Paper Exit Rules — Slippage, Stops, and Stale Data

Paper exits are scenario rules, not evidence that a real exit was available. Define them before evaluation and apply them to timestamped snapshots. Every exit requires an eligible open paper position, a fresh validated quote, and a policy reason. Missing data produces ABSTAIN_EXIT, never an invented price.

Create separate rules for scheduled review, data-quality stop, hypothesis invalidation, synthetic adverse threshold, and final resolution. A “stop” is a paper control evaluated at captured times; it is not a guaranteed protection. Price can move between snapshots, spreads can widen, and no quote may exist at the threshold.

The fill assumption uses the adverse visible side plus predeclared slippage stress and latency. Record quote time, evaluation time, assumed delay, quote fields, spread, stress increment, policy version, and result. If age exceeds the limit, append STALE_NO_EXIT and keep the position unresolved in the report.

Avoid intraperiod lookahead. With hourly snapshots, you cannot claim an exit at a price briefly visible between them unless that observation exists in the dataset. Do not use daily high/low to simulate an exact stop without an explicit conservative ordering rule. When both a stop and favorable threshold appear in the same coarse interval, mark ordering ambiguous and run adverse/best scenarios separately.

Run sensitivity analysis across latency and slippage values fixed in advance. Show paper outcome, blocked-exit count, and maximum stale duration. A robust report highlights when performance depends on optimistic exit assumptions.

Also compare scheduled-review exits with threshold exits on the same sealed dataset. The comparison must use identical eligibility and cost policies, and it must include cases where neither policy could produce a synthetic fill. Publish the event-family contribution table so one unusual case cannot create a misleading difference between policies.

Idempotency links one exit scenario to position ID and policy version. A corrected snapshot creates a new scenario, never edits the old result. Resolution exits remain separate from mark-based scenarios.