Read the setup scorecard in Context

The short answer

No rule passed. Context’s level rules were re-run on 1,195 back-tested trades in nine stocks, 2018–2024, with each level rebuilt at the 5-minute mark (recomputed after the fact from stored option data) as the wall in force at 15:45. Three rules were scored and none cleared the screen; a fourth had too few trades to score. The earlier verdict, none of eight, stands.

That says “not detected”, not “no edge”. Each scored average carries a 95% range about ±0.10R to ±0.23R wide, so a real average smaller than that would not have shown.

The rebuilt level is not the earlier one: on the 6,601 sessions where both exist they name the same strike for 60.8% of walls. The numbers moved; the verdict did not.

• One clock only, 15:45; the earlier rules that need an overnight wall step are not rebuilt.
• Nine of twelve stocks, 45 of 84 stock-years; GPRO, IBM and VZ are not rebuilt.
• The one accuracy check, on SPY only, found the most error late in the day (15:45 is late) and in quarterly months.
• Not fresh evidence: the same sessions and prices as the earlier version.

What changed from the earlier version

The earlier version set each level once, from the prior close, and held it all day. The rebuilt version reads the wall in force when the decision is made: at 15:45, the rebuild from the 15:40 mark, ready at 15:41. These are different constructs, not a correction of one by the other. On the 6,601 sessions where both exist, the earlier call wall matched the rebuilt one on 63.6% and the put wall on 57.9%; both matched on 36.5%, and the gamma sign (whether dealers are net long or short gamma) on 85.5%.

So the rules fire on different sessions. The call-wall breakout long went from +0.129R to +0.114R and its gap over the naive trade from +0.060R to −0.037R; the put-wall hold long’s gap went from −0.014R to −0.101R; and the put-wall reclaim long after a close below, left out of the earlier tests for too few trades, now has 305 trades and is scored.

Part of each move is the level and part is the sample (nine stocks, 45 of 84 stock-years, four rules). The earlier rule on the same sessions separates the two: only the level source differs between that column and the rebuilt one.

• The arms are not the same trades: 140 of the call-wall breakout long’s 312 rebuilt trades are also earlier-rule trades. Of 120 rebuilt-minus-earlier-rule comparisons, 7 have a range that excludes zero, about what chance gives, and none is an average R; that is not evidence the rebuild changed nothing.

Where the panel is

Context is a tab in the Prism by GEXfreak workspace (sign in to open it), for the study’s stocks only. The tab still shows the earlier version of the study, eight rules on twelve stocks; the figure above shows the rebuilt version beside it, and “The rule, in full” under each rule gives its trigger, stop, target and exit.

All four rebuilt rules are long: three around the put wall, one around the call wall. Each enters at the open of the 15:45 one-minute bar and holds up to five sessions.

Read a row in sixty seconds

Five checks, in this order, on any row. Worked on the call-wall breakout long, the best-looking rule:

• 1. Badge. Did it qualify? If not, was it scored and failed, or set aside as too thin? Here: “Did not qualify”, scored and failed.
• 2. Bar against zero. Does the 95% range cross zero? If so, the data cannot tell the rule from zero. A hollow dot has no range: none was measured. Here: the dot is +0.114R and the bar runs −0.120R to +0.348R: it crosses zero.
• 3. Trades. How many, in how many stocks? Few trades give a wide bar. Here: 312 trades, 4 stocks with 30 or more.
• 4. Naive tile. Did the rule beat what the stock paid anyway? The tile is the rule’s average minus the naive trade’s, before any correction. A tile smaller than the bar’s half-width is noise. Here: −0.037R against a half-width of about ±0.24R.
• 5. Stock list. Does one stock, or one stretch of years, carry the average? Multiply each stock’s average R by its trades to get its total R, and compare it with its share of the trades. Here: NVDA’s +0.339R × 71 trades is about +24R of the rule’s +35R (68%) on 23% of the trades.
• A row that fails gives you nothing to act on. A row that passed every check would be a lead to track on paper, not a clearance.

What each badge means

“Did not qualify” (three rules) means scored and failed at least one test. “Too few trades” (one rule) means left out of the tests: a rule needs 30 or more trades in at least four stocks and 100 or more in all. The put-wall undercut and reclaim long fired 62 times and reached 30 in no stock; its average was −0.175R. Like the tab, the figure shows it with no range and no naive comparison.

Failing does not always mean “no result”. All three scored averages have a range that crosses zero. One gap over the naive trade does not: the put-wall hold long’s, −0.101R (−0.188R to −0.012R), so on these trades it earned less than the naive trade. The earlier rule on the same sessions showed no such gap (−0.012R, −0.136R to +0.117R), and it is one of many ranges looked at, so it describes these trades and is not a finding about walls.

Read R, not dollars

An average of +0.114R means the average back-tested trade finished 0.11 times its risk distance ahead: about 11 cents gained per dollar risked. A win is any trade that finishes above zero. A stop can be jumped when a stock gaps, so R is no cap on a loss: across all trades R ran from −6.67R to +4.41R, and 6.0% lost more than 1.2R after the study’s modeled costs.

One back-tested trade, start to finish: the most recent call-wall breakout long in the study, shown because it is the latest, not because it is typical. On 12 December 2024 at 15:45 ET, MSFT entered at $451.930 with a stop at $447.090 and a target at $464.902; the call wall in the 15:40 rebuild was $450.000. Entry minus stop is 1R: $4.840, 1.07% of the price, close to the 1.02% typical for MSFT in this rule. The target sat 2.68R above entry. On 13 December 2024 at 11:26 ET the stop was hit: −1.000R at the stop, −1.039R once the study’s modeled commission and slippage are charged.

Read the bar before the dot

The dot is the average R and the bar is the 95% range. The call-wall breakout long has the highest dot, +0.114R, but its bar runs −0.120R to +0.348R, so the figure cannot call it above zero. The study’s ranges are wider than the spreadsheet formula’s in “Score your own trades” because trades overlap in time, which the formula cannot see: on the same 312 trades the study says −0.120R to +0.348R and the formula says −0.092R to +0.319R. A spreadsheet range that only just clears zero has not cleared it.

One number is not enough, so the figure and the table split each headline by price period and month type. The put-wall reclaim long after a close below averaged −0.051R over 222 trades through 28 March 2023, when option prices are estimates, and +0.232R over 83 from 29 March 2023, but the 83-trade half alone has a range of −0.100R to +0.534R, so the halves cannot be told apart from each other or from zero. In quarterly months the call-wall breakout long averaged −0.195R over 102 trades and in other months +0.264R over 210; every range includes zero, and the accuracy check found more error in quarterly months, so read the split as a warning about level quality, not a pattern.

Read the naive tile and the stock list

Each scored row has a “Versus a naive trade” tile: the rule’s average minus what the same trade in the same stocks paid without the signal. The call-wall breakout long’s naive trade paid +0.151R, more than the rule’s +0.114R, so its tile is −0.037R (−0.273R to +0.198R): noise. The put-wall hold long’s, −0.101R, is the one range wholly below zero.

Under a row, the stock list shows where the average comes from. The put-wall hold long’s stocks run from −0.229R (NKE) to +0.225R (AAPL) around a pooled −0.010R, but each stock’s own range is ±0.21R to ±0.55R wide, so most of that spread is what chance produces. AAPL’s, +0.012R to +0.435R, is the one of 21 stock ranges (three rules, 20 or more trades each) wholly above zero; with that many, about one is expected by chance.

Call-wall breakout long: NVDA, MSFT and PFE supply 140 of its 312 trades (45%) but 124% of its total R. Where the R came from:

• Those three stocks: 140 trades, +43.95R. The other six: 172 trades, −8.53R (INTC alone, −10.99R over 29 trades).
• 2020 and 2021: 117 trades, +62.46R. The other five years: 195 trades, −27.04R (2022 alone, −30.52R over 37 trades).
• Take out its best 20 trades (+78.48R, from seven stocks) and the other 292 average −0.147R; the whole rule is 312 trades, +35.42R.
• An average carried by a few stocks or two years is a warning, not evidence of an edge. The study’s screen would cap any one stock at 30% of a rule’s total R (GEXfreak’s own number, not a standard); NVDA alone is 68%.

Score your own trades

Five steps in a spreadsheet, one row per closed trade of one written rule. Mixing rules hides which one works. Columns: A ticker, B date, C entry, D stop, E exit, F R. Steps 1 to 3 are the same routine as in “How many trades until you know?”; step 4 is unique to this page.

• 1. Write the stop before you enter; a trade with no stop written down beforehand is left out. For an option bought outright the premium is the most you can lose, so R = profit ÷ premium paid. The study measured shares, so its numbers are no yardstick for options.
• 2. In F2, for a long: =(E2-C2)/(C2-D2). For a short: =(C2-E2)/(D2-C2). The trade above: =(447.090-451.930)/(451.930-447.090) = −1.00 before costs.
• 3. Then =COUNT(F2:F41) is your trades and =AVERAGE(F2:F41) your average R. The low end of your 95% range is =AVERAGE(F2:F41)-1.96*STDEV(F2:F41)/SQRT(COUNT(F2:F41)); use + for the high end. At a swing near 1.85R, 40 trades leave a range about ±0.57R wide; about 1,000 narrow it to ±0.11R, the size of the rebuilt study’s best average.
• 4. Check concentration. A ticker’s share of the R is =SUMIF(A2:A41,"NVDA",F2:F41)/SUM(F2:F41) and its share of the trades is =COUNTIF(A2:A41,"NVDA")/COUNT(F2:F41). A share of R far above its share of trades is a warning: the call-wall breakout long’s three biggest stocks supplied 124% of its R on 45% of its trades. If the sum is zero or negative, compare each name’s average instead.
• 5. Read the answer. A range wholly above zero is an average that luck alone is unlikely to explain, if you chose the rule before looking and the trades are independent. A range that crosses zero, which most logs will show, is unproven: keep logging.

What the six tests ask

To qualify, a rule had to pass every one of six tests. Each is a way the average could be an illusion.

• 1. Average R. The average’s 95% range had to sit entirely above zero.
• 2. Gap over a naive trade. The gap needed a range above zero and an adjusted p-value of 0.10 or less. The p-value is the chance of a gap this big by luck if the rule had no real edge; “adjusted” allows for having tried three rules.
• 3. Beat placebo trades. A placebo is the same trade with no real wall behind it: a phantom level at the same distance in implied moves on other stocks the same day or, when none fits, on the same stock 6 to 20 sessions away. It is scored on the market-adjusted average of test 4, with the same two conditions as test 2.
• 4. Market-adjusted average. After subtracting the market’s move over each trade (the stock’s beta to SPY times SPY’s move), the 95% range had to sit above zero.
• 5. Both periods. The average had to be above zero through 28 March 2023 (fallback option prices) and after it (native prices).
• 6. The wider screen. Eight more checks had to pass together, among them a t-score (the average divided by its standard error) of at least 1.5, positive years, at least four stocks with 30 or more trades and none carrying more than 30% of the R, an average above zero at doubled costs, and little overlap with the other rules. The earlier version never computed it, because every rule had already failed; its “older inputs” test is gone, since it applied only to the 09:35 rules, which are not rebuilt.

The six tests, with the numbers

The table gives the numbers behind every verdict, in R per trade with 95% ranges. Every scored rule failed test 1, and none passed tests 2, 3, 4 or 6. The call-wall breakout long was above zero in both periods (5), which cleared nothing else.

All three had an adjusted p of 0.99 on test 2 against the 0.10 needed. On test 3 the closest was the put-wall hold long: +0.076R (−0.052R to +0.203R), adjusted p 0.37. A further check asked whether the option level beat a price level taken the same way (the prior session’s low for the put-wall hold long, its high for the call-wall breakout long): −0.035R (−0.135R to +0.060R) and +0.073R (−0.146R to +0.299R). Neither was rejected, and it is not one of the six.

Limits

These levels are rebuilt at each five-minute mark; Prism by GEXfreak keeps no five-minute history of its levels. A tier-0 rebuild uses that morning’s open interest and the prior close’s option prices, re-evaluated at that mark’s market price. Only SPY has a true intraday map (from August 2025), used just to check accuracy. Within $1 of it, tier-0 matched the call wall 77.5% of the time, the put wall 83.3%, the daily flip 68.9%, the weekly flip 73.7% and the monthly flip 66.9%. QQQ and stock accuracy is unmeasured.

• Through 28 March 2023 an option’s price is a stored last trade, or a modeled value where none traded (the fallback-premium era); after that date prices are native.
• The accuracy check above ran on SPY sessions from October 2025 to August 2026, after this study’s window. In its by-hour columns the call wall matched 74% of the time at 15:00 and 57% at 16:00, and the put wall 76% and 61%; 15:45 falls between those marks and was not itself checked. A rebuilt level is not a record of what a member saw.
• Nine stocks, 45 of 84 stock-years: BAC (2020–2022), PFE (2021–2022) and DVN (2022) sit in the fallback period only, so the native period rests on six stocks (AAPL, MSFT, NVDA, TSLA, INTC and NKE). The result does not speak for the twelve stocks of the earlier version, and the stocks were all listed when the study was built, so delisted companies are missing.
• One clock, 15:45, on single stocks; ETFs, indexes and futures are not covered. The earlier rules that read an overnight wall step are not rebuilt, because the site does not show a wall’s history.
• Same-date phantom levels are scarce, so most placebo matches use the same stock 6 to 20 sessions away, and 41%, 31% and 17% of the scored rules’ trades had none. The put-wall undercut and reclaim long has no usable placebo test.
• Not independent evidence: this re-scores the same 2018–2024 sessions as the earlier version, and GEXfreak’s earlier research had already looked at these years, and at 2025 for put-wall long and call-wall breakout rules, so 2025 is not a clean test for them. Nothing dated 2025 or later was read. The level family has had 21 counted looks, 9 of them here, and the adjusted p-values count only this study’s tests, so they are more lenient than the full count. The frozen plan and single run are stamped in the figure’s data receipt, and an independent audit passed all eight checks with five minor defects, none changing a number.
• A back-test of a rule on rebuilt levels is not a signal and not a promise of a future return.