How near is near a key level? I measured the answer

Re-evaluating channel and trendline sensitivity

A reader recently pointed out that price action rarely respects a perfectly thin line. Because charts feature delays, quick touches, and false breakouts, my previous rejection of certain channel and trendline strategies might have been premature. They argued that these lines require a “buffer zone” or a tolerance band to function correctly. I have already implemented this via ATR (Average True Range) multipliers in the engine, but I had only stress-tested these bands for horizontal levels. I had not systematically swept the band widths for channels and trendlines. To fix this, I ran a series of forward-testing sweeps to see how adjusting these tolerances changes the outcome.

Channels: Sensitivity to width

When I widened the proximity band for channels on the H1 timeframe from 1.0 to 1.5 ATR, the cumulative performance shifted from -6.8% to +15.7%. This suggests my previous finding in research 75 was an artifact of using a band that was simply too narrow. However, the results show signs of over-optimization. Even with the wider band, consistency remains limited to 3 out of 6 years, whereas horizontal levels consistently perform in 5 out of 6 years. Furthermore, increasing the “touch” bands (merge/break) within the engine actually backfired, resulting in a PF of 1.08, a 14% DD (drawdown), and a chaotic spike in trade frequency. While channels can be made profitable through band tuning, they remain less robust than horizontal levels.

Trendlines: Revisiting the “deployment impossible” verdict

My previous analysis of trendlines (research 60) concluded they were unfit for deployment. After testing, I found that the band width was not the primary cause of failure. In fact, H1 and H4 trendlines are quite robust across a wide range of retest bands (0.3 to 1.2 ATR).

H4 performance: 5/6 years profitable, +8% to +12% cumulative return.
H1 performance: 4/6 years profitable, +21% to +53% cumulative return.
D1 performance: Collapsed across all widths (-2% to -4% return, 1 to 2/6 years). The D1 failure is structural, likely due to a lack of swing data and thin statistics. The “deployment impossible” label from research 60 was driven by the inclusion of short-side trades, M1 intraday issues, and the D1 collapse. When restricted to long-only trades, the trendline strategy shows genuine, robust edge on H1 and H4.

Final verdict: Why it won’t replace the core system

I performed a final gate test on the long-only trendline strategy using data from 2015 to 2019 for selection and 2019 to 2025 for out-of-sample (OOS) validation.

Metric	H1 Trendline	H4 Trendline	Combined
Cumulative Return	+21.9%	+0.4%	+22%
PF	1.08	1.01	1.06
DD	-18.7%	-	-21%
MC Pass Rate	56%	-	-
While the strategy is technically “real” and passes M1 intraday safety checks, it is not strong enough to replace the current core system. The PF is thin (~1.0 to 1.1), the DD is high (-18% to -21%), and the MC (Monte Carlo) pass rate is hindered by a 23 to 30% failure rate in maximum loss scenarios.
Furthermore, the strategy shows a 0.42 to 0.51 correlation with my core system. It is essentially just “rediscovering” trends already captured by existing tools. I am correcting my previous assessment: the lines themselves are valid, and the band-tuning logic will be kept as a permanent asset. However, the strategy does not offer enough diversification to justify adding it to the live portfolio. The core remains at version 1.2.0.

How this connects

This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).

Code to reproduce

You can reproduce this with the following scripts (see repo).

scripts/research/study_trendline_long_gates.py