RESEARCH LOG · OPEN RECORD
NINE STRATEGIES MEASURED, NINE DISCARDED
This page does not show winners. It shows the ones that failed, with the numbers behind each and the date they were dropped. It is the part of the work nobody publishes, and it is the part that tells you whether a method is any good.
HOW IT IS MEASURED, AND WHY THE ORDER MATTERS
- The pre-registration is written and frozen BEFORE the code that measures. It fixes the hypothesis, the grid, the criteria and what happens on failure. It is never edited afterwards.
- The data is split in two: one part to build on, one part left untouched. The second is looked at once, and only if the first passes every criterion.
- The gate lives in the code, not in good intentions: if the in-sample fails, the out-of-sample is never computed.
- Costs are always subtracted: commission on both legs plus slippage. Most of what looks like an edge stops being one once the toll is paid.
THE NINE CANDIDATES
| CANDIDATE | WHAT IT CLAIMED | WHAT IT MEASURED | VERDICT |
|---|---|---|---|
| TD-01 2026-09-21 | Follow the daily trend in BTC and ETH: buy the N-day high, exit on the M-day low. | 23 ops · PF 8.61 · Sharpe 0.69 · sin las 5 mejores −0.444R · DD 60.6% | Discarded in-sample |
| TD-02 2026-09-22 | The same, with the selection rule fixed and funding cost included. | 73 ops in-sample (pasa los 7) · 57 ops fuera: +0.150R · PF 1.31 · Sharpe 0.38 | Discarded out-of-sample |
| E-01 2026-09-22 | Scalping ETH: false breakout of the accumulation range on one-minute candles. | 0 de 16 configuraciones positivas · de −0.126R a −0.407R por operacion | Discarded in-sample |
| E-02 2026-09-23 | Buying the bounce right after a liquidation cascade. | 0 de 12 configuraciones positivas | Discarded in-sample |
| F-01 2026-09-23 | Liquidity at the extremes: price sweeps the 40-session low, structure turns, and entry is on the pullback. | 22 ops · +0.357R · PF 1.92 · Sharpe 0.70 · minimo exigido: 50 ops | Discarded in-sample |
| F-02 2026-09-24 | The same, with a selection rule that cannot pick what fails the minimum sample. | 77 ops · +0.078R · PF 1.18 · Sharpe 0.33 · sin las 5 mejores −0.026R | Discarded in-sample |
| F-03 2026-09-26 | Is it fractal? The same rule, the same numbers, on four-hour candles instead of daily. | 132 ops · +0.018R · PF 1.04 · Sharpe 0.11 · en BTC −0.085R y en ETH −0.060R | Discarded in-sample |
| G-01 2026-09-26 | Long-only BTC: price above the 50-day average, three green 4h candles and three green days. | 22 ops · +1.035R · PF 4.09 · Sharpe 0.97 · sin las 5 mejores −0.03R | Discarded in-sample |
| G-02 2026-09-26 | The same G-01 rule, without touching a single number, applied to ETH. | 42 ops · +0.289R · PF 1.63 · Sharpe 0.43 · sin el 10% mejor −0.368R | Discarded in-sample |
WHAT IS LEFT AFTER NINE ATTEMPTS
Eight died on the first sample and never touched the reserved data. One, TD-02, passed all seven criteria and failed outside: that is where its reserve was spent and the line was closed.
A negative result measured properly is worth more than a positive one measured any old way. The first can be discarded with confidence; the second can be believed for years.
What comes in 2027: keep measuring in the open, publish every pre-registration before running it, and touch no reserved data until something deserves to spend it.
Palace.red does not guarantee returns. The figures on this page are historical test results, not promises: a backtest does not predict the future and every strategy carries risk of loss.