Benchmark Lab · Metrics
Metric definitions.
Honest definitions, not marketing labels. Every metric on the Benchmark Lab overview is defined here, with notes on what it does and does not capture.
Execution gap (naive − realistic)
Average naive (paper-leg) PnL minus average realistic (execution-adjusted) PnL over the same benchmarked trades, in percentage points (pp).
JSON historic exports used the label “paper − live”; realistic means modeled execution stress, not a guarantee of wallet-level live results. Not Sharpe or risk-adjusted.
Win rate (naive vs realistic)
Share of trades with positive PnL under naive assumptions vs under realistic execution adjustments.
Median entry / exit slippage (%)
Difference between intended and realized prices on entry and exit legs, expressed as a percentage of price.
Published as medians to reduce outlier impact. Naive leg does not carry the same execution stress as the realistic leg.
Failed exit rate
Share of trades where the intended exit did not complete in the benchmark sense (route, depth, timeout bucket, etc.).
This is an operational execution statistic, not a legal settlement guarantee.
Average latency
Mean round-trip operational latency in milliseconds for the benchmark window.
ATA rent (cumulative)
Sum of associated-token-account rent debits included in the benchmark cost model for the window.
Market regime rows
Each row aggregates benchmark trades whose entry-time features mapped to that public-safe bucket (label only; rule internals and raw feature JSON are not published).
Strategy family row
Opaque family_id, human-neutral display_name, and aggregate win rate / median PnL (realistic leg) over trades attributed to that family.
No parameters, no thresholds, no entry/exit logic are published.