Actionable insights — All The Pros All Buying These Companies

3:41 1. Before reading a sale as a signal, test whether it was forced

1. Look at the direction mix of the whole filing first. All-reductions-and-no-buys is the shape of a cash need, not of a re-underwriting; a genuine change of mind normally shows up as a sale on one side and a purchase on the other.
2. Check the prices sold into. Selling names that have already fallen and now screen cheap is the opposite of what a discretionary seller does — it's what someone raising money does.
3. Check the manager's stated cash policy (letters, interviews, past filings). A fully invested manager with no cash buffer must sell shares to fund a withdrawal.
4. Confirm with the performance record: redemptions cluster after a prolonged stretch of underperformance, when clients start asking for liquidity.
5. Then read the residual signal — the uneven cut sizes. Even a forced seller chooses which positions absorb the hit, so relative trim percentages reveal preference even when the sale itself doesn't.

• A quarter with zero purchases; sales concentrated in the most liquid names; a fund whose AUM fell more than its performance explains; the untouched position (that one is the real conviction).

6:53 2. Audit a portfolio by risk factor, not by company count

1. List every holding and, for each, write down the two or three macro variables that actually drive its revenue — not its sector label.
2. Group by shared variable. Count how many distinct groups you have; that number, not the number of tickers, is your real diversification.
3. Name the Achilles' heel explicitly: the single variable that, if it goes the wrong way, takes down the largest share of the book at once.
4. Fix it by substitution, not by adding names — swap a duplicate exposure for an uncorrelated one, and refuse to hold two near-identical businesses just because both are excellent.
5. Re-run whenever a position grows: a winner compounding inside an already-crowded factor quietly re-concentrates the book.

• Two holdings whose charts move together on the same news day; a "diversified" list where every name is sensitive to one rate or one commodity; the odd-one-out position being sold for valuation reasons (you may be selling the only real diversifier).

8:50 3. Explain the untouched position — the uncorrelated holding earns a valuation exemption

1. In any forced-sale quarter, find the position that was not reduced, and ask why that one specifically.
2. Test the obvious explanations in order: is it the cheapest, the highest conviction, or the least correlated with everything else?
3. If it's the least correlated, treat its retention as a portfolio-construction decision rather than a valuation call — and don't infer that the manager thinks it's cheap.
4. Apply the same test to your own book before trimming a stretched winner: check whether it is also your only genuine diversifier, and price that role into the trim decision.

• A manager holding an "expensive" name through a cash crunch; a diversifier that has grown so large it has become the concentration; the moment the uncorrelated name starts correlating (then the exemption expires).

11:44 4. Sell on moat erosion, not on price — and state the mechanism

1. For each holding, write down the specific thing that makes it hard to replace — the certification burden, the network, the workflow lock-in, the distribution.
2. Then name what would have to become true for that to weaken, and monitor the outskirts of the business for it continuously, not at earnings.
3. When the erosion mechanism becomes concrete rather than hypothetical, exit fully; don't wait for it to show in the numbers, because by then the multiple has already gone.
4. Trace the second-order damage before sizing the decision: which other revenue line depends on the eroding one?
5. Balance it against the reverse discipline: be equally willing to declare an earlier erosion thesis wrong when the data refuses to confirm it.

• A complementary product quietly becoming the interface customers use first; pricing power softening before volumes do; a dependent business line nobody is modelling as exposed; and, on the other side, a feared disruption that never shows up in the reported numbers.

14:02 5. Split a great investor's argument into the part you accept and the part that changes your position

1. Restate the opposing thesis in its strongest form and concede every part of it that is factually true.
2. Separate direction from magnitude: agreeing that value is shifting is not the same as agreeing that the moat is now too narrow to own.
3. Set the threshold that would actually force your sale, in terms of the moat rather than the price, and check whether the argument clears it.
4. Where two credible managers take opposite sides of the same pair, consider owning both and removing the need to be right about the pair.
5. Score the disagreement later against the price path — not to gloat, but to calibrate how much weight to give that manager's next moat call.

• An argument you can only rebut with price action (weak); a moat threshold you can't state in advance (weaker); the same manager's next reversal, which tells you whether the vigilance is discipline or churn.

17:39 6. Read weighting, not the buy list — the top position is the recommendation

1. Ignore the count of new names. Sort by portfolio weight, and separately by percentage added this quarter.
2. The name at the top of the weight column is the manager's answer to "what would you buy today" — concentrated managers cannot hide that.
3. Cross-check with the add rate: a new position climbing fast toward the top carries more information than a legacy holding that grew by appreciation.
4. Read the same names across several managers' filings; convergence across independent books is a stronger signal than any single fund's action.
5. Finally, discount for lag — the trade is at least six weeks old — and re-underwrite at today's price before acting.

• A big headline "new position" that is only 0.5% of the fund; a starter "watcher position" (~1.4%) being reported as conviction; the same two or three names appearing at the top of unrelated managers' books.

19:34 7. Trim strength slowly — an expensive price is not a finished move

1. When a position doubles, resist the reflex to take the whole gain; decide instead what fraction to sell and over what period.
2. Distinguish the two reasons to trim: valuation stretch (sell gradually) versus thesis damage (sell fully). Only the second justifies exiting quickly.
3. Where sentiment is strongly positive and the fundamental story is intact, assume the move overshoots your fair value — "they travel up a lot further than you'd think."
4. Keep the residual position large enough that being wrong about the trim doesn't cost you the thesis.
5. Record the counterfactual afterwards (where it traded after your trim) as calibration for the next one.

• Momentum plus improving fundamentals (trim slower); momentum without fundamentals (trim faster); your own urge to sell purely because the number is large; a trim that quietly turns into an exit.

25:30 8. Price a scandal with cohort math and a probability tree, not with the headline

1. Identify precisely which cohort or product line the legal action concerns — not the company as a whole.
2. Pull that cohort's share of users, then its share of revenue, which is usually smaller. Ask who actually pays: the user or someone else.
3. Build the remediation timeline: list the changes the company has already made, year by year. That is what a court weighs and what caps the incremental cost.
4. Write two or three explicit outcome branches with probabilities — the survivable one (a fine plus tighter controls) and the business-changing one (a forced product/algorithm overhaul extended beyond the cohort).
5. Ask whether the extreme branch is even coherent for a regulator to impose, given the competitive market it would leave the company in.
6. Compare the market's implied reaction to your weighted outcome, and act on the gap rather than on the coverage.

• A framing analogy ("big tobacco moment") doing the work that numbers should; an accusation type nobody will publicly defend, which suppresses the counter-argument and widens the mispricing; remedies quietly extending from the named cohort to the whole user base — that is the branch that matters.

29:33 9. Calibrate on the last identical panic in the same company

1. Find the prior episode where the same company faced comparable existential coverage, and pull the actual headlines from that window — not the retrospective summary.
2. Note what was predicted (advertisers gone, users fleeing, permanent impairment) and check each claim against what subsequently happened.
3. Mark the price then and now to get the realised cost of having sold the fear.
4. Use it as base rate, not proof: ask what is genuinely different this time, and require a specific answer.
5. Pre-commit to the behaviour — keep buying on the schedule you set before the news broke.

• The specific difference that would break the analogy (a structural remedy rather than a fine); coverage volume peaking while the fundamentals hold; your own willingness to keep buying on schedule while it's loudest.

31:17 10. Assume the documents you feed a model are adversarial

1. Treat any third-party document you pass to an AI system — a filing, a transcript, a PDF, a webpage — as potentially containing instructions aimed at the model rather than at you.
2. Strip formatting before analysis: read the extracted plain text, where invisible tricks (white or micro-sized type) become visible.
3. Where the output matters, verify the model's conclusion against the numbers yourself rather than accepting a summary of a document you never read raw.
4. Expect the frequency to rise as more institutional processes route through models, and build the check into the workflow rather than doing it ad hoc.

• Documents whose plain-text extraction is longer than what renders on screen; summaries that assert a conclusion the numbers don't support; any workflow where a model reads a document no human has opened raw.