Skip to content
TotalFinance

backtest · wire operation

Bench a baseline policy over one trading-environment episode

Run one of the six maintained baseline policies through the deterministic trading environment over a catalogue episode (or a named declarative definition) at a seed, and return Agent Bench's report for that episode: operational conformance (look-ahead, replay equality, duplicate orders under the retry drive, external-order attempts, reconciliation; the mask and limit counts) apart from strategy quality (return, volatility, Sharpe, Sortino, drawdown, turnover, cost, exposure, violations) — no single score by design. With artifact: 'embed', the recorded episode travels as an `environment` run artifact that replays to its run hash. Declarative inputs only: a TypeScript policy or a reward goal callback is SDK-only. Opt-in (backtestPack).

totalfinance.backtest.environment_episode
MCP tool name
totalfinance_backtest_environment_episode
Pack / profiles
backtest / backtesting, full
Default discovery
Opt-in (select its pack or profile)
Required capabilities
None
Effects / authorization
none / none
Cost / cancellation
job / Do not assume inline interruption
Handle-eligible inputs
episode.definition.marketData
Idempotency
not-applicable

Input schema

FieldTypeMeaning
episodeRequired"trending" | "mean-reverting" | "range-bound" | "volatility-expansion" | "volatility-contraction" | "overnight-gaps" | "limit-moves" | "stale-data" | "missing-data" | "duplicated-data" | "corrected-data" | "out-of-order-data" | "wide-spreads" | "thin-liquidity" | "option-expiration" | "margin-pressure" | "retry-storm" | "corporate-actions" | "multi-currency" | "model-market-disagreement" | objectA catalogue id, or a named declarative definition
policyRequiredobject | object | object | object | object | objectA maintained baseline; a TypeScript policy is SDK-only
seedOptionalintegerDefault 42
artifactOptional"none" | "embed"Default 'none'; 'embed' returns the recorded episode as an `environment` run artifact
Complete JSON Schema

Loading…

Output schema

FieldTypeMeaning
policyRequiredstring—
episodeRequiredobjectThe bench report for the one episode: operational and strategy blocks
traceRequiredobjectThe recorded episode: runId, engineRunId, traceHash, steps, rewardTotal, terminated, truncated, reason
artifactRequiredobject | nullThe `environment` run artifact when artifact is 'embed'
assumptionsRequiredobjectApplied conventions
diagnosticsRequiredobjectEngine, method, warnings
Complete JSON Schema

Loading…