fix(class-a-p0c): MIN_HOLD_TARGET → ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451] (adaptive)
Per Class A audit: MIN_HOLD_TARGET=30.0f hardcoded was creating a
deterministic gradient pushing trades toward 30-bar holds regardless of
edge expiry. User's trading frequency is between HFT-MFT and varies by
regime; a 30-bar fixed target kills MFT-frequency alpha when the
optimal hold for current data is shorter (or longer).
The producer slot ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451] already exists
from SP14 Layer C Phase C.4b (commit 3b71d2183) — Pearl-A-bootstrapped
Welford EMA of observed winning trade hold times. Wiring fix only.
Cold-start fallback: when slot still at sentinel (no winning trades
observed yet), use MIN_HOLD_TARGET=30.0f as safety floor. Once a
winning trade closes and the EMA bootstraps, the adaptive value
takes over.
Validity window: isv_hold_target > 0.0f && < 240.0f; outside window
falls back to min_hold_target param (= MIN_HOLD_TARGET=30).
Added #define AVG_WIN_HOLD_TIME_BARS_INDEX 451 to state_layout.cuh
(C-side mirror of sp14_isv_slots.rs:97).
Per feedback_isv_for_adaptive_bounds: every adaptive bound in ISV.
Fixes the third Class A P0 hardcoded constant.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
@@ -2032,11 +2032,16 @@ extern "C" __global__ void experience_env_step(
|
||||
float* __restrict__ popart_component_per_sample,
|
||||
/* SP12 v3 (2026-05-04): per-trade event-driven reward composition.
|
||||
*
|
||||
* `min_hold_target` (bars): patience requirement on voluntary trade
|
||||
* exits. Phase 1 default 30 bars (Invariant-1 numerical anchor in
|
||||
* state_layout.cuh::MIN_HOLD_TARGET). Voluntary exits with
|
||||
* `hold_time < min_hold_target` get a soft penalty subtracted from
|
||||
* `r_popart` via the deficit/(deficit + T) saturation formula.
|
||||
* `min_hold_target` (bars): cold-start fallback for the patience
|
||||
* requirement on voluntary trade exits. Class A P0-C (2026-05-08):
|
||||
* the actual target used inside the kernel is ISV-driven from
|
||||
* ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451] — EMA of winning-trade hold
|
||||
* times. This parameter is the fallback when ISV[451] is still at
|
||||
* sentinel (0.0 = no winning trades observed yet) or when
|
||||
* isv_signals_ptr is NULL. Default 30 bars (MIN_HOLD_TARGET in
|
||||
* state_layout.cuh). Voluntary exits with effective hold_time below
|
||||
* the target get a soft penalty subtracted from `r_popart` via the
|
||||
* deficit/(deficit + T) saturation formula.
|
||||
*
|
||||
* `min_hold_penalty_max`: max penalty magnitude when the deficit is
|
||||
* infinite (factor → 1). Phase 1 default 3.0 (60% of REWARD_POS_CAP).
|
||||
@@ -3111,7 +3116,7 @@ extern "C" __global__ void experience_env_step(
|
||||
r_trail = capped_pnl;
|
||||
} else {
|
||||
/* SP12 v3 Change 2: min-hold soft penalty for voluntary exits.
|
||||
* Encourages MFT-leaning trade duration via patience requirement.
|
||||
* Encourages patience-matched trade duration via a soft penalty.
|
||||
* Soft saturation formula: factor = deficit/(deficit + T) — bounded
|
||||
* [0,1], smooth transition (no cliff), monotone increasing in
|
||||
* deficit, monotone decreasing in T. Temperature anneals over
|
||||
@@ -3120,9 +3125,9 @@ extern "C" __global__ void experience_env_step(
|
||||
* enforces commitment to longer holds.
|
||||
*
|
||||
* Trail-fire exits exempted (the surrounding `else` already
|
||||
* guarantees `!trail_triggered`). When `min_hold_target <= 0`
|
||||
* (defensive guard against malformed configs) the entire block
|
||||
* collapses to no-op since `deficit = max(0, -hold) <= 0`.
|
||||
* guarantees `!trail_triggered`). When `effective_min_hold_target
|
||||
* <= 0` (defensive guard against malformed configs) the entire
|
||||
* block collapses to no-op since `deficit = max(0, -hold) <= 0`.
|
||||
*
|
||||
* Voluntary-exit only by construction: this lives in the
|
||||
* `else` arm of the `trail_triggered` branch, which itself is
|
||||
@@ -3138,12 +3143,32 @@ extern "C" __global__ void experience_env_step(
|
||||
* (oracle tests in `tests/sp12_reward_math_tests.rs`). The
|
||||
* device function returns the penalty MAGNITUDE (positive) and
|
||||
* subsumes the `hold_time >= target` early-exit; preserving the
|
||||
* outer `if (segment_hold_time < min_hold_target)` guard would
|
||||
* be redundant since the function already returns 0 in that
|
||||
* case and `capped_pnl - shaping_scale * 0 == capped_pnl`. */
|
||||
* outer `if (segment_hold_time < target)` guard would be
|
||||
* redundant since the function already returns 0 in that case
|
||||
* and `capped_pnl - shaping_scale * 0 == capped_pnl`.
|
||||
*
|
||||
* Class A P0-C (2026-05-08): target is ISV-driven from
|
||||
* ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451] — Pearl-A-bootstrapped
|
||||
* Welford EMA of observed winning trade hold times. Adapts to
|
||||
* whatever duration the model actually finds profitable in the
|
||||
* current regime rather than a hardcoded MFT-leaning constant.
|
||||
* Cold-start fallback: sentinel = 0.0 (no winning trades yet);
|
||||
* guarded by the `> 0 && < 240` window below. Outside that
|
||||
* window (cold-start or anomalous ISV) falls back to
|
||||
* `min_hold_target` (= MIN_HOLD_TARGET=30 from Rust config).
|
||||
* Per feedback_isv_for_adaptive_bounds. */
|
||||
const float isv_hold_target = (isv_signals_ptr != NULL)
|
||||
? isv_signals_ptr[AVG_WIN_HOLD_TIME_BARS_INDEX]
|
||||
: 0.0f;
|
||||
/* Validity window: > 0.0 excludes sentinel; < 240.0 caps at
|
||||
* ~1 hour of bars (sanity guard against ISV corruption). */
|
||||
const float effective_min_hold_target =
|
||||
(isv_hold_target > 0.0f && isv_hold_target < 240.0f)
|
||||
? isv_hold_target
|
||||
: min_hold_target; /* cold-start / NULL fallback */
|
||||
const float min_hold_penalty = compute_min_hold_penalty(
|
||||
segment_hold_time,
|
||||
min_hold_target,
|
||||
effective_min_hold_target,
|
||||
min_hold_temperature,
|
||||
min_hold_penalty_max
|
||||
);
|
||||
|
||||
@@ -299,10 +299,13 @@ pub struct ExperienceCollectorConfig {
|
||||
|
||||
/// SP12 v3 (2026-05-04): per-trade event-driven reward composition.
|
||||
///
|
||||
/// Patience target on voluntary trade exits (in bars). Voluntary exits
|
||||
/// with `hold_time < min_hold_target` get a soft penalty subtracted
|
||||
/// from `r_popart` via the deficit/(deficit + T) saturation formula.
|
||||
/// Phase 1 default = `MIN_HOLD_TARGET=30` (state_layout.cuh).
|
||||
/// Cold-start fallback for the patience target on voluntary trade exits
|
||||
/// (in bars). Class A P0-C (2026-05-08): the kernel uses
|
||||
/// `ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451]` (EMA of winning-trade hold
|
||||
/// times) as the effective target; this field is only used when ISV[451]
|
||||
/// is still at sentinel (0.0 = no winning trades observed yet) or when
|
||||
/// `isv_signals_ptr` is NULL. Default = `MIN_HOLD_TARGET=30`
|
||||
/// (state_layout.cuh cold-start fallback).
|
||||
/// Trail-fire forced exits exempted — see kernel docstring.
|
||||
pub min_hold_target: f32,
|
||||
/// SP12 v3: max penalty magnitude when min-hold deficit is infinite
|
||||
@@ -5645,12 +5648,16 @@ impl GpuExperienceCollector {
|
||||
// SP12 v3 (2026-05-04): per-trade event-driven reward
|
||||
// composition parameters. The kernel applies a soft
|
||||
// min-hold penalty on voluntary exits (deficit/(deficit+T)
|
||||
// saturation) using these three scalars; constants live
|
||||
// in state_layout.cuh (MIN_HOLD_TARGET, MIN_HOLD_PENALTY_MAX,
|
||||
// MIN_HOLD_TEMPERATURE_{START,END,DECAY}). The temperature
|
||||
// is recomputed in Rust per epoch via the annealing
|
||||
// schedule before this launch (see
|
||||
// `training_loop.rs::min_hold_temperature_for_epoch`).
|
||||
// saturation). Class A P0-C (2026-05-08): the effective
|
||||
// min-hold target inside the kernel is ISV-driven from
|
||||
// ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451]; `min_hold_target`
|
||||
// here is the cold-start fallback only (used when ISV[451]
|
||||
// is still at sentinel 0.0 or isv_signals_ptr is NULL).
|
||||
// Constants in state_layout.cuh (MIN_HOLD_TARGET,
|
||||
// MIN_HOLD_PENALTY_MAX, MIN_HOLD_TEMPERATURE_{START,END,
|
||||
// DECAY}). Temperature recomputed per epoch via the
|
||||
// annealing schedule in `training_loop.rs::
|
||||
// min_hold_temperature_for_epoch`.
|
||||
.arg(&config.min_hold_target)
|
||||
.arg(&config.min_hold_penalty_max)
|
||||
.arg(&config.min_hold_temperature)
|
||||
|
||||
@@ -168,6 +168,17 @@
|
||||
#define ISV_SEED_FRAC_EMA_IDX 84 // == SEED_FRAC_EMA_INDEX — adaptive EMA of (1 - done/target) ∈ [0, 1] (Plan 3 Task 8 B.3; consumed by Task 9 CQL ramp)
|
||||
#define ISV_EVAL_THOMPSON_TEMP_IDX 339 // == EVAL_THOMPSON_TEMP_INDEX — eval Thompson selector temperature (SP10 / Fix 38 2026-05-03; ISV-driven temperature blend on direction-branch Thompson sample)
|
||||
|
||||
// ────────────────────────────────────────────────────────────────────────────
|
||||
// === SP14-C.4b SLOT INDICES === (2026-05-05, aux trunk adaptive horizon)
|
||||
// Mirror of crates/ml/src/cuda_pipeline/sp14_isv_slots.rs slot 451.
|
||||
// Pearl-A-bootstrapped Welford EMA of observed winning trade hold times (bars).
|
||||
// Sentinel = 0.0 (SENTINEL_AVG_WIN_HOLD_TIME_BARS). First winning trade close
|
||||
// replaces sentinel directly (first-observation bootstrap). Consumer here
|
||||
// (Class A P0-C): replaces MIN_HOLD_TARGET hardcoded constant in
|
||||
// `compute_min_hold_penalty` call site — see experience_kernels.cu.
|
||||
// ────────────────────────────────────────────────────────────────────────────
|
||||
#define AVG_WIN_HOLD_TIME_BARS_INDEX 451 // == AVG_WIN_HOLD_TIME_BARS_INDEX — EMA of winning-trade hold duration (bars); Pearl-A bootstrap sentinel 0.0
|
||||
|
||||
// ────────────────────────────────────────────────────────────────────────────
|
||||
// === SP13 SLOT INDICES === (2026-05-04, redefine success for predictive skill)
|
||||
// Mirror of crates/ml/src/cuda_pipeline/sp13_isv_slots.rs. Slots [372..383)
|
||||
@@ -229,10 +240,13 @@
|
||||
// experience_kernels.cu line ~2758 (segment_complete vol-normalized P&L).
|
||||
// Per pearl_audit_unboundedness_for_implicit_asymmetry.
|
||||
//
|
||||
// MIN_HOLD_TARGET — patience requirement on voluntary trade exits, in bars.
|
||||
// 30 bars = MFT-leaning but not pure MFT (full MFT ≈ 100 bars,
|
||||
// HFT-tolerant ≈ 10 bars). Matches the user-profile target band
|
||||
// "between HFT and MFT".
|
||||
// MIN_HOLD_TARGET — cold-start fallback for the patience requirement on
|
||||
// voluntary trade exits, in bars. Used when ISV[AVG_WIN_HOLD_TIME_BARS_
|
||||
// INDEX=451] is still at sentinel (0.0 = no winning trades observed yet).
|
||||
// Once the first winning trade closes and the EMA bootstraps, the
|
||||
// adaptive ISV value takes over in experience_kernels.cu. Value 30 bars
|
||||
// = MFT-leaning fallback ("between HFT and MFT" user-profile target band).
|
||||
// Class A P0-C (2026-05-08): per feedback_isv_for_adaptive_bounds.
|
||||
//
|
||||
// MIN_HOLD_PENALTY_MAX — max penalty magnitude when voluntary exit at hold=0.
|
||||
// 3.0 = 60% of REWARD_POS_CAP. Meaningful gradient pressure but not
|
||||
|
||||
@@ -7770,3 +7770,43 @@ EMA blend logic directly in the kernel body and writes to ISV in-place. Fixed
|
||||
- C.7-C.8: ISV-driven aux trunk Adam β1/β2/ε/LR/grad-clip.
|
||||
- C.9: synthetic-data smoke + audit close-out + memory pearls.
|
||||
- C.10: L40S 30-epoch validation.
|
||||
|
||||
---
|
||||
|
||||
## 2026-05-08 — Class A P0-C: MIN_HOLD_TARGET ISV-driven from AVG_WIN_HOLD_TIME_BARS_INDEX[451]
|
||||
|
||||
### Problem
|
||||
|
||||
`MIN_HOLD_TARGET=30.0f` (state_layout.cuh:254) was a hardcoded MFT-leaning constant applied to every voluntary exit's min-hold soft penalty. Creates a deterministic gradient pushing all trades toward 30-bar holds regardless of actual edge expiry or current regime. A profitable 12-bar scalp can be net-negative after penalty — kills MFT-frequency alpha. User's trading frequency is between HFT-MFT and varies by market regime.
|
||||
|
||||
### Fix
|
||||
|
||||
Wired `ISV[AVG_WIN_HOLD_TIME_BARS_INDEX=451]` — Pearl-A-bootstrapped Welford EMA of observed winning trade hold times — as the effective target in `compute_min_hold_penalty`. Added `#define AVG_WIN_HOLD_TIME_BARS_INDEX 451` to state_layout.cuh (C-side mirror of sp14_isv_slots.rs:97). The Rust `min_hold_target` parameter (default 30.0) becomes cold-start fallback only.
|
||||
|
||||
### Cold-start behavior
|
||||
|
||||
- Sentinel = `0.0f` (SENTINEL_AVG_WIN_HOLD_TIME_BARS in sp14_isv_slots.rs).
|
||||
- Fallback condition: `isv_hold_target > 0.0f && isv_hold_target < 240.0f`. Outside window (or NULL ptr): uses `min_hold_target` = 30.0 from config.
|
||||
- First winning trade close: Pearl-A bootstrap replaces sentinel with observation directly; adaptive path takes over immediately.
|
||||
- Upper bound 240 bars = sanity cap (>= ~1hr of bars, protects against ISV corruption).
|
||||
|
||||
### Sites modified
|
||||
|
||||
| File | Lines | Change |
|
||||
|------|-------|--------|
|
||||
| `state_layout.cuh` | ~169 (add) | `#define AVG_WIN_HOLD_TIME_BARS_INDEX 451` block |
|
||||
| `state_layout.cuh` | ~232-239 | MIN_HOLD_TARGET comment → cold-start fallback role |
|
||||
| `experience_kernels.cu` | ~2034-2044 | `min_hold_target` param doc → cold-start fallback |
|
||||
| `experience_kernels.cu` | ~3113-3165 | ISV read + fallback + pass `effective_min_hold_target` to `compute_min_hold_penalty` |
|
||||
| `gpu_experience_collector.rs` | ~300-310 | struct field doc update |
|
||||
| `gpu_experience_collector.rs` | ~5645-5658 | launch-site comment update |
|
||||
| `docs/dqn-wire-up-audit.md` | here | this entry |
|
||||
|
||||
### Verification
|
||||
|
||||
- `SQLX_OFFLINE=true CUDA_COMPUTE_CAP=86 cargo check -p ml --lib` — clean (19 pre-existing warnings, 0 new).
|
||||
- `SQLX_OFFLINE=true CUDA_COMPUTE_CAP=86 cargo test -p ml --test sp14_oracle_tests --release -- --ignored --nocapture` — 4/4 pass.
|
||||
|
||||
### Consumer audit
|
||||
|
||||
Only one consumer of `MIN_HOLD_TARGET` was found in the CUDA call path (the `compute_min_hold_penalty` call in `experience_kernels.cu`). No additional consumers discovered. `trade_physics.cuh` takes `min_hold_target` as a parameter — no constant reads — no change needed there. Per `feedback_no_partial_refactor`: complete.
|
||||
|
||||
Reference in New Issue
Block a user