Gate stop-entry fills at the open under margin; decline insolvent frozen opens cleanly

[luisleo526]

Summary

Two probe-pinned admission fixes (batch-4 clean-room TV probes):

KI-62 margin gate. Under active margin simulation TradingView checks every stop-entry fill at the fill moment against the fill bar's open (decline iff qty * open * margin% > available equity, side-symmetric; intrabar extremes ignored) and cancels declined orders outright — arm-once entries die silently while Pine-level reissue revives them, so re-posted shorts fill at the first bar whose open crosses the stop. Pinned at 99.17% on a 6-cell dedicated probe (this branch scores 99.9%, 1,436/1,438); independently cross-confirmed on over-allocated fixed-qty stops by a second probe (canonical 93.8 → 100.0). Declined fills bypass the 1×-long affordability trim; MARKET/LIMIT/RAW/exit orders and margin=0 paths untouched by construction (byte-identical sentinels prove it).

KI-72. A frozen percent-of-equity order sized at non-positive equity slipped past the sizing-equity guard as a negative quantity and opened a corrupt negative-qty position (negative trade rows, blown cumulative PnL, healthy internal sums). Such opens now decline cleanly and symmetrically, with declined-reversal close-legs suppressed via the existing ledger-restore path.

Verification

• New suites test_margin_stop_admission + test_short_reversal_emission; REDs stash-proven (5 suites RED on base). ctest 98/98.
• Two expectation flips justified in-test by second-probe evidence (over-allocated stop declines; the LIMIT sibling keeps admit+trim).
• Corpus 252: 240/11/1 exact. Byte-identical sentinels: lukeborgerding (MARKET+exit-stops under margin), freddycandle/taro $1M, KI-34 set, keepscratch/coof probes, margin=0 activation probe 2,410/2,410.
• pct=100 cohort (83 slugs): zero DOWNs. Full 412-slug private sweep: 2 up / 0 down (two weak→strong movers at 99.7/99.8 canonical; one anomaly-group slug 72.5 → 100.0).
• Fresh-context adversarial review: PASS (gate scoping, decline+cancel lifecycle, dust-trim reconciliation, stage-revert completeness all independently verified).

Known unpinned cells (recorded, non-blocking)

• Over-allocated FIXED-qty stops: evidenced by the second probe, no dedicated probe cell — candidate future probe.
• Stop-entry reversal admission uses pre-close realized-only equity (theoretical over-decline on profitable reversals) — validated basis, unpinned cell.
• Frozen qty exactly 0 now declines (was a fill-price refloor path) — extreme edge, zero corpus impact.

🤖 Generated with Claude Code