The Eightfold Rules: Version 1.0

In force from November 1, 2026. My record starts that month too. Every rule below was written and dated before the first trade.

What this is

Every rule for the portfolio, published before any results exist, so that anyone can watch whether I follow them or not. Every breach of these rules is counted in the Breach Count, published every month and never reset.

The values follow one principle: conventional and defensible beats optimized. Where a convention exists, I start from it, and the table at the end marks each value that departs from one as Judgment. Where none exists, the value errs toward less activity, because the pull in this kind of investing is almost always toward more. Several rules were changed in October 2026 after tests on a backtest from 2016 to 2026, so that backtest flatters these rules, and none of its results is a forecast.

The portfolio

The Core: a long-only portfolio of US-listed stocks, tilted toward the characteristics of the Fama-French five-factor model. It holds market exposure deliberately and isn't hedged. These rules cover this one account and nothing else. I've designed an options account too, but it's not in this version. It may be launched later, with rules published before its first trade.

Which companies are eligible

I only rank companies that pass every one of these gates:

Market value. The floor is where Fama and French draw the line for microcaps: the 20th percentile of market value among NYSE stocks. Kenneth French publishes it every month, a month or two behind, because it's built from data that arrives after each month closes. On each monthly date I use the newest month he's published, and the Ledger says which month that was. If his site can't be reached, I use the last copy I downloaded. If his newest month is more than six months old, I don't run the screen until it's fixed.

Depositary receipts are eligible, with one extra gate, so companies such as Shell, Sanofi, and Unilever can be held. Owning them costs extra: a small depositary fee, foreign withholding tax on dividends, and a currency spread. The bank behind a depositary receipt will charge a fee of up to about 5 cents per receipt each year, so I set the floor at $5 to keep that fee under 1%. Beyond that, I accept these costs and don't score them.

Partnerships taxed as partnerships and royalty trusts aren't eligible. I skip those partnerships because their K-1 tax forms can mean filing a tax return in every state where they do business. A partnership that has chosen to be taxed as a corporation sends the usual 1099, so it's eligible. A royalty trust collects royalties and pays them out; it doesn't run a business, so its book value and asset growth say nothing a ranking can use.

Reconciliation. The whole ranking depends on a handful of numbers from a single data provider, so I check them against each other and against the year before, and a company whose numbers don't agree is out:

For the two market value checks, a factor of two can't be ordinary price movement; it can only be a unit, ratio, or currency error. The share count gets two-thirds, since no company buys back a third of its shares between one annual report and the next. Total assets get a factor of ten, since a real company can lose three quarters of its assets in a year through a sale or a write-down.

How companies are ranked

Every eligible company gets scored on three characteristics:

1. Rank each from 0 to 100
2. Average the three at equal weight
3. Rank by that average

An industry with fewer than 10 eligible companies is ranked at sector level for value. All three scores are required: a company missing one isn't ranked, and a company with negative or zero book equity has no value score. I run the screen every month, on the latest closing prices, except that value can use the price from a year earlier, as explained below. Nothing in selection uses a valuation model.

Why the larger market value. With today's price, a stock that fell this year would look cheaper by exactly its fall, so the ranking would lean toward recent losers, and stocks that have fallen over the past year have tended to keep falling for a while. Last year's price leaves out the latest fall, so a stock that's just dropped doesn't look cheaper yet. A rise is the opposite case: on last year's price, a stock that has doubled looks as cheap as it did before it rose, though the market now values it at twice as much. So value uses whichever market value is larger, today's or the one a year earlier. A fall waits a year to make a company look cheaper, and a rise counts at once. Fama and French do much the same, their value measure uses a market value that's about a year old on average. Market value a year earlier means today's share count times the share price on the monthly date 12 months back, adjusted for splits. A company without a price from a year back uses today's.

Why total assets. Fama and French divide operating profit by book equity, and debt flatters that: a company funded mostly with borrowed money has little equity, so an ordinary profit becomes a large return on it. In October 2026 Organon's equity was 6% of its assets, which made its profitability 151% and put it near the top of the ranking. Dividing by total assets measures what the whole business earns however it's financed, as Ball, Gerakos, Linnainmaa and Nikolaev proposed in 2015. Interest is still subtracted, so expensive debt still counts against a company.

Lease-to-own companies. A lease-to-own company depreciates the merchandise it leases out, and that depreciation is the cost of what it sells. Added back, it makes the company look far more profitable than it is: in October 2026 PROG Holdings' profitability came out at 113% of its assets, against 12% without it. So depreciation and amortization isn't added back for a company whose business description from the data provider says it offers lease-to-own or rent-to-own, once its depreciation and amortization is more than a quarter of its total assets. Ordinary depreciation is almost never that large, and below it the add-back changes little. The screener checks every company each month, so a new one is covered the month it appears.

The trap guard

Cheap companies are sometimes cheap because they're failing. A company isn't bought if its Piotroski F score is 4 or below, or if it has no score. The F score is nine yes-or-no tests of whether a company's profitability, cash flow, debt, liquidity, share count, and efficiency are improving or deteriorating. The guard blocks purchases, including further purchases of a company already held.

Known tilts, not corrected

• Buybacks. Companies shrinking their share count are over-represented. Nothing asks for this; profitability and investment produce it.
• Negative or zero book equity is excluded, because such companies can't be scored on value.
• Recent listings wait. The history gate keeps a new listing out until it has three years of annual statements, the minimum its F score needs.
• Asset sales and spin-offs. A company that sells or spins off a large business shows its assets shrinking, which the investment score reads as restraint. In October 2026 Sony (assets down 55%) and Harley-Davidson (down 32%) scored near 100 on investment for that reason. A fix tested in October 2026 didn't help, so the January revision will look at it again.
• Smaller companies. Each month's money brings the companies bought toward the same position size, whatever their market value, and the eligible list reaches down to the NYSE 20th percentile of market value (about $1.2 billion in 2026). So the portfolio leans toward small and mid-size companies compared with the world market (VT), and the quarterly factor loadings show this as a positive size loading. The lean costs the Core in years when the largest companies lead: in the backtest from 2016 the Core trailed VT in three of the six years they led, one of them by a fifth of a point.

Buying

Every month I work down the ranked list from the top:

1. Take the top 15 eligible companies not blocked by the trap guard
2. Skip any company that's agreed to be acquired, is in bankruptcy, has said its financial statements can't be relied on, or says it's a passive foreign investment company, as well as any I can't buy that day. Also skip any already at 10% of the portfolio, any in an industry already at 15% or a sector already at 25%, and any in a country already at its cap or in an emerging or frontier market when those together are at theirs (see Caps). Skip, too, any I don't already hold whose caps leave it room for less than 0.5% of the portfolio, since the housekeeping rule would close a position that small. A skipped company passes its place to the next one down, so fifteen are always being bought
3. Split the contribution across those fifteen toward making them equal in size, so the smallest receives the most

None of the month's money is left in cash. When the caps stop money from reaching the fifteen, it goes to the next companies down the buyable list that aren't skipped or at a cap, split among them toward equal size and added one at a time until it's all placed. Here too, a company I don't already hold is passed over when its caps leave it room for less than 0.5% of the portfolio; only if nothing on the list has more room does the money go to the smaller openings.

A company entering the top 15 starts at zero and receives the most that month. A company that stays in the top 15 keeps every share I own unless a cap trims it, and new money goes first to the ones I hold least of, so no single company comes to carry the portfolio. A company leaving the top 15 is still held, and gets money only when the caps send it down the list. Holdings accumulate.

After a market fall. On each monthly date I check the US market's total return over the previous 24 months, dividends included, from VTI. When it's negative, that month's purchases come from a second ranking that's the same in every way except that value uses today's price, so new money goes to the companies the fall made cheapest. The steps above work the same way on that ranking. Holdings, the sell line, and the rule for winners keep using the main ranking, so a false alarm never forces a sale. Over the past century, recent losers have beaten recent winners in these months, the reverse of ordinary times (Daniel and Moskowitz, 2016). A negative 24-month return has come about one month in six since 1927.

A company that has agreed to be acquired. Once a company announces a signed agreement to be bought, its price follows the deal instead of the business, so new money would buy only the small gap to the offer and the risk that the deal fails. It isn't bought until the deal closes or gets called off. A holding in that position is kept, and every other rule applies to it as usual. On each monthly date a program searches the SEC filings of the companies in line to be bought for merger proxies, tender offers, and announcements, and I confirm each company it flags from the announcement itself; the record links to it. Talks and offers the board hasn't agreed to don't count.

A company in bankruptcy, or whose statements can't be relied on. The ranking reads a company's latest financial statements, but two events can make those statements the wrong guide. A company in bankruptcy isn't bought from its filing until it emerges or the case is dismissed: shareholders are paid last, and are often left with nothing. A company that has said its past financial statements can't be relied on isn't bought until it files corrected ones. The program that looks for acquisitions searches the same SEC filings for both, and I confirm each company it flags; the record links to the filing.

A passive foreign investment company. US tax law treats a foreign company whose income or assets are mostly passive (cash and investments, for example) as a passive foreign investment company, or PFIC. Held in a taxable account, its gains are taxed at the highest ordinary rate plus an interest charge unless the owner makes an election, and it needs an extra IRS form every year. A company whose latest annual report says it was or expects to be a PFIC isn't bought, and a holding whose report says so is sold. The program that looks for acquisitions also reads the latest annual reports, and I confirm each company it flags; the record links to the report. A warning that a company might become a PFIC doesn't count.

A security that can't be traded. Now and then a company in line can't be bought on the monthly date: the broker doesn't offer it, trading is halted, or a sanction forbids buying it. It's skipped like a company at a cap, and the record says why. A sale that can't be placed waits for the next monthly date on which it can be.

Fractional shares are used where the broker offers them. Otherwise a purchase rounds to the nearest whole share, and what the rounding leaves goes down the list the same way. Less than a dollar waits for the next contribution.

Withdrawals. Money leaves the Core for only two reasons: once, toward buying a home, and for a medical emergency. If I take money out toward a home, I'll do it on a monthly date, but a medical emergency can't wait, so that money can come out on any day. Lots are sold in the order that costs the least tax: losses first, then long-term gains, then short-term gains, and within each group the largest loss or the smallest gain goes first, skipping, where possible, lots in companies bought in the preceding 30 days. A withdrawal is published the month it happens, with its reason and its size as a share of the portfolio. A permitted withdrawal isn't a breach, and the one-year hold doesn't apply to it. A withdrawal for any other reason is a breach.

Caps

Money that a cap stops goes to the next companies down the list. Anything over the cap is trimmed using lots held for more than a year. Anything over one and a half times the cap is trimmed regardless.

A trim only takes a company or a group back down to its cap. A group trim is spread across the group's holdings in proportion to their value. The sector cap stops a broad lean across many industries. The industry cap stops a narrow bet disguised as diversification.

Countries. A company's country is where its headquarters is, as the screener reports it; Puerto Rico counts as the US. Companies based in one country share its law and its currency, so a sanction or a currency crisis hits them together. Emerging and frontier markets, meaning every country outside FTSE Russell's developed markets, share those risks more often. The US cap is the US share of world stock markets and the emerging-markets cap is theirs: the market value of the US companies, and of the emerging-market companies together, in FTSE's Global All Cap index (the one VT tracks), as a share of the whole index, rounded to the nearest whole percent. They're set on the first monthly date of each year from FTSE's figures for the end of the year before, so they follow the world's markets without a change to these rules, and only once a year, so a small move in a single month doesn't force a trim. If I can't find the new figures, the last ones stay in use. The Rules page shows the figures in use and their date. Until January 2027 they're 62% and 10%, from the end of September 2026 (62.35% and 10.02%). Bermuda, the Cayman Islands, and other offshore centers each get the usual 10% cap, but I don't count a company based there as an emerging market, because the address says little about where its business is.

Selling

Six triggers create a sale. Losses aren't sold just for the tax: a cheap company that has fallen 20% is often near the bottom of its cycle, and selling there gives up the recovery.

1. The ranking falls. I sell a holding once its rank drops below the top half of the ranked list, however big the position is. The middle of the list is a plain place to draw the line: a company stays while it still ranks above the typical eligible company. In the backtest from 2016, companies bought earlier that slipped out of the top 30% but stayed in the top half went on to do at least as well as the top 15 bought that month, so a tighter line would mostly have added trading. A large position can slide to the middle of the list before it's sold, but the company cap keeps any one company at 10% or less.

Winners are held. I don't sell a holding below the line while its momentum is in the top 30% of ranked companies, the cutoff Fama and French use for momentum winners. Momentum is the stock's total return, dividends included, from the monthly date 12 months back to the monthly date one month back; the latest month is left out, as in the research. Stocks that have risen over the past year have tended to keep rising, so a winner whose rank slips is kept until its momentum fades. This only changes the ranking trigger, so a winner can still be trimmed at a cap. Every other trigger works as before.

A holding with no rank (it no longer passes a gate, or a score is missing) isn't bought, and is held while it stays unranked. At the 2nd consecutive monthly date without a rank, its lots held for more than one year are sold, at a gain or a loss. The one month's wait keeps a gap in a single month's data from forcing a sale. A lot I've held for less than a year remains held, and I sell it on the first monthly date after its one-year mark if the company still has no rank. Every other sell trigger still applies.

2. A company exceeds its cap. Trimmed as above.

3. A group exceeds its cap. An industry, a sector, a country, or emerging and frontier markets together: trimmed as above.

4. Housekeeping. A position under 0.5% of the portfolio, outside the top 15, holding only lots owned for more than a year, is closed.

5. Bankruptcy. A holding that files for bankruptcy is sold on the next monthly date, all of it, whatever its lots' age.

6. A PFIC. A holding whose latest annual report says it was, or expects to be, a passive foreign investment company is sold on the next monthly date, all of it, whatever its lots' age.

The one-year hold. I don't sell a lot at a gain until I've held it for more than a year, so the gain is taxed at the long-term rate. A sale that would break this waits for the first monthly date after the lot qualifies, and happens then only if the trigger still holds. Only the qualifying shares are sold; the rest wait. When the ranking falls, a lot at a loss is sold however long I've held it; cap trims, unranked holdings, and housekeeping use only lots held for more than a year. A loss sale within 30 days of buying the same company, before or after, is a wash sale: the loss moves into the newer shares' cost, as the broker reports it, and counts when those shares are sold. There are three exceptions: a holding above one and a half times any cap, which is trimmed regardless, a holding in bankruptcy or that's a PFIC, and a permitted withdrawal.

When a deal changes a holding. A deal isn't a decision, so it isn't one of the six triggers. A holding bought for cash is paid out on the day the deal closes, whatever day that is, and the cash is invested with the next contribution. Shares received in a merger or a spin-off are held like any other holding under the same rules, so a company with no rank is sold under the rule for unranked holdings. When a holding offers a choice, such as tendering shares into an offer or choosing between cash and stock in a merger, I take whatever happens if I do nothing.

The routine

Contributions, the screen, purchases, and any required sales all happen on the 1st of each month, or the next trading day, and at no other time. The one exception is a withdrawal for a medical emergency, which can be made on any day. The Core isn't checked between monthly dates; checking anyway is logged as a temptation. Opening the account to make a permitted withdrawal isn't a temptation.

Every quarter, the Core's returns are measured against the Fama-French factors Kenneth French publishes for developed markets (the US is one of them), and the loadings are published. The first come after the quarter ending in June 2027, when there are seven months of returns to measure. The target is a positive loading on value, profitability, and investment. Value and investment overlap, so the value loading can absorb the investment one. A third or more of the Core is usually based abroad, and the US factors alone would count the gap between US and foreign markets as part of its result. Emerging markets aren't in French's developed-markets factors, so the Core's few holdings based in one are measured against them anyway. A drift from target is published, not quietly corrected. Long-only portfolios capture only part of the published factor returns, which are measured long and short, so I'll see how these numbers play out.

The Premium (not part of this version)

I've also designed an options account that would sell defined-risk put spreads on the S&P 500. It isn't in force and nothing trades in it. If I start it, I'll publish its rules as a new version at least a month before the first trade. It will be reported separately and never blended with the Core.

Governance

Breaches, errors, and temptations

Three records, kept separately, published monthly, never edited or removed.

• The Breach Count is cumulative and never reset. Each breach records the date, the principle it broke, what was done, and what the rule said.
• Errors record what happened and why, and are corrected at the next monthly date. They're counted apart so the Breach Count measures behavior, not competence.
• A temptation is logged when it reaches a concrete step: opening the brokerage account between monthly dates, looking up a holding's price, drafting a trade, or drafting a change to a rule. Each entry records the date, the step, what prompted it, the principle involved, and whether it became a breach.
• No reversals. A breach is never undone by another deviation. The portfolio returns to the rules at the next monthly date.
• Timing. If the 1st isn't a trading day, the routine runs on the next one. If I'm late beyond that for a genuine reason, it counts as an Error. A deliberate delay is a breach.
• Not breaches: a rule turning out to be wrong, a month with no activity, a loss, a sale deferred by the one-year hold, a purchase rounded to a whole share, a permitted withdrawal.

The principle tested each month

Every monthly entry names one principle and whether it held: the one attached to the month's most significant breach, temptation, or hard decision, in that order. If there was none, the entry says no principle was tested. A principle can be broken without a breach, for instance by judging the strategy on one month of data. You'll find each principle's test in The Eight.

Changing these rules

I can fix a factual error, or a rule that's become impossible to follow, at any time. Any other change outside January is itself a breach. Adding the Premium account (selling options credit spreads) is the one exception: its rules may be published at any time as a new version, at least one month before its first trade, and that version may only add rules for the new Premium account.

Never changes

The Core's index level from 100, time-weighted. The Breach Count never being reset. Every version of this document stays published. There will never be trade alerts or signals. A second account, if added, is reported separately and never blended with the Core.

What is published

• Every month: the Core's index beside VT, a fund that tracks FTSE's index of the world's stock markets, companies of every size included (the one the country caps are set from), with dividends reinvested and the difference shown even when negative; contributions as a share of income; any withdrawal, with its reason and its size as a share of the portfolio; the Breach Count, the Error count, and the temptation log; which principle was tested and whether it held; the number of companies held (a count, not their names); and whether the month's purchases came from the main ranking or the one used after a market fall.
• Every quarter: factor loadings against French's published factors for developed markets.
• A month later: every holding by name with its weight, each sector's exposure against its cap, and the screen's full ranked list (both rankings, in a month after a market fall).
• Never: balances, net worth, or income in dollars; my broker or any account identifier; anything in real time.
• Never done: editing or deleting a past entry, or accepting any arrangement that pays according to trading activity.

AI assistance. The standing commitment: no AI system makes, recommends, or influences a live position decision. The rules are written by me and carried out by me.

The rules at a glance

Convention means an established practice is followed. Derived means the value follows from another rule. Judgment means it was chosen within a defensible range and could reasonably be different.

Nothing trades until this document is complete, dated, and published. That's the whole claim.