A model for each strategy that learns, from thousands of past trades and every live result, which of its signals tend to do better than usual, and gives each new signal a score.
Verdict of the year-by-year test
Not proven yet: it informs, it does not filter
From 2013 to 2026, each year's signals were scored by a model trained only on trades that had closed before that year. Its stronger half averaged +2.37% a trade against +2.51% for all signals, and beat them in 10 of 14 years. The pass rule, set before the test, needed both a better pooled average and at least 60% of years; the pooled average fell short, so the score is shown next to each signal but never used to drop one.
"Higher volume against normal" means signals with more of it beat the strategy's usual trade more often. Weights are on standardised inputs; bigger = stronger pull.
How it learns
1. Every signal is scored. Ten facts at the signal close (1- and 3-month return, distance above the 200-day average, volatility, volume, RSI, place in the 52-week range, and the market's trend and breadth) go into that strategy's model.
2. Every trade is followed to its exit with the backtest's rules, so a live result means the same as a backtest one (track record).
3. Every week it retrains on all 19,457 backtest trades plus every closed live trade, with live trades counting three times, and keeps each version below.
4. It is checked in public. Once live trades close, the table below compares the ones it called stronger with the weaker ones. If that gap does not show up live, we say so.
The model is a small statistical model (logistic regression), not a chatbot, and it does not change the strategies' rules. A first version that predicted plain wins failed this test: it preferred small, frequent wins, and its picks returned less than all signals. The current design was fixed before its first test.
Versions
Version
Trained
Trades
Live trades
Live: stronger
Live: weaker
v1
16 Oct 2026
19,457
0
-
-
v0
backtest only
19,457
0
-
-
Year by year
Year
Signals
All
Stronger half
Weaker half
2026
792
−1.79%
−1.06%
−2.84%
2025
1576
+1.48%
+1.98%
+0.79%
2024
1568
+4.44%
+3.24%
+5.84%
2023
1100
+17.24%
+16.42%
+18.12%
2022
582
−0.50%
+0.37%
−1.49%
2021
1202
−0.44%
+0.07%
−1.06%
2020
1435
+3.42%
+1.61%
+6.07%
2019
530
−2.26%
−2.09%
−2.37%
2018
838
−1.43%
−1.23%
−1.69%
2017
1037
−1.15%
−0.70%
−1.59%
2016
1311
+3.88%
+4.94%
+3.15%
2015
1362
+0.95%
+1.20%
+0.76%
2014
1518
+3.68%
+2.47%
+4.45%
2013
1385
+1.05%
+1.56%
+0.51%
The score is a probability from past patterns, not a forecast. Danio AI does not know about news, results or anything outside these price facts.