[Roadmap] External signals and auditable AI entry filters

Problem

Users want deterministic strategies to consult news, sentiment, or model output before opening a position without making exits dependent on an unavailable network service or introducing look-ahead bias.

Desired outcome

Create a point-in-time external signal contract that strategies can use as an optional, auditable entry gate.

Initial scope

• Ingest timestamped external factors outside the strategy sandbox.
• Persist source, model/provider version, observation time, availability time, score, and explanation metadata.
• Expose deterministic point-in-time reads to research, backtest, paper, and live strategies.
• Let users choose fail-open or fail-closed behavior for new entries.
• Keep risk exits and explicit close signals independent by default.

Acceptance criteria

• An RFC defines timing, provenance, retention, and fallback semantics.
• Backtests cannot read a factor before its recorded availability time.
• Every allow or veto decision can be audited after the run.
• Provider outages and stale signals are covered by tests.
• The strategy sandbox does not gain unrestricted network access.

Out of scope

• Installing arbitrary libraries during strategy execution.
• Letting a non-deterministic model silently override liquidation or safety controls.

Contributing

Comment with one contract, storage, provider adapter, fixture, policy, UI, or documentation task and a short plan. A maintainer will confirm scope before assignment.