The Desk Playbook — Main St. Investors

Framework 01

The Alpha Signal test

Most "ideas" are observations wearing a costume. An observation says a thing is cheap. An idea says why the market has it wrong, what will correct it, when, and what it costs you if you're wrong. Five questions. If you can't answer all five in writing, you don't have a position — you have an opinion.

1. Variant perception — what does the market misunderstand? Not "this is a good company." What specifically does consensus have wrong, and why does it have it wrong? If your answer is something anyone could read in a headline, it's already in the price.
2. Catalyst — what specific event closes the gap? A date, a print, a decision, a filing. "Eventually the market will notice" is not a catalyst. Without one, you're paying to wait with no idea how long.
3. Timeline — 1–3 months, 3–6, or 6–12? Commit to a bucket up front. It decides your sizing, your patience, and whether a drawdown is noise or evidence you were wrong.
4. Downside — bear case level and maximum loss? Where does this trade if you're wrong, and what does that cost in dollars? Written before entry, not rationalised after.
5. Risk/reward — upside versus downside, in dollars and percent? Do the arithmetic. A lot of exciting ideas die quietly at this step, which is the entire point of the step.

Question 1 tells you whether there's an edge. Question 4 tells you whether you can afford to be wrong about it.
— The Desk

Framework 02

Position sizing bands

Nobody blows up on a bad idea. They blow up on a bad size. Bands remove the single most emotional decision in investing — how much — by making it a rule you set on a calm day rather than a feeling you have on a red one. These are percentages of your total portfolio, not of your cash.

Standard positionThe idea clears all five Alpha Signal questions.2–4%

High convictionClears all five, plus a dated catalyst and a defined downside you'd accept twice.5–8%

Maximum single nameA hard ceiling. Not a target, and not something conviction can argue its way past.10%

Short positionsAlways price the borrow cost and days to cover before sizing — a crowded short is a different risk from a crowded long.+ borrow

The honest test for which band an idea has earned: could you write the bear case out loud, to someone who disagrees with you, without getting defensive? Standard band if it's a maybe. The 5–8% band is rarer than it feels in the moment.

The cap is the point. A 10% ceiling means no single name can end your year. Every rule above it exists so that the one idea you were most certain about — and they are always the ones — can't take the whole book with it.

Framework 03

The macro tiers

Financial media treats every data release as urgent. Almost none of them are. Sorting releases into three tiers tells you which prints are worth repositioning around, which are worth reading, and which are noise with a timestamp.

Tier 1 — market-moving

Non-farm payrolls. CPI and core CPI. PCE and core PCE. Advance GDP. FOMC decisions. ISM Manufacturing. These re-rate the rate curve, and everything priced off it. Know the date, know what's expected, and know your exposure before the print — not after.

Tier 2 — important, rarely decisive

Retail sales. Industrial production. Housing starts. Jobless claims. PPI. Consumer confidence. These confirm or complicate the Tier 1 story. They move sectors more often than they move the index.

Tier 3 — supplementary

JOLTS. Durable goods. Trade balance. Productivity. Unit labour costs. Useful for building a picture over months. Almost never a reason to change a position on the day.

Plain version: if you only have ten minutes a week for macro, spend all of it on Tier 1. A Tier 3 release that contradicts your thesis is information. It is not a signal to trade.

Framework 04

Scenario weighting

A forecast with no probability attached can't be wrong, which means it can't be useful. Every macro view we publish gets three scenarios with weights that sum to 100, each carrying the trade that expresses it and what that trade does if the scenario lands.

BULL · 20–25%The upside case. The macro state that has to be true, the expression that benefits most, and the profit if it happens.

BASE · 50–60%What you're actually positioned for. The majority of the weight, and the scenario your book should survive comfortably.

BEAR · 20–25%The one that costs money. Named explicitly, with the loss quantified and the hedge — or the decision to wear it — stated in advance.

Two disciplines make this work. First, the weights are set before you pick the trade, not reverse-engineered to justify one you already like. Second, when a scenario's preconditions stop being true, the weights get rewritten and the position changes with them. A scenario table you never revise is decoration.

Watch the bear weight. If yours is consistently under 20%, you're not forecasting — you're rooting. The bear case is the one that decides whether the position survives long enough for the base case to pay you.

Framework 05

The mistakes log

This is the least glamorous framework here and the one that compounds hardest. Keep a written log of expensive errors, sorted by the kind of error rather than by the stock. Not a trade journal — a lessons file. The categories that matter: sizing, entry timing, stop placement, exit discipline, and claiming an edge from too small a sample.

Before you log a mistake, ask: would I make this again if I forgot I'd made it once? If yes, the note isn't sharp enough yet.
— The test we apply to every entry

Vague lessons don't prevent recurrence. "I got greedy" teaches nothing. "I sized a 3-month idea like a 12-month one because the catalyst felt certain, and the catalyst slipped a quarter" is specific enough to catch yourself next time.

The ritual: read the log on a fixed day — we use Saturday. Re-read three entries at random before committing to the following week's positions. That's the whole practice. It takes ten minutes and it's the reason a desk gets better over a decade while a retail account tends to repeat the same year.

Framework 06

All five, on one real trade

Frameworks are easy to nod at and hard to apply. Here they are run against a trade we've written up in full — the AI power trade, which was right for eighteen months and then wasn't. It's the clearest example we have of a thesis surviving while the position stopped working.

Worked example · The AI power trade

What the frameworks caught, and what they missed

Variant perceptionGenuine. The market was pricing compute; the binding constraint was generation. This was the part we got right, and it produced the entire move.

CatalystWeak — "data centre demand keeps growing" is a trend, not a dated event. A thesis without a catalyst gives you no schedule for being paid, and no trigger for reassessing.

TimelineNever committed to a bucket. That's the root failure: a 6–12 month idea held on a multi-year clock, with no rule for what happens in between.

DownsideUnder-modelled. The crowding risk didn't show up in the bear case, and the one-day test in January 2025 priced it at −28.3% — worse than the crowded chip trade it was meant to be safer than.

Bear weightToo low. A serious bear case would have named "consensus has already moved" as the risk, which is exactly what the drawdown turned out to be.

Logged asEntry timing / exit discipline. Signal-loud means trade-crowded — the recurring trap, now written down where we'll read it again.

The full teardown, with the price charts and the eighteen-month scoreboard, is on the case study page. It's worth reading alongside this guide — the frameworks make more sense against a trade that cost money than one that didn't.

Read the case study →

Five frameworks. Use them by name.