Exit stop and limit matching now consumes concrete activation bounds attached to a position-cycle identity. Each bound names the earliest script bar where that leg may be evaluated for that owner. A mismatch cannot reuse another cycle's constraint; an unbound value imposes no lower-bound constraint. This is one exposure-cycle scope, not a physical lot or from_entry claim. Existing quantity/entry ownership and other path/trailing rules remain separate.

The lower bounds also apply when a dormant bracket is revived after a margin partial and when a prearmed bracket selects a stop or limit at the open. Revival can restore the bracket without executing an unready stop. Open-gap selection checks each leg before applying stop/limit precedence, so an eligible sibling retains its own execution and slippage semantics. Risk liquidation itself is independent of a bracket's activation time. The historical chart-extreme rounding fallback checks the same per-leg bounds; a normal no-fill result cannot become an early fill through that fallback.

The transitions are connected to the current book. New or replaced EXIT construction binds against live exposure. Both fresh-position producers bind retained exits after establishing the new cycle, before later matching or callbacks. Going flat clears bindings while retained policy evidence stays with the order. Partial reductions and same-cycle adds do not refresh a deadline. Order replacement creates fresh placement evidence and activation; cancellation and OCA erasure remove the values with their owning order.

For Pine compatibility, immutable ExitPlacementEvidence retains the original callback marketability inputs: placement cycle/entry bar, physical exposure direction, cursor price, raw stop/limit levels and any named limit continuation. Current trigger fields can be neutralized or modified without rewriting those facts. LaterSameOpen and FirstHighRecross continuations remain explicit Pine exceptions; their existing predicates are evaluated in the adapter. Their recorded observed fill sequence is the sequence visible at policy selection; the separate OrderBirth still owns the actual callback interval.

At binding, the adapter resolves the retained original decision to concrete bounds for the new owner. A held leg receives owner_entry_bar + 1; the other leg receives owner_entry_bar. The arithmetic uses a wider bar coordinate. The matcher never refreshes these numbers from the current position. In particular a surviving same-ID exit can bind a new cycle and receive another hold, even though its original birth-bar deadline has already passed. New marketability is not recalculated from the replacement owner's opposite side.

The two native coof_suppress_*_on_entry_bar booleans are removed. Their old C mirror fields are read-only projections from the Pine evidence; no native decision reads those outputs. All 128 pre-existing mirror fields and the full old byte prefix remain unchanged, and new activation/evidence fields append. Every new owner, coordinate, evidence value and optional-presence discriminator is hashed and mirrored. There is no side map or second pending-order book.

This slice moves the two decisions out of the matching masks, but does not eliminate the Pine policies or detach their producer from existing native callback use. The Pine adapter is still invoked by the current exit producer. PineHistoricalBirthReach, cascade leg/gap permissions, dormant-leg revival and other historical execution rules remain separate work. The effective-level accessor still reports resolved levels rather than masking eligibility.

Historical internal object layouts require matching C++ headers/library. L3b removes the compatibility order object and projects the frozen pending-row ABI from native requests plus adapter facts. Public C ABI 4, stream API 1, and pending mirror version 1 are preserved.