AI Adaptive Money Flow Index (Clustering) (AlgoAlpha): tested

Edge too small for costs TradingView indicator · tested 11 Sep 2026 · figures re-checked 8 Oct 2026

What it claims

Billed as an AI, clustering-based Money Flow Index, it labels every bar Up-Trend, Down-Trend or Ranging and marks bullish and bearish reversals. The name suggests the machine-learning clustering drives those labels. It makes no performance claim.

What we tested

Its own output on gold 5-minute bars (10,976 bars, 19 July to 11 September 2026) at 25 settings of its two inputs, traded three ways fixed in advance: follow its trend label, follow it with a 2x ATR stop, or trade its reversal marks with a 2x ATR stop; entry at the next bar's open, 1.9 basis points per round trip.

1.1 bpswhat its trend label is worth, before coststhe move after Up-Trend bars minus after Down-Trend bars, 8 hours ahead; a second sample gave 1.0

1.9 bpscost of one round trip on goldthe label earns about half of what acting on it costs

25 of 25settings that made money before costsfollowing the label with a stop

0 of 75settings profitable in both halves after costsall three ways of trading it

Before trading costs unless stated. R = the amount risked on a trade.

The backtest

one dot = one of its 25 settings, following its trend label with a 2x ATR stopEach dot is the average result per trade, in %. 25 of 25 were above zero before costs; 0 after.

After a 1.9 basis-point round trip every trend-following setting lost money; the best reversal-mark setting made +0.010% a trade overall, split into -0.023% and +0.043% across the two halves of the test.

Why

The Up-Trend and Down-Trend labels lean the right way slightly and consistently: on two separate samples, the move after an Up-Trend bar beat the move after a Down-Trend bar by about 1 basis point. One round trip on gold costs about 1.9 basis points, so acting on the label costs roughly twice what it earns. No setting held up in both halves of the test, so tuning does not fix it. And the 'AI' part is not what drives it: the money-flow length changed the labels on about 41% of bars, the clustering window on only 5 to 13%.

Limits of this test

One market (gold) over 55 days of 5-minute bars; the second check used a 1-minute sample.
No single setting reached statistical significance on its own; the case that the label carries a little information rests on two samples agreeing at about 1 basis point.
A rule that trades only some bars within a trend label was not tested; it would need to beat 1.9 basis points a trade from a label worth about 1.1 on average.

Measured results, not advice. We have no commercial link with AlgoAlpha. If you made this tool and think a figure is wrong, write to corrections@trackedsignals.com; corrections are published within 7 days. How we test.