- Delta is the volume that hit the ask minus the volume that hit the bid. Positive means buyers were the ones crossing the spread.
- Divergence: price makes a new high for the session but cumulative delta does not — so sell. Mirror for a new low — buy.
- Absorption: a bar whose delta is in the top tenth of the session so far, while its range is below the session's median — somebody passive is soaking up the aggression. Fade whoever was aggressive.
- Entry at the close of the bar that fired. Stop 1.0x the median bar range so far, target 2.0x. Flat at the close of the 15:55 bar (the 16:00 price).
- Thresholds use only bars that have already printed. Nothing is scaled by numbers from later in the session.
| trades | win | a trade | longs only | shorts only | |
|---|---|---|---|---|---|
| With the delta feed | |||||
| Nasdaq · divergence | 2,364 | 31.9% | −$7.92 | −$5.28 | −$9.85 |
| Nasdaq · absorption | 373 | 31.6% | −$13.88 | −$24.55 | −$4.04 |
| Gold · divergence | 3,307 | 33.7% | −$40.90 | −$41.37 | −$40.53 |
| Gold · absorption | 1,606 | 37.4% | −$28.04 | −$19.98 | −$35.75 |
| Same trigger, no delta feed — price only | |||||
| Nasdaq · new extreme | 3,414 | 33.5% | −$4.22 | −$2.34 | −$5.88 |
| Nasdaq · quiet bar, big body | 65 | 35.4% | −$5.07 | $11.52 | −$10.49 |
| Gold · new extreme | 4,186 | 33.9% | −$32.24 | −$35.89 | −$28.98 |
| Gold · quiet bar, big body | 159 | 36.5% | $31.57 | $30.00 | $32.66 |
What the feed is worth. Take the paid version and subtract the free one: −$3.70 a trade on Nasdaq divergence, −$8.81 on Nasdaq absorption, −$8.66 on gold divergence, −$59.61 on gold absorption. All four negative. Adding the flow condition to a losing rule made it lose more, on every market and both claims.
Every setting, not the worst one. Three stops crossed with four targets, for each claim on each market — 48 versions in all. 1 of them made money. The table below changes with the market and claim you have selected, so you can check that yourself.
Why this one is worth testing at all. Every other strategy in this lab can be checked by anyone with a price chart. This one cannot — it needs the aggressor side of every trade, which is a paid feed. That is exactly why the claims about it go unchecked, and why so much is sold on top of them. We have the data on 99.7% of Nasdaq bars and 99.4% of gold bars, so it can be checked.
Careful with the one green row. Gold's price-only absorption makes money, but on 159 trades, and its Nasdaq twin runs to only 65. That is not a strategy, it is a small sample. The comparison that carries weight is divergence, where both sides have thousands of trades and the free version still wins by $3.70 and $8.66 a trade.
What would have made us believe it. A feed worth paying for should beat the same rule without it, somewhere. It does not beat it anywhere. Not one of the 48 paid settings makes money either, and that is not a sample-size problem — divergence alone fires 3,307 times on gold.
Everything else we tested
We made the test harder than the research version. That version sized the stop from the median bar range of the whole session, which is not known when the trade is taken. Using only the bars that have already printed — the honest version — the losses get slightly bigger, because a look-ahead volatility estimate places the stop better than anyone could have in real time.
Both directions are reported separately for every cell. This is the check that catches a fake edge: over this period Nasdaq and gold both rose a long way, so any rule that mostly buys will look good for reasons that have nothing to do with the signal. Splitting long from short removes that excuse. Here it does not matter — 8 of the eight direction-books are negative.
One position at a time. Divergence fires on a few per cent of all bars with no cooldown, so an earlier version of this test let a single trend leg open several trades that all resolved against the same forward path. That inflates the trade count without adding independent evidence. Every number here takes one position at a time, as the rest of the lab does, which cut Nasdaq divergence from 4,300 trades to 2,364.
The signal definitions are the standard ones, not strawmen. Divergence compares the session's running extremes against the running extremes of cumulative delta, and fires only on a genuinely new price extreme. Absorption uses the top decile of one-sided delta against a below-median bar range. Both are what the platforms draw for you.
What this test does not do
We tested two named, mechanical claims about delta. We did not test order flow as a discipline, which is a way of reading a market rather than a rule you can code.
- Five-minute bars, not the footprint. Real order-flow traders read volume at each price inside the bar. We use the bar's net delta. A footprint reader would say that throws away the detail that matters.
- Delta is an approximation. It is inferred from which side of the spread a trade printed on. It is the standard measure, and it is what the platforms show, but it is not a record of intent.
- No context filter. Every signal is taken. Most people who trade this say it only counts at a level that already mattered.
- The free absorption arms are small. 65 trades on Nasdaq and 159 on gold. They are shown because leaving them out would be selective, not because a number that thin proves anything.
- Two markets, regular hours only. More markets would make the result stronger.
- The replay is a sample. Every number on this page is scored over all 1,047 Nasdaq and 1,663 gold sessions. The chart ships the most recent 220 of each (2025-10-17 to 2026-09-25) so the page is not a five-megabyte download.
If you think a rule here is wrong, say which one and what it should be. We will test it against the same gates and publish the result, including if it beats ours.
Order flow has no single author. This is the version every delta platform ships with, tested as it is normally stated.