What a credit spread has to earn, and how far under it we drifted

Win often, lose big

A credit spread sells an option and buys a further one as insurance. It collects the difference up front, and that credit is the most it can ever make; the rest of the width is what it can lose. So a spread that collects a fifth of its width risks four fifths of it, and one loss undoes four wins. Our 120 settled credit spreads went 102 and 18, a win rate of 85.0%, and the losers lost 0.80 of their maximum risk on average, measured on 17 of the 18 losses for which the ledger publishes a modeled payoff. In the ledger's modeled dollars a win on these spreads earns $286 on average and a loss costs $899.

A win here is the ledger's own label: the spread's modeled payoff at settlement came out positive. That is not quite the same as the price finishing beyond the short strike, because a close between the strikes can go either way depending on the credit, and the grading reads the credit.

The arithmetic

Call the credit a share of the width, c. The spread earns c when it wins and loses the giveback times (1 minus c) when it loses. At a win rate of 85.0% and a giveback of 0.80, the expected result is zero when c is 12.4% of the width. Below that, the spread loses money on our own record before a single commission or a nickel of slippage. The giveback is read off each loss's payoff as the ledger graded it, so the line is a reading of the record, not a formula solved for every credit. The median credit our own notes stated is 16.3% of width, above that line.

Two things about that number. It is a floor on the payoff, not a forecast: a spread at exactly break-even wins small 85.0% of the time and loses large the rest of it, and a reader who mistakes the win rate for the edge is reading the wrong column. And it moves. Both inputs are recomputed every week from the settled ledger, so the line on this page is this week's line.

What the notes actually collected

Only 29 of the 120 notes state the credit the spread was expected to collect; the other 91 describe the strikes and the thesis and leave the premium to the reader. Where a note did state one, this is what it said.

The credits our notes stated fell: a median of 30.0% of width in May 2026 against 16.8% in October 2026, and nothing in the pipeline was watching the figure.

6 of the 29 stated credits sat at or under the break-even line; the week of 7 September 2026 alone wrote 2 of them.

The floor, and why it is not a quality filter

The fix is a floor: a credit spread now has to collect at least 20.0% of its width or it is not published. That sits above the break-even line. It is a floor on the payoff, not on the thesis. The tempting alternative was to ban the profile that prompted the review, cheap stocks with weak ratings. That would have been a filter on the stock, and three days out and a few percent from the money a credit spread is a bet on distance, not on quality. When the floor binds, the rule is to narrow the width rather than walk the short strike in toward the price, which buys credit by spending the cushion the implied move gives you. Where neither clears it, the week publishes a No Trade with the arithmetic stated.

There is one exception, and it is carded on the ledger. In a week where fewer than two plays clear every gate, at most two bull put spreads may run at a 15.0% floor if they are at least $2.50 wide and sit at or beyond the implied move. That too sits above the break-even line. The lane follows a reading of our own settled spreads by how far the short strike sat from the price, which this page does not recompute; the width floor is the fill haircut, since on a very narrow spread a nickel of slippage is a large share of the width. Those plays carry a fallback label in their row.

What the dollar column assumes

The ledger's dollar column is modeled, and the track record says so on its own page. Where a note stated its credit, the spread is graded on it. The 91 spreads whose notes stated nothing are graded at a flat 32.0% of width, and they are most of the ledger's credit spreads. That is above the 16.3% median our notes stated. So the published total is a modeled figure and a generous one for those rows, and this page would be dishonest to quote it without saying so.

Re-pricing those spreads at the 16.3% the notes did state, instead of the flat assumption, turns it negative: the modeled total on the track record reads $9,446 as published and would read minus $3,992 under that re-fit. 2 published outcomes change label under that re-fit: DOCU in the week of 1 June 2026 from Win to Loss, CI in the week of 27 July 2026 from Win to Loss.

That re-fit is not applied to the track record. A published call is never edited, and the ledger changes how it grades only with a dated entry in its Corrections block stating the effect on the total, as every grading change before it has. This page describes the change; it does not make it.

What this study cannot tell you

Nothing here was filled with a broker. Outcomes are graded at settlement against published strikes, the giveback is read off the same modeled payoff, and there are no commissions or slippage in any figure. Costs push the break-even line up, not down. 9 of the 120 rows carry strikes the ledger's rebuild refused to print; they count in the record above and are left out of every dollar and giveback figure, as the track record withholds their dollar too.

The stated credits are 29 notes out of 120. The notes that stated a credit may differ from the ones that did not, so the median is the best reading we have of what the rest collected, not a measurement of it.

Distance from the price is not machine-readable in the notes, so the fallback lane's distance rule is reviewed by a reader of the document and not measured on this page.

A bull put spread has an easier time in a rising market than in a falling one, and this page does not split the record by market regime. Nothing here is advice.