# Corrected Phase 2 — Canonical LPH4 versus A0

## Validity and correction

The corrected comparison is valid for the requested 2022–2024 development and 2025 validation periods. The only source change was:

`ResearchOrderPrefix(): LPH_A0_SIMPLE_ -> A0_` for A0; Branch A still returns `LPH_TEST1_`.

This retains timeframe, direction, signal timestamp, and `_P1/_P2`, while keeping comments below the platform limit. Compilation completed with **0 errors and 0 warnings**. The defective A0 signal ledgers and corrected signal ledgers are byte-identical: 2,407 rows in development and 791 in validation. Thus signal selection did not change. Execution changed after TP2 breakeven became functional.

The first intentional metadata difference is the first A0 pending-order comment (`A0_H4_B_1735675200_P1/P2`). The first behavioral lifecycle difference is **2025-01-02 03:28:29**, when corrected ticket 3 moved to breakeven with intact live-position comment `A0_H4_B_1735675200_P2`; the defective run had no modification.

Canonical v15 remains unchanged:

- Source SHA-256: `9F3DA74DA26B905A8FF15D9AD442A65D8648E9BAD3AE6E5D574BB09605B4D666`
- Binary SHA-256: `2C02881D219C3A3FB76F3DD9DB0D9FA949B569EFD81539F167D3C1CEFD73480D`
- Branch A preset SHA-256: `26226E5A8B8F945ED66AA6D1E39E4F2C21472A6984462747171133BB89495FC1`

The archived Branch A control reports, logs, cache, and ledgers were not changed. Branch A was not rerun because the one-line conditional change alters only the A0 return value; the Branch A return value and all shared execution/management code are unchanged.

## Main comparison

Final balance/net profit are exact tester figures including simulated transaction costs. PF, gross outcome, expectancy in money, and drawdown are reconstructed from actual fills/exits before commission and swap because the headless tester did not emit its HTML report. Drawdown is closed-group realized drawdown, not intratrade equity drawdown.

| Period | Metric | A canonical | A0 corrected |
|---|---:|---:|---:|
| 2022–2024 | Net profit / return | $1,761.58 / 17.62% | **$4,094.57 / 40.95%** |
| | Valid signal states | 121 | 2,407 |
| | Pending groups | 89 | 462 |
| | Completed groups | 75 | 373 |
| | Price-only PF | **1.593** | 1.223 |
| | Group win rate | **60.00%** | 53.89% |
| | Price-only expectancy/group | **$26.94** | $14.17 |
| | Expectancy | **0.304R** | 0.099R |
| | Realized drawdown | **$461.65** | $1,500.10 |
| | Maximum consecutive losses | **3** | 8 |
| | TP1 / TP2 exits | 45 / 31 | 203 / 110 |
| | BE moves / BE-stop exits | 45 / 14 | 201 / 90 |
| | Hard-stop leg exits | 58 | 342 |
| 2025 | Net profit / return | $1,818.90 / 18.19% | **$9,023.16 / 90.23%** |
| | Valid signal states | 49 | 791 |
| | Pending groups | 32 | 153 |
| | Completed groups | 31 | 123 |
| | Price-only PF | **1.7284** | 1.7267 |
| | Group win rate | **64.52%** | 60.16% |
| | Price-only expectancy/group | $65.00 | **$78.54** |
| | Expectancy | **0.381R** | 0.265R |
| | Realized drawdown | **$666.85** | $2,132.00 |
| | Maximum consecutive losses | **2** | 4 |
| | TP1 / TP2 exits | 20 / 13 | 74 / 46 |
| | BE moves / BE-stop exits | 20 / 7 | 74 / 27 |
| | Hard-stop leg exits | 22 | 98 |

## Long and short performance

| Period | Branch | Long: groups / P&L / expectancy | Short: groups / P&L / expectancy |
|---|---|---|---|
| 2022–2024 | A | 40 / $2,177.60 / 0.545R | 35 / −$156.75 / 0.030R |
| | A0 | 201 / $6,787.20 / 0.218R | 172 / −$1,501.25 / −0.041R |
| 2025 | A | 29 / $1,736.80 / 0.338R | 2 / $278.35 / 1.001R |
| | A0 | 89 / $9,428.55 / 0.331R | 34 / $231.60 / 0.092R |

## Selection attribution

Branch A selected 121/2,407 A0 signal states in development and 49/791 in validation, removing **94.97%** and **93.81%**. These are completed-bar state counts; repeated qualifying A0 states can occur while a group is active.

Among corrected A0 groups that were actually submitted and completed:

| Period | Also selected by A | Removed by A | Retained win rate | Removed win rate | Removed price P&L / expectancy |
|---|---:|---:|---:|---:|---:|
| 2022–2024 | 14 | 359 | 64.29% | 53.48% | $5,005.95 / 0.096R |
| 2025 | 9 | 114 | 55.56% | 60.53% | $9,287.10 / 0.264R |

The filters selected a higher-win-rate subset in development, but not in validation. They did **not consistently remove proportionally more losers than winners**, and removed signals remained profitable as a group. The matched retained sample is small because active-order timing differs materially between branches, so this attribution should be treated as descriptive rather than causal.

## Verdict

1. **Risk-adjusted trade quality:** Yes. Full LPH4 retains higher R expectancy, equal-or-better PF, lower drawdown, and shorter losing streaks in both samples.
2. **Does A0 frequency compensate?** In total profit, decisively yes. On risk-adjusted efficiency, not consistently: canonical wins clearly in development; in 2025 A0's PF nearly matches canonical and its net-profit-to-realized-drawdown ratio is higher.
3. **Do the extra filters justify removing ~94–95%?** They justify a conservative, low-drawdown profile, but the evidence does not justify that degree of filtering as profit-efficient. The filters discard substantial positive expectancy, especially in validation.
4. **Carry A0 into production?** Do not replace canonical LPH4 with A0 and do not add A0 logic based on total profit alone. Carry A0 forward only as a research benchmark and as evidence that individual canonical filters may be overly restrictive. The next valid action is component ablation, particularly on short-side filters, not production integration.

The 2026 final OOS period remained sealed. No Momentum, Stochastic, RSI, or ablation branch was run.