Measured on data it never saw.

• Two strategies beat simply buying, with the same exits: buying pullbacks in uptrends, and breakouts. Most of the 'edge' in rising markets is just being long - these clear that bar.
• Choosing markets by their own past record - going forward, never in hindsight - cut drawdowns by 64% on daily and 84% on 4-hour charts.
• The chance-of-a-win figures are calibrated: on unseen data they were off by about 4 points on average on 4-hour and 1-hour charts.
• No short-selling strategy survived testing, so the system doesn't show SELL signals. If one ever passes, it will.
• Predicting the next candle from indicators is close to a coin flip on every timeframe, so Radarline doesn't try: it trades tested setups and uses the model only for the odds.
• Charts under 1 hour lost money after costs in every strategy tested, so there are no 15-minute signals.
• A 90% win rate is easy to get and loses money: tiny targets, big losses. Nothing tested earned money at 85%+ wins.

No settled paper trades yet. Signals are logged from the moment they appear (last market scan 13/11/2026, 6:03:38 am); results fill in here as each trade reaches its exit. Nothing is back-filled.

Radarline optimises for expected profit per trade and smaller drawdowns, not an impressive win rate. Its pullback signals win about 2 in 3; its breakouts about 1 in 3 but with much bigger winners.

• Daily pullbacks now exit when the 2-period RSI closes above 65 (the rule from Larry Connors' original research) instead of a close above the 5-candle average. 4-hour pullbacks keep the average - the RSI exit didn't hold up there.
• On 4-hour charts, hold at most 2 open trades per asset class (e.g. two crypto, two FX). Markets in one class tend to move together, so a third position adds risk more than return. The limit had to give both a smaller worst drop and a better return per unit of drop: on daily charts the drop got deeper, and on hourly charts the ratio got worse, so it applies to 4-hour only.
• A "chandelier" trailing exit for breakouts made less per trade in 4 of 5 windows on every timeframe, so it wasn't adopted.

Sources: Deflated Sharpe ratio (Bailey & López de Prado) · Probability of backtest overfitting · Volatility targeting · Time-series momentum (CME) · RSI(2) exits · Chandelier exit

These are development-period results, not a new unseen test. The final-test figures further up were run once, on the version before these two changes, and can't honestly be re-run. The next independent check of the changes is the paper record since launch.

• Banking half at +1R lifts breakout "wins" from about 1 in 3 to about 1 in 2 - and cuts what each trade makes by half or more, because it cuts off the big trends that pay for everything.
• Showing only the most confident signals lifted daily and 4-hour win rates by 5-6 points while profit per trade fell by about two thirds: the model is most confident about small, safe bounces.
• Holly's rule (60%+ wins and winners twice the losers) left no market at all - on these charts a strategy has one or the other.
• Three filters beat the current rules in 4 of 5 windows, but not by enough to rule out luck after 38 ideas tested (t = 1.3 to 2.1; the usual bar is about 3). They don't change any signal: every signal already logs its confidence and trend, so the live record will show whether they hold.

Past and backtested results do not predict future results. General information only, not financial advice - see the General Advice Warning.