Research question
Which transformations of quoted prices and book imbalance remain useful across parameter choices rather than only in a single favourable configuration?
Context
This was a high-frequency factor-research exercise in a SQL/Python competition environment. It is retained as a method record, not as a live strategy claim.
Method
The explored family includes microprice, order-book imbalance, and toxicity-adjusted variants. Parameter sweeps compare construction choices before long–short validation.
Validation
The supported workflow compares signs, horizons, parameter grids, and long–short separation. Diagnostics matter more than a single headline statistic.
Current result
The archive supports that the factor family was implemented and evaluated. It does not publish a stable return claim, competition rank, or production status.
Limitations
Microstructure signals are sensitive to sampling, market regime, transaction costs, queue position, and implementation details that a competition environment may not reproduce.
Public artifact
Only the factor taxonomy and validation approach are described; code, private data, and unsupported metrics are omitted.
Next question
How much of the apparent signal survives cost-aware validation and tests across distinct liquidity regimes?