How the book is built, and how it is scored.

Every evening the models rank. They do not forecast

After the close on every trading day, the models take about a thousand US names. The field is chosen by how many dollars of each name trade in a day, with a floor on price and on daily volume.

The models sort that field from strongest to weakest on the day's data. Every name gets a place in the queue. Nobody adds a name or removes one by hand.

A place in the queue is not a forecast. It says where a name stands against the other names that evening. It does not say which way a price will go, how far, or when.

The models also emit a handful of figures that read like a forecast or a confidence. None of them varies from one name to the next in any useful way, so the desk never shows them.

When a measurement is missing for a name, the models fill it with the middle value across the field. That keeps the field whole. It also means some places in the queue rest on fewer real inputs than others.

One set of models produces every ranking and every book. There is no second model kept aside for research and quoted in marketing.

Three time frames, and what agreement means

The current model scores every name three times, once for each time frame: ten, twenty and forty five trading days.

Within each time frame the field is ranked on its own. A name near the top leans bullish, a name near the bottom leans bearish, and a name in the middle band carries no side.

Agreement is a count. It is how many of the three time frames put a name on the same side. Three of three means all three pointed the same way.

Agreement is not confidence and it is not a promise. It only means the three readings were less scattered for that name than for another.

A time frame is a span, never an instruction. It says how far ahead the models looked when they ranked. It does not tell anyone how long to hold anything.

How a name becomes a signal or a book entry, and how it leaves

Every evening the desk publishes tonight's signals and one book. Each is a fixed rule applied to that evening's ranking, and the same page goes to every reader.

• Signals. A name carries a signal when it sits in the top or bottom five percent of the forty five day ranking. It also qualifies in the top two percent of the twenty day ranking. Either way, its realised volatility must be over 25 percent.
• The measured record beside each signal. Every ranked name shows how often its own rank slice has worked in the past, beside the same measurement over any stock.
• The model book. Five names long and five short, at equal weight, taken from the top and bottom of the twenty day ranking. It is rebuilt every evening, and it is the book the record leads with.

Three more books, the ladders, were published until 22 September 2026 and then retired. Their published books stay on the record exactly as they went out.

Before any book is drawn, names a book cannot hold are taken out. Leveraged and inverse funds are the common case. They stay in the ranking and out of the book.

The field also has to be nearly whole. If too few of the thousand names could be scored that evening, no book goes out. A late book is acceptable and a short one is not.

The book that is published is the book that is scored. It is not a longer list the models wrote first, and not a book rebuilt afterwards. The published file is hashed and time stamped as it goes out.

Every entry is at the next morning's open, as a market on open order. That is the only price a reader could have got, so it is the price the record uses.

The model book is rebuilt every evening, so a name leaves when it is no longer among the five at its end of the ranking.

The record then scores every published name under fixed rules, stated before the fact. The strict rule holds a name ten trading days and closes it at the close, with no target and no stop. The headline rule closes a name the first time it trades five percent in its favour inside ten trading days. Otherwise it closes at the close of the tenth trading day.

Those are the two rules the record page leads with today. It names its rule beside every figure, and the ten counting rules it scores are all listed there. Nothing is pulled early by hand, and no rule changes after a result is seen.

What the coin flip is, and why every win rate carries one

A win rate on its own means nothing. Hold any large basket of US shares for a couple of weeks and more than half of them finish up, because the market drifts upward. A rule with a small target and a long clock can print a very high win rate on names picked at random.

So every win rate on the record carries its coin flip in the same row. For every published name, twenty five random liquid names are drawn on the same trading day and given the same side. They come from the same volatility band and run through the same exit rule.

The gap between our names and the coin flip is what matters, never the win rate alone. Names held for days overlap, so the gap carries a range built from whole weeks. A range that clears zero is the bar. A range that straddles zero is not weak evidence. It is no evidence.

The draw is seeded, so a rebuild draws the same coin flips. The ten rules were chosen from a research grid of 179. Out of that many, a few clear their coin flip by luck alone. That is why the verdict column is read before the win rate column, every time.

The options readings, and how a signal becomes an option strategy

The options readings are taken from each name's option chain after the close. They show what is open at each strike, what changed hands, and the move already priced in. They are observations of what the options market shows, not forecasts of what will happen.

Every trading evening the desk reads these figures from that day's chain capture.

• IV rank. Where tonight's implied volatility sits against the name's own weekly record, 0 to 100. It never compares one name with another.
• The priced move. The one standard deviation move the option quotes carry to each expiry. The down side and the up side are read apart.
• Gamma positioning. Dollar gamma by strike: each contract's gamma times its open interest, on a dollar scale. The net is calls minus puts, a convention, never a claim about who holds the contracts.
• The call wall and the put wall. The strikes carrying the most net call gamma and the most net put gamma. The flip strike is where the net changes sign.
• Put/call and changed hands. Puts over calls, and how many contracts traded in each option beside the open interest already standing. The two are never merged.
• Unusual volume. Tonight's options volume at twice the name's own average or more. It is a count, not a verdict.

A signal becomes an option strategy by one fixed rule, read off its side and its IV rank. Rich options, at IV rank 50 or more, sell premium. Cheap options, under 50, buy it.

• Bullish. A cash-secured put at IV rank 50 or more, and a bull call spread below it.
• Bearish. A bear call spread at IV rank 50 or more, and a bear put spread below it.
• No side. An iron condor at IV rank 50 or more. A long straddle below it, or when IV rank is not held for the name.

The expiry leaves time after the signal's hold ends: the hold plus 21 days to expiry. A time stop closes the position at the end of the hold or with 21 days left, whichever comes first.

The limit sits at the mid of the combination's own bid and ask, with the natural price beside it. When a report date falls before expiry, the card names the date.

Two probabilities sit on every card, and they are never the same number. The model's is how often names ranked like this one finished the hold on its side, beside any stock over the same hold. The option market's probability of profit is read off the short strike's delta.

The index funds run their own rule. SPY, QQQ, IWM and DIA sell the put nearest 0.25 delta and buy the put nearest 0.05 delta below it. The expiry is 21 to 45 days out, held to expiry.

The options readings are never delayed on either plan. The listed options market publishes that data, not Ohey, so there is nothing for us to hold back.

What is not known, and what is not published

Paper only. No client account has traded any of this, and no real money has. Every figure on the record is simulated paper trading on published names and daily prices.

A short record. The record begins on 16 July 2026. That is weeks of evenings, and weeks cannot separate skill from luck. Read every figure on it with that width in mind.

No backtested return. The models were run over years of history before they were published. We publish none of those figures as a return behind any book, because our own audits found them flattered by things a live account cannot do. They stay off until they pass our own checks again. The only performance figures Ohey publishes for its books are on the record page, and every one of them is paper trading.

No fills and no costs. The record is prices only. No commission, slippage, borrow cost or financing is taken off anywhere on it. Every figure on it is therefore an upper bound on what a reader would have kept.

Paper accounts, with gaps. Two simulated brokerage accounts trade the books as well. Their orders did not always fill, their commissions are recorded but not always deducted, and neither charges a borrow fee. The record page names each gap.

The stamp is ours. Every book is hashed and time stamped by our own process when it is published. That proves our order of operations. It is not a third party's countersignature.

Nothing here is presented as skill. We measured how well pure luck does through our own machinery, and a record this short can neither clear that bar nor fail it.

How a reader can check it

The record page. Every published name is listed with its entry price, its exit price and its move, under every rule, with the misses left in. It is rebuilt after every trading day and needs no account. Open The Record.

The receipts. Each published book carries a hash and a time stamp, and the latest ones are printed on the record page. A book that was stamped after the open it was traded at is named there as late.

Your own rows. Once you have an account you can read the rows we served you and the hashes over them. You also get the exact text each hash covers and the command to recompute one. That proves the rows were not edited afterwards. It cannot prove that no row was withheld.

Your AI. The connector hands the same figures to Claude, ChatGPT, Cursor and other clients, read only, with a receipt on every answer. Ask it to quote the stamp and check it against the page. How the connector works.

Us. If anything on this page is wrong, [email protected] is the most useful email you could send.

Investing puts your money at risk and you can lose money, including all of the money you put in.