The Microcap Minute Classroom. Module 5, chapter 6: Audit your own work: the pre-mortem checklist **What you will learn** - What a pre-mortem is, and why you run one on your own analysis - The six checks we ran on the Apple dive, with real catches - How to mark figures you cannot verify instead of hiding them - Why the notes to the accounts beat every screen
You know the last ten minutes of an exam, checking your own paper? Part of you does not want to know. But that is where the marks are rescued. A deep dive needs the same ritual, with a name: a pre-mortem. Assume your work is wrong, and find where, before anyone else does. The checklist we ran on the Apple dive, and what each caught. **1. Trace every figure.** Each number gets a source: a filing, or a named page. No exceptions, no "everyone knows". Facts come from filings, because a filing is signed; framing may come from the company's website, but figures never do, because marketing copy is not. If a figure cannot be traced, it leaves the document or stays with a visible flag. **2. Cross check wherever two sources cover the same number.** Apple's cash and marketable securities in June 2026: the market data page said $146.52 billion; adding the three lines from the filing gave $146.5 billion. Debt: $84.34 billion against $84.3 billion. Employees: 166,000 in both. When independent routes agree, relax. When they do not, the filing wins. **3. Mark your own arithmetic.** Some figures are computed, not reported. Apple never publishes a September quarter table, so our fourth quarter is the full year minus the nine months: computed, and labelled. The early EPS figures are adjusted for the 2020 four for one stock split: computed, and labelled. A computed figure is not a sin; an unlabelled one is. **4. Flag what you cannot verify, out loud.** Two figures in the dive carry flags. The roughly $20 billion a year that Google reportedly pays Apple to stay the default search engine appears in no Apple filing; it surfaced in a court case (about Rs 1.7 lakh crore a year, if true). The "2.35 billion active devices" claim came from an earnings call, not a filing. Both stayed, because both matter, and both carry a "not verified" label. An honest flag makes a document more trustworthy, not less. **5. Record disagreements instead of smoothing them.** The market value on the statistics page, $4.57 trillion, implies a slightly higher share count than the 14.594 billion on the 10-Q cover. The gap is under 2%, probably share based pay; it goes in the audit file, not silently averaged away. Small unexplained gaps are where big mistakes hide. **6. Hunt the one offs.** Fiscal 2024 profit fell, which looks like decline until Note 7 explains it: a one time tax charge of $10.2 billion after a European court confirmed an Irish tax ruling. Without it, profit rose. The dispute began in 2016, and the money sat in escrow, disclosed in the notes, for eight years before the charge landed. Anyone reading the contingency notes knew. Every screen missed it, because screens do not read notes. Notice how much a company must confess: tariffs, single source suppliers, App Store regulation. The pre-mortem reads these first: the company's own list of what could break your thesis. Notice the search box: every filing of every US company, searchable. Type "State Aid" or "tariff" and you read what the professionals read, free. That is the whole ritual: trace, cross check, label your arithmetic, flag the unverifiable, record disagreements, hunt one offs. One rule sits above them all: **print the line that moves against you**. Every honest analysis has a figure arguing with its thesis, usually the most informative number in the document. A write up showing only the figures pointing one way is marketing with a spreadsheet. Chapter 7 shows a finished dive ending on such a line. It takes an hour, it is boring, and it separates analysis from content. The market will audit your work eventually, and it charges tuition. Better to fail your own exam first. **Try it yourself** Pick one number from any summary page, say Apple's 166,000 employees, and find it in the [fiscal 2025 10-K](https://www.sec.gov/Archives/edgar/data/320193/000032019325000079/aapl-20250927.htm) (search "full-time equivalent"). Then take a figure from a company's website, say the "about 30,000 employees" one big insurer still claims, and check it against the signed annual report's figure: 34,000. Anything you cannot trace gets a "not verified" label. Your first audit, done in ten minutes.
**Key takeaways** - A pre-mortem: assume your analysis is wrong, and find where before the market does. - Trace every figure; where two sources disagree, the filing wins. Websites are for framing, never for figures. - Label computed figures; flag unverifiable ones; never smooth over a disagreement. - Print the line that moves against you; it is usually the most informative number there. - The notes to the accounts hold the stories screens cannot show.
. I am Ritu, a synthetic voice, and the words I am reading are Ayush Agrawal's.
# Audit your own work: the pre-mortem checklist **What you will learn** - What a pre-mortem is, and why you run one on your own analysis - The six checks we ran on the Apple dive, with real catches - How to mark figures you cannot verify instead of hiding them - Why the notes to the accounts beat every screen
You know the last ten minutes of an exam, checking your own paper? Part of you does not want to know. But that is where the marks are rescued. A deep dive needs the same ritual, with a name: a pre-mortem. Assume your work is wrong, and find where, before anyone else does. The checklist we ran on the Apple dive, and what each caught. **1. Trace every figure.** Each number gets a source: a filing, or a named page. No exceptions, no "everyone knows". Facts come from filings, because a filing is signed; framing may come from the company's website, but figures never do, because marketing copy is not. If a figure cannot be traced, it leaves the document or stays with a visible flag. **2. Cross check wherever two sources cover the same number.** Apple's cash and marketable securities in June 2026: the market data page said $146.52 billion; adding the three lines from the filing gave $146.5 billion. Debt: $84.34 billion against $84.3 billion. Employees: 166,000 in both. When independent routes agree, relax. When they do not, the filing wins. **3. Mark your own arithmetic.** Some figures are computed, not reported. Apple never publishes a September quarter table, so our fourth quarter is the full year minus the nine months: computed, and labelled. The early EPS figures are adjusted for the 2020 four for one stock split: computed, and labelled. A computed figure is not a sin; an unlabelled one is. **4. Flag what you cannot verify, out loud.** Two figures in the dive carry flags. The roughly $20 billion a year that Google reportedly pays Apple to stay the default search engine appears in no Apple filing; it surfaced in a court case (about Rs 1.7 lakh crore a year, if true). The "2.35 billion active devices" claim came from an earnings call, not a filing. Both stayed, because both matter, and both carry a "not verified" label. An honest flag makes a document more trustworthy, not less. **5. Record disagreements instead of smoothing them.** The market value on the statistics page, $4.57 trillion, implies a slightly higher share count than the 14.594 billion on the 10-Q cover. The gap is under 2%, probably share based pay; it goes in the audit file, not silently averaged away. Small unexplained gaps are where big mistakes hide. **6. Hunt the one offs.** Fiscal 2024 profit fell, which looks like decline until Note 7 explains it: a one time tax charge of $10.2 billion after a European court confirmed an Irish tax ruling. Without it, profit rose. The dispute began in 2016, and the money sat in escrow, disclosed in the notes, for eight years before the charge landed. Anyone reading the contingency notes knew. Every screen missed it, because screens do not read notes. Notice how much a company must confess: tariffs, single source suppliers, App Store regulation. The pre-mortem reads these first: the company's own list of what could break your thesis. Notice the search box: every filing of every US company, searchable. Type "State Aid" or "tariff" and you read what the professionals read, free. That is the whole ritual: trace, cross check, label your arithmetic, flag the unverifiable, record disagreements, hunt one offs. One rule sits above them all: **print the line that moves against you**. Every honest analysis has a figure arguing with its thesis, usually the most informative number in the document. A write up showing only the figures pointing one way is marketing with a spreadsheet. Chapter 7 shows a finished dive ending on such a line. It takes an hour, it is boring, and it separates analysis from content. The market will audit your work eventually, and it charges tuition. Better to fail your own exam first. **Try it yourself** Pick one number from any summary page, say Apple's 166,000 employees, and find it in the [fiscal 2025 10-K](https://www.sec.gov/Archives/edgar/data/320193/000032019325000079/aapl-20250927.htm) (search "full-time equivalent"). Then take a figure from a company's website, say the "about 30,000 employees" one big insurer still claims, and check it against the signed annual report's figure: 34,000. Anything you cannot trace gets a "not verified" label. Your first audit, done in ten minutes.
**Key takeaways** - A pre-mortem: assume your analysis is wrong, and find where before the market does. - Trace every figure; where two sources disagree, the filing wins. Websites are for framing, never for figures. - Label computed figures; flag unverifiable ones; never smooth over a disagreement. - Print the line that moves against you; it is usually the most informative number there. - The notes to the accounts hold the stories screens cannot show.
That was chapter 6 of module 5. The text, the pictures and the exercise are on the lesson page. Thank you for listening.