Suppress the close leg of a fill-price-declined reversal

[luisleo526]

What

Pine reversal idiom: strategy.entry(opposite) then strategy.close(current) on the same signal. When the percent-of-equity admission gate declines the reversal entry at fill (unaffordable at the slipped fill price), TV refuses the reversal atomically — neither leg fires, the position is held. The engine declined the entry but still filled the co-queued close: flat where TV holds, then re-entry on a later signal TV no-ops. Ten all-in strategies carried ~960 such extra round-trips.

Fix, scoped to the reversal-decline site only (same-direction, gap-reject, and fixed-sizing role-change decline sites explicitly excluded): flag matching pending close legs (same-direction FULL closes with explicit target, created on the declined entry's signal bar after it), remove them via classify_order_eligibility + an apply-time guard (the calc_on_order_fills kernel pre-classifies its candidate set, so mid-segment flags need the apply-time check), and re-credit the close's call-time id-ledger debit exactly once so later closes on the held position still fire. close_all and partial closes keep current behavior (characterization-frozen).

Evidence

• 0-mismatch predictive rule across 208 traced reversal bars (55 declines / 153 admits, both directions); every traced decline gaps exactly +1 mintick over the sizing close.
• TV fired the paired close once in ~960 decline events across ten strategies (budgeted as a documented residual).
• Creation-order binding pinned by a strategy whose sell leg creates the close BEFORE the entry — TV fires that close 115/115; it remains unsuppressed.

Gates (author + independent fresh-context R7, both green)

• REDs by stash-cycle: 27/12 pre-fix → 39/39 post-fix (suppression rows fail pre-fix; characterization rows pass both sides).
• Full ctest 89/89; corpus 252/252 — 240 excellent / 11 strong / 1 anomaly, zero movement, cap/reversal/GB2 probe families byte-neutral; ledger re-credit proven by a follow-up-close fixture.
• Private validation sweeps: targeted 13 UP / 0 DOWN, full 412 +5 more UP / 0 DOWN (18 total, 11 straight to excellent), error inventory stable; all down-sentinels held.

🤖 Generated with Claude Code