Update from SolTrace

SolTrace update: integrated strategy verification

The latest development build connects Predictive Aiming Lab, strategy replay and the Decision Ledger in one optional verification workflow.

The aim is to make the contribution of predictive features measurable and explainable across the complete decision process.

For a selected strategy, SolTrace runs a paired comparison: predictive features enabled versus neutralized forecast inputs. Both runs use the same frozen policy, recorded market observations, capital, modeled costs and execution delays.

This makes it possible to examine how predictions affect entries, exits, trade frequency, drawdown and returns under consistent conditions.

The workflow connects three components:

1. Predictive Aiming Lab

A dedicated research stage examines market and freshly calculated forecast features using only the training portion of the selected recordings. It creates a frozen forecast-error calibration for the subsequent comparison. Validation and holdout data do not feed this research stage.

2. Strategy replay

The selected policy is replayed chronologically with the predictive contribution enabled and disabled. Training, validation and data withheld within the run are reported separately. Recorded forecast values are discarded and recalculated from observations available at each decision point.

The disabled branch uses neutral forecast inputs. If a policy requires positive predictions before entering, neutralization may prevent trades entirely. The report explicitly accounts for this limitation when interpreting the result.

3. Decision Ledger

Rule values, thresholds and decision reasons are linked to simulated execution outcomes, including fills, rejections and cancellations. This provides a way to investigate why the two branches behaved differently.

The combined report includes:

• Paired performance and execution results. • Separate training, validation and holdout comparisons. • Decision examples linked to their execution outcomes. • Downloadable evidence containing frozen inputs, model information and SHA-256 checksums. • Explicit inconclusive results when calibration support is insufficient, execution is incomplete or positions—including Moonbags—remain open.

The existing standalone modules remain independently available. The combined workflow does not automatically publish or activate a trading policy.

Next: validate the build on my server using real recordings, followed by newly collected data.

The goal is reproducible evidence of where predictive features help, fail or make no difference—and an explanation of the decisions behind those results.