The rules

Every pick is opened, reviewed and closed under these rules, published before any pick they govern. A change to a rule is a dated row on the changelog.

1. Three sleeves, each against its own fund

   The board has three sleeves: Growth, measured against QQQ; Defensive & Quality, measured against USMV; and Income, measured against SCHD. SPY is the reference for all three, and BIL stands in for cash. Each sleeve holds four to six names, or three while the board is being built.

   Version r1 · adopted 31 Oct 2026

2. How a Growth name is chosen

   We rank companies on two things, averaged as percentiles: how many analysts have raised rather than cut their earnings estimates for this year and next over three months, and the stock's total return over the past twelve months leaving out the latest month. Only the top 30% go forward. When SPY has lost ground over 24 months and its one-month swings are in the wildest fifth of the past ten years, the twelve-month return is dropped from the ranking for new Growth names.

   Every name must also clear four gates: gross profit against assets in the top half; free cash flow above zero over the past year, or a disclosed path to it within four quarters; net debt under three times EBITDA; and a valuation at or below its own 70th percentile of the past five years on forward earnings or on sales, with its forward earnings multiple, set against expected three-year earnings growth, below its sector's median. A nine-part screen must confirm the name. Clusters of insider trades and changes in filing language can break a tie or rule a name out; they never lift it.

   Version r1 · adopted 31 Oct 2026

3. How a Defensive & Quality name is chosen

   We rank companies on a quality score that averages profitability, financial safety, and how they pay out or issue shares. Only the top 30% go forward. A name must have moved no more than 0.85 times as much as the market over three years, sit in the calmest 40% over the past year, and be valued at or below its own ten-year median plus one standard deviation. No more than two names may come from utilities, real estate and consumer staples together.

   Version r1 · adopted 31 Oct 2026

4. How an Income name is chosen

   We rank dividend growers on free cash flow as a share of market value. A name must have raised its dividend for at least five years in a row, by at least 5% a year over five years; pay out no more than 65% of earnings (80% for regulated utilities) and no more than 75% of free cash flow; and carry net debt no more than three times EBITDA, adjusted for its sector. A name is ruled out when any two warning signs show together: dividends above free cash flow over the past year, a yield more than twice its sector's median, this year's earnings estimate down more than 10% in 90 days, debt and payout both rising, or a Piotroski score of four or less. Up to two equity REITs may be held, judged on adjusted funds from operations. Mortgage REITs, business development companies, preferred shares, partnerships that issue K-1s, closed-end funds and covered-call funds are never held.

   Version r1 · adopted 31 Oct 2026

5. What never moves a ranking

   Dividend yield, a single insider purchase, the drift after an earnings report, and the next day's mood are never used to rank a name.

   Version r1 · adopted 31 Oct 2026

6. The valuation ceiling

   Each pick names one valuation measure on the day it is added: forward earnings, operating earnings against enterprise value, free cash flow, or, for REITs, adjusted funds from operations. Its ceiling is the stock's own 85th percentile on that measure over ten years, and never more than 1.3 times the measure at entry (1.6 times for Growth). We check it on every close. The first time a close reaches the ceiling, the pick leaves at the next open, unless estimates are still rising on balance over 90 days and the stock is in the top third on twelve-month return; then the ceiling is raised once, by 15%. The second time, the pick leaves.

   Version r1 · adopted 31 Oct 2026

7. What would prove a pick wrong

   Each pick names up to three disconfirmers on the day it is added, chosen only from this list: a named business measure missing for two quarters; this year's earnings estimate down more than 15% from entry; a dividend cut, new shares above 5% of the count, or a debt-funded deal above a size named at entry; an auditor change, a restatement, a material weakness, or an unusually large change in filing language; or at least three insiders reducing their holdings outside routine plans within 60 days. We read every filing and earnings release against them. When one is met, the pick leaves at the next open after the release. In the Income sleeve a dividend cut always ends the pick.

   Version r1 · adopted 31 Oct 2026

8. The yearly review

   Every twelve months each pick must earn its place again: it has to pass the screens as they stand that day and rank above the middle of the names that would replace it. If it does not, it leaves at the next open.

   Version r1 · adopted 31 Oct 2026

9. No price stops

   A falling price alone never ends a pick. When a Growth pick trails its sleeve's fund by 25 points from entry, or a Defensive or Income pick by 15 points, we publish a note within five trading days. The note can only keep the pick or apply a disconfirmer it already carries. It can never add a new reason to leave.

   Version r1 · adopted 31 Oct 2026

10. Ninety days out

    A name that leaves the board cannot return for 90 days.

    Version r1 · adopted 31 Oct 2026

11. Weights and limits

    Every name starts at the same weight. There is no calendar rebalancing: a name is brought back to its set weight only when it grows past 1.75 times or shrinks below half of it, at the next open. No more than three names, or 25% of the board, from one sector, and no more than two from one industry. A new name cannot move with any name already held more closely than a correlation of 0.75 over the past year. At most one name is added to each sleeve in a week, and at most 24 names are replaced in a year. Each month we flag the board if it moves more than 1.2 times or less than 0.7 times as much as the market.

    Version r1 · adopted 31 Oct 2026

12. How the board is scored

    The board is scored against SPY and against a blend of QQQ, USMV and SCHD at the sleeves' actual weights, with BIL for the cash. Scoring of the whole board starts once it holds twelve names. Results run from the start and over three, six and twelve months, and nothing is annualized before twelve months. Alongside them: the swings, the worst fall from a peak, how much the board moves with the market, how far it strays from the blend, the return per unit of straying, how it fares in rising and falling markets, and a 90% range on the excess return drawn by resampling. These figures describe; they do not prove. With about 8% tracking error, a true 3%/year edge needs about 28 years of data to reach t = 2. A 6–12 month record cannot distinguish skill from luck.

    Version r1 · adopted 31 Oct 2026

13. What the record keeps

    For every pick: the screens it passed with their values, its ceiling, its disconfirmers, and the model and data cutoff used to draft it. The share of picks ahead of their fund after three, six and twelve months, and the size of the average gain against the average shortfall. How much of the result came from the mix of sleeves and how much from the names. Every pick that leaves is followed for twelve months against its fund and against a name still held, drawn at random on the day it left. The share of exits a rule caused (the goal is all of them), the number of rule changes (the goal is none), and the days from a rule being met to the exit.

    Version r1 · adopted 31 Oct 2026

14. How a pick is written

    Every case is drafted with the base rate stated: newsletter picks show no average skill after costs. The model that drafts it is never asked for a price or a return; the ceiling is set by these rules, not by the model. Before a pick is added, a second reader checks every number against the filing, confirms the entry valuation clears the gate and the ceiling and disconfirmers are in place, and reads the case against first, answering it in one paragraph.

    Version r1 · adopted 31 Oct 2026