AI Trading Platform: The v1.1 Pivot — Admitting Failure, Finding the Fix

Introduction: Honesty About Failure

In trading system development, the hardest thing is not building a strategy. The hardest thing is admitting when your strategy does not work.

Today we share a story of failure, discovery, and correction. Our v0.4 momentum-decay exit strategy produced negative expectancy after 1,486 paper trades. Instead of hiding this result, we analyzed it, found the structural flaw, and designed a new hypothesis with pre-registered rules.

This is not about being smarter. This is about being honest.

Part 1: The v0.4 Failure

What v0.4 Was Designed to Do

The v0.4 strategy used momentum decay as an exit signal. The idea was simple:

1. Enter on a volatility compression breakout
2. Ride the move while momentum stays strong
3. Exit when momentum weakens (before it reverses)

This sounds reasonable. Many traders use momentum indicators to time exits.

What Actually Happened

After 1,486 paper trades over several months, the numbers told a different story:

The 96% timeout rate was the clearest signal. Our exit hypothesis was wrong. Momentum decay was triggering too early. We were cutting winners and holding losers.

This is not a profitable strategy. This is a strategy that slowly loses money.

The Hard Question

At this point, we had a choice:

1. Tune the parameters — adjust thresholds, change timeouts, hope for better results
2. Question the hypothesis — ask if momentum decay was ever the right exit signal

We chose option 2. Parameter tuning without structural analysis is just curve fitting.

Part 2: Finding the Problem with MFE Analysis

What is MFE?

Maximum Favorable Excursion (MFE) measures how far a trade moved in your favor before it ended. If you enter a long trade, MFE is the highest point price reached during the trade, measured in R-multiples.

MFE tells you what was possible, not what you captured.

The Discovery

We ran MFE analysis on 68 historical compression breakout trades. Here is what we found:

This was the key insight. Our old 3R target was structurally misaligned:

• 3R target requires 25% win rate to break even (1 win pays for 3 losses)
• Actual 3R reach rate: 23.5%
• Gap: -1.5 percentage points below break-even

No amount of parameter tuning can fix this. If your target reach rate is below break-even, you will lose money. The math is simple and final.

The 2R Sweet Spot

But look at the 2R numbers:

• 2R target requires 33% win rate to break even
• Actual 2R reach rate: 36.8%
• Margin: +3.8 percentage points above break-even

This is structural edge. Not optimization. Not curve fitting. Basic math alignment.

Part 3: Designing v1.1 with Pre-Registered Rules

The Core Hypothesis

v1.1 tests a specific mechanical hypothesis:

During volatility compression, liquidity thins and stops cluster. Breakout triggers a cascade effect that produces asymmetric price movement.

This is not “momentum continuation.” It is a liquidity vacuum event.

Locked Parameters

Before any paper trading, we locked every parameter:

Why Binary Exits?

This is important. v1.1 uses only two exit types:

• Stop hit: -1R (exactly)
• Target hit: +2R (exactly)

No trailing stops. No timeouts. No partial exits.

Why? Because we are testing a specific claim: Compression breakouts produce 2R moves often enough to be profitable.

Adding trailing stops or timeouts contaminates this test. If we use trailing stops and lose money, we do not know if the problem is:

1. Compression does not produce expansion
2. Our trailing stop parameters are wrong

Binary exits give a clean answer. Either the mechanism works or it does not.

Part 4: Backtest Validation

Before paper trading, we validated v1.1 against historical data:

This is not proof that v1.1 works. This is evidence that v1.1 is structurally viable.

The real test is paper trading with real-time data.

Part 5: Kill Criteria (Pre-Registered)

We committed to specific failure conditions before starting:

Primary Kills (300 trades)

Early Termination (150 trades)

Mechanism Falsification

Even if v1.1 is profitable, we test if the mechanism is real:

If median MFE is below 1.0R, this means trades are not even trying to expand. The mechanism would be falsified even if we happen to be profitable by luck.

Part 6: Gap Handling (Technical Detail)

One important technical decision: how to handle price gaps.

Stop Fill Rule

If price gaps through your stop level on bar open:

Fill at WORSE price (bar open, not stop level)

This is realistic. In a real gap-down, you get filled at the open, not at your stop. This means some losses exceed -1R.

Target Fill Rule

If price gaps through your target level on bar open:

Fill at TARGET level (cap at +2R)

This is conservative. We do not take credit for lucky gaps beyond our target.

This asymmetry is intentional:

• Pessimistic on stops — assume worst-case fills
• Conservative on targets — no windfall gains

Part 7: What We Learned

Mistakes We Made

1. Wrong exit hypothesis — Momentum decay exits too early
2. Target misalignment — 3R target was structurally below break-even
3. No pre-registration — v0.4 was built without locked parameters
4. No mechanism tests — We only looked at profitability, not structural validity

What We Changed

1. Binary exits — Clean test of the expansion hypothesis
2. Target calibration — 2R aligns with actual MFE distribution
3. Pre-registered spec — All parameters locked before trading
4. Mechanism tests — MFE and target hit rate test the underlying thesis

The Meta-Lesson

Parameter optimization is not research. If you tune parameters until backtests look good, you have not learned anything. You have just found numbers that fit historical noise.

Real research requires:

1. A specific hypothesis about market structure
2. Locked parameters before testing
3. Kill criteria that falsify the hypothesis
4. Honest reporting when things fail

v0.4 failed. We said so publicly. v1.1 might also fail. If it does, we will report that too.

Current Status

v1.1 paper trading starts now. Progress checkpoints:

All results will be shared transparently.

FAQ

Why did v0.4 fail?

Two reasons: (1) Momentum decay exits cut winners too early, and (2) the 3R target was above the structural reach rate. The strategy was mathematically losing.

Is v1.1 guaranteed to work?

No. v1.1 is a hypothesis under test. Historical backtests show positive expectancy, but paper trading will confirm or reject this.

Why not use trailing stops?

Trailing stops add complexity and ambiguity. If v1.1 fails with binary exits, we know the compression hypothesis is wrong. If it fails with trailing stops, we don’t know if the problem is the hypothesis or the trailing parameters.

How is v1.1 different from v0.4?

What happens if v1.1 fails?

We archive it and publish the failure. No re-tuning. No “v1.2 with adjusted parameters.” The hypothesis would be falsified.

Conclusion

Failure is information. Our v0.4 momentum-decay strategy lost money. Instead of hiding this, we analyzed it, found the structural flaw, and designed a new test with pre-registered rules.

v1.1 may succeed or fail. Either outcome teaches us something real about market structure. That is the point of research.

The dashboard is live. The rules are locked. The experiment begins now.