Where we start
By now you can compute owner earnings from the filings, normalise them, discount them, reverse a test price into the growth it needs, and check that growth against base rates and fade. Each of those steps produced a number, and each number came with a choice you made. A memo is where the choices are written down in an order that stops you from cheating: the business before the numbers, the numbers before any price, and the conditions for being wrong before you know whether you would like to be right.
A memo about a business, not a recommendation. The worked example uses Procter & Gamble's figures because its filings are public and long. The kill criteria written for it are teaching examples, not a view on the company, and no market price appears anywhere: every price on this page is a test price you type, taken from the source map's half-or-double results.
★ What you walk out with tonight. A one-page memo template with six blocks in a fixed order, three kill criteria for a real filer that a stranger could check against next year's 10-K, a short Python step that prints your filer's range and implied growth straight from the source map, and one honest paragraph on the value-versus-growth evidence that you can paste into any memo without overclaiming.
A memo is a test you set yourself
5 min liveA good memo fits on one page and is read top to bottom in the order it was written. Six blocks: the business in four sentences, the starting point and why you chose it, the range, what a price implies, the risks, and the kill criteria. Five of the six can be written without any price at all, and they should be, because the moment a price is in your head every block after it starts leaning towards it.
LiveSix blocks, and what each one is allowed to contain3 min▶
1 · The business. Owner session 1's owner's paragraph, four sentences: what it sells, why customers come back, what it must spend just to stay where it is, what would break it. If any sentence is blank, stop: no number below can be trusted, because you do not yet know what it measures.
2 · The starting point. Which rung you start from, last year's earnings, the five-year mean or the ten-year mean, and one sentence of evidence for the choice. The ladder in the source map is that evidence: the five-year rung had the smallest miss at the 3- and 5-year hold-outs, last year's earnings at the 4-year one, and every honest rung undershot because owner earnings rose sharply after 2020.
3 · The range. The central value and the nine-case range from Analyst session 6, with the assumptions printed next to it. Buffett's 1986 line is the standard for this block: "I would rather be vaguely right than precisely wrong." A range that looks precise is a warning sign.
4 · What the price implies. The only block that uses a price, and it is a test price you type. The reverse DCF from Analyst session 5 turns it into a growth rate, and the base rates from Analyst session 7 say how often growth like that has lasted.
5 · The risks. Two kinds: what the numbers get wrong (the all-capex approximation, one-off years, a tag that leaves something out) and what the business could lose. Analyst session 8's fade belongs here.
6 · Kill criteria. Part 2.
Self-studyWhy write it before you know any price3 min▶
Owner session 6 taught anchoring, from Tversky and Kahneman's 1974 paper (reported): an estimate made after hearing a number drifts towards that number, even when you know it is irrelevant. A price is the most powerful anchor in investing because it arrives with a crowd's confidence attached. Write the range after seeing a price and the range quietly moves to contain it. Write the kill criteria after seeing a price and they are set just outside wherever the price would leave you comfortable.
The fix is mechanical, not heroic: write blocks 1, 2, 3, 5 and 6, put a date on them, and only then type a price into block 4. If the price changes your mind about block 3, you will at least see that it did, because block 3 already exists in writing. The full treatment of anchors, narrow ranges and the pre-mortem is in Owner session 6: The mistakes made in your own head.
Buffett's 1977 test as a memo's spine. Berkshire's 1977 letter says it picks marketable stocks much as it would evaluate a whole business for acquisition, and wants four things: a business it can understand, with favourable long-term prospects, run by honest and competent people, available at a very attractive price. Three of the four are about the business; only the last needs a price. The six-block memo keeps the same proportion: the price waits until block 4, and it cannot rescue a page whose first block is blank.
Kill criteria
5 min liveA kill criterion is a statement that, if it comes true, means your thesis was wrong. It has three properties: anyone can observe it, it has a date or a window, and it names a specific number. "Owner earnings below the ten-year average for three years" passes. "If the business deteriorates" fails all three, which is exactly why people like writing it.
LiveObservable, dated, specific, and calibrated to the filer4 min▶
Observable means the evidence will appear somewhere you and a stranger can both read it: a 10-K tag, a line in the cash flow statement. Not "management loses focus". Dated means a window: three fiscal years, by FY2031, in each 10-K from FY2027. A criterion with no window can always be postponed. Specific means a number you wrote down before the outcome was known.
The fourth property is easy to miss: calibrate to the filer's own history. A sensible-sounding generic rule is "capex rising above D&A without growth". For Procter & Gamble that rule fires on day one, because its ten-year capex to D&A ratio is already 1.19, the highest of the four filers (Caterpillar 0.57, Exxon Mobil 0.82, Johnson & Johnson 0.57). A criterion that is already true is not a test. So the P&G version says "above 1.19, for three years, while owner earnings do not grow": the same idea, set against what the business has actually done.
And state what you will do. Not "sell", which is a trading decision and outside the scope of a memo, but "re-open the memo and redo blocks 2 and 3 from scratch". The criterion's job is to force the rewrite.
Self-studyPre-commitment, and why criteria written afterwards bend3 min▶
Pre-commitment is deciding now what a future observation will mean, so that the future you, who will want the thesis to be right, does not get a vote. It works for the same reason the memo order works: once you own a view, every new fact gets read in the light most kind to it. Owner session 6 calls this confirmation; a criterion written in advance is the cheapest defence against it.
A criterion written afterwards bends in three predictable ways. The number moves ("three years" becomes "three years, excluding the restructuring year"). The window stretches ("by FY2031" becomes "once the cycle turns"). Or the claim it was attached to is quietly swapped for a different claim that still holds. Each step sounds reasonable on its own, which is why the original must be dated and kept.
Buffett's 1989 letter lists twenty-five years of his own mistakes, and several are exactly the kind a dated criterion would have flagged early: a department store, Hochschild Kohn, bought below book value that returned little, and the general lesson that good managers cannot fix bad economics. A cheap price was the thesis; nothing in it said what would show the business itself was the problem. A memo forces that sentence to exist.
Procter & Gamble on rung 3, worked through. Its ten-year mean owner earnings are 5.00 a share. At the defaults (a 9 percent discount rate, 10 years of 5 percent growth, then 2.5 percent), the central value is 95.16 a share and the nine-case range runs from 70.84 to 132.39. A test price at double the value, 190.33, implies 14.17 percent growth a year for ten years; a test price at half implies -4.31 percent. Block 4 then writes the sentence that matters: the double price asks for 14.17 percent a year from a business whose owner earnings per share grew 3.1 percent a year, nominal, from the mean of its first five years (3.84) to the mean of its last five (5.90). All figures from the source map; the test prices are typed, not quoted.
Teach the disagreement you cannot settle
5 min liveEvery memo eventually meets a question bigger than the business: does buying cheap businesses pay, as a strategy? The honest memo does not answer it, because the research has not. What it can do is state both readings, give each its source and how firmly it is known, and then say why the memo's own conclusion does not depend on which side is right.
LiveTwo readings of the same decades4 min▶
The premium got smaller. Fama and French's 2020 paper "The Value Premium" (reported, via Chicago Booth Review) splits US returns at 1991. As summarised there, the big-stock value premium was 4.3 percent a year from 1963 to 1991 and 0.6 percent from 1991 to 2019; the small-stock premium went from 7 percent to 4 percent. Then the authors' own caution: monthly returns are volatile enough that they cannot reject equal expected premiums in both halves, so the declines look large but are statistically indistinguishable from zero.
The yardstick moved. Arnott, Harvey, Kalesnik and Linnainmaa (2021, reported) argue that much of value's weak run came from value stocks getting cheaper relative to growth stocks, a widening spread rather than worse businesses, and that intangible assets distort book value, the measure most factor studies use to call a stock "value".
Neither reading says value "works" and neither says it "is dead". The first says the measured premium shrank by an amount the data cannot distinguish from chance. The second says the measurement itself is suspect. Both are reported here, not read at source, and a memo that cites either should say so.
Self-studyWhat a memo should say about evidence it cannot resolve3 min▶
Three sentences are enough, and they go in block 5. First, what the disagreement is, with both sources and their tiers. Second, that it is about a factor, the average return of portfolios sorted on cheapness, usually measured with book value. Third, why the memo's conclusion does not depend on it: the memo values one business from its owner earnings, not from its book value, and its kill criteria are about that business.
What a memo must not do is borrow certainty. "Value has always outperformed" overclaims one side. "Value is dead, so the cheapness of this business is meaningless" overclaims the other. Both turn an unresolved literature into a premise. A reader who later learns the literature is disputed will trust nothing else on the page.
The same rule applies to any contested input: say there is a disagreement, give its shape, and show that your decision survives either answer. If it does not survive, that dependence is itself a risk and belongs in block 5 in plain words.
One number, two honest sentences. The big-stock value premium in Fama and French's split fell from 4.3 percent a year to 0.6 percent (reported, via Chicago Booth Review). A memo that stops there is misleading. The honest version adds the authors' own next line in substance: the decline is statistically indistinguishable from zero. And then the counter-reading: Arnott and co-authors argue the period's weakness owed much to value getting cheaper and to intangibles distorting book value. Three clauses, two sources, both tiers stated, no verdict.
Write the memo for one filer
★ 22 min · everyone buildsA worksheet with one short Python step. The Python does not compute anything new: it prints your filer's starting point, range and implied growth exactly as the source map records them, so the numbers in your memo are the ones the bench will reproduce. The prose is yours. Pick one of the four filers, or carry one you built yourself through sessions 1 to 8 and use your own figures, labelled as yours.
Step 1 · Choose a filer and write the header. One business, one date, one line on why you are writing about it. If you are using one of the four filers, write which fiscal year your numbers end in.
Step 2 · Block 1, the owner's paragraph. Four sentences, in the shape from Owner session 1. If you cannot write the fourth sentence, what would break it, you are outside your circle of competence and the honest memo says so and stops.
Step 3 · Block 2, the starting rung and its justification. The ladder, as the source map records it, mean absolute error across the four filers. Pick a rung and write one sentence that cites this table, including its awkward parts.
Step 4 · Block 3, the range, printed from the map. Copy the numbers in; do not recompute them. Rung 1 has no range in the map and rung 4 equals rung 3 at the same settings, so the table holds rungs 2 and 3. Type any test price; the script says where it sits and, for the two map results, what growth it implies.
Step 5 · Block 4, what the test price implies. Run it twice, once at double the central value and once at half, and write one sentence for each that places the implied growth against something outside the model: the filer's own history, or the base rates from Analyst session 7. For P&G the half case, a test price of 47.58, sits below the whole range and implies -4.31 percent a year.
Step 6 · Block 6, three kill criteria, then block 5. Write the criteria before the risks, so the risks cannot soften them. Each one names a number, a window and the filing that will show it, and is calibrated to this filer's own history. Then list the risks, including the three known errors from Analyst session 1 and one line on the value-versus-growth disagreement from Part 3.
File it with the date and leave it alone. Keep the memo exactly as written, even after you change your mind. The value of a kill criterion is entirely in the fact that it was set before the outcome; a memo edited later to match what happened is a diary, not a test.
Before Analyst session 10
30 min- Finish the memo for your filer, all six blocks on one page, dated, with the price-seen box in the header filled in honestly.
- Swap kill criteria with someone and grade each of theirs on the three properties: observable, dated, specific. Rewrite any that fail one, and check that none is already true today for that filer.
- Write the same memo from rung 2 instead of your chosen rung. Which blocks change, and do any of your kill criteria now need a different number?
- Write the three-sentence block on contested evidence in your own words: the disagreement, both sources with their tiers, and why your memo's conclusion does not depend on which side is right.
Three questions
Clear all three and Analyst session 9 gets stamped in your passport.
1 · Which of these is a good kill criterion for a memo on Procter & Gamble's rung-3 thesis?
A is observable in the filing, has a window and names a number. B is none of those. C is about the price, not the thesis about the business. D sounds right but is already true: P&G's ten-year capex to D&A ratio is 1.19, so the criterion fires on day one and tests nothing until it is calibrated to the filer's history.
2 · Why are blocks 1, 2, 3, 5 and 6 of the memo written before any price is looked at?
Anchoring (Tversky and Kahneman, 1974, reported) pulls an estimate towards a number heard first, even an irrelevant one. Writing the range and the kill criteria first, and dating them, makes any later drift visible. Block 4 still uses a price, a test price you type; the order is the point, not a ban.
3 · What does the value-versus-growth evidence in Part 3 support saying in a memo?
Both sources are reported, and they disagree about what the weaker period means. A memo states the disagreement with its tiers and shows that its conclusion, about one business's owner earnings, does not rest on either side.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Analyst session 9 cheat sheet · pin this
You have a memo with a range, a test price and three criteria that could prove you wrong. Analyst session 10 puts every rung of the ladder on one bench: the four filers, the hold-outs, the anti-lever that misses by two orders of magnitude, the break button that shows why terminal growth cannot reach the discount rate, and the one sentence you should be able to say about any intrinsic value you compute.