What Doesn't Work

This page is a graveyard of ideas that looked promising and failed when tested properly. It matters because a trading system gets stronger not just from what it adds, but from what it refuses to keep.

If you want the vocabulary behind terms like profit factor, RSI, delta, or order block, see the Glossary.

Everything below was tested and rejected. These are closed doors.

1. OB Confluence (Order Block Gate)

Think of an order block as a price zone where large buyers or sellers previously showed their hand. The idea was to only take H9 entries when they lined up with those zones.

Tested: Use order block proximity as an entry filter for H9.
Result: Look-ahead biased. Without bias, all variants hit 99%+ max drawdown.
Status: Removed from live code and strategy.

2. Kelly Sizing

The Kelly criterion is a bet-sizing formula that can be powerful in clean mathematical models. In live options trading, costs make it dangerous.

Tested: Kelly criterion for position sizing.
Result: Costs inside the Kelly loop create a death spiral. 97%+ max drawdown.
Fix: Fixed-fraction (6%) is the correct approach.

3. PM Session

This tested whether Plan H could simply keep trading into the afternoon. More hours sounds good in theory. The numbers said otherwise.

Tested: Trading H9 signals in the afternoon session.
Result: Net -$14K over 10 years across all variants. AM-only is strictly better.
Status: PM session permanently disabled. Force close at 11:28 ET.

4. sig=3+

Here, sig means the number of signals allowed per day. The question was whether taking the third, fourth, or later setup would add profit.

Tested: Allowing 3 or more signals per day.
Result: Extra signals are net losers. They dilute edge and increase drawdown.
Fix: sig=2 is optimal.

5. OB Quality Scoring

Instead of just checking whether an order block existed, this version tried to grade each one.

Tested: Scoring order blocks by quality metrics to filter entries.
Result: Inverse correlation (r=-0.33). Higher quality = worse performance.
Status: Rejected.

6. 1-Min Chart Filters

This was an attempt to use finer-grained 1-minute price patterns to improve 5-minute entries.

Tested: Using 1-minute chart patterns to filter 5-minute entries.
Result: No discriminating power between winners and losers at entry.
Status: Rejected.

7. Midday Skip

The hypothesis was simple: midday is usually quiet, so skip it. The data showed those quiet periods were not damaging enough to justify the filter.

Tested: Skipping signals during the midday dead zone.
Result: Hurts P&L. Dead zone losses are tiny — not worth filtering out.
Status: Rejected.

8. J1 Gap Reversal

A gap reversal tries to trade the market snapping back after a strong move between yesterday's close and today's open.

Tested: Gap reversal signal as a complement to H9.
Result: 286 trades, -\(51K net. H9-only (\)556K) beats H9+J1 ($512K).
Status: Permanently parked.

9. TSLA in Plan H

This tested whether the H9 intraday logic could be ported from index ETFs to a single high-volatility stock.

Tested: Running H9 on TSLA during bull regimes.
Result: H9 was designed/optimized on index ETFs, not single stocks. TSLA options have wider spreads and the $33K bull backtest was never validated against realistic execution costs.
Status: Removed 2026-03-19. Bull regime capital goes to Plan M.

The Lesson

Every rejected idea seemed promising in theory. Data killed them. This is why we backtest everything before going live.