Rank Not Return
Last updated 2026-09-25
This is the reading primer that every model-output pane assumes. Read it once and the signals side of the desk makes sense. The options suite is a different kind of pane. It shows plain market data rather than a ranking, and it has its own primer: Reading The Options Suite.
At a glance
- What the models do. Every evening they look at about a thousand stocks. They rank each one at 45 trading days and at 20 trading days.
- What a rank is. A place in that evening's queue, 0 to 100, 100 the top. Think of a teacher ranking a class from top to bottom without predicting anyone's test score.
- Nothing here is a return. There is no price target and no expected move at all. A number that looks like a percentage return is a rank, a chance or a measured move, never a return.
- A signal. A name in the top 5 percent at 45 trading days, or the top 2 percent at 20. Its realized volatility also has to be over 25 percent a year. A quiet stock at the top of the ranking is not a signal.
- Confidence. High, Medium, Low or no edge. It says how rare a name's band in the ranking is, not how sure the model feels about the outcome.
- Chance. How often names ranked like this one finished the horizon on the signal's side, measured on trading days the model never trained on. Any stock, the coin flip, sits beside it.
- The middle band. A rank from about 20 to 80 sits too close to the middle of that evening's queue to call a side. It counts toward neither bullish nor bearish.
- The 10 day rank. Shown for context only. That ranking failed its own holdout test, so it never carries a signal.
- Held. A count of trading days behind a signal, not a position. See the Held section below.
- Bullish and bearish. The two sides of the book. Bullish is names to hold. Bearish is names not to hold, and in the model book that side is a short.
Key difference: a rank compares one name to that evening's other names. It never says what a price will do.
The fields the desk does not draw
The models rank tonight's thousand stocks against each other. They do not forecast a return, and they emit a couple of fields that carry no real information on their own. The desk drops them before anything is drawn.
| Field | Why it is never drawn |
|---|---|
| expected move | The models rank, they do not forecast a move on the ranking side. The Watchlist and Read-out state the model's own expected size of move instead, from the heads described below. |
| the model's own confidence output | This is a different number from the Confidence word shown on screen. The model's raw confidence barely moves across a thousand names, so the desk drops it and derives Confidence from where a name sits in the ranking instead. |
So a rank, a chance and a confidence word are what you see. Nothing here is ever drawn as money.
Why only the rank is drawn
The raw scores carry an offset that changes with the horizon. A 45 trading day score and a 20 trading day score are not on the same scale. Plotted directly, comparing them as numbers would be meaningless.
Only the rank, a place from 0 to 100 in that evening's queue, is comparable across horizons. When a pane's axis says "rank", it means this percentile, never a raw score.
The words on screen
- Rank (
RankPct). Where a name sits inside one horizon's ranking, 0 to 100, 100 the top. Market Scan prints it as Rank 45, Rank 20 and Rank 10. - Side. Bullish or bearish, read off the same ranking. A bar tints green on the bullish side and red on the bearish side.
- The 45 and 20 day ranks. The two horizons a signal can come from. Rank 45 in the top or bottom 5 percent, or Rank 20 in the top or bottom 2, is what the signal rule reads.
- The 10 day rank. A third rank, shown beside the other two. It failed its own holdout test, so it decides nothing and is drawn quieter.
- A dash. That horizon produced no reading for that name tonight. It is not a low reading. Unknown is not zero.
Held
Held counts the trading days in a row a name's signal state has run, tonight included. It prints as "Signal held 7 trading days". A name with no signal prints "No signal for at least 14 trading days".
Held is not a position. Nothing is bought, sold or kept in an account, and the count says nothing about a price.
Key difference. Held counts the trading days behind a signal. It never turns that count into a probability of its own; the Chance figure beside a name is measured separately.
Do the two ranks agree
Timeframe Agreement asks a plain question: do the 45 and 20 trading day ranks put a name on the same side tonight. It also states how many trading days running. This is a fact about the readings, stated once. It is never a score added across more horizons than the model serves.
An earlier model scored seven horizons and counted how many pointed the same way, printed as a fraction such as "3 of 7". That count is gone. The current model ranks two horizons that carry a signal, plus the 10 day rank for context. A fraction built from seven no longer describes anything on screen.
Bullish and bearish, the two sides of the book
- Bullish. Names to hold. In the industry this side is called long, which just means owning the shares. Green, with an up triangle.
- Bearish. Names not to hold. In the industry this side is called short, which means betting against the shares, and in the model book that side is a short. Red, with a down triangle. See Bearish List for how the desk reads it.
The model book holds five bullish and five bearish names, equal weight. See Words the desk uses for how the book itself is picked and dated.
What the model panes are, and what's delayed
Some panes take their numbers from Ohey's models. That is the whole signals suite, plus Trade Cards. Each one is a visual of what our models computed.
On the free plan, those same model-output panes show the prior trading day's read. A delayed-data chip names both dates. The panes are:
- Trade Cards (
todays-book). - Market Scan (
sig-scan). - Watchlist (
sig-watch). - Star Field (
sig-galaxy). - Signal History (
sig-map). - Book History (
sig-bookreplay). - Sector Split (
sig-flux). - Bearish List (
sig-avoid). - Rank by Timeframe (
sig-cone). - Timeframe Agreement (
sig-council). - Market Conditions (
sig-regime).
The options suite is never delayed. It restates public market data rather than the firm's own work. Holding it back would mean withholding someone else's data to sell early access. That is a different and worse thing than delaying Ohey's own output. See Reading The Options Suite and the Plans and Billing guide.
How much to trust the whole thing
Two facts sit under every number on these panes.
- Some of the model's inputs are missing and stood in for. When a piece of information is missing, the model substitutes a typical value instead of leaving a hole. See Words the desk uses for what "median filled" means.
- The volatility feed behind Market Conditions can run stale. When it does, a banner on that pane says how many days old the reading is.
Treat a rank as a place to start looking, not an answer. It is what the models computed, on the data they had that evening.
Frequently asked questions
Why is a name ranked where it is? The models read the same measurements for every name in the scan, then rank the whole list from strongest to weakest at each horizon. A name is ranked where it is because of how it compared with the other names that evening, not because of any judgement about the company. Ranks move night to night as that comparison changes. A rank is not a forecast and not a price target.
Does a higher rank mean the stock is more likely to go up? Not in the way people usually mean. A high rank only means the model placed that name near the top tonight, against about a thousand scanned stocks. The Chance figure, measured against any stock, is the desk's own read on how often names ranked like this finished on that side. It is still not a promise about this one name.
What happened to Overall rank? It was an average taken across an earlier model's seven horizons and is gone from every current pane. One exception: the Record page's By reading tab groups older, pre-cutover paper trades by that earlier model's own rank and agreement. Those trades were picked under that model, so regrouping them by the current model's ranks would credit a read it never made. See The Record.
Why is a horizon sometimes shown as a dash instead of a number? That horizon produced no reading for that name that evening. It does not mean the name ranked last. RingMaster never draws "unknown" the way it draws "ranked last", because those are different facts.
Why do I sometimes see yesterday's data instead of today's? On the free plan, every pane built from Ohey's own models shows the prior trading day's read rather than tonight's. Both dates are on the label. The list of those panes is above, under "What the model panes are, and what's delayed". The options suite is not delayed, because it is public market data rather than Ohey's own output.
What happened to "conviction", "certified" and steadiness? All three were retired for the same reason: they read as claims about outcomes when the models only rank. "Conviction" was the label later renamed Overall rank, and it is gone with it. "Certified" was a per-horizon grade that only restated a horizon's own role and was withdrawn in 2026. Steadiness was a word like Rock steady or Drifting. Checked against what happened next it was found unreliable, and it is gone from every pane, replaced by the measured Held count. Older material using any of these words describes the same things under their current names.
Related panes
- Reading The Options Suite is the same kind of primer for the options panes. Those show observed market data rather than a ranking.
- The Panes is the full map of all forty panes, grouped by workspace.
- The Plans and Billing guide sets out what the free tier shows a trading day later.