HOW-TO: test whether a price level is doing anythingE-mini S&P 500 FuturesCME_MINI_DL:ES1!SignalPilotLabsLevels are the easiest thing to draw and the hardest thing to check. A chart with twenty lines on it always looks like the lines mattered, because price is always near one of them. Here is the method, and the control that most level tests leave out. The setup SP Janus Atlas on ES1! daily, drawing three independent families. Current-period extremes (weekly, monthly, quarterly, yearly high and low). Settled prior-period extremes (previous week, month, quarter). Volume-derived levels (monthly POC and value area, quarterly POC, naked weekly POC). Where several land on the same price, the label carries the count, and several do on the chart above. One instance, checked in bar replay: on 18 January 2026 three of them converged on 6,994.00, labelled mH, qH and yH, all inside 0.15% of each other. The distance table scored them four stars and placed them 165 points above the last close of 6,829.50, which is where price finished after breaking down through them. One instance proves nothing. The numbers below are 54,000 bars of them. The question Does price behave differently at a price where several levels agree than at a price where only one does? The control, which is the whole argument Comparing levels against "the rest of the chart" measures nothing. A level is only ever tested on a bar that reached it, and bars that travel far tend to keep travelling. So the control was a random price drawn inside the same bar. Same bar, same volatility, same market state, and touched by construction. That separates whether the price is special from whether the bar is special. What was measured On every bar where price touched a level that had already been in place at least three bars: Penetration. How far through the level price went over the next five bars, measured in ATR. Closed back. Whether price closed on the side it had approached from, five bars later. Thirteen instruments: ES1!, NQ1!, SPX, AAPL, NVDA, TSLA, MSFT, BTCUSD, ETHUSD, EURUSD, GC1!, CL1!, DXY. Roughly 54,000 daily bars, most reaching back to 2009. The result random price, same bar ... penetration 1.048 ATR ... closed back 50.2% one level ............... penetration 1.071 ATR ... closed back 50.3% two levels .............. penetration 1.110 ATR ... closed back 49.5% three or more ........... penetration 1.215 ATR ... closed back 45.4% The relationship is monotonic, and it runs opposite to the usual assumption. Price cut deeper through the prices where more levels agreed, and closed back on the approach side less often. It held on 13 of 13 instruments for closed back, and 12 of 13 for penetration. Two checks, because a single number is not a finding Compression would inflate a penetration measured in ATR. Measured at the moment of touch, ATR against each instrument's own median was 0.999 where three or more levels agreed, against 1.000 across all bars. No compression. A result concentrated in one episode is not a result. Those touches spread across 18 years, with no single year holding more than 7.6% of them. What this does not say It is measured across all available history with no holdout. It is daily bars only. It covers the price-derived families; the volume-derived ones were deliberately left out, because reproducing their calculation would have meant grading a reimplementation rather than the indicator. It also reconstructs the period extremes from open, high, low and close rather than reading them out of the indicator. So it measures what happens where period extremes converge, which is the general question, and not literally what one indicator plots. The chart shows the indicator; the numbers describe the structure. And it describes what price did on arrival. It says nothing about what anyone should do with that. Why publish it Because the useful version of a level test is the one where the answer is not the one you were hoping for. The method above is the part worth keeping. Run it on any levels, from any source, including these. Earlier in this series, on my profile: HOW-TO: check whether an indicator repainted HOW-TO: judge an indicator by its misses HOW-TO: tell if two indicators are saying the same thing HOW-TO: measure what waiting for confirmation costs