Actionable insights — Your 10 Favorite Stocks

1. Close every idea list with the same four numbers, and let the table do the judging

1. Pick four columns that between them cover profitability, capital efficiency, price and growth — here net profit margin, ROIC, forward PE and long-term expected EPS growth.
2. Apply them identically to every name, including the ones you are enthusiastic about and the ones you would never own.
3. Colour-code against your own thresholds rather than against the list's internal range, so a list of bad businesses cannot grade on a curve.
4. Publish the table without commentary. If the numbers contradict the prose, that is the useful part.

• Stated ROIC that penalises a business model rather than the business — insurance float (KNSL 10.5%) and acquisition accounting (CSU.TO 11.8%) both understate it.
• Blank cells. Three names carry no long-term EPS growth estimate at all, which is itself a signal about forecastability.

2. Compress a thesis to one claim you could be publicly wrong about

1. Ask what single measurable fact the case depends on, and write it as a sentence with a number in it.
2. Source that number to a primary document — an investor presentation, a filing — with the date, so it can be rechecked.
3. Show the trend across at least two points, so the claim is an extrapolation rather than an assertion.
4. Commit to abandoning the position if the number stops moving, and say so in advance.

• Two data points extrapolated as a trend. Three years of share gain in a hardening insurance market may not survive a soft one.
• A "one simple fact" that is actually a compound claim — market share can rise while margins fall.

3. Build the case that works even if growth is zero

1. State the deterioration plainly first, without hedging, so the case is not resting on denying it.
2. Compute shareholder yield — dividends plus net buybacks — as a percentage of market value at today's price.
3. Ask whether that yield alone clears your required return. If it does, growth becomes optionality rather than a requirement.
4. Name the one assumption still doing work, usually that current earnings hold — and monitor that, not the growth rate.

• A double-digit shareholder yield funded from a shrinking earnings base. The yield falls with the earnings, so it is not a floor.
• Buybacks that only offset stock compensation. Check shares outstanding, not the money spent.

4. Read the pay package as a business input, not a governance footnote

1. For any founder- or owner-led business, find what the chief executive is paid and what triggers it.
2. Ask what behaviour the package rewards — a share price on a date, a revenue milestone, nothing at all.
3. Match that against the behaviour the business model needs. A permanent-capital acquirer needs patience; a target-linked award buys urgency.
4. Weight publicity-seeking the same way: a chief executive who has promised a number in public has a reason not to walk away from a bad deal.

• Low pay used as a proxy for good capital allocation. It is evidence, not proof — check the deals.
• Milestone awards that are unreachable. They stop being an incentive and become a free option.

5. Haircut management's own targets before building a return on them

1. Convert multi-year targets into implied annual growth rates so their aggressiveness is visible.
2. Apply your standard discount to forecasts — the house habit elsewhere in this archive is 30-40% off analyst estimates.
3. Rerun the valuation at the haircut number and see whether the case survives.
4. Check the metric the targets are set in, and whether it is the one that reaches shareholders.

• EBITDA targets for a serial acquirer. They exclude the amortisation of what was bought, which is most of the cost.
• A metric that adds back charges and also sits above the minority interests owed to partner-shareholders.

6. Discount a "wisdom of crowds" result by how correlated the crowd is

1. Ask where the respondents got their information. The cow-weighing result requires independent guesses.
2. Check the overlap between the crowd's answers and the surveyor's own published positions.
3. Treat high overlap as evidence of reflection, not of consensus — the list is measuring readership, not the market.
4. Use it as a sentiment gauge on your own holdings rather than as idea generation.

• Your own book coming back to you as external validation. That is the failure mode this format is most exposed to.
• Vote counts read as conviction quality. On the 2025 evidence the correlation ran the wrong way.