The Five Questions & the Source Stack

Phase 8 — The Rapid-Analysis System (Signature Skill) · Week 64 · ~20 hours (≈8 h teaching content, ≈4 h practice set, ≈6 h timed navigation drills + mini-project, ≈1 h mastery check)

Where it sits. This is the first module of Phase 8, the phase the whole program has been serving. Behind you: fifty-six weeks. You can read a filing (Phase 1), distrust it (Phase 2), value it (Phase 3), judge its moat (Phase 4), match it to its sector lens (Phase 5), know your own temperament (Phase 6), and read the macro weather (Phase 7). You have also run the company of the week since week 4, roughly fifty scaffolded reads, each a little less guided than the last. What changes now is that the method becomes explicit and fast. Not more knowledge: compression. It takes everything you can already do slowly and teaches you to do the first, decisive 20% of it in under an hour, so that the teardown (M8.02) has a foundation and the deep dive (M8.03–M8.04) has a spine.

Two things are finalized here: the five questions as the automatic shape of your thinking, and the source stack, the exact set of documents you open, in the exact order, for any company on Earth. The reusable 45–60 minute statement-read checklist (seeded as v1 in M1.10) is hardened into its final field form as the artifact you carry out.

Learning objectives

By the end you can:

1. State the five questions from memory and, for each, name what it really asks, which prior-phase skills answer it, and what "good" looks like, so that any business you meet is immediately five questions rather than a blank page.
2. Explain the ROIC-above-WACC-plus-reinvestment master idea as the single organizing principle the five questions serve, and show why growth without that spread destroys value.
3. Assemble the four-layer source stack in order (the company's own story, the transcripts, ten years of numbers, the competitor-and-bear view) for both a US and an Indian company, knowing what each layer is highest-signal for and where to get it free.
4. Skim a US 10-K and an Indian annual report in the professional read order in 45–60 minutes: Business → Risk Factors → MD&A → the three statements → segment note → cash-flow → related-party note → contingent-liabilities note → auditor's report (incl. India CARO) → India shareholding-and-pledge table, extracting the right thing at each stop.
5. Apply the skim/ignore triage (read ~30 pages deeply, skim ~50, ignore the rest knowing what you are ignoring) and defend why boilerplate, generic risks, MD&A adjectives, and property lists are ignorable.
6. Run the Ctrl-F tripwire list ("related party," "pledge," "contingent," "going concern," "restated," "auditor," "exceptional," "one-time," "guarantee," "receivable") and explain why each word is a tripwire and what it triggers.
7. Do a rough ROIC-vs-WACC read and a rough "what's priced in" read in seconds from screener/aggregator numbers, as the two gut-checks that orient the whole teardown.
8. Navigate a named real filing under time pressure to answer targeted questions ("find X in this filing in N minutes"), and produce the 8-line first-impression output every time.

Prerequisites & connections

Builds on. M0.01 first stated the five questions and the ROIC>WACC idea; here they get their full toolkit. M1.10 taught the anatomy of both filings, the spreading template, and seeded the statement-read checklist v1 (§4.14) that becomes this module's artifact. All of Phase 2 lives inside the read order (the red-flag Ctrl-F list is the ★ pass of the quality-of-earnings checklist). Phase 3 supplies the "what's priced in" reverse-multiple read; Phase 4–5 supply the moat and sector lenses you apply in seconds; Phase 6 gives you the discipline to read the bear case honestly. The ~50 weekly reads are the reps that make the read order feel like walking, not spelling.

Feeds into. M8.02 (the 1–2 hour teardown) runs on the source stack and fast read you build here, which is its first thirty minutes. M8.03–M8.04 (the 2–3 day deep dive) expand every layer: the ten-year numbers get rebuilt by hand, the transcripts multiply to 4–8, the bear note becomes a full steelman. M8.05 certifies the whole system under time pressure. M8.06 takes the source stack past the annual report and the 10-K into the six documents the stack keeps naming but never opens. Those are the draft red herring prospectus, the quarterly result, the earnings release with its deck and transcript, the 8-K, the ownership filings and the rating rationale. The artifact (the 45–60 minute checklist) becomes a permanent field tool you run weekly for decades.

Competency map. This is the foundation of C10, the signature skill (run the teardown and deep dive on a novel company). It directly serves C1 (read the language of business, now fast) and C3 (understand a business model cold, now from the right sources in the right order).

Core teaching content

4.1 Why speed is a skill, not a shortcut

Thirty pages out of three hundred. That is what a professional first read of a filing actually covers, and the other 90% is ignored on purpose.

There is a lazy fast and a skilled fast, and they look nothing alike. The lazy fast skims everything equally, reads the glossy, quotes the adjusted number, and forms an opinion from the investor deck. The skilled fast reads the right thirty pages, in the right order, knowing exactly what each page can and cannot tell it. Speed, for an analyst, is not reading faster. It is knowing what to ignore, and being right about it.

You could not have started here. The whole program was the slow path: spreading by hand, deriving CFO from balance-sheet deltas, building WACC from a blank sheet. That order exists so that speed, when it arrives, rests on understanding rather than replacing it (Working Agreement 2). A radiologist reads a scan in seconds because she spent years reading them in minutes; the seconds are earned. When you skim the cash-flow statement now and your eye stops on a CFO that trails PAT for the third year, that stop is not luck. It is Phase 2 compressed into a reflex.

Phase 8 is where fifty-six weeks become reflexes. The compression has a shape, and the shape is five questions.

4.2 The five questions: the spine, with its full toolkit

Every technique in this program is machinery for answering five questions about a business faster and more reliably than the market. You memorized them in week 1. Now you own the tools behind each one. Say them until they are the automatic shape of your thinking, the thing your mind does the instant it meets a company, before it knows anything else:

1. What does it do, and how does it make money?
2. Is it a good business — is ROIC durably above the cost of capital?
3. Is it run by good, honest, aligned people who allocate capital well?
4. What can kill it?
5. What is the price implying, and is that beatable?

Here is what each question really asks, and which prior-phase skills answer it:

Notice the architecture. Q1 and Q2 are the business. Q3 is the people. Q4 is the survival. Q5 is the price. A complete analyst answers all five; most amateurs answer only Q1 (a story) and a distorted Q5 (a P/E they didn't reverse). The skill this phase builds is running all five, fast, in a fixed order, from a fixed set of documents.

4.3 The master idea: ROIC > WACC and reinvestment

The five questions are not an arbitrary list. They are the five things you must know to judge one number and its durability: the spread between the return a business earns on the capital it employs and the cost of that capital, and whether the business can pour more capital into that spread.

Write it as the organizing principle of the entire program:

Value is created only when  ROIC > WACC  AND the company can reinvest at that spread.

• ROIC (return on invested capital) = NOPAT / Invested Capital, the after-tax operating profit a business earns divided by the capital tied up to earn it (M2.03).
• WACC (weighted-average cost of capital) is the blended return debt and equity providers require to fund it (M3.03).
• The spread ROIC − WACC is the value created per rupee of capital. Positive spread → each rupee employed is worth more than a rupee. Zero spread → the business runs to stand still. Negative spread → growth destroys value: every rupee reinvested comes back worth less than a rupee, and the faster it grows, the faster it burns (this is the single most counter-intuitive truth in finance, and M3.07 proved it with the key-value-driver formula Value = NOPAT·(1 − g/ROIC)/(WACC − g)).
• Reinvestment is the multiplier. A 20% ROIC business that can redeploy only a trickle is a cash cow you must reinvest yourself; a 20% ROIC business with a long runway to reinvest half its earnings at 20% compounds internally at ~10% before you lift a finger. g = reinvestment rate × ROIC (M3.04). The rare, wonderful machine is high spread and a long runway.

Now re-read the five questions and see them for what they are: a disciplined way to judge that spread and its durability faster than the market.

• Q1 finds where the return is earned (the profit engine).
• Q2 measures whether the return beats the hurdle, and Phase 4's moat work asks how long it can last. The durability of the spread is the moat, expressed in numbers.
• Q3 asks whether management will reinvest the spread wisely or squander it (capital allocation is the reinvestment decision).
• Q4 asks what could collapse the spread (a killed moat, a cycle, a fraud).
• Q5 asks what durability and reinvestment the market has already paid for, because a wonderful business at a terrible price is a bad investment, and the spread only helps you if you don't overpay for it.

Every ratio, every model, every sector KPI is machinery for judging that spread and its durability faster and more reliably than the market. Keep this sentence in front of you for the rest of Phase 8. When a teardown feels like a checklist of disconnected tasks, it isn't: every task is aimed at the spread.

4.5 How to skim a 10-K / Indian annual report fast: the exact read order

Inside Layer 1's filing, you do not read top to bottom. You follow a fixed read order that front-loads signal and back-loads (or skips) boilerplate. This is the beating heart of the artifact you are building. The order, and what you are hunting at each stop:

Two notes on the order. First, it is not strictly linear in practice (a KAM at stop 9 will send you back to the note behind it), but you start in this sequence every time, so nothing high-signal is left to chance. Second, the US and India stops are the same analytical questions in different document furniture: "where is the profit engine" is the segment note in both; "who is being paid and how" is the DEF 14A in the US and the Board's Report + shareholding pattern in India. The furniture changes; the questions never do.

The skim/ignore list: what you deliberately do not read. Ignoring the right things is the whole game. On a first read you ignore:

• Legal and CSR boilerplate: the CSR annual report, the "value creation model" infographics, the sustainability prose. (BRSR is awareness-level, M1.10; the s.135 CSR spend is one number in the Board's Report, not a page of reading.)
• Generic risk factors: "competition may adversely affect results," "we depend on key personnel," "economic conditions may change." These are legal cover, identical across ten thousand filings. Read Item 1A only for the risks that carry a specific number, name, or date.
• MD&A adjectives: "robust," "resilient," "industry-leading," "well-positioned." A paragraph with no number is filler at best, camouflage at worst. Read MD&A for the numbers and the promises, not the prose.
• Properties lists: the table of factories and their square footage. (Exception: a Ctrl-F on "title deeds" in India's CARO clause (i), because promoter-asset entanglement hides there.)
• The chairman's glossy letter. Read it once, last, for tone and promises you will test, never for information.

Reading ~30 pages deeply, skimming ~50, and ignoring 200 on purpose is not laziness; it is the professional's triage. The amateur reads everything with equal attention and therefore reads nothing with enough.

4.7 Timed navigation walkthrough: one US 10-K

Nothing teaches the read order like doing it against a clock on a real document. Below is a full navigation of a US 10-K, minute by minute. The company is Costco Wholesale (ticker COST; files a 10-K with the SEC, fiscal year ending early September). Figures are illustrative, modeled on ~FY24 scale; pull the actual 10-K from EDGAR (sec.gov/cgi-bin/browse-edgar, or EDGAR full-text search). Costco is chosen for one beautiful reason you will meet at minute 12.

Minutes 0–2: Ctrl-F sweep. Open the 10-K; run the ten tripwires. "Related party" → routine (some real-estate JVs, small). "Guarantee," "contingent" → standard commitments note, nothing large vs equity. "Going concern," "restated," "resign" → no hits. "Exceptional/one-time" → minimal. Clean sweep. Two minutes well spent: you now know there is no landmine hiding, so you can read for business quality rather than fraud.

Minutes 2–7: Item 1, Business. Costco operates membership warehouse clubs. It sells a limited assortment (~3,800 SKUs vs a supermarket's 30,000+) at razor-thin markups, in no-frills warehouses, to members who pay an annual fee. Segments: US, Canada, Other International; plus e-commerce. Note the numbers Item 1 volunteers: ~890 warehouses; membership renewal rate ~90%+ in the US/Canada. Write the one-liner: "A membership retailer that sells goods near cost and makes its money on the membership fee — a scale-economies-shared machine."

Minutes 7–12: Item 1A, Risk Factors, hunted not read. Skip "competition is intense," "we depend on key personnel," "economic conditions." Stop on the specific: dependence on a smooth relationship with a few large suppliers; the concentration of membership income (if renewal rates fell, the profit model bends); international/FX exposure. Three specific risks logged in ninety seconds; twenty pages of boilerplate ignored.

Minutes 12–18: Item 8, the statements + the segment/income detail. The teaching moment. Here is why Costco. Total revenue ≈ $254bn, of which merchandise sales ≈ $249bn and membership fee income ≈ $4.8bn. Operating income ≈ $9.3bn. Do the arithmetic that separates an analyst from a reader: membership fees (~$4.8bn) are roughly half of all operating income (~$9.3bn), on ~2% of revenue. The merchandise business, 98% of the top line, runs at a wafer-thin margin on purpose, to drive traffic and renewals. The revenue engine (merchandise) is not the profit engine (memberships). That is Q1's real question answered in one calculation, and you would have missed it entirely reading only the headline revenue line. This is the pattern the read order is built to catch, and it recurs everywhere (a bank's fee income vs spread income; a carmaker's financing arm; a mall's anchor vs specialty rents).

Minutes 18–24: cash flow + a rough ROIC read. CFO comfortably exceeds net income (a retailer collecting at the register while paying suppliers later, which is negative working capital, the best kind); capex is real (new warehouses) but FCF is solidly positive. From stockanalysis.com's ten-year sheet (Layer 3, open in a second tab): ROIC durably in the high teens to low-20s%, well above a ~7–8% US WACC. Q2: yes, a good business, and the spread has persisted for a decade. Share count drifting slowly down; occasional special dividends. Balance sheet net-cash-ish.

Minutes 24–28: auditor + governance. Auditor's report: unmodified. CAM (recent filings): typically none or a single low-drama matter. For a business this clean, the CAM section is short, which is itself information. Part III points to the DEF 14A for compensation and ownership, noted for the teardown (Q3), not opened now.

Minutes 28–30: write the 8-line output. Business: membership warehouse retailer, ~$254bn revenue. Shape: mid-single-digit-plus revenue growth, stable thin merchandise margin. Cash: CFO > PAT, FCF positive, negative working capital. Balance sheet: net cash, buying back slowly. Profit engine: membership fees, ~half of operating income on ~2% of revenue; renewals ~90%+. Quality: clean auditor, clean Ctrl-F sweep. Risks that matter: renewal-rate erosion; supplier/FX shocks. Questions I can't yet answer: what is priced in at today's multiple (Q5, reverse it in the teardown), and how durable is the membership moat internationally. Thirty minutes; you understand Costco better than 90% of the people who own it.

4.8 Timed navigation walkthrough: one Indian annual report

Now the same discipline on the other document furniture. The company is Asian Paints (NSE/BSE; files an Indian annual report under Ind AS, fiscal year ending March). Figures are illustrative, modeled on ~FY24 scale; pull the actual annual report from the company's investor-relations page or BSE/NSE. The Indian AR is a different animal from the 10-K: 300+ pages, glossy front, standalone and consolidated statements, a by-nature P&L with no gross-profit line, and a shareholding pattern you will not find in any US filing.

Minutes 0–2: Ctrl-F sweep. Ten tripwires. "Pledge" → promoter pledge ~0% (the founding families hold cleanly, a good baseline). "Related party" → routine (some dealings with associates; total modest vs turnover). "Contingent" → disputed indirect-tax and income-tax matters, small vs net worth. "Exceptional" → minimal. "Going concern/restated/resign" → none. Clean. Note: on an Indian AR you always run "pledge" first. It is the single most important India-specific search.

Minutes 2–8: MD&A + corporate overview (India's "Item 1"). Asian Paints is India's largest paints company; decorative paints (the paint you buy for your home) dominate the mix, with smaller industrial coatings, an international business, and a newer home-improvement/kitchen-and-bath push. The moat, in the MD&A's own hints: an unrivalled distribution network (tinting machines in tens of thousands of dealer outlets) and a formidable supply-chain/technology backbone. One-liner: "India's dominant decorative-paints franchise; the moat is distribution density and brand, expressed as high, persistent ROCE."

Minutes 8–13: the segment note (the profit engine). Indian segment reporting splits the business into paints/coatings vs home improvement vs (sometimes) international. The read: decorative paints in India is the overwhelming profit engine; the newer segments are strategically interesting but small and lower-margin. Q1 answered: don't be distracted by the home-improvement story in the glossy. The money is in Indian decorative paint, and its margin is the number that matters.

Minutes 13–20: the statements, Schedule III style. Remember the format traps (M1.10): the P&L is by nature, so you construct gross profit and EBIT yourself from "cost of materials consumed + purchases + changes in inventory" and the rest; and "other income" (treasury on a cash-rich balance sheet) sits inside total income, so split it out before quoting an operating margin. Illustrative shape: revenue ≈ ₹35,000 crore; EBITDA margin ~18–21% (sensitive to crude-linked input costs: titanium dioxide and monomers); PAT converting well to CFO; near-zero net debt. Analyze the consolidated; glance at standalone for the dividend and any group loans (M1.10, Rule 2).

Minutes 20–24: ROCE read + cash. From screener.in's ten-year sheet (Layer 3): ROCE durably ~25–30%, far above a ~12–13% India WACC. Q2: an excellent business, with a spread that has persisted through input-cost cycles. That persistence is the moat, in numbers (Phase 4). CFO tracks PAT; FCF positive; dividends steady; share count flat.

Minutes 24–28: auditor, CARO, shareholding. Auditor's report: unmodified. CARO 30-second scan over clauses (ii)(b) stock-vs-bank statements, (iii) group loans, (vii) statutory dues, (ix) defaults, (xi) fraud, (xvii) cash losses, (xviii) auditor resignation: "no exceptions noted" across the board for a clean company (half a minute). Shareholding pattern (Reg 31): promoter (the four families) ~52%, pledge ~nil; healthy FII/DII holding. Contrast this with the US 10-K, which has no equivalent promoter/pledge table. That stop simply does not exist on EDGAR, and its presence is why India analysis starts with ownership.

Minutes 28–30: write the 8-line output. Business: India's #1 paints company, ~₹35,000 cr revenue, decorative-led. Shape: steady double-digit-ish revenue growth, margin swinging with input costs. Cash: CFO ≈ PAT, FCF positive, net cash. Balance sheet: pristine, dividend-paying. Profit engine: Indian decorative paints (distribution + brand moat). Quality: clean auditor, clean CARO, ~0 pledge, promoters ~52%. Risks that matter: crude-linked input-cost spikes compressing margin; new well-funded entrants attacking distribution. Questions I can't answer yet: what growth the premium multiple is pricing in (Q5), and how much the new competitor will actually dent decorative share. Thirty minutes; you have a structured, defensible first impression of a company you'd never seen.

4.9 Two thirty-second gut-checks that orient everything

Before you leave Layer 3, run two rough reads. They are not the valuation; that is the teardown (M8.02) and the deep dive's reverse DCF (M8.03). They are the orientation: are we looking at a good business, and is the market already excited about it?

Gut-check 1: ROIC vs WACC (Q2 in one line). Read ROIC/ROCE off the ten-year sheet; compare to a rough WACC (~7–9% for a stable US large-cap, ~11–13% for India; you build the real number in the teardown). If ROIC sits several points above WACC and has for years, you have a good business and the rest of the work is about durability and price. If ROIC hugs or trails WACC, no amount of growth story rescues it. Growth is destroying value, and Q5 had better be pricing in a turnaround you can defend.

Gut-check 2: reverse the multiple (Q5 in one line). You cannot do a full reverse DCF in thirty seconds, but you can do the crude version. For a business that pays out most of what it earns, P/E ≈ 1/(r − g), so the market's implied perpetual growth is roughly g ≈ r − (1/PE). A P/E of 25 at a 10% required return implies 10% − 4% = ~6% perpetual growth; a P/E of 50 implies ~8%; a P/E of 12 implies ~1.7%. Say the number out loud ("at this price the market is paying for ~6% growth forever") and ask whether that's obviously too high, obviously too low, or plausible. (Caveat, loudly: this ignores reinvestment and ROIC, because a high-ROIC compounder deserves a higher multiple at the same growth, which is exactly why the real reverse DCF, P/E ≈ (1 − g/ROIC)/(r − g), comes in M3.05. The crude version is a gut-check, not a valuation. Never quote it as one.)

These two reads, together, are the entire teardown in miniature: good business? and already priced as one? Everything in M8.02 is the careful version of these two sentences.

Common mistakes & how experts think differently

1. Starting with the numbers instead of the business. Opening the ten-year screener before you can narrate what the company does gives you ratios with nothing to hang them on. Experts read Layer 1 first, the story, precisely so that every ratio in Layer 3 means something. A 25% ROCE is a fact; "a 25% ROCE from distribution density that would cost a rival a decade and a fortune to replicate" is an insight. The narrative makes the number legible.
2. Reading the filing front-to-back (glossy first). The chairman's letter and the value-creation infographics are the front of the document and the back of the priority list. Experts read in signal order (Business, risks, MD&A, statements, notes, auditor) and read the glossy last, if at all, as a record of claims to test.
3. Rebuilding the numbers by hand in a teardown. Spreading a filing by hand is a Phase 1–2 discipline and a deep-dive step (M8.03), and the wrong tool for a 60-minute read. The learner who insists on hand-building a ten-year model in the teardown has confused rigor with slowness. You earned the aggregator; in the teardown, use it. Rigor returns in the deep dive.
4. Treating the investor deck as truth, or dismissing it as spin. Both are lazy. The deck is simultaneously the highest-signal summary of the business and the most curated document management produces. Experts mine it for the model and the KPIs, then verify every number against Layer 3 and the filings, and never quote a deck figure they haven't confirmed.
5. Confusing fast with shallow. Speed does not come from reading every page quickly; it comes from reading the right 10% deeply and ignoring the rest on purpose. The amateur skims 300 pages evenly and retains nothing; the expert reads 30 pages hard and ignores 270 with full awareness of what's in them.
6. Skipping the transcript. The single highest-signal-per-minute source, and beginners skip it because it isn't a "filing." Experts read the latest Q&A before almost anything else. It hands you the soft spots (what analysts push on), the framing (what management leads with), and the tone (do they answer straight) that would otherwise take days to surface.
7. Reading only the company's own story. A read built entirely from the 10-K and the deck is a bull case wearing a lab coat. Experts deliberately fetch one bear note (Layer 4) to import the disconfirming view, then verify its claims against the filings rather than believing them. The point is not to agree with the bear; it is to have met the bear before you commit.
8. Over-reading the risk factors. Item 1A / the risk report is 90% legal boilerplate designed to be un-sue-able, not informative. Experts hunt it for the 2–3 risks carrying a specific number, name, or date and ignore the generic scaffolding, the opposite of the anxious beginner who reads all forty risks with equal dread.
9. Missing the segment note, and so missing the profit engine. The biggest revenue line is frequently not where the money is made (Costco's memberships; a bank's fee vs spread income; a conglomerate's one crown jewel funding four also-rans). Experts go to the segment note early, every time, and ask "which segment earns the profit," not "which is biggest."
10. (India) Reading consolidated only and skipping the shareholding pattern. Consolidated is the economic entity, but dividends come from standalone profits and group loans hide in standalone notes, and the promoter-and-pledge table has no US analog and no substitute. Experts run "pledge" first on any Indian AR and read the encumbrance trend before the board page.
11. Not writing the output. A read with no written artifact evaporates by the next morning. Experts force the 8-line output every time, no exceptions, because the discipline of compressing an hour of reading into eight lines is where the understanding actually consolidates, and because next quarter you will want to diff it.

The meta-difference: amateurs read a filing to extract facts; experts read it to answer five questions about a spread and its durability, in a fixed order, from a fixed stack, fast enough to do it every week. The facts are raw material; the five questions are the point.

Worked examples

Worked example 1: Mapping the five questions to their sources (a US compounder)

Task. You are handed Visa Inc. (illustrative, modeled on ~FY24 scale; pull the 10-K). In ten minutes, map each of the five questions to the one source that best answers it and give a one-line answer. This is the skill of knowing where to look, which is half of being fast.

The lesson. Notice that no single document answers more than one question well. The 10-K's Item 1 nails Q1 but is useless for Q5; the ten-year sheet nails Q2 but says nothing about Q4's specific killers. Being fast is knowing which of the five doors each question lives behind. The source stack is simply those doors, arranged in opening order.

Worked example 2: Assembling the source stack for an Indian company

Task. Build the complete source stack for Titan Company (India; jewellery, watches, eyewear; illustrative). List the exact documents, in order, where to get each free, and the one thing each is highest-signal for.

1. Layer 1a, investor presentation (Titan IR page): the segment mix (jewellery dominates) and the KPIs management watches (studded ratio, store adds, same-store growth). Highest signal for: the business model on one page. Believe none of the numbers yet.
2. Layer 1b, MD&A + segment note (latest annual report, IR page or BSE): jewellery vs watches vs eyewear economics; the profit engine (jewellery). Highest signal for: Q1, the real profit engine.
3. Layer 2, latest 1–2 concall transcripts (screener.in "concalls" tab, free; or Titan IR): what analysts push on (gold-price volatility, competitive intensity in jewellery, margin), and management's tone on studded-jewellery mix. Highest signal per minute.
4. Layer 3, ten-year numbers (screener.in, free): revenue growth, EBIT margin, ROCE (is the spread there and durable?), CFO vs PAT, working-capital intensity (jewellery ties up gold inventory), net debt, share count.
5. Layer 4a, one competitor's deck (a listed jewellery peer): the shared KPIs and how Titan differs (brand, trust, organized-retail share gain from the unorganized market).
6. Layer 4b, one bear note (a broker "sell" note or a ValuePickr thread): the strongest case against: gold-price risk, competition, valuation. Borrow the questions; verify against filings.

Then the two gut-checks. ROCE off the screener vs a ~12% India WACC (Q2); reverse the multiple to state the implied growth (Q5). Total assembly time: ~15 minutes, mostly downloads. You are now ready for the teardown (M8.02) with every door already open.

Worked example 3: A Ctrl-F tripwire hunt (synthetic)

Setup. A synthetic Indian mid-cap, "Vikram Infra Ltd." You run the ten tripwires on its annual report PDF. Five hits; here is the read on each:

The cluster rule (Phase 2). One flag is a question; these five align. Pledge + RP loans + group guarantees + bloated contingents + serial exceptionals all point the same direction (a stressed promoter using the listed company as an ATM). That is not five questions; it is one position decision: this needs deep governance work before it earns another minute, and likely fails the L4 screen. Two minutes of Ctrl-F did what an hour of the glossy never would.

Worked example 4: The 60-second business-quality read

Setup. Three companies, ten-year averages off the aggregator (synthetic, clean numbers). WACC ≈ 11% for all. Which are good businesses, and what does each imply for Q5?

Step-by-step for B (the trap). ROIC = 180/2,000 = 9%. WACC = 11%. Spread = 9% − 11% = −2%. Every ₹1 reinvested returns 9% but costs 11%, so it comes back worth less than ₹1. At 15% growth, B is reinvesting heavily into a negative spread: the income statement shows a growing company; the value is shrinking. This is the single most important thing the 60-second read protects you from: the seductive fast-growing value-destroyer. Company A grows slower on the top line but creates value with every rupee; Company C creates value but has nowhere to put it. Growth is only good above the hurdle. That sentence, applied in sixty seconds, is Q2 and half of Q5.

Worked example 5: Reconciling a deck's "record margin" against the filing (India)

Setup. Rohan Specialty Chemicals Ltd (synthetic). The latest investor presentation (Layer 1) headlines "Record EBITDA of ₹1,100 crore — a 22% margin." Revenue for the year is ₹5,000 crore. This is exactly the moment the source stack exists for: Layer 1 makes a claim; you reconcile it against the notes before you believe a single digit. (All numbers verified in Python.)

Step 1: the headline, taken at face value. ₹1,100 cr ÷ ₹5,000 cr = 22.0%. Impressive, and the number you would quote if you stopped reading at the deck (the amateur's whole analysis).

Step 2: strip the "other income." The Ctrl-F sweep and the P&L show ₹150 cr of the ₹1,100 cr is other income: treasury yield on the cash pile, not the chemicals business (the Asian Paints trap of §4.8: on an Indian AR you split "other income" out before quoting an operating margin). Operating EBITDA so far: 1,100 − 150 = ₹950 cr.

Step 3: re-include the recurring "exceptional." The deck's EBITDA is struck after adding back ₹120 cr of "exceptional items." Ctrl-F "exceptional" → the notes show this line has appeared four years running, an operating cost in a costume (§4.6). Put it back: 950 − 120 = ₹830 cr of clean operating EBITDA.

Step 4: the reconciled number and the size of the gap. Clean margin = 830 ÷ 5,000 = 16.6%. The deck's 22.0% overstates the real operating margin by 5.4 percentage points; in rupees, ₹1,100 − ₹830 = ₹270 cr of "EBITDA" is non-operating income plus a serial charge dressed as a one-off. The deck is flattering operating profit by about a third (270 ÷ 830 ≈ 32.5%).

The lesson. Two tripwires ("other income," "exceptional"), each worth thirty seconds, moved the headline margin down a third of its distance to nothing. The deck is highest-signal for what the business does and most advocacy-laden for how well. Mine it for the model, then verify every number against the filing. The reconciled 16.6% is the figure that enters Q2; the 22.0% never does.

Worked example 6: Quantifying a source-conflict: what is really priced in (US)

Setup. Cedar Payments Inc. (synthetic; US $). Off the ten-year sheet (Layer 3): NOPAT $480m on invested capital $2,000m; WACC ≈ 9%. The stock trades at a P/E of 25. A sell-side "buy" note (Layer 4) argues: "At 25×, the market is pricing in only ~3% perpetual growth — undemanding for a franchise this good." Two questions: is it a good business, and is the note's "3%" a number you can use? (Arithmetic verified in Python.)

Step 1: Gut-check 1, ROIC vs WACC (Q2). ROIC = 480 ÷ 2,000 = 24.0%; spread = 24.0 − 9.0 = +15.0 pts, and (stipulated) durable → an excellent business. The note's premise ("a franchise this good") checks out.

Step 2: Gut-check 2, reverse the multiple yourself (Q5). Using the crude payer approximation g ≈ r − 1/(P/E) with your own required return r = 9%: g ≈ 9% − 1/25 = 9% − 4% = ~5.0%. The market is paying for ~5% growth forever, not 3%.

Step 3: back out the note's hidden assumption. For the note's "3%" to be right, 3% = r − 1/25 = r − 4%, so it silently used r = 7%. That is the entire disagreement: the note never disclosed its required return, and a 2-point difference in r is a 2-point difference in implied growth. A borrowed "what's priced in" number is meaningless without the r behind it.

Step 4: the honest caveat, both ways. The crude formula ignores reinvestment and ROIC: a 24%-ROIC compounder justifies a higher multiple at the same growth, so the real reverse DCF (P/E ≈ (1 − g/ROIC)/(r − g), M3.05) would show even 5% is not especially demanding. The note's conclusion ("undemanding") may well be right, but you reached it with your own arithmetic and your own r, and you would say "the market is pricing ~5% growth at my 9% hurdle," never quote the note's 3%.

The lesson. Layer 4 hands you a hypothesis ("cheap for the quality"); you take the question and compute the answer yourself. The trap is quoting a reverse-multiple figure whose assumed r you never saw. It is the same discipline that forbids quoting a deck's adjusted margin.

Practice set

Work each in full before reading the solution (Working Agreement 3). Problems ramp guided → independent → timed.

P1 (guided). Match each of the five questions to the prior-phase skill(s) that most directly answer it: (a) reverse DCF / expectations; (b) ROIC build + WACC + moat durability; (c) business-model anatomy + sector lens; (d) capital-allocation record + proxy/governance read; (e) red flags + capital cycle + sector risks.

Solution. Q1 → (c); Q2 → (b); Q3 → (d); Q4 → (e); Q5 → (a). If you hesitated on Q2 vs Q4, the distinction is: Q2 asks is the spread there and how durable (moat = durability of the spread); Q4 asks what could collapse it (the threat to durability). They are two sides of the same coin: the moat and the thing that kills the moat.

P2 (guided). State the master-idea inequality and explain, in two sentences, why a company growing revenue 20% a year can still be destroying value.

Solution. Value is created only when ROIC > WACC and the company can reinvest at that spread. If ROIC < WACC, every additional rupee the company invests returns less than it costs, so it comes back worth less than a rupee; growing 20% simply means it is destroying value faster, because it is shovelling more capital into a negative spread. Growth is a multiplier on the spread's sign — wonderful when positive, ruinous when negative.

P3 (independent). For each Ctrl-F hit, state what the word warns of and your next action: (a) "going concern"; (b) "restated"; (c) "unbilled" (rising faster than revenue); (d) "auditor" near "resigned."

Solution. (a) Going concern: management/auditor doubt about surviving 12 months; read the full paragraph and management's plan; treat as near-fire-alarm. (b) Restated: previously-reported numbers were wrong; find what/how much/why and check for CFO/auditor turnover around it. (c) Unbilled/contract assets rising faster than revenue: possible premature or percentage-of-completion revenue; cross-check DSO and read the revenue-recognition policy. (d) Auditor resigned: highest-signal event; read the exit reason; "information not provided" ≈ fire alarm; a scheduled Companies Act rotation is routine and benign.

P4 (independent). Put these filing stops in the correct read order: contingent-liabilities note; Business (Item 1 / MD&A); auditor's report; the three statements; Risk Factors; segment note; cash-flow statement; related-party note; MD&A; shareholding & pledge (India).

Solution. (1) Business → (2) Risk Factors → (3) MD&A → (4) the three statements → (5) segment note → (6) cash-flow statement → (7) related-party note → (8) contingent-liabilities note → (9) auditor's report → (10) shareholding & pledge (India). The logic: understand the business, then its self-declared risks and story, then the shape of the numbers, then where the profit is (segments) and whether it's cash (CFO), then the two governance landmines (RP, contingents), then the gatekeeper (auditor/CARO), then (India only) ownership.

P5 (independent). In a US 10-K, where (which Item) do you find: (a) the auditor's opinion and CAMs; (b) the monthly buyback table showing price paid; (c) executive compensation metrics; (d) segment profit; (e) "material weakness" in internal controls?

Solution. (a) Item 8 (financial statements + auditor's report). (b) Item 5 (Market for Common Equity). (c) Not in the 10-K; the DEF 14A proxy (Part III of the 10-K incorporates it by reference). (d) Item 8, the segment note. (e) Item 9A (Controls and Procedures). Knowing the furniture cold is what makes the read fast.

P6 (independent). In an Indian annual report, where do you find: (a) promoter pledge %; (b) the full related-party ledger; (c) the auditor's clause-by-clause answers on group loans and defaults; (d) the ratios that changed >25% and why; (e) whether dividends can be paid?

Solution. (a) The shareholding pattern (SEBI LODR Reg 31, quarterly; also in the AR's corporate-governance section). (b) The related-party note in the financial statements, plus AOC-2 in the Board's Report for non-arm's-length/material RPTs. (c) CARO 2020 annexure to the standalone auditor's report: clause (iii) for loans, clause (ix) for defaults. (d) MD&A, the LODR Schedule V mandated ratio table (and the Schedule III ratio note). (e) The standalone statements; dividends are declared out of standalone profits (Companies Act s.123).

P7 (numeric, ±0.5 pt). A company reports NOPAT of $420m on invested capital of $2,800m. Its WACC is 8.5%. (a) Compute ROIC. (b) Compute the spread. (c) It plans to grow revenue 10% a year by heavy reinvestment. Value-creating or value-destroying, and why?

Solution. (a) ROIC = 420 / 2,800 = 15.0%. (b) Spread = 15.0% − 8.5% = +6.5 pts. (c) Value-creating: every dollar reinvested earns 15% against an 8.5% cost, so it comes back worth more than a dollar; growing 10% by reinvesting into a +6.5-pt spread compounds value. (Contrast: had ROIC been 7%, the same 10% growth would destroy value.) The growth is good because the spread is positive — the growth rate alone told you nothing until you knew the spread's sign.

P8 (numeric, ±0.5%). Using the crude payer approximation g ≈ r − 1/(P/E), estimate the perpetual growth the market is implying for: (a) P/E 20, r = 10%; (b) P/E 40, r = 10%; (c) P/E 10, r = 9%. Then state the one-sentence caveat that keeps this honest.

Solution. (a) g ≈ 10% − 1/20 = 10% − 5% = ~5%. (b) g ≈ 10% − 1/40 = 10% − 2.5% = ~7.5%. (c) g ≈ 9% − 1/10 = 9% − 10% = ~−1% (the market is pricing decline). Caveat: this ignores reinvestment and ROIC (a high-ROIC compounder justifies a higher multiple at the same growth), so it is a gut-check on optimism, never a valuation; the real reverse DCF (M3.05) uses P/E ≈ (1 − g/ROIC)/(r − g).

P9 (timed, 10 minutes). For HDFC Bank (India) and Moody's (US), list the exact source stack you would assemble: name each document/tool and where you'd get it free, Layers 1–4.

Solution (model). HDFC Bank: L1: investor presentation + MD&A + the segment/business-line note (bank AR, IR page/BSE); note banks report under old Indian GAAP, not Ind AS (M1.10). L2: latest 2 concalls (screener.in concalls tab). L3: ten-year sheet on screener.in, watching NIM, GNPA/NNPA, ROA/ROE, CASA, cost-to-income, CRAR (the bank KPIs, M5.01; not ROIC, which is meaningless for a bank). L4: a peer bank's deck (ICICI/Axis) + one bear/broker note. Moody's: L1: 10-K Item 1 + 1A + investor deck (moodys.com IR / EDGAR). L2: latest 2 call transcripts (IR site / stockanalysis.com / Seeking Alpha). L3: ten-year numbers on stockanalysis.com (ROIC, margins, the ratings-vs-analytics segment split). L4: S&P Global's deck (the direct competitor) + one bear note on the ratings duopoly's regulatory risk. Grading: full marks require the right jurisdiction-specific moves: banks use bank KPIs not ROIC; India needs the pledge/CARO check; both need transcripts in Layer 2.

P10 (timed: allocate 60 minutes). You have exactly 60 minutes for a first read of a company you've never seen. Allocate the minutes across the read order and justify the allocation.

Solution (model allocation). 0–2 Ctrl-F sweep (cheap insurance). 2–10 Business/Item 1 or MD&A (you cannot judge anything you can't narrate; 8 min). 10–14 Risk Factors, hunted (4). 14–22 the three statements, shape only (8). 22–27 segment note, the profit engine (5). 27–32 cash-flow, profits-to-cash (5). 32–38 related-party + contingents (6). 38–44 auditor + (India) CARO scan (6). 44–48 (India) shareholding & pledge, or (US) a glance at the proxy's comp metrics (4). 48–52 the two gut-checks, ROIC vs WACC and reverse the multiple (4). 52–60 write the 8-line output (8). Justification: the biggest blocks go to the business (you must narrate it) and the output (writing is where understanding consolidates); the governance notes get real time because that's where permanent losses hide; the "read" of adjectives and boilerplate gets zero.

P11 (timed). Classify each filing section as deep-read / skim / ignore on a first pass: (a) Item 1 Business; (b) the CSR annual report; (c) the segment note; (d) generic risk factors; (e) the cash-flow statement; (f) the properties list; (g) the auditor's KAM/CAM section; (h) the "value creation model" infographic.

Solution. Deep-read: (a) Business, (c) segment note, (e) cash-flow, (g) KAM/CAM. Skim: none essential here, though you might skim specific risk factors for the non-generic ones. Ignore: (b) CSR report, (d) generic risk factors, (f) properties list, (h) value-creation infographic. The through-line: deep-read anything that reveals where and whether the spread is earned or threatened; ignore anything drafted by the marketing or legal-cover teams.

P12 (timed). From this synthetic concall Q&A snippet, extract the three things a transcript gives you (framing, what analysts push on, tone):

Analyst 1: "Your gross margin fell 180bps — is this pricing or input costs?" CFO: "Largely transient input inflation; we expect normalization next year." Analyst 2: "You said that last year too — what's different now?" CFO: "Fair point. Honestly, we underestimated competitive pricing pressure; we're resetting our margin guidance down 150bps." Analyst 3: "And the new segment — still targeting breakeven by FY-next?" CFO: "We're pushing that out a year."

Solution. Framing: management first reached for "transient input inflation" (the comfortable explanation) before conceding. What analysts push on: margin (two of three questions) and the new segment's path to breakeven, so margin and the new-segment burn are the market's real worries; put both on your question list. Tone: mixed but ultimately candid. The CFO, when pressed on last year's identical promise, owned the miss ("we underestimated competitive pricing") and reset guidance rather than deflecting. That candor is a positive governance signal (Q3), even though the news is bad; a dodging CFO who kept blaming "input costs" would be the worse sign. The lesson: the Q&A hands you the soft spots and the management's character in five exchanges — nothing in the filing does that.

P13 (guided). Meghdoot Textiles' (synthetic) investor deck claims "the balance sheet is net-cash positive." The filing shows gross borrowings ₹1,200 cr and cash & bank balances ₹1,400 cr. But Ctrl-F turns up two things: ₹300 cr of that cash is margin money / deposits pledged against guarantees, and the related-party note shows ₹250 cr is a loan outstanding to a promoter-owned entity. (a) Compute the deck's net debt. (b) Recompute using only usable corporate cash. (c) State the swing and the one-line read.

Solution. (a) Deck net debt = 1,200 − 1,400 = −₹200 cr (i.e., ₹200 cr net cash, as claimed). (b) Usable cash = 1,400 − 300 (restricted) − 250 (RP loan) = ₹850 cr; adjusted net debt = 1,200 − 850 = +₹350 cr of net debt. (c) The claim swings by ₹550 cr, from ₹200 cr net cash to ₹350 cr net debt, once you strip cash that cannot be spent. Pledged/restricted balances and loans parked with promoter entities are not corporate liquidity; the deck's "net cash" is a phantom, and two tripwires ("guarantee," "related party") exposed it in a minute.

P14 (guided). A US company's deck leads with "Adjusted EPS $5.00 — just 24× at today's $120 price." GAAP EPS is $3.80. The $1.20 bridge is $0.70 of stock-based compensation and $0.50 of "restructuring" charges booked for the third consecutive year. (a) Confirm the bridge. (b) Which add-backs do you reject, and why? (c) Compute the P/E on the earnings you actually trust and compare.

Solution. (a) 5.00 − 3.80 = $1.20 = 0.70 + 0.50 ✓. (b) Reject both: stock-based comp is a real, recurring economic cost (paying people in shares dilutes you), and a "restructuring" that recurs three years running is a run-rate operating cost in a costume, and neither is a genuine one-off. (c) On GAAP $3.80, P/E = 120 ÷ 3.80 = ~31.6×, not the flattering 24× (120 ÷ 5.00). The "cheap" 24× is ~7.6 turns of multiple manufactured by add-backs. Serial "one-time" items and SBC add-backs are the two commonest EPS-inflation levers; price the multiple on numbers you trust, not the deck's.

P15 (independent). From a ten-year sheet (synthetic; US $): NOPAT $270m, invested capital $1,500m, WACC 8%; the stock trades at a P/E of 22 (take r = 8%). (a) ROIC and the spread. (b) The crude implied perpetual growth. (c) One sentence combining the two gut-checks.

Solution. (a) ROIC = 270 ÷ 1,500 = 18.0%; spread = 18.0 − 8.0 = +10.0 pts → a good business. (b) g ≈ 8% − 1/22 = 8% − ~4.5% = ~3.5%. (c) "Good business (+10-pt spread, sustained); the market is pricing only ~3.5% growth forever, which for an 18%-ROIC compounder is undemanding — so the over/under is on the durability of the spread, and that is where the teardown goes next." (Caveat as always: the crude g ignores ROIC/reinvestment; a gut-check on optimism, not a valuation.)

P16 (independent). On the call (Layer 2), management calls demand "strong" and points to revenue up 20% (₹4,000 cr → ₹4,800 cr). The balance sheet (Layer 3) shows trade receivables of ₹600 cr rising to ₹960 cr. (a) Compute revenue and receivables growth. (b) Compute DSO for both years. (c) Does Layer 3 confirm or challenge the Layer-2 story, and what do you do?

Solution. (a) Revenue +20%; receivables +60%: receivables grew 3× faster than sales. (b) DSO = receivables ÷ revenue × 365: year 1 = 600 ÷ 4,000 × 365 = 54.75 days; year 2 = 960 ÷ 4,800 × 365 = 73.0 days, a jump of ~18 days. (c) It challenges it: receivables outrunning sales while DSO stretches is exactly where premature or fictitious revenue settles (the "receivable" tripwire). The transcript's "strong demand" is now a hypothesis, not a fact. Pull the revenue-recognition policy, check for a change in credit terms or channel stuffing, and do not accept the growth number until the DSO is explained. When Layer 3 and Layer 2 conflict, the audited numbers win.

P17 (independent). Rank the source hierarchy audited filings > management transcripts > channel checks (scuttlebutt) > sell-side/bear notes, then assign the right tier to settle each claim and say why: (a) "The new plant will reach 30% margins by FY-next." (b) "Fair value is ₹2,400 (broker target)." (c) "Dealers tell me inventory is piling up." (d) "Related-party sales were ₹840 cr this year."

Solution. The hierarchy tracks verifiability and independence, both falling as you descend. (a) A promise from the transcript: log it and test it against future filings; it settles nothing today. (b) Sell-side, the weakest tier: borrow the reasoning, never the target; you compute value yourself. (c) A channel check: a useful hypothesis (is demand softening?) that you verify against reported inventory and DSO (Layer 3), not act on directly. (d) The filing's related-party note: primary and dispositive; this figure is simply true as reported, and your job is to judge its size and direction, not to confirm the number. The rule the whole stack obeys: borrow questions from every tier; take answers only from the filings.

P18 (timed: 12 minutes, all four parts). Kaveri Consumer Ltd (synthetic; India ₹). Ten-year sheet, latest year: revenue ₹8,000 cr; reported EBIT ₹1,360 cr (of which ₹160 cr is "other income"); NOPAT ₹900 cr; invested capital ₹4,500 cr; net cash; P/E 45; WACC ≈ 12% (take r = 12%). The deck claims "operating margin 17%." (i) Reconcile the operating margin. (ii) ROIC and spread (Q2). (iii) Crude implied growth (Q5). (iv) Write the two gut-check sentences and name the binding question.

Solution. (i) The deck's 17% = 1,360 ÷ 8,000 includes the ₹160 cr of other income; clean operating margin = (1,360 − 160) ÷ 8,000 = 1,200 ÷ 8,000 = 15.0%, overstated by 2.0 pts. (ii) ROIC = 900 ÷ 4,500 = 20.0%; spread = 20.0 − 12.0 = +8.0 pts → an excellent business. (iii) g ≈ 12% − 1/45 = 12% − ~2.2% = ~9.8%. (iv) "Good business: +8-pt ROIC-WACC spread, sustained. But at 45× the market is pricing ~9.8% growth forever — extremely demanding even for a 20%-ROIC compounder." Q5 (price) is the binding constraint, not Q2. This is a wonderful business whose price already assumes near-perfection. Scoring: all four parts correctly reconciled in 12 minutes = fluent; if the clock beat you, the bottleneck is almost always the reconciliation (redo Worked example 5).

Applied mini-project

The task. Run the full source-stack assembly and 45–60 minute first-read on one Indian and one US company, alternating jurisdictions per Working Agreement 5. Suggested pairs (pick one from each, avoiding any you've deep-dived): India: Pidilite, Marico, or Nestlé India; US: TJX, Automatic Data Processing, or Church & Dwight. (Clean compounders, so the method is what's tested, not landmine-spotting.)

Exactly what to pull, and from where.

1. US: the latest 10-K (EDGAR full-text search, efts.sec.gov); the latest DEF 14A; the two most recent earnings-call transcripts (IR site, or the 8-K Ex-99.1 press release + Seeking Alpha); the ten-year numbers (stockanalysis.com); one competitor's investor deck; one bear/skeptical note.
2. India: the latest annual report (company IR page or BSE); the two most recent concall transcripts (screener.in "concalls" tab); the ten-year data (screener.in, download the Excel); the latest shareholding pattern (BSE/NSE or screener); one peer's deck; one bear note (a broker "sell" note or a ValuePickr thread).

The deliverables. For each company: (1) a source-stack manifest: the six documents, with links/locations, and one line on what each was highest-signal for; (2) the 8-line first-impression output (the artifact's template); (3) a timed log: start time, stop times per read-order block, total (target ≤60 min); (4) the two gut-checks written as two sentences (ROIC-vs-WACC verdict; "the market is pricing in ~X% growth").

Scoring rubric (0 / 1 / 2 per criterion; 20 points total; ≥16 to pass):

File both manifests and outputs in your company files (M0.04); they become the opening move of any future teardown or deep dive on these names.

Reading & resources

• Michael Mauboussin, Measuring the Moat (Counterpoint Global; free PDF online): the source-stack-to-industry-map logic and the "how to read a business" scaffold compressed here. [Free] [Intermediate]
• Alfred Rappaport & Michael Mauboussin, Expectations Investing (rev. ed.; free companion site expectationsinvesting.com): the discipline behind "what is the price implying," which Q5 and Gut-check 2 gesture at and M3.05 built. [Paid book / Free site] [Intermediate]
• Aswath Damodaran, Narrative and Numbers: turning the story you read in Layer 1 into the numbers you test in Layer 3; the connective tissue of the whole method. [Paid] [Intermediate]
• SEC EDGAR full-text search (efts.sec.gov): free, searches the text of every filing since 2001; the fastest way to the exact 10-K, DEF 14A, or 8-K. [Free] [Beginner]
• screener.in (India): the ten-year data sheet, ratios, downloadable Excel, and the "concalls" tab with transcript links; the Indian analyst's default Layer 3 + Layer 2. [Free / cheap Pro] [Beginner]
• stockanalysis.com (US): ten years of statements and ratios, free, no login; the default US Layer 3. [Free] [Beginner]. Companions: TIKR, Koyfin (free tiers), Tijori/Trendlyne (India). [Free/Paid]
• Earnings-call transcripts: company IR pages (always free, always primary); Seeking Alpha (large free transcript library); the 8-K Ex-99.1 press release for the numbers behind the call. [Free/Paid] [Beginner]
• Philip Fisher, Common Stocks and Uncommon Profits: the fifteen points and scuttlebutt; Layer 4's spirit, and the backbone of M8.04's Day-3 qualitative work. [Paid] [Intermediate]
• Zerodha Varsity, Fundamental Analysis (India): free, and its "how to read an annual report" chapter mirrors this read order for the Indian AR. [Free] [Beginner]
• Podcasts: Business Breakdowns and Acquired. Hour-long teardowns of individual businesses; excellent for hearing the five questions answered aloud on companies you can then check. [Free] [Beginner–Intermediate]
• Named filings to practice on: Costco 10-K and Asian Paints annual report (this module's walkthroughs); then anything from the starter bank (M0.05 / §5.2 of the master map).
• Going deeper: Howard Schilit, Jeremy Perler & Yoni Engelhart, Financial Shenanigans (4th ed.) [Paid] [Advanced]. The field guide to the exact games the reconciliation step above reverses: serial "one-time"/"exceptional" charges, non-operating "other income" booked as profit, and receivables that outrun sales. Optional now; it is Phase 2's forensics compressed into cases you can run the Ctrl-F list against.
• Going deeper: Michael Mauboussin & Dan Callahan, Calculating Return on Invested Capital (Counterpoint Global / Morgan Stanley, free PDF) [Free] [Intermediate]. How to build the ROIC that Gut-check 1 reads off the screener, and why the ROIC-vs-WACC spread, not the growth rate, is the number the entire read is aimed at.

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand — that is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "The five questions and their toolkits, the four-layer source stack, the ten-stop read order, the skim/ignore triage, the ten Ctrl-F tripwires, and the two thirty-second gut-checks",
  "use": "This module is where the modern layer is most obviously worth having and most easily abused, because the clerical half of the source stack collapses to a structured request while the analytical half does not move at all. Four legitimate jobs. (1) LOCATION, not comprehension: EDGAR full-text search answers 'which of this filer's documents contains this phrase, and when did it first appear' as a query rather than an afternoon — the natural instrument for the §4.6 tripwires run across a decade instead of one document. (2) ASSEMBLY of Layer 3: the ten-year panel arrives from the XBRL endpoints below as tagged facts with accession numbers attached, so the forty minutes §4.4 spends getting numbers onto one screen becomes a fetch, and the thirty seconds §4.4 spends reading them stays exactly where it was. (3) RETRIEVAL over documents you have already downloaded — this filing, its predecessor, the transcript — where every answer comes back with the page or note number, so you can go and read the sentence yourself. (4) TRANSLATION of document furniture: 'which note in this Ind AS annual report answers what Item 8's segment footnote answers in a 10-K' is a genuine question with a stable answer, and the furniture is what makes a first Indian AR slow. The three bans are the module's own content. Never let it perform the §4.5 skim/ignore triage: deciding what to ignore, and being right about it, is the skill this module exists to install (§4.1). Never let it write the eight-line output — §4.7's closing line says the compression is where the understanding consolidates, and a compression someone else performed consolidates nothing. Never accept a tripwire VERDICT: a hit buys thirty seconds of your attention, and WE3's cluster read is a judgement about a promoter's behaviour, not a text-classification result.",
  "tools": ["EDGAR full-text search and the data.sec.gov XBRL endpoints — for locating language across a filer's history and pulling the Layer-3 panel as tagged facts", "Document/PDF AI with retrieval over your own downloaded set — the filing, the prior year, the transcript — used only with a page or note citation on every claim", "Chat assistants — Claude, ChatGPT for document-furniture translation between a 10-K and an Ind AS annual report, and for drafting the fetch code below", "screener.in, Tijori, stockanalysis.com — the aggregators §4.4 already licenses for Layer 3, which you earned by spreading a dozen filings by hand"],
  "prompt": "Here is the related-party note, the contingent-liabilities note and the shareholding pattern from <COMPANY>'s FY<YY> annual report, pasted verbatim: <PASTE>. For each of these ten terms — related party, pledge, contingent, going concern, restated, auditor, exceptional, one-time, guarantee, receivable — list every occurrence with its page or note number and the full surrounding sentence, quoted exactly. Do not summarise, do not paraphrase, do not characterise anything as a concern or as routine, and do not rank them. If a term does not appear, say 'not present' for that term. I am doing the reading.",
  "verify": "The citation IS the verification, and it has to be structural rather than advisory: an extraction with no page number is not weak evidence, it is unusable, because you cannot tell a quotation from a fabrication without going to the page — and going to the page is the whole point of the exercise. So set the rule as a property of the output format, not as a hope about the model: every retrieved claim carries a locator, you open at least a sample of the locators before you believe any of them, and a claim whose locator does not resolve invalidates the whole extraction rather than just that line. Then verify the numbers the ordinary way. Re-run the two gut-checks yourself with your OWN required return (WE6 is the whole lesson: the note's '3% priced in' and your '5% priced in' differ only because the note never disclosed that it used r = 7%, and a borrowed implied-growth figure with an undisclosed r is not a number). Recompute any margin you were handed against the filing — WE5's deck margin of 22.0% became 16.6% on two tripwires, a 5.4-point gap, and no tool volunteers that reconciliation because the deck and the notes are different documents. And treat the tripwire list as the battery it is: §4.6 says most hits turn out innocent, and the Quant lens below puts a number on how innocent.",
  "diy": "The gate is unaided and it is a navigation test, not a recall test: given a filing you have never opened and no tool in the room, run the ten-stop read order in the fixed sequence, defend what you chose to ignore, execute the ten tripwires from memory and say why each is a tripwire, do the ROIC-versus-hurdle and reverse-the-multiple gut-checks on paper, and write the eight lines in thirty minutes. Nothing in this callout is admissible in that room, and the reason is §4.1: the speed is only earned if the understanding underneath it is yours."
}

Modern Data Analysis

By hand first. You navigated a 10-K to a profit engine in thirty minutes (§4.7), ran the same discipline over Indian document furniture that has no Item 1 and does have a pledge table (§4.8), stripped other income and a serial exceptional out of a headline margin (WE5), and reversed a multiple with your own required return rather than a borrowed one (WE6). Every one of those must stay in your hands. What the modern layer changes is not the reading. It is that the retyping stops, and only the retyping.

Today's workflow. The source stack has a machine-readable half and a PDF half, and the line does not fall where a beginner expects. Classify §4.5's own ten stops by a single criterion (does the number you go to that stop for arrive as a tagged fact you never retype?) and the answer for a US filer is three stops fully (the three statements, the segment note, the cash-flow statement), two partly (related-party amounts and contingent totals carry monetary tags, but the counterparty, the direction and the narrative that make them mean anything are text blocks), and five not at all (Business, Risk Factors, MD&A, the auditor's report, and the India-only shareholding-and-pledge stop that has no US analog at all). Read that list again and notice what it says: the stops that automate are the stops that were never the hard part. The three that automate are exactly the ones §4.4 tells you to pull from an aggregator anyway; the five that do not are Q1, Q3, Q4 and the whole governance layer.

For India the count is different in a way that should change how you budget the hour. Indian companies do file in XBRL, and the data exists. But there is no free public JSON endpoint in the shape of data.sec.gov, so zero of the ten stops arrive as a fetched fact. The Indian half of this workflow is a PDF-and-portal parse: you download an annual report, you scrape an exchange announcement page, you pay per document at the Registrar for an unlisted subsidiary. The US half is an API call. That asymmetry is not a complaint, it is a planning input. On a US name the clerical minutes of §4.4's Layer 3 nearly vanish and you spend the saved time on Layer 4, while on an Indian name Layer 3 costs what it always cost and the compression you get comes from the aggregator's export button, not from an endpoint.

Tools & sources (IN + US). US, machine-readable: www.sec.gov/files/company_tickers.json maps a ticker to the CIK everything else is keyed on; data.sec.gov/submissions/CIK##########.json is the filing history with form types, dates and accession numbers (the feed that answers "when did the auditor change"); data.sec.gov/api/xbrl/companyfacts/CIK##########.json returns every fact the filer has ever tagged, which is Layer 3 in one request; data.sec.gov/api/xbrl/companyconcept/CIK##########/us-gaap/<Tag>.json returns one concept's full history when you want a single row rather than the whole panel; data.sec.gov/api/xbrl/frames/us-gaap/<Tag>/USD/<period>.json returns one concept across all filers for a period, which is the peer benchmark of §4.4 without assembling peers by hand. EDGAR full-text search (the interface at sec.gov/edgar/search/, with a JSON backend) is the tripwire list run across a decade rather than a document. The CIK is zero-padded to ten digits in all of these. The SEC requires a declared User-Agent identifying you with a contact address on every request, and publishes a fair-access rate policy. Read the current policy on their site rather than any number quoted in a course, because the policy is stable and the number is not. India, portal-and-PDF: BSE (bseindia.com) and NSE (nseindia.com) corporate-announcement feeds and filing archives, scrapeable but not a documented public API, and the payload is usually a PDF; MCA21 (mca.gov.in) for the AOC-4 financial statements and MGT-7 annual returns of the unlisted subsidiaries that explain a consolidation you cannot otherwise read, at a per-document fee; SEBI's LODR disclosure regime, which is what makes §4.5's stop 10 exist at all: Reg 31 for the shareholding pattern and the pledge percentage, Reg 23 for the half-yearly related-party disclosures, Reg 30 for material events; screener.in for the ten-year sheet, the Excel export and the concalls tab, which is the highest-yield unstructured Indian source there is; Tijori and Trendlyne alongside it. Run any of this in your notebook, not against a live portal you have not read the terms of.

# R3 — recomputed in-session. Two arithmetics: what the source stack automates, and what a tripwire hit is worth.
# 1. Section 4.5's own ten stops, classified by ONE criterion: does the number arrive as a tagged
#    fact you never retype?   2 = fully, 1 = amounts tagged but the meaning is a text block, 0 = prose.
STOPS = {"1 Business": (0, 0), "2 Risk factors": (0, 0), "3 MD&A": (0, 0),
         "4 Three statements": (2, 0), "5 Segment note": (2, 0), "6 Cash flow": (2, 0),
         "7 Related party": (1, 0), "8 Contingents": (1, 0), "9 Auditor": (0, 0),
         "10 Shareholding & pledge": (0, 0)}                       # (US, India-free-structured)
us, ind = [ [s[i] for s in STOPS.values()] for i in (0, 1) ]
print(us.count(2), "of", len(STOPS), "US stops fully tagged;", us.count(1), "partly;",
      us.count(0), "prose |  India:", ind.count(2), "fully")        # 3 of 10 ; 2 partly ; 5 prose | 0

# 2. The 4.5 illustrative panel is 22 rows, of which 7 are pure formulas over the other 15.
PRIMARY, DERIVED, YEARS = 15, 7, 10
print(PRIMARY * YEARS, "figures typed by hand vs", DERIVED * YEARS, "computed")   # 150 vs 70

# 3. Ten tripwires are a BATTERY, so a single hit is weak evidence by construction.
K = 10
for q in (0.05, 0.10, 0.15, 0.20):
    print(f"{q:.0%} per-term false alarm -> {1-(1-q)**K:.1%} of CLEAN filings trip at least one")
    #  40.1%          65.1%          80.3%          89.3%

# 4. WE3's cluster rule as odds. ASSUMED, not measured: 1 in 20 impaired, each flag 70% sensitive,
#    15% false-alarm rate, flags conditionally independent given state (the assumption that bites).
prior, sens, fa = 1/20, 0.70, 0.15
odds = prior / (1 - prior)                                          # 0.0526
for k in (1, 2, 3, 5):
    o = odds * (sens / fa) ** k
    print(k, "aligned flag(s) ->", f"{o/(1+o):.1%}")                # 19.7% / 53.4% / 84.2% / 99.1%

Verify. Prove any fetched panel against the module's own worked figures before it goes near a company: WE5's reconciliation must reproduce a deck margin of 22.0%, an operating EBITDA of ₹950 cr after other income and ₹830 cr after the serial exceptional, a clean margin of 16.6%, a gap of 5.4 percentage points, and a deck that flatters clean operating profit by 32.5%; WE6 must give ROIC 24.0%, a +15.0-point spread, implied growth of 5.0% at your 9% hurdle, and the note's undisclosed r = 7% backed out of its "3%"; WE4's three companies must give 25.0%, 9.0% and 22.0% ROIC against the 11% hurdle; and §4.9's crude inversion must give 6.0%, 8.0% and 1.67% at P/Es of 25, 50 and 12 with r = 10%. Then verify the one thing a fetch cannot: that the tagged fact means what you think it means. A companyfacts payload gives you a concept, a value, a unit, a period and an accession number, and it does not tell you that the filer re-cut its segments in year six or that "Revenues" and "RevenueFromContractWithCustomerExcludingAssessedTax" both appear in the same history with a definitional break between them. That check is §4.5's stop 5 read by a human, and it is why the endpoint saves you the typing and not the reading.

Quantitative lens

Two questions the module poses and answers in words can be answered in numbers, and both change how you run the hour. Every figure below comes from the block above, on inputs stated there; they are properties of that arithmetic and its assumptions, not measurements of any filing.

The tripwire list is a battery, and that is why a single hit means almost nothing. §4.6 says most hits turn out innocent. Quantify it: ten independent searches at a 15% per-term false-alarm rate on a genuinely clean filing trip at least one tripwire 80.3% of the time. Even at a stingy 5% per term the figure is 40.1%, and at 20% it is 89.3%. So "I found a hit" carries roughly the information content of "I ran the list." Cast as odds it is worse than that sounds: on an assumed prior of one impaired company in twenty, with each flag 70% sensitive and 15% false-alarming, one flag lifts the probability of impairment from 5% to only 19.7%, a number that would be irresponsible to act on and irresponsible to ignore, which is exactly the "thirty seconds of attention" §4.6 prescribes and not one second more. The picture inverts at two. Two aligned flags reach 53.4%, three reach 84.2%, and WE3's five reach 99.1%. That is the cluster rule of Phase 2 with a number attached, and it says something operationally useful: the correct response to one hit is to finish the list, because the second hit is worth vastly more than the first was.

But the cluster arithmetic overstates itself, and the reason it overstates is the reason clusters matter. The multiplication above assumes flags are conditionally independent given the company's state. For a stressed promoter they are emphatically not: pledge, related-party loans, group guarantees and bloated contingents in WE3 all flow from one cause, so they co-occur even among impaired companies, and correlated evidence multiplied as though independent inflates the posterior. Treat the 99.1% as the ceiling of a range whose floor is meaningfully lower, and read the honest lesson: five aligned flags are decisive not because the arithmetic says 99% but because a single mechanism explains all five, which is a narrative claim you must be able to state. If you cannot say in one sentence what would have to be true for all five to occur together, you have five coincidences rather than one diagnosis.

And the automation lands where the work was cheapest. Three of §4.5's ten stops arrive as tagged facts in the US, five are prose, and none arrive that way for free in India; §4.5's illustrative panel is 22 rows of which 7 are formulas, so a ten-year build is 150 figures typed by hand and 70 computed, and it is exactly those 150 that an endpoint removes. Set the two facts next to each other and the conclusion is uncomfortable for anyone hoping the tools do the job: the machine-readable half of the source stack is the half §4.4 already told you to buy from an aggregator, and the five prose stops it cannot touch are where Q1's profit engine, Q3's governance and Q4's real killers live.

Honest limits, and they are load-bearing. The sensitivities, false-alarm rates and the one-in-twenty prior above are assumed for the arithmetic and are not measured base rates. I have no panel of Indian filings scored for governance impairment, and neither, probably, do you; the correct use of these numbers is to re-run them with your own estimates and watch how fast the answer moves, not to quote 19.7%. The independence assumption fails in the direction described. The ten-stop classification is a judgement about the module's own table under a stated criterion, not a survey of what filers tag: a specific company may tag more or less, tag it inconsistently across years, or use a custom extension element that no standard concept query will find. And the whole exercise is silent on the thing that actually decides a first read — whether you noticed that the biggest revenue line is not the profit engine. No endpoint has ever noticed that for anyone.

Do it in code: map ticker to CIK once, pull companyfacts for the ten-year panel and keep the accession number beside every figure, run the ten tripwire terms through full-text search across a decade rather than one document, and hold every retrieved sentence to a page citation you actually open. Then close the laptop and do §4.5's read order on the prose. See G1/QM1.01, G2/CS2.05 and G5/AI0.04.

Where this goes next: galaxy cross-links

• G2 · CS2.05 APIs, HTTP, JSON & Web Scraping: how the endpoints above actually work: request headers (including the declared User-Agent the SEC requires), pagination, rate limiting, and the difference between a documented API and a portal you are scraping, which is precisely the US/India asymmetry this addendum turns into an hour budget.
• G5 · AI0.04 Retrieval & RAG — Grounding AI in Filings & Documents: the machinery behind the one defensible AI pattern here: chunking a filing, retrieving over it, and returning a locator with every claim so the citation requirement is a property of the pipeline rather than a request in a prompt.
• G1 · QM1.01 Probability, Distributions, Sampling & Estimation: the battery arithmetic above: why ten tests at a 15% false-alarm rate produce a hit on most clean filings, what a likelihood ratio is, and why correlated evidence multiplied as though independent overstates a posterior.
• G2 · DA1.02 pandas I: Data Wrangling: turning a companyfacts payload into the ten-year panel §4.4 wants on one screen, including the concept-name breaks and re-segmentation that make a naive join silently wrong.
• G5 · AI0.06 Failure Modes, Verification & the Primary-Source Guardrail: the general form of the three bans above: what a fluent, confident, unsourced extraction looks like, and why the primary filing is the only thing that ever settles a disagreement.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "How a professional understands a company in one hour." You must convey, without jargon: that there are only five questions worth asking about any business (what it does and how it makes money; whether it's actually good; whether the people are honest and smart with money; what could kill it; and whether the price already assumes all the good news); that "good" means it earns more on its money than that money costs, and can keep doing it; that you read a company in a fixed order from a fixed set of documents, starting with the company's own story and ending by trying to prove yourself wrong; and that being fast means ignoring the right 90%, not reading everything quickly. If your page needs the words "ROIC," "WACC," or "reverse DCF," you don't own it yet. Say "the return it earns on its money," "what that money costs," and "working out what the price already assumes."

Journal prompt. Two entries. (1) Pick the last company you read (company-of-the-week or otherwise) and, honestly, which of the five questions did you actually answer and which did you skip? Most people answer Q1 (a story) and a distorted Q5 (a P/E they never reversed) and quietly ignore Q3 and Q4 — is that your pattern? Write the two questions you habitually under-weight and one concrete step to fix it next week. (2) Time yourself on your next company-of-the-week using the read order above and the 8-line output, and log the actual minutes per block. Where did the clock run away from you: the business, the notes, the writing? Your bottleneck this week is your training target next week; name it.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.