, ai-capex-mid-s-curve, ai-capex-low, uncorrelated). When the sum of ai-capex-high + ai-capex-mid-s-curve position sizes exceeds 15% of portfolio, the individual position cap on each affected ticker drops by 2 percentage points (e.g., a 5-7% Hold-with-sizing position becomes a 3-5% effective cap). The intent is to surface correlation at sizing time, not to embed it in each tree's narrative.

Worked example: holding NVDA at 5% + TSM at 4% + MSFT at 4% + GOOGL at 3% = 16% in correlated bucket → trim each by 2pp (4% + 3% + 3% + 2% = 12% effective). The 2-of-4 trigger fire on ai_capex_cycle_overlay.md's T-Capex-{1,2,3,4} set is the operational signal to apply this rule defensively even before the correlated bucket reaches 15%.

Status (2026-05-19): rule is documented, not yet enforced in decisions.jsonl schema. Defer enforcement until one cycle of overlay use produces feedback on threshold calibration.

K.5 — When NOT to buy (anti-patterns)

A stock is NOT automatically a good buy because:

1. People online are talking about it. Reddit/X/finance Twitter sentiment is a contrarian indicator more often than a leading one.
2. The stock went up fast. Fast-rising stocks have often baked in good news already; new buyers are momentum chasers.
3. It has "AI" attached to it. Theme labels are not investment thesis. Many "AI stocks" have minimal AI revenue or moat.
4. It looks "cheap" because the stock price is low. A $5 stock is not cheaper than a $500 stock. Share count + earnings + growth determine value, not price per share.
5. You remember the brand. Brand familiarity (Nokia, GE, Sears) ≠ business quality. Often the opposite.
6. Someone says "this is the next [Nvidia/Apple/Tesla]." Pattern-matching to past winners is poorly correlated with finding new winners.

K.6 — Pre-purchase checklist (12 questions)

Before buying any individual stock, answer these (write the answers; if you can't, don't buy):

1. What does the company actually do? (one paragraph, no buzzwords)
2. Why is the stock interesting now? (catalyst or undiscovered thesis)
3. What is the bull case? (specific mechanisms, not vibes)
4. What is the bear case? (steelmanned, not strawmanned)
5. What valuation is it trading at? (forward P/E, EV/Sales, FCF yield)
6. Is revenue growing? (3-year CAGR + recent quarter)
7. Are profits growing? (operating EBIT trend)
8. Is free cash flow positive and growing? (TTM FCF + 3-year trend)
9. Does it have too much debt? (Net Debt / EBITDA; interest coverage)
10. Who are the strongest competitors? (and what's their relative position?)
11. What would make you sell? (specific exit conditions, not "if it goes down")
12. What would prove your thesis wrong? (falsification conditions — same as the StockNews FF1-FF7 framework)

For long-term holds (5+ years), add 3 more from the durability test:

1. Will this business model still matter in 10 years?
2. Is the moat widening or eroding? Name the mechanism.
3. Has management deployed capital at ROIC > WACC over the past 10 years?

K.7 — Tax basics (US retail investor)

Stocks generate two types of taxable events:

Practical rules:

• Hold winners ≥1 year before selling (long-term cap gains rate)
• Realize losses to offset gains (tax-loss harvesting), but mind the wash sale rule (30 days)
• Use tax-advantaged accounts (Roth IRA, 401k, HSA) for highest-growth holdings
• Foreign withholding on ADR dividends can sometimes be credited; check W-8BEN status

This is a US-resident framework. Consult a tax professional for your specific situation.

K.8 — Best learning order

If starting from scratch, learn in this sequence:

1. P/E, forward P/E, market cap (Part B)
2. Revenue, earnings, free cash flow (Parts B-C)
3. Debt and balance sheet (Part C)
4. Margins (gross, operating) (Part C)
5. Competitive advantage / moats (K.5 + Mauboussin 2x2 in Part E)
6. Industry trends + business model (K.1, K.6)
7. Valuation vs competitors (Parts D + E)
8. Position sizing + portfolio rules (K.3)
9. ETFs vs individual stocks (K.3)
10. Tax basics (K.7)
11. Risk taxonomy (K.4)
12. The StockNews tree methodology (Parts A + E)

K.9 — Decision integration with the StockNews tree

When reading any StockNews tree report, the Section XII Investment Scorecard at the end synthesizes the H-0 thesis + durability test + 12-question checklist into a structured buy/hold/skip recommendation. That section is the place to look for the bottom line. Use this Part K to understand WHY the scorecard answers are what they are.

K.10 — Reading guide for "AI-sleeve" theses (AJNMY + TOTDY archetype)

A growing number of legacy industrial / consumer companies have been re-rated upward in 2025-2026 because a previously-ignored sub-segment of their business sells into the AI semiconductor capex cycle. The two canonical examples in this library are Ajinomoto (AJNMY) — best known as a Japanese seasonings company but holds ~95% global share of Ajinomoto Build-up Film (ABF), the dielectric film inside every advanced CPU/GPU substrate (NVIDIA Blackwell, AMD MI300, Intel Xeon, Apple M-series via TSMC, every hyperscaler custom silicon) — and TOTO (TOTDY) — best known as a bathroom-fixtures maker (Washlet bidet) but is the #2 global supplier of electrostatic chucks (ESCs) for cryogenic 3D-NAND etching tools sold to Lam Research and Tokyo Electron and ultimately consumed by SK Hynix / Samsung / Kioxia.

When you see this archetype, the analytical question always decomposes into the same four components:

1. Sleeve magnitude. What % of company revenue, OP, and market cap is the AI sleeve actually accounting for? At AJNMY the Functional Materials sub-segment is ~25-29% of FY25 BP and growing; at TOTDY Advanced Ceramics is now >50% of consolidated OP for the first time ever. The sleeve has to be material (>15% of OP run-rate trajectory in 3 years) for the SoTP rerating to be worth pursuing.

1. Sleeve durability. Is the AI exposure structural (multi-decade qualified position with switching costs) or cyclical (riding a particular capex wave)? ABF is structural — AI accelerator substrates have multi-year qualification cycles and end-customer co-funded capex confirms no practical alternative. ESCs for cryogenic NAND are partly cyclical (NAND capex pauses periodically) and partly structural (1988 mass-production vintage at TOTO; not easily displaceable on short timescales).

1. Mispricing mechanism. Why is the sleeve under-priced today? Two most common causes (per the StockNews mispricing taxonomy in Part E):

• Structural blindness on segment reporting — company doesn't break the sleeve out as a standalone segment, so generalist investors literally cannot see its economics. AJNMY is the canonical case; Functional Materials is buried inside "Healthcare & Others" alongside the unrelated bio-pharma CDMO. Resolution: standalone segment disclosure, often pushed by activist investors.
• Cognitive bias from peer-group anchoring — sell-side analysts who covered the stock for 20 years as a food / fixtures / seasoning company anchor on those peer multiples and adjust slowly. Resolution: re-classification by buy-side allocators, a slow multi-quarter process that activist campaigns accelerate.

1. Disclosure/pricing catalysts. The thesis only delivers if the sleeve's economics get converted from implicit to explicit. Watch for:

• Standalone segment disclosure (the easiest grant; costs management nothing operationally)
• Price increases on the sleeve product (medium difficulty; substrate-makers may absorb the increase to defend share rather than pass through)
• Capacity-expansion announcements (pure proof-of-belief from management)
• Spin-offs / partial-IPOs / strategic stakes (rare but most catalytic)

The "activist as catalyst" frame. Both AJNMY and TOTDY have public Palliser Capital activist campaigns (March 31 2026 and Feb 17 2026 respectively). When you encounter an activist campaign, treat it as both signal and catalyst: signal because professional analysts at Palliser have done the SoTP work and concluded the gap exists; catalyst because the public engagement raises the probability that management discloses / reprices the sleeve. Do not buy a stock SOLELY because an activist is involved — the campaign may stall, management may resist, and 6-12 months can pass with no resolution. Weight the activist's published thesis as ~Tier B evidence on the underlying mispricing claim, not a buy signal.

Glass substrate threat (specific to AJNMY). Glass-core IC substrates are an emerging 2026-2030 alternative to organic FR-4 cores. Critical clarification frequently missed by generalist investors: glass replaces only the substrate core, not the build-up dielectric layers above and below. The ABF film is laminated between copper wiring planes in the build-up stack regardless of whether the core is glass or organic. So glass substrates do NOT eliminate ABF demand; they only change the structural center plate of the substrate. The real long-tail existential risk to ABF is TSMC's CoPoS (panel-level glass) which targets 2028+ and could replace both the core AND the build-up layers — a 2029-2030 watch but not a 5-year-horizon material risk.

Cryogenic NAND etching (specific to TOTDY). 3D NAND now stacks 200+ layers (Samsung V8 → V9, SK Hynix 321L → 400L in 2026). Channel-hole etch through hundreds of layers requires Lam Research / Tokyo Electron cryogenic etchers operating below room temperature. TOTO's electrostatic chucks must hold flatness, particle counts, and clamping force at sub-zero temperatures while exposed to plasma. The "5-year competitive moat" Palliser describes for TOTO rests on (a) 38 years of mass-production process IP since 1988, (b) tool-platform qualification cycles that take 2-3 years to displace, (c) Japan's deep ceramic-process industrial base. Kyocera (6971.T) launched a competing high-durability ESC in June 2024 — credible long-term threat. NAND capex is famously cyclical; TOTO is ramping into the next NAND peak in 2027-2028 and depreciation drag in 2029-2030 is the cleanest bear case.

Position-sizing for AI-sleeve theses. Both AJNMY and TOTDY are recommended at 2-5% portfolio weight (per Section XII), with the upper end reserved for post-catalyst-confirmation. The AI-sleeve archetype has a mean-reversion-on-disappointment risk that punishes oversize positions: if the sleeve story unwinds (management refuses disclosure, activist exits, hyperscaler capex pauses), the multiple compresses fast because you effectively bought a "specialty chemicals + AI sleeve" story that the market reverts to pricing as "Japanese food / fixtures." A 5% maximum on first conviction protects against the 30-40% downside in the bear scenario.

ADR mechanics for AI-sleeve Japanese names. AJNMY is sponsored Level I (1 ADR ≈ 1 underlying common share, BNY Mellon depositary; ~$673K/day USD ADV). TOTDY is unsponsored (1:1 also; thinner liquidity, broker-route conversion). Both face 15% Japan dividend withholding under the US-Japan tax treaty. Both have ADV in the $200-700K/day range — fine for personal-account sizing, painful for institutional blocks. Yen weakness eats ADR returns even when the underlying TSE share rises (e.g., over the 1Y to May 2026, ~21pp of 2802.T outperformance vs AJNMY was FX).

Last updated: 2026-05-02. Maintained by the StockNews system + owner (Ming).