US execution evidence

简体中文

Existing US daily caches still lack the evidence required for next-session replay. The adapter has passed synthetic checks, but real-data replay remains blocked. It submits no orders.

The US next-session adapter is an offline price-NAV diagnostic. Synthetic contract tests pass through the existing USD cash ledger. Existing daily caches still need execution evidence before they can support this replay. No orders are submitted and no total-return or execution eligibility is inferred.

The adapter is separate from the five-market next-open experiment. Its model is us-next-session-price-nav.v1. It requires USD instruments, raw signal and execution prices, an explicit XNYS calendar version, time-zone-aware signal, availability and decision times, and the first regular open strictly after information availability. The decision must precede that open. A late decision or missing first open blocks the replay rather than moving the trade. The signal field records the source cutoff. The ledger’s signal clock uses the later availability time, before the declared decision. Daily close valuation availability remains a modeled date-lag assumption.

Corporate-action coverage must span every input session and instrument. Known events block this price-only model. Missing evidence also blocks. Real event-free evidence requires an inspected, source-controlled digest in TRUSTED_ACTION_REVIEWS. This registry is initially empty. Setting a JSON flag cannot establish that a real interval is event-free. Synthetic action evidence is accepted only with synthetic input identity and remains labeled synthetic.

Initial cash is USD 100000. Fractional sizing is explicitly modeled. Integral sizing requires a positive integer lot size. --fee-bps is a finite nonnegative commission scenario below 10000, with zero additional slippage and FX cost. The default is 20 bps. Fees are not an estimate of any particular broker’s tariff. The bundle’s configuration_sha256 binds input configuration bytes and effective fee/model parameters, so different fee scenarios have distinct fingerprints.

From the standalone research/platform-compare environment:

/absolute/... paths are placeholders. On Windows use drive-qualified absolute paths resolved from the project configuration.

python -m platform_compare.us_next_session.run --config /absolute/us-inputs.json --output-root /absolute/new-output --fee-bps 20 --json

The configuration schema is competition.us_next_session_inputs.v1. It binds CSV and action-evidence files by SHA-256. CSV columns are date, open, close. All inputs and outputs stay outside the checkout. The CLI rechecks all bound files after calculation and writes a diagnostic bundle plus run.json.

The root CLI dispatches without importing the accounting package:

python -m competition_tools.team run us-next-session --config /absolute/team.toml --experiment-config /absolute/us-inputs.json --output-root /absolute/new-output
python -m competition_tools.team execution-readiness --config /absolute/team.toml --spec /absolute/readiness-spec.json --output-root /absolute/new-readiness

The readiness spec contains schema set to competition.us_execution_readiness_spec.v1 and manifests, a list of path/sha256 records. This command inventories declared daily-manifest metadata and missing execution evidence. It does not verify a complete price dataset, grant trading permission or calculate performance. A raw-price declaration alone does not establish first-regular-open, availability, listing, lot or action coverage. Original study engines and published results remain unchanged.