Lower generated Pine execution onto the native kernel and retire the legacy loop (R4 slice C) - #254

…gin configuration provider to the native kernel (R4-D L1)

Implement the fenced generic lowering prerequisites:
- native v5 cohort ownership/authority, live NoTarget deferral, deterministic roster resolution, receipts, and continuation hashing;
- native v2 owned lower-timeframe intrabar paths and v5 driver decision context/delivery;
- native host v17 pre-open and pre-begin provider hooks plus staged ingress;
- v7 native-consumer matching, validation, lifecycle, hash/ABI fences, relocation matrix, and native witnesses.

Contract rules: §3.1; P7a; P8a/P8b; P11; P15; P-DA1/P-DA3; amendments A1/A3/A5/A7/A8.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

…NativeBeginArgs (R4-D L1c)

Pass the rich overload borrowed SymInfo through the native pre-begin provider, preserve null semantics on the other begin entries, and pin the additive value member in the ABI manifest/checkers. Extend the native fixture witness to verify by-value metadata projection and hash distinction.

Contract rules: section 3.1; P8a/P8b; P11; A8; A12.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

WIP: add the dead-code PineExecutionAdapter, PineScheduler, PineNativeHost fixture, source-state hashing, A11/A12 provider copies, and exact native deferred-ANY oracle twin. The twin remains intentionally failing at the pre-command native timestamp admissibility boundary; no literal oracle assertion was weakened.

Contract rules: §0, §2 P1/P3/P5/P7/P8/P13/P15, §3.2–§3.5; amendments A2–A8, A11, A12; deferred-ANY pin P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… and audit every native-only refusal (R4-D L1d)

Add the hashed Canonical/LegacyTolerant slot-label policy and the explicit strict-by-default legacy batch structural tolerance. Preserve raw timestamp decision labels under the compatibility policy, retain the legacy timestamp-overflow refusal, and pin the spec/driver ABI checker and relocation-manifest rows.

Add pure-native witnesses for the L0 1000/2000/3000/4000 ms shape, canonical refusal, raw decision coordinates, market fill progression, structural tolerance, and policy hash coverage. Contract rules: §0, A11, A12, A13, A4, A5.

Verification: ci_verify release and native each have only test_native_oracle_deferred_any_witnesses_l2 failing: known WIP failure owned by L2, fixed when the source provider opts into the A13 policies. All L1d-owned witnesses and guards are green.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Extend the L2 fixture-only adapter, scheduler, PineNativeHost, source-state hash coverage, and exact native-route oracle twins. The deferred-ANY, magnifier, >64-fill, frozen-size, and reversal witnesses pass; the POOC immediate twin remains blocked by a native-vs-legacy execution allowance/close-size divergence recorded in the L2 report.

Contract rules: §0, P1–P9, P13, P15, §3.2–§3.5; amendments A2, A4–A13; P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Repair fixture-host configuration retention, percentage close snapshots, immediate-close lowering, per-origin bracket lifecycle, source pyramiding accounting, terminal receipt handling, source-state hashing, and the exact deferred-birth oracle twin. POOC/immediate and deferred-birth literals are now green; remaining exact-twin repair ledger is recorded in LANDING-L2-REPORT.md.

Contract rules: §0, P1–P9, P13, P15, §3.2–§3.5; amendments A2, A4–A13; P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Extend the fixture-only lowering with POOC ordering, deferred relative exits, source mintick retention, frozen priced reversals, and paired later-same-tick reversal semantics. Add exact native fixture twins for POOC freeze, relative exits, close-only reversals, and later-tick reversals.

Contract rules: §0, P1–P9, P13, P15, §3.2–§3.5; amendments A2, A4–A13; P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Extend fixture-only COOF callback cadence with first-open and later Applied recalculations, retained source context, tail command deferral, and fully hashed source scheduler/adapter state. Keep the existing native fixture oracle set green while the remaining exact COOF rows are recorded as WIP evidence.

Contract rules: §0, P1–P9, P13, P15, §3.2–§3.5; amendments A2, A4–A13; P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

…intrabar matching to the native kernel (R4-D L1e)

Allow copied account-currency FX configuration to be staged before a run and after completion or cooperative abort, with staged FX and driver facts folded into continuation identity. Add IntrabarPath::SampleEligibility and generic decision-context driver statistics so hosts can select discrete distribution samples or the default continuous segments. Update the native ABI manifest and ownership guards, with native fixture coverage for FX replacement/clear/abort and distribution fill splits.

Contract: §0.2, §3.1 staged configuration ingress and intrabar path; P8b; P15; amendments A5, A15, A16.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Opt source fixture magnifier paths into L1e DistributionSamples eligibility and project generic driver statistics into the legacy-compatible fixture diagnostics. The exact source magnifier oracle then exposed the remaining LegacyTolerant raw-partition intrabar grouping gap, recorded in the L2 report without altering literals.

Contract rules: §0, P8b, P15, §3.2–§3.5; amendments A13, A15–A17.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Freeze default stop-entry sizing at the directionally snapped stop level and add fixture-native stop-snapshot and ShortSeed role witnesses for fixed and percent books. ShortSeed plan handles are run/incarnation-qualified, invalidated on terminal receipt, and source-hashed.

Contract rules: §0, P5, P9, P13, P15, §3.2–§3.5; amendments A2, A4–A17; P-DA1–P-DA7.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Reproduce staged-FX frozen admission, broker-open FX rollover liquidation, and the source scheduler refusals on the fixture-only adapter route.\n\nContract rules: §0, §3.2–§3.5, P2, P5, P6, P13, P15; amendments A13, A15, A16, A17.\n\nCo-Authored-By: Claude Fable 5.1 <[email protected]>\nClaude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Restore fixture-source timeframe/day-partition/report lifecycle; add exact day-key and replacement-reversal twins; and carry COOF tail/open behavior through source lowering.\n\nContract rules: §0, §3.2–§3.5, P2, P4–P9, P13, P15; amendments A13, A15, A16, A17.\n\nCo-Authored-By: Claude Fable 5.1 <[email protected]>\nClaude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Implement the source-owned frozen same-bar MARKET transaction batch, targeted-close artifact, and gross placement admission census; register the unchanged exact fixture twin. Hash all added placement and batch state.

Contract rules: §0, P4, P6, P9, §3.2, §3.3, §3.5, P13, P15; amendments A13, A17.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Add exact fixture-native frozen-size coverage and source lowering for frozen lot sizing, default strategy.order, OCA sibling semantics, source stop placement/admission, and tick-projected market fills. Preserve unregistered exact STOP twins and derived fixture views for the remaining direct-owner cases; no legacy owner is reintroduced.

Contract rules: §0, P2, P4, P5, P6, P9, §3.2, §3.3, §3.4, §3.5, P13, P15; deferred-ANY P-DA1–P-DA5; amendments A13, A17, A18, A19.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Refine the derived fixture pending projection so a targeted close exposes its truthful request label while preserving the source-owned batch facts. This advances the exact fixed ShortSeed oracle evidence without restoring a legacy pending book.

Contract rules: §0, §3.4, §3.5, P9, P13; amendment A17.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

…r feed exists (R4-D L1f)

Add the generic synthesized IntrabarPath variant, validation, FNV path digest, driver delivery, and statistics coverage.

Advance intrabar decision floors at emitted sub-bar points so applied-callback requests remain eligible within the active script bar. Preserve lower-feed and no-path behavior, and update native ABI manifests/checkers and native witnesses.

Contract: §0.2, §3.1, P15; amendments A13, A16, A18, A19.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

… host without switching the generated route (R4-D L2)

Complete the L1f source projection, fill-bar margin slice, and variable ShortSeed lowering on the fixture-only native route.

Add exact full fixture coverage for stop snapshots and fixed/percent ShortSeed books; retain all added adapter state in the source hash.

Contract: §0, §2 P2/P4-P9/P13/P15, §3.2-§3.5, P-DA1-P-DA5; amendments A15-A20.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

…egister the A24 targets (R4-D L3a-fix)

Copy map-backed request handles and placement snapshots before adapter erasure or submission can invalidate them.

Re-register the A24 behavioral targets and rewrite the syminfo index fixture to observe the switched public route with unchanged literals.

Contract: R4-D §0; A20; A23; A24.

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

…ve scheduler hooks and copy request handles before retiring adapter placements (R4-D L3a-fix)

Restore the legacy source security/timeframe feed cadence through the switched scheduler and provider, including raw input visibility, source-series/index projection, split-feed ordering, and quiet source abort reporting.

Add the permitted generic accepted-input hook and abort-reporting run-spec policy with continuation hash, ABI relocation, and native witness coverage; retain strict native defaults.

Replace the market-admission owner-book harness with the A25 fixture/public projection twin and preserve A23 lifetime-copy coverage from the WIP parent.

Contract: §0, P7b/P7c, P8a, P11, P13-P15; amendments A13, A20, A23-A25.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

Claude-Session: https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

PineExecutionAdapter::exit's submit_leg lambda ($_11) resolved the short-exit
limit/stop trigger with two unbounded while loops that stepped one ULP at a
time via nextafter until nearest_tick crossed the level. On an on-grid level
k*mintick lands one ULP above the decimal grid point, so the walk crossed a
whole tick (~1e9 steps) on every exit call -- alby1611-bollinger (POOC, 3761
trades) never terminated. Replace both loops with the existing
source_trigger_threshold(level, tick, is_buy=true, is_limit), which resolves
the same half-tick boundary on the grid index with a bounded 16-step walk and
already sets the other exit thresholds. (ab9714b include/pineforge/engine.hpp
:1268-1312 tick_grid_price k/(1/mintick))

Verified: alby1611 owner 1881 / candidate 1881 diffrows=0 (was infinite hang);
guards joat/traderhayz 0; ctest 492/492; hash guard OK. src/source only.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit a65f4132558872a39381c5ba7176836f5b117731)

…F aux-security checkpoint

Three below-strong parity fixes (src/source only):
- entry(): every entry limit, explicit-qty included, tests against the
  tick-quantized bar; removed the explicit-fixed bypass so an off-grid limit
  triggers on the half-up rounded print (drop-trades-orb 66->0).
  (ab9714b pine_policy_members.cpp:11-17, engine.hpp:1374-1381)
- on_applied(): a released entry on the closed position's own side books its
  own slippage off the shared open print, not the close's slipped price
  (3commas-triple 2->0). (ab9714b pine_fills.cpp:3848-3856, engine.hpp:1207-1210)
- scheduler_feed_aux_security(): under calc_on_order_fills, restore the script
  checkpoint, feed the auxiliary request.security slice, then re-snapshot, so
  the fed values survive the per-bar restore (fran-pineda-470 2->0).
  (ab9714b pine_scheduler.cpp:1705-1708, 467)

Verified by executor: 3 targets + guard diffrows=0, ctest 492/492, hash guard OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 98dfe87eb641d910ddbc3bf55ba9dbca3107694d)

…r_close trim, zero-offset trail threshold

Five exit fill-price parity fixes (src/source/pine_adapter.cpp only; PARTIAL:
5 of 8 targets, the other 3 need generic kernel features not in scope):
- Legs from the same strategy.exit no longer count as surviving siblings, so a
  filled percent-qty exit is marked used and does not re-fill
  (joat-concordance 586->0, jlall7892 18->0). (ab9714b pine_fills.cpp:7012-7024)
- A used-up id's re-issue is blocked only when itself partial against the live
  position (keeps the pinned exit-bracket tests green).
  (ab9714b pine_strategy_commands.cpp:1661-1692)
- A commissioned explicit-qty market short at 100% margin trims at the fill
  price first, then the bar's adverse extreme (mdfe3757 1530->0).
  (ab9714b pine_fills.cpp:1266-1340, 5998-6005, 6168-6179)
- A stop beside an armed zero-offset trail takes the zero-offset trail price;
  an unarmed zero-offset trail uses the half-tick boundary on the rounded
  price path (boztilkiserhan 270->0).
  (ab9714b engine_path_resolve.cpp:620-705, 292-308, 746-766)

Verified by executor: 5 targets + 2 guards diffrows=0, ctest 492/492, hash OK.
Remaining (kernel-gated, tracked): francescodimichele (trail slippage vs limit
rejection), 3commas xlm/btcusdt-p grids (near-zero close UnrepresentableQuantity).

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 2e2fbf06e8e0b67aa7fe955ee4da77a571315283)

…gaps at open

Two below-strong parity fixes (src/source/pine_adapter.cpp only):
- close(): under process_orders_on_close (and not calc_on_order_fills), an
  ordinary close_all fills at the call, ahead of any same-bar market entry the
  script placed before it, so an under-cap add issued earlier survives the flat;
  the script keeps its pre-close position view (freeze_script_position_view,
  KI-64). Fixes 3commas-rsi-dip-ema on both 1D datasets (38->0, 46->0).
  (ab9714b pine_strategy_commands.cpp:789-801, 2437-2446)
- resolve_terms(): a stop armed with its pending market parent and already
  breached at the parent's fill open is an opening gap, not a path cross, so it
  books bar.open (matches TV PnL 0). Guarded to ExitStop + Confirmed +
  raw==script-bar open + no intrabar + not POOC/COOF. Fixes bhajneek12 (2->0).
  (ab9714b pine_fills.cpp:4136-4240, 7795-7800)

Verified by executor: 3 targets (bhajneek12, 3commas-rsi-dip x2) diffrows=0 via
1D dual-feed; stevenygabbyperez already 0; guards + 33 extra pairs no new
regression; ctest 492/492; hash guard OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 2216507a92950585714aae5e9161dc14cf7d658f)

…lipped-touch trail (ruling A-STOPFILL)

Fixes francescodimichele-gold-ai-strategy-v2-0 (258->0). A one-shot trail with a
sub-tick offset is a favourable-direction touch: the level gates WHEN it fills
but the owner applies slippage so the fill lands WORSE than the level. A native
Stop fires on an adverse move (wrong direction), and a bounded Limit rejects a
worse-than-level fill; neither models this.

Generic kernel capability (ruling A-STOPFILL, user-approved D180):
- native_order::Limit gains `bool fill_through = false` (market-if-touched): the
  level still gates executability, but the fill is not bounded by the level.
  Default false => every existing request is unchanged; folded into hash_trigger
  ONLY when set, so bounded limits keep their digest. native_execution_consumer
  skips the InvalidTerms level guard and protect_limit when the flag is set.
  No Pine/TV/probe references, no new constants -- a plain order-execution option.

Adapter (src/source) decides when to use it, owner-cited:
- exit(): a zero-distance one-shot trail leg with slippage>0 is sent as
  Limit{activation, fill_through=true} (touch at the unslipped activation, so it
  does not fire early). (ab9714b engine_path_resolve.cpp:984-987, pine_fills.cpp:4396-4411)
- resolve_terms(): the fill moves slippage ticks against the exit, snapped to the
  tick grid. (ab9714b pine_policy_members.cpp:45-53)
- precommit excursion: the trail peak uses the pre-slippage fill; an open-gap
  fill adds no peak. (ab9714b pine_fills.cpp:5760-5769, engine_path_resolve.cpp:620-631)

Verified: francescodimichele 0, guards apkw82/mdfe3757/joat-aureate 0, ctest
492/492, hash guard OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 5dedf5a7e9f57a7a619278b8467b49a68e003677)

Fixes stevenygabbyperez-fast-scalper-with-stops (excellent->weak, visible only
under the Cloud Run readback config: trimmed start-of-window feed + chart-warmup).
At a margin call, revive_brackets_after_margin revived a stale OLDER placement of
Exit Short (an earlier bar's stop) even though a later same-id strategy.exit had
re-priced it in place with a new stop. The candidate then closed the remainder at
the stale stop instead of the current one.

The owner keeps ONE pending exit per (id, from_entry); a later-bar re-issue
replaces it in place and its revival scans only live pending orders
(ab9714b pine_fills.cpp:2069-2094). A re-issue over a dormant predecessor
carries no projection link, so match by identity: an older placement is
superseded when a later placement with the same leg family, id, from_entry and
position cycle was issued on a later script bar with a different stop. A same-bar
re-issue still revives the originally armed stop (pine_fills.cpp:2088-2094); a
same-level re-materialization keeps the predecessor.

Verified: steven Cloud-Run-config 5360->0 (2680/2680); full-history and --window
both 0; guards apkw82/mdfe3757 0; ctest 492/492; hash guard OK. src/source only.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit ed67c8a8aea9686aa29e8496bda78c82b6a72496)

…vation (near-zero close)

Fixes the two 3commas grid probes from rejected/diffrows=1170 to diffrows=2 (only
the dust-row fee remains, see below), in src/source. Three owner-cited changes:
- flush_pending_closes: when a partial close uses up only whole lots it becomes a
  Flatten of exactly those lots, but ONLY when a plain FIFO reduce would leave or
  take a dust fragment (clean drains stay a reduce); later closes in the same
  flush see the lots left after earlier ones. Lots are picked at the fill, not at
  submission, so a same-bar strategy_order trim no longer stales them
  (fixes test_integration_l4d). (ab9714b pine_orders.cpp:29-70, 340-347)
- off-grid close target: the owner closes its raw (unrounded) target, so an
  off-grid target goes through a host-sized close resolved at the fill; on-grid
  targets keep the fixed-quantity reduce. (ab9714b pine_strategy_commands.cpp:1397-1408)
- reservation: reserve qty_before - position_qty_ (the owner's pre-fill held
  units) via a transient precommit_held_units_ cache, cleared at the applied
  boundary; the id close ledger keeps the settled-units floor.
  (ab9714b pine_strategy_commands.cpp:1575, 1557-1575)
precommit_held_units_ waived in broker_state_hash_waivers.txt (transient cache,
same class as excursion_*).

Verified: grids diffrows 1170->2 (dust-row fee only), guards apkw82/joat/mdfe3757
0, ctest 492/492, hash guard OK. src/source only. Remaining dust-row fee is a
kernel fee-allocation matter (tracked).

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit fe7aab8de51be79942c7cdc2f87c39554c774d8b)

…review blocker)

sort_same_bar_exit_trades reordered the Trade objects but not the parallel
trade_exit_phase_ (keyed by trade index), so is_open_phase_exit read stale
phases on the next per-bar sort and the same-bar exit group could flip order
(nondeterministic export order; latent, introduced by aea775d). Add
PineExecutionAdapter::permute_exit_phases(start, indices) and apply the sort's
permutation to trade_exit_phase_ right after the trade write-back, so each
trade's phase moves with it. The sort's adapter parameter becomes non-const
(both callers pass the non-const adapter_); trade_exit_phase_ stays hashed
source-state (not made mutable). Also removes a dead duplicate return in exit().

Verified: ctest 492/492 (test_l10m_corpus_parity, test_l10ab, test_l10ak pass);
hash guard OK; two runs byte-identical (determinism); guards apkw82/joat 0.
src/source only. Fixes the Opus final-review blocker.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 09a24feecb82dcb54430b1538b0925d57f45cc0d)

The engine is two layers: a Pine-agnostic generic kernel (matching, fills,
sizing, margin, magnifier, indicators, request.security, time) and a
source-adapter parity runtime (src/source/: PineStrategyHost,
PineExecutionAdapter) that maps Pine/TradingView execution semantics onto the
kernel; codegen owns Pine->C++ translation, not execution. Adds an
"Architecture: kernel vs. parity" section (kernel changes only for ruled
generic capabilities), splits the libpineforge.a description, and corrects the
stale Repository layout (26->48 .cpp, adds src/source and src/compat/pine) and
the ABI-count number (32->65).

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

Two hard-band probes regressed vs the legacy baseline (gate hard.regression):
- market-logic-india-low-lag: on a bar that does entry Long; close Long; entry
  Short; close Short, the candidate made close("Long") wait on the same-bar
  capped re-entry, which the open-boundary admission then retired, cancelling
  the close with it; the short then only flattened. Fix: a close no longer waits
  on a capped re-entry unless an earlier opposite same-bar entry exists.
  (ab9714b pine_fills.cpp:2991-3009). data-OANDA-XAUUSD 10352->0, BTCUSDT
  19364->0, EURUSD 11460->0, NYSE-F 3938->0, AAPL 2206->0.
- stockhunter2025-btcusd-4h: (A) a margin slice sized on the pre-fill position
  liquidated dust left by same-bar priced exits; re-size any still-live margin
  slice from the reduced book after a priced exit fills (guarded !POOC/!COOF/
  !FX/!1x-long). (ab9714b pine_scheduler.cpp:267-282,177-183, pine_fills.cpp:2224).
  (B) the deferred margin slice missed the exit-side slippage the current-point
  route already applies; add it (zero-slippage unchanged). (ab9714b
  pine_fills.cpp:1711-1726). BTCUSDT 208->0, data 1366->0.

Verified: both probes diffrows=0; margin-path guards apkw82/joat/mdfe3757/
traderhayz 0; ctest 492/492; hash guard OK. src/source only.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT
(cherry picked from commit 0ac768491aec67b2d89d7d895e1d984288267b01)

…ry POOC

The 6 process_orders_on_close probes timed out on the Cloud Run gate's
full-history broker-active rung: the source adapter kept full history and
several hot paths rescanned it every bar/order, so a long feed ran O(N^2)
(21-798s vs the legacy 0.6-2.2s). Add derived indexes/counts (per-id
command-sequence minima, per-id entry position index, per-family bracket
member counts, per-from_entry reversal cache) so those paths are O(1)/O(k).
Behavior-preserving: all 42 before/after trade CSVs byte-identical, guards
diffrows=0, ctest 492/492, hash guard OK (indexes are derived caches, not
hashed state). src/source only, kernel untouched.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…emes

Class A of the target.not-positive gate blocker. A calc_on_order_fills
cascade order that targets a bar extreme must trigger at the RAW high/low
and book at that price rounded to the tick (owner ab9714b
pine_scheduler.cpp:543-606, booking pine_fills.cpp:7962-7966). R4-D armed
these orders at the rounded price, so an order that should touch the raw
extreme missed and slipped to the close (pine_adapter.cpp market closes
6110-6115, market-entry Limit 4397/4402), coof_next_waypoint (3798) mis-read
a mid-bar fill as at-the-low, and an exit limit held for the next extreme
(6919) armed at the raw low but booked the rounded low so the kernel limit
check (native_execution_consumer.cpp:2880) rejected it. Fix (src/source
only): extreme-aimed orders trigger at the raw extreme with the booked price
unchanged; limits whose booked price is past the level use the kernel's
existing fill-through; an off-grid extreme counts as reached only when the
fill landed exactly on a bar point, tracked by a new coof_fill_cursor_t_
(live and hashed only during a recalculation); close-aimed orders keep the
rounded level. Restores joat NYSE-F 303->304 and joat NASDAQ-AAPL 348->350,
all trades identical to legacy; guards diffrows=0, ctest 492/492, hash OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

flush_pending_bracket_legs dropped a from_entry take-profit that the owner
(ab9714b) still books when a position re-enters on the same bar it exits,
under process_orders_on_close. Restores vasudevshenoy-manoj-betrayed-me
(BINANCE-BTCUSDT) 818->821, diffrows 1564->0, matching legacy. src/source
only, no new state; guards diffrows=0, ctest 492/492, hash OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…t add

At the bar open the owner (ab9714b pine_fills.cpp:3799-3845
samebar_add_exit_first) ranks buy-market(1) > sell-market(2) > gapped
limit(3), so a same-id market add is judged against the still-open book
(pyramiding cap + margin) before the gapped stop closes the leg. R4-D's
kernel breaks the same-point tie by smaller incarnation, so the older gapped
leg filled first and the add opened from flat. New
reaccept_gapped_bracket_behind_same_id_add (pine_adapter.cpp, from on_bar_open)
cancels and resubmits the gapped from_entry leg so the fresh incarnation puts
the add first; resubmitted limit legs are forced to the open print. Restores
colasbreugnon-good-luck (373->372, the Class-B extra 836-lot long removed)
and fixes masayanfx's two KI-62 sites. src/source only, no new state; Class A
(joat) and apkw82 guards diffrows=0, ctest 492/492, hash OK. masayanfx has a
separate remaining on-grid-limit tick-rounding cause handled elsewhere.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…ed exits

Class C of the target.not-positive gate blocker, two owner-cited causes in
src/source (kernel untouched):
1. Margin-call size / skipped trim (pridarasx): on_applied's priced-exit
   preemption check (short_preempted_by_priced_exit) and
   opening_slice_precedes_priced_exit_fill counted a stale Exit-Long bracket
   whose from_entry the reversal had already closed, so R4-D skipped the
   fill-price trim step and mis-sized the margin call (0.00016 vs 0.00004).
   Owner removes exits whose from_entry has not filled in the current cycle
   (ab9714b pine_fills.cpp:7671-7674, cycle_filled_entry_ids_). Added
   from_entry_filled_this_cycle(id) and applied it at both sites.
2. Reversal admission (shiroi): validate_precommit checked explicit-qty MARKET
   reversals against post-close marked_equity (includes the closing fee and
   the carried side's gap) and refused them; the owner costs a MARKET reversal
   against the placement-time equity snapshot, fees excluded (ab9714b
   pine_fills.cpp:5564-5663), admits it, then closes on the fill-price
   shortfall. Affordability branch now uses the owner's placement-equity rule.
Restores pridarasx BTC 869->824, pridarasx XAUUSD, shiroi ES1 952->966,
shiroi NQ1 568->577, all diffrows=0 vs legacy; guards + Class A (joat) 0,
ctest 492/492, hash OK, no new durable state.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…arity

Three src/source causes on winthetrade-ema-9-vwap (EURUSD), none Class B/C/vasudev
(present since the native adapter port 271032d):
- Skipped pre-exit margin slice: defer_rounded_pooc_short_margin_until_close
  counted a leftover strategy.exit whose from_entry never filled this cycle as a
  second live order and never scheduled the slice; owner removes that exit
  (ab9714b pine_fills.cpp:7672-7675) and books the slice before the trail fill
  (pine_fills.cpp:378-384,2150-2305). Count now skips unfilled-this-cycle exits.
- Extra 1-unit long money-margin call under POOC: schedule_tv_money_long_margin_before_trail
  fired whenever a trail was resting; owner fires only when the owned trail fills
  the same bar, at a path point before the fill (pine_fills.cpp:1780-1822,1935-1937).
  Now walks the trail over the bar waypoints from the open and schedules only if
  the trail fills after the fire point.
- MFE drift: margin_call_sample_bar falls back to tick-rounded waypoint match;
  zero-offset trails take the peak from the tick-snapped fill with no open-gap peak
  (owner engine_internal.hpp:187-198, engine_path_resolve.cpp:620-631).
Restores winthetrade EURUSD 1746->1762 diffrows=0; interakktive/itzkarmakyo were
never regressed (phantom diff from an unbuilt dylib). Guards + Class A (joat) 0,
ctest 492/492, hash OK, no new state.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

An exit limit whose level is exactly on the tick grid was never reached when
the bar's raw high/low stopped half a tick short, because R4-D sent on-grid
levels to the kernel raw while the owner tests triggers on the tick-quantized
bar (ab9714b pine_policy_members.cpp:11-17, engine.hpp:1374-1381: NYSE:F high
10.175 -> 10.18 fills a 10.18 sell limit). On-grid exit limits now also get the
half-tick source_trigger_threshold, except under calc_on_order_fills where the
owner compares the raw bar. Booked price unchanged (source_limit_fill still
books the level). Restores latibonit15 (408) and ygd (111) diffrows=0, and a
109-replay sweep changed only latibonit15/ygd/joat-concordance, all to 0.
No new state, no kernel edits; guards + Class A (joat) 0, ctest 492/492, hash OK.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…h_pending_bracket_legs

Sibling of the on-grid exit-limit fix (5fe4753), at a different site. When a
re-issued exit leg waits until after the bar's calculation,
flush_pending_bracket_legs tested the touch with policy_script_bar_.high >= level
against the RAW high. The owner tests priced exits on the entry bar against the
tick-quantized bar (ab9714b pine_fills.cpp:7711-7713 -> broker_trigger_bar,
pine_policy_members.cpp:11-17 -> broker_tick_bar, engine.hpp:1374-1381), so a
raw high half a tick short still fills (NYSE:F 13.455 -> 13.46 fills a 13.46
take-profit). The touch test now uses the tick-quantized high/low via
source_decimal_tick, staying raw under calc_on_order_fills or without a valid
tick (same exemption as 5fe4753). Restores vasudevshenoy-manoj-betrayed-me
NYSE-F 180->0 (426/426); BTC lane and guards diffrows=0, ctest 492/492, hash OK,
no new state, kernel untouched.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

… script

The 17th target excellent->strong. caldera's 35/35/30 qty_percent exit legs
split against the wrong base after a 2026-01-21 margin call. R4-D's
schedule_margin_call_path placed the carried-short margin order at the day's
high so it filled BEFORE the close-time script, so the script sized its exits
against 0.90596 (0.31708/0.31708/0.27178). The owner processes that margin call
AFTER the script (ab9714b pine_scheduler.cpp:258 script then :278
process_margin_call; the early pine_fills.cpp:1172-1196 check is ruled out by
commission/slippage :1189, and the before-priced-exit hook :2150 needs a resting
order, none on 01-21), so the script sizes against the full 1.0 (0.35/0.35/0.30),
matching TV. Fix: in the process-orders-on-close block a carried short skips the
early margin order when there is no calc-on-order-fills/stream, commission or
slippage is nonzero, and the book is empty apart from margin orders; the existing
post-script carried-short checkpoint then books the same slice at the same price
after the script. caldera now byte-identical to legacy (bar the internal
entryIncarnation id), sizes 0.35/0.35/0.20596; guards diffrows=0, ctest 492/492,
hash OK, no new state, kernel untouched.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…ervation, limit open-gap)

The last 3 target excellent->strong, all diffrows=0 under plain --window but
strong under the gate readback config (trim + chart_ema_na_warmup), each a
distinct src/source cause (kernel untouched):
- interakktive: a carried 100%-margin long's at-open money margin slice was
  flagged fill-only unconditionally, losing the bar's favorable excursion
  (MFE 0.000760 vs owner 0.002750). Owner exempts it and books at the open
  before sampling only when a priced exit is resting (ab9714b
  pine_fills.cpp:2544-2548,164-218,1350). Flag now applies only then.
- itzkarmakyo: a negative 'SL Long' sell stop (-0.00002 on the trimmed feed)
  was skipped by exit(), so the later partial take-profits filled; owner stores
  the negative stop and it reserves 100% of the position so the partials get no
  qty (ab9714b pine_strategy_commands.cpp:1929-1953). Negative sell stop now
  rests at 0.0 and keeps its reservation.
- forexver: an exit limit already marketable at the market parent's open gap
  booked at its level; owner closes both stop and limit legs at the open
  (ab9714b pine_fills.cpp:4136-4240,7795-7800); the earlier 6c1f431 fix covered
  only the stop leg. source_limit_fill now gets the same guarded open-gap rule.
All three now byte-identical to legacy under the gate config; winthetrade guard
identical; plain --window guards diffrows=0; ctest 492/492; hash OK; no new state.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT

…kyo pad parity

Recovers the two problems ship3's gate revealed, both from the gate3 change:
A) jenmb timed out on the Cloud Run afull ladder rung (full history, trading
   before window). Root cause: gate3 rests jenmb's negative sell stops at 0.0
   (correct per ab9714b pine_strategy_commands.cpp:1929-1953), so under
   calc_on_order_fills three exit legs are re-issued every bar and placement_ /
   the bracket-family lists grow ~1/reissue (~300k); three walks inside exit()
   made it quadratic (afull 97s). Made those O(1)/O(k): the consumed-leg family
   walk runs only when deciding to skip; the wrong-side-limit placement_ walk
   runs only for a wrong-side limit; the origin list is built from the origin
   roster index. afull 97s->9.5s, trades byte-identical. Also narrowed gate3's
   at-open guard to the owner-exact carried_long_money_precedes_priced_exit
   (pine_fills.cpp:164-218, exemption 2544-2548).
B) interakktive/itzkarmakyo strong under the gate's PAD window. interakktive
   #325: the owner narrows its 1x-long rounded-money-call tolerance to
   8*eps*money_scale + net_profit_roundoff_bound_ (pine_fills.cpp:1929-1965) and
   books the slice at the fire price, not the bar close (1985-2020); ported both.
   itzkarmakyo #303: a 6-decimal PnL tie decided by the buy-limit last bit;
   keep the owner's floor(r+eps)*tick (pine_policy_members.cpp:53-58).
interakktive/itzkarmakyo/winthetrade/forexver IDENTICAL to owner under pad AND
gate; jenmb/apkw82/joat/pridarasx --window diffrows=0; ctest 492/492; hash OK;
no new state; kernel untouched.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01CSmYixVAp4NqWpBZxkrLPT