Summary

PR#95 gave calc_on_order_fills ENTRY orders cascade eligibility on the historical 4-tick path, but it never reached strategy.exit: the engine rolled 100% of exits placed by a mid-bar fill recalc — even those whose level sits at a remaining extreme. TradingView does not.

A clean-room probe (pf-probe-ki67-exitlim-midseg, 4,851 TV trades) pins the rule, Model S "R-cascade-gapjump" (4,848/4,850 = 99.96% event-level): after an intrabar fill at price p triggers a coof recalc, the emulator resumes the O→W1→W2→C proximity path from p. A priced strategy.exit placed at that recalc gap-fills same-bar at the in-flight leg-end waypoint when its level lies in the leg remainder (limits AND stops); fills same-bar at the exact level on subsequent legs; a terminal in-flight leg or off-path level rolls to next bar. The in-flight leg is read from the dispatch cursor's real position (price re-derivation is ambiguous at waypoint fills). Market/close and trailing cascade exits keep PR#95's extreme-waypoint reach; scope is narrowed to OrderType::EXIT (opposing raw strategy.order brackets keep PR#95 behavior — unpinned class, per review). Bar-magnifier path untouched.

Verification

• New suite test_cascade_exit_gapjump (34 checks; 6 RED on base + 1 corrected POOC assertion whose old expectation encoded PR#95's own flagged residual, refuted by the probe's FRACSTP cells). ctest 96/96. REDs independently reproduced by fresh-context review.
• pf-probe-ki67-exitlim-midseg: 36.8% → 98.8% raw, canonical excellent 100.0, exit/pnl P90 → 0.
• pf-probe-coof-refill-depth: 85.7% → 99.8%, P90s → 0.
• Coof cohort (fran/jenmb/delta/mrwick/wayward) byte-identical vs clean base (md5, verified twice independently); masayanfx + both KI-62 keepscratch probes byte-identical; 3commas standard sentinels byte-identical.
• Corpus 252: 240/11/1 exact. Full 412-slug private sweep: 0 up / 0 down.

Known residuals (recorded, non-blocking per review)

• 56-trade probe residual isolated to WP-control cells whose exit level is a dynamic bar-extreme (high/low) series re-arming intrabar — distinct mechanism, tracked as its own KI.
• Cascade-exit × same-bar add coverage (PR#97) is a new combination with no dedicated probe; mechanism is membership-based and unit-tested.
• Marketable-at-placement stays on the coof-suppress roll path (cohort-validated; pinned in test M1).

🤖 Generated with Claude Code