Fade-the-failure — the counter-direction trade that never existed

Intuition says a failed signal reverses. We tested it across HRR 07:30 and 08:09 setups. Movement post-failure is a coin flip; occasionally it's still directional the original way. No counter-direction edge anywhere. Decay, don't flip.

The intuition

If an 80% signal doesn’t work by minute 60, the trader narrative says something changed. “The setup failed, now go the other way.” It’s the kind of rule that sounds right at a trading-desk level and can feel right on a handful of cherry-picked charts.

We wanted to know: does it hold in the data?

The test

We took HRR (Herman’s Range Reversion) signals that failed by 60 minutes past their anchor — specifically 07:30–08:30 and 08:09 zones — and measured the next hour’s movement from t=60m forward. Crucially we measured from t=60m, not from entry, to kill survivor bias: we needed to see what happens in a 10:30–11:30 window on days where the morning signal already didn’t work.

The result

Four cohorts, zero counter-direction edges. Two of them drift with the original thesis (especially the 07:30–08:30 LONG at 68%). The 07:30–08:30 SHORT at 41% up suggests the bearish thesis is still playing out — it just takes longer than 60 minutes to resolve.

Why the intuition was wrong

A failed signal is a noisy sample of the same underlying distribution, not a clean negative read. If a bull thesis has a 60% base rate, failed bull signals at 60m are drawn from the remaining 40% — many of which are chop or slow resolutions, not outright bear days. The idea that “if bull failed, bear must be right” is a classic gambler’s fallacy dressed up as risk management.

The decision and what we shipped instead

No counter-direction lines. The correct treatment of a signal that hasn’t worked yet is time-decay on probability, not a flip of direction. We built time-decay curves for every major flagship that has one (AMD, 1H CRT, 0809 SEQ, HRR, LON SWEEP) — these dial probability down as the window burns without ever flipping the line’s direction.

A faded Tier 1 line at 45% probability drops out of the top-weighted slot on the chart HUD naturally. It does not get replaced by a contra line.

Lessons

• Failure is noise, not signal. You can’t cleanly invert a probabilistic edge. The sample of failed bull days isn’t a sample of bear days.
• Decay, don’t flip. Expiring confidence is the right response to a stale setup. Opposite-direction conviction needs its own independent edge.
• Measure from the decision point, not from the entry. The survivor-bias trap here is subtle: if you measure from entry, failed signals by definition haven’t moved favorably — your forward-measurement inherits that. Starting the clock at t=60m cleans it up.

This was a three-line fix in the code (don’t add counter lines; keep decay) and a permanent ban on “fade-the-failure” as a heuristic on this stack.