plan(sp18 v2): 22 fold-reset registry entries + dispatch arms (PP.3)
Adds FoldReset registry coverage for all 22 SP18 ISV slots [483..505)
landed in PP.2 (10 D-leg + 12 B-leg), in lockstep with matching match
arms in `reset_named_state` per the existing
`every_fold_and_soft_reset_entry_has_dispatch_arm` contract test (which
catches the "add registry entry, forget dispatch arm → runtime panic
at fold boundary" pattern surfaced twice historically — SP5 #281, SP7
T7 commit 6e479c55c).
Sentinels match `sp14_isv_slots.rs` constants (single source of truth):
D-leg POS/NEG caps → 5.0 / -10.0 (SP14 P0-A REWARD_*_CAP_ADAPTIVE pattern)
D-leg HRC Welford → 0.0 (SENTINEL_WELFORD_ZERO)
D-leg DIAG/FIRE → 0.0 (Pearl-A direct-replace)
B-leg HEALTH_DIAG → 0.0 (Pearl-A)
B-leg POPART_RESET → 1.0 (one-shot per B-DD11 — see below)
B-leg TDB Welford → 0.0 (SENTINEL_WELFORD_ZERO)
B-leg RESERVED → 0.0
Slot 497 POPART_RESET_FLAG semantics: each fold start, FoldReset writes
1.0; on the first epoch of that fold the Phase 5 PopArt-reset consumer
reads 1.0, resets PopArt slot 63 EMA to identity normalization, and
writes 0.0 back — one-shot per fold. Cheap insurance against the 1+
epoch PopArt adaptation lag when switching `q_next` source from
rewards-distributed to Q-distributed in the Phase 5 atomic refactor.
New lock test `sp18_fold_reset_entries_present` asserts all 22 entries
exist with FoldReset category and a description containing "SP18 D-leg"
or "SP18 B-leg" prefix marker. Existing
`every_fold_and_soft_reset_entry_has_dispatch_arm` catches drift in
either direction (registry adds without dispatch arms, or dispatch arms
without registry entries).
Audit doc updated per Invariant 7 with PP.3 entry + full sentinel
mapping table.
Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
@@ -1892,6 +1892,134 @@ impl StateResetRegistry {
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[ADVANTAGE_CLIP_BOUND_INDEX=482] — SP17 adaptive |A_centered| upper bound, producer-tracked from `p99(|A_centered|) × ADVANTAGE_CLIP_SAFETY_FACTOR=1.5` (mirrors SP14 P0-A REWARD_POS_CAP producer pattern). FoldReset sentinel SENTINEL_ADVANTAGE_CLIP_BOUND=1.0 — Pearl-A bootstrap interpreted as cold-start `no effective clipping` (1.0 ≫ typical |A_centered| early in training when A → 0; clip is a no-op until the first valid observation lands a real bound). α=ADVANTAGE_CLIP_EMA_ALPHA=0.01 slow per-fold cadence — the bound shouldn't track per-batch noise. Bilateral clamp to `[ADVANTAGE_CLIP_BOUND_MIN=0.1, ADVANTAGE_CLIP_BOUND_MAX=100.0]` per `pearl_symmetric_clamp_audit.md` — Category-1 dimensional safety floors per `feedback_isv_for_adaptive_bounds`: floor 0.1 below which centered advantage is essentially noise (clipping would erase the action signal); ceiling 100.0 above which a single adversarial outlier dominates per-batch EMAs. NEVER hardcoded — adaptive bound. Producer + consumer wire in Phase 1-3 of SP17 (atomic) — not yet wired. Plan: docs/superpowers/plans/2026-05-08-sp17-dueling-q-network.md.",
|
||||
},
|
||||
// ── SP18 v2 Pre-Phase Task PP.3 (2026-05-08): combined Hold-attractor fix ──
|
||||
// 22 ISV slots [483..505) split across two atomic legs per the v2
|
||||
// combined plan. D-leg [483..493) — 10 slots driving the structural
|
||||
// Hold opportunity-cost reward chain (replaces SP13 P0a + SP16 P2 +
|
||||
// SP16 T3 reactive per-bar Hold-cost chain per DD7=c). B-leg
|
||||
// [493..505) — 12 slots driving the TD(λ) Q(s') bootstrap chain
|
||||
// (replaces self-bootstrap q_next = rewards at gpu_experience_
|
||||
// collector.rs:4143 with real per-branch DDQN target-Q via target-
|
||||
// net forward pass on next_states). Producer + consumer kernels
|
||||
// land in Phases 2-7 atomic per `feedback_no_partial_refactor`;
|
||||
// this Pre-Phase commit is additive infrastructure only. All
|
||||
// sentinels are Pearl-A first-observation bootstrap except slot
|
||||
// 497 POPART_RESET_FLAG (sentinel 1.0 — one-shot per B-DD11,
|
||||
// gates per-fold PopArt slot 63 EMA reset at SP18 deployment
|
||||
// boundary). Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md.
|
||||
RegistryEntry {
|
||||
name: "sp18_hold_reward_pos_cap",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HOLD_REWARD_POS_CAP_INDEX=483] — SP18 D-leg adaptive upper bound on the structural Hold opportunity-cost reward (Hold during a losing bar). Producer-tracked from `p99(|step_ret|)` over Long/Short trade closes × HOLD_REWARD_CAP_SAFETY_FACTOR=1.5 (DD3=b) by `hold_reward_cap_update_kernel` (lands in Phase 3) with Wiener-optimal α from Welford accumulators in [487..493) per `pearl_wiener_optimal_adaptive_alpha`. FoldReset sentinel SENTINEL_HOLD_REWARD_POS_CAP=5.0 — Pearl-A first-observation bootstrap matching the SP14 P0-A REWARD_POS_CAP_ADAPTIVE pattern for bit-identical cold-start before the cap producer's first valid launch. Consumer: `compute_sp18_hold_opportunity_cost` device function in `trade_physics.cuh` (added Phase 2) consumes pos_cap as the upper bound on the opportunity-cost reward returned to the 3 Hold-cost subtraction sites in `experience_kernels.cu`. Bounds [HOLD_REWARD_POS_CAP_MIN_BOUND=0.5, HOLD_REWARD_POS_CAP_MAX_BOUND=50.0] are Category-1 dimensional safety floors per `feedback_isv_for_adaptive_bounds`. Producer + consumer wire in Phase 2-3 of SP18 (atomic 3-site migration) — not yet wired in this Pre-Phase commit. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hold_reward_neg_cap",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HOLD_REWARD_NEG_CAP_INDEX=484] — SP18 D-leg adaptive lower bound on the structural Hold opportunity-cost reward (Hold during a winning bar — the missed-winner case). Producer writes NEG = -2 × POS (DD5=b mirrored 2:1 loss-aversion ratio matching SP14 P0-A REWARD_POS/NEG_CAP — preserves the Kahneman/Tversky prospect-theory asymmetry: missing a winner is the bigger pain, drives the policy toward engagement). FoldReset sentinel SENTINEL_HOLD_REWARD_NEG_CAP=-10.0 — Pearl-A first-observation bootstrap matching SP14 P0-A REWARD_NEG_CAP_ADAPTIVE for bit-identical cold-start. Same producer kernel as slot 483 (`hold_reward_cap_update_kernel`) writes both slots in one launch. Consumer: `compute_sp18_hold_opportunity_cost` device function applies the asymmetric clamp via `compute_asymmetric_capped_pnl(-2 × vol_normalized_log_return, neg_cap, pos_cap)`. Producer + consumer wire in Phase 2-3 of SP18 (atomic) — not yet wired. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hold_reward_decomp_diag",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HOLD_REWARD_DECOMP_DIAG_INDEX=485] — SP18 D-leg HEALTH_DIAG observability slot for the per-action-bin reward decomposition diagnostic introduced in Phase 0 (per-bar EMAs of r_micro / r_opp_cost / r_popart / fire_rate per Hold/Long/Short/Flat action bin, surfaced by the new `reward_decomp_diag_kernel.cu`). FoldReset sentinel SENTINEL_HOLD_REWARD_DIAG=0.0 — Pearl-A direct-replace on first valid observation. Diagnostic only — does NOT modify any consumer path. Phase 0 KILL CRITERION evaluator reads this slot to abort D-leg if the per-step Hold-cost magnitude is already disproportionately large pre-fix (the controller is winning). Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § Phase 0.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hold_opp_cost_fire_rate_ema",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HOLD_OPP_COST_FIRE_RATE_EMA_INDEX=486] — SP18 D-leg engagement canary: fraction of bars on which the new Hold opportunity-cost reward fires (non-zero) — should be ≈ Hold% itself once the structural reward is wired. FoldReset sentinel SENTINEL_HOLD_OPP_FIRE_RATE=0.0 — Pearl-A direct-replace on first valid observation. Sanity-checks the structural-reward producer is actually emitting non-zero on Hold bars (pre-Phase-3 sentinel-bootstrap pre-fires the per-bar zero, post-Phase-3 should track the realised Hold rate). Producer kernel embeds the fire-rate EMA update into the same launch that consumes the new opportunity-cost reward. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § Phase 3.",
|
||||
},
|
||||
// ── HRC_* Welford accumulators (D-leg, mirrors SP16 T3 HCS_* layout)
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_target_mean",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_TARGET_MEAN_INDEX=487] — SP18 D-leg Welford running mean of target_cap for the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; first valid observation initialises via Welford increment. Combined with the cold-start fallback (sample_count<3 → α=1.0 REPLACE), the first 3 epochs of each fold force REPLACE; from epoch 3 onward Wiener-optimal α emerges from Welford state per `pearl_wiener_optimal_adaptive_alpha`. Mirrors the SP16 T3 HCS_TARGET_MEAN_INDEX=462 pattern.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_target_m2",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_TARGET_M2_INDEX=488] — SP18 D-leg Welford ΣΔ² (sample-variance accumulator) for target_cap in the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; tracks variance of `target_cap` series across each fold. Used to compute `sample_var = M2 / (n - 1)` once `n ≥ 3`. Mirrors HCS_TARGET_M2_INDEX=463.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_diff_mean",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_DIFF_MEAN_INDEX=489] — SP18 D-leg Welford running mean of consecutive `target_cap` one-step diffs in the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; first diff observation lands when sample_count reaches 2. Mirrors HCS_DIFF_MEAN_INDEX=464.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_diff_m2",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_DIFF_M2_INDEX=490] — SP18 D-leg Welford ΣΔ² for the consecutive-diff series in the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; provides `diff_var` half of the Wiener formula α = diff_var / (diff_var + sample_var + ε). Mirrors HCS_DIFF_M2_INDEX=465.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_prev_target",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_PREV_TARGET_INDEX=491] — SP18 D-leg previous-epoch `target_cap` for the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; written every launch from the freshly-computed `target_cap` so the next launch's diff calculation has a basis. Mirrors HCS_PREV_TARGET_INDEX=466.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_hrc_sample_count",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[HRC_SAMPLE_COUNT_INDEX=492] — SP18 D-leg Welford observation counter for the `hold_reward_cap_update_kernel` producer. Sentinel 0.0; incremented every launch (excluding cold-start sentinel-guard early-return). The cold-start fallback (sample_count < WELFORD_MIN_SAMPLE_COUNT=3 → α = 1.0 REPLACE) keeps the producer responsive in epochs 1-2 of each fold while Welford state stabilises. Mirrors HCS_SAMPLE_COUNT_INDEX=467.",
|
||||
},
|
||||
// ── B-leg observability + reset gate slots [493..505) ─────────────
|
||||
RegistryEntry {
|
||||
name: "sp18_td_error_mag_ema",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TD_ERROR_MAG_EMA_INDEX=493] — SP18 B-leg HEALTH_DIAG observable: `avg(|TD-error|)` computed post-fix from the per-batch L1 norm of `td_errors` (already computed for PER priority recomputation at training time). FoldReset sentinel SENTINEL_TD_ERROR_MAG_EMA=0.0 — Pearl-A direct-replace on first valid observation. The B-DD9 ratio gate compares post-fix / pre-fix at HD0–HD1 (target [0.5, 5.0]) — ratio outside that range = bootstrap change blew up loss magnitude. Producer + consumer wire in Phase 0 (diagnostic emit) + Phase 8 (combined HEALTH_DIAG emit) of SP18 — not yet wired in this Pre-Phase commit. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § B-DD9 + B-DD12.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_q_next_target_p99",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[Q_NEXT_TARGET_P99_INDEX=494] — SP18 B-leg HEALTH_DIAG observable: p99 of the new target-Q bootstrap value (the per-branch DDQN sum-of-argmaxed target-Q across dir/mag/ord/urg branches that replaces `q_next = rewards`). FoldReset sentinel 0.0 — Pearl-A direct-replace. Bound check on the new path (catches blow-up if target net diverges). Producer + consumer wire in Phase 4-5 of SP18 — not yet wired. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § B-DD12.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_q_next_minus_reward_p99",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[Q_NEXT_MINUS_REWARD_P99_INDEX=495] — SP18 B-leg HEALTH_DIAG observable: p99 of `|q_next_target − rewards|`. FoldReset sentinel 0.0 — Pearl-A direct-replace. Sanity check that the new bootstrap is doing real work — should be O(1) post-fix (NOT O(0) which would mean q_next_target ≈ rewards, i.e. the fix is doing nothing; NOT O(100) which would mean it's blowing up). Pre-fix this slot would be ~0 because q_next *was* rewards. Producer + consumer wire in Phase 4-5 of SP18 — not yet wired. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § B-DD12.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_v_share_trend_diag",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[V_SHARE_TREND_DIAG_INDEX=496] — SP18 B-leg HEALTH_DIAG observable: slope of the SP17 V_SHARE EMA (slots [478..482)) across the last 4 epochs, computed as `(EMA[now] - EMA[now-4]) / 4`. FoldReset sentinel 0.0 — Pearl-A direct-replace. The B-leg synergy hypothesis (pearl_TBD: bootstrap-bias drives v_share elevation) predicts v_share monotone HD0→HD4 pre-fix; post-fix the slope should *flatten* or *reverse*. Slope < 0 OR slope ≤ 0.05 with v_share[HD3] > 0.7 = synergy hypothesis confirmed; otherwise B-leg KILL CRITERION fires. Producer + consumer wire in Phase 0 (diagnostic emit) of SP18 — not yet wired. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § Phase 0.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_popart_reset_flag",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[POPART_RESET_FLAG_INDEX=497] — SP18 B-leg per-fold PopArt EMA reset gate per B-DD11. FoldReset sentinel SENTINEL_POPART_RESET_FLAG=1.0 (NOT 0.0 — this is the only non-zero diagnostic sentinel in the 22-slot SP18 block). Each fold start, the FoldReset machinery writes 1.0 here; on the first epoch of that fold the PopArt-reset consumer kernel (Phase 5) reads 1.0, clamps PopArt slot 63 EMA back to identity normalization (μ=0, σ=1), and writes 0.0 back to slot 497 — making it one-shot per fold. Switching `q_next` source from rewards-distributed to Q-distributed (Phase 5 atomic refactor) shifts the input distribution to PopArt; PopArt's α-EMA will adapt but may take 1+ epochs. The reset is cheap insurance. Producer + consumer wire in Phase 5 of SP18 — not yet wired in this Pre-Phase commit. Plan: docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md § B-DD11.",
|
||||
},
|
||||
// ── TDB_* Welford accumulators (B-leg, mirrors HRC_* layout) ──────
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_target_mean",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_TARGET_MEAN_INDEX=498] — SP18 B-leg Welford running mean of q_next_target for the post-fix Wiener-α blend on the TD-error magnitude EMA. Sentinel 0.0; same Welford algorithm as the D-leg HRC counterpart at slot 487 but tracks q_next_target instead of target_cap. Combined with the cold-start fallback (sample_count<3 → α=1.0), the first 3 epochs of each fold force REPLACE; from epoch 3 onward Wiener-optimal α emerges. Producer wire in Phase 4 of SP18 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_target_m2",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_TARGET_M2_INDEX=499] — SP18 B-leg Welford ΣΔ² for q_next_target. Sentinel 0.0; mirrors HRC_TARGET_M2_INDEX=488 and HCS_TARGET_M2_INDEX=463 modulo the producer it serves. Producer wire in Phase 4 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_diff_mean",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_DIFF_MEAN_INDEX=500] — SP18 B-leg Welford running mean of consecutive q_next_target diffs. Sentinel 0.0; mirrors HRC_DIFF_MEAN_INDEX=489 and HCS_DIFF_MEAN_INDEX=464. Producer wire in Phase 4 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_diff_m2",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_DIFF_M2_INDEX=501] — SP18 B-leg Welford ΣΔ² for the consecutive-diff series. Sentinel 0.0; provides `diff_var` half of the Wiener formula. Mirrors HRC_DIFF_M2_INDEX=490. Producer wire in Phase 4 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_prev_target",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_PREV_TARGET_INDEX=502] — SP18 B-leg previous-epoch q_next_target. Sentinel 0.0; mirrors HRC_PREV_TARGET_INDEX=491. Producer wire in Phase 4 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_tdb_sample_count",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[TDB_SAMPLE_COUNT_INDEX=503] — SP18 B-leg Welford observation counter. Sentinel 0.0; cold-start fallback (sample_count < 3 → α = 1.0) ensures the first 3 epochs of each fold force REPLACE so the Welford-derived α takes over from epoch 3 onward. Mirrors HRC_SAMPLE_COUNT_INDEX=492 and HCS_SAMPLE_COUNT_INDEX=467. Producer wire in Phase 4 — not yet wired.",
|
||||
},
|
||||
RegistryEntry {
|
||||
name: "sp18_b_leg_reserved",
|
||||
category: ResetCategory::FoldReset,
|
||||
description: "ISV[SP18_B_LEG_RESERVED_INDEX=504] — SP18 B-leg RESERVED slot for follow-up if needed (e.g. distributional flavour of B-DD8 if 5-epoch smoke shows atom-distribution drift). Sentinel 0.0. No producer or consumer in the Pre-Phase scope — preserves the 12-slot B-leg block aligned and the layout fingerprint stable.",
|
||||
},
|
||||
];
|
||||
Self { entries }
|
||||
}
|
||||
@@ -2035,4 +2163,68 @@ mod tests {
|
||||
missing
|
||||
);
|
||||
}
|
||||
|
||||
/// Lock SP18 v2 Pre-Phase Task PP.3: 22 fold-reset entries land for the
|
||||
/// combined Hold-attractor fix slots [483..505). 10 D-leg + 12 B-leg.
|
||||
/// Asserts each by name is present with FoldReset category and a
|
||||
/// description containing "SP18 D-leg" or "SP18 B-leg" (sentinel /
|
||||
/// description content checked separately for clarity).
|
||||
#[test]
|
||||
fn sp18_fold_reset_entries_present() {
|
||||
let r = StateResetRegistry::new();
|
||||
let by_name: std::collections::HashMap<&str, &RegistryEntry> =
|
||||
r.all().iter().map(|e| (e.name, e)).collect();
|
||||
|
||||
// 22 expected SP18 entries — D-leg 10 + B-leg 12.
|
||||
let expected: &[(&str, &str)] = &[
|
||||
// D-leg [483..493) — 10 slots
|
||||
("sp18_hold_reward_pos_cap", "SP18 D-leg"),
|
||||
("sp18_hold_reward_neg_cap", "SP18 D-leg"),
|
||||
("sp18_hold_reward_decomp_diag", "SP18 D-leg"),
|
||||
("sp18_hold_opp_cost_fire_rate_ema", "SP18 D-leg"),
|
||||
("sp18_hrc_target_mean", "SP18 D-leg"),
|
||||
("sp18_hrc_target_m2", "SP18 D-leg"),
|
||||
("sp18_hrc_diff_mean", "SP18 D-leg"),
|
||||
("sp18_hrc_diff_m2", "SP18 D-leg"),
|
||||
("sp18_hrc_prev_target", "SP18 D-leg"),
|
||||
("sp18_hrc_sample_count", "SP18 D-leg"),
|
||||
// B-leg [493..505) — 12 slots
|
||||
("sp18_td_error_mag_ema", "SP18 B-leg"),
|
||||
("sp18_q_next_target_p99", "SP18 B-leg"),
|
||||
("sp18_q_next_minus_reward_p99", "SP18 B-leg"),
|
||||
("sp18_v_share_trend_diag", "SP18 B-leg"),
|
||||
("sp18_popart_reset_flag", "SP18 B-leg"),
|
||||
("sp18_tdb_target_mean", "SP18 B-leg"),
|
||||
("sp18_tdb_target_m2", "SP18 B-leg"),
|
||||
("sp18_tdb_diff_mean", "SP18 B-leg"),
|
||||
("sp18_tdb_diff_m2", "SP18 B-leg"),
|
||||
("sp18_tdb_prev_target", "SP18 B-leg"),
|
||||
("sp18_tdb_sample_count", "SP18 B-leg"),
|
||||
("sp18_b_leg_reserved", "SP18 B-leg"),
|
||||
];
|
||||
|
||||
assert_eq!(expected.len(), 22, "expected 22 SP18 entries (10 D-leg + 12 B-leg)");
|
||||
|
||||
for (name, leg_marker) in expected {
|
||||
let entry = by_name.get(name).unwrap_or_else(|| {
|
||||
panic!(
|
||||
"SP18 v2 PP.3: registry missing entry '{}'. Add a `RegistryEntry` to \
|
||||
`StateResetRegistry::new()` AND a dispatch arm in \
|
||||
`trainer/training_loop.rs::reset_named_state` (else \
|
||||
`every_fold_and_soft_reset_entry_has_dispatch_arm` fires too).",
|
||||
name
|
||||
)
|
||||
});
|
||||
assert_eq!(
|
||||
entry.category,
|
||||
ResetCategory::FoldReset,
|
||||
"SP18 entry '{}' must be FoldReset (got {:?})", name, entry.category
|
||||
);
|
||||
assert!(
|
||||
entry.description.contains(leg_marker),
|
||||
"SP18 entry '{}' description must contain '{}'; got: {}",
|
||||
name, leg_marker, entry.description
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -8903,6 +8903,270 @@ impl DQNTrainer {
|
||||
);
|
||||
}
|
||||
}
|
||||
// ── SP18 v2 Pre-Phase Task PP.3 (2026-05-08): combined Hold-attractor fix ──
|
||||
// 22 ISV slots [483..505) split across two atomic legs per the v2
|
||||
// combined plan. D-leg [483..493) — 10 slots driving the structural
|
||||
// Hold opportunity-cost reward chain. B-leg [493..505) — 12 slots
|
||||
// driving the TD(λ) Q(s') bootstrap chain. All FoldReset with
|
||||
// Pearl-A first-observation bootstrap sentinels except slot 497
|
||||
// POPART_RESET_FLAG (sentinel 1.0 — one-shot per B-DD11, gates
|
||||
// the per-fold PopArt slot 63 EMA reset at SP18 deployment
|
||||
// boundary; the Phase 5 reset consumer reads 1.0, performs
|
||||
// the PopArt reset, and writes 0.0 back, making it one-shot per
|
||||
// fold). Producer + consumer kernels land in Phases 2-7 atomic
|
||||
// per `feedback_no_partial_refactor`; this Pre-Phase commit is
|
||||
// additive infrastructure only. Plan:
|
||||
// docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md.
|
||||
"sp18_hold_reward_pos_cap" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HOLD_REWARD_POS_CAP_INDEX, SENTINEL_HOLD_REWARD_POS_CAP,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HOLD_REWARD_POS_CAP_INDEX,
|
||||
SENTINEL_HOLD_REWARD_POS_CAP,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hold_reward_neg_cap" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HOLD_REWARD_NEG_CAP_INDEX, SENTINEL_HOLD_REWARD_NEG_CAP,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HOLD_REWARD_NEG_CAP_INDEX,
|
||||
SENTINEL_HOLD_REWARD_NEG_CAP,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hold_reward_decomp_diag" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HOLD_REWARD_DECOMP_DIAG_INDEX, SENTINEL_HOLD_REWARD_DIAG,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HOLD_REWARD_DECOMP_DIAG_INDEX,
|
||||
SENTINEL_HOLD_REWARD_DIAG,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hold_opp_cost_fire_rate_ema" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HOLD_OPP_COST_FIRE_RATE_EMA_INDEX, SENTINEL_HOLD_OPP_FIRE_RATE,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HOLD_OPP_COST_FIRE_RATE_EMA_INDEX,
|
||||
SENTINEL_HOLD_OPP_FIRE_RATE,
|
||||
);
|
||||
}
|
||||
}
|
||||
// HRC_* Welford accumulators — D-leg, mirrors SP16 T3 HCS_* layout.
|
||||
// All 6 sentinels = 0.0 (SENTINEL_WELFORD_ZERO).
|
||||
"sp18_hrc_target_mean" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_TARGET_MEAN_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_TARGET_MEAN_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hrc_target_m2" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_TARGET_M2_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_TARGET_M2_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hrc_diff_mean" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_DIFF_MEAN_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_DIFF_MEAN_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hrc_diff_m2" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_DIFF_M2_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_DIFF_M2_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hrc_prev_target" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_PREV_TARGET_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_PREV_TARGET_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_hrc_sample_count" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
HRC_SAMPLE_COUNT_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
HRC_SAMPLE_COUNT_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
// ── B-leg observability + reset gate slots [493..505) ──
|
||||
"sp18_td_error_mag_ema" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TD_ERROR_MAG_EMA_INDEX, SENTINEL_TD_ERROR_MAG_EMA,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TD_ERROR_MAG_EMA_INDEX,
|
||||
SENTINEL_TD_ERROR_MAG_EMA,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_q_next_target_p99" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
Q_NEXT_TARGET_P99_INDEX, SENTINEL_Q_NEXT_TARGET_P99,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
Q_NEXT_TARGET_P99_INDEX,
|
||||
SENTINEL_Q_NEXT_TARGET_P99,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_q_next_minus_reward_p99" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
Q_NEXT_MINUS_REWARD_P99_INDEX, SENTINEL_Q_NEXT_MINUS_REWARD_P99,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
Q_NEXT_MINUS_REWARD_P99_INDEX,
|
||||
SENTINEL_Q_NEXT_MINUS_REWARD_P99,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_v_share_trend_diag" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
V_SHARE_TREND_DIAG_INDEX, SENTINEL_V_SHARE_TREND_DIAG,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
V_SHARE_TREND_DIAG_INDEX,
|
||||
SENTINEL_V_SHARE_TREND_DIAG,
|
||||
);
|
||||
}
|
||||
}
|
||||
// POPART_RESET_FLAG sentinel = 1.0 (one-shot per B-DD11). At
|
||||
// each fold start the sentinel is rewritten to 1.0; the Phase 5
|
||||
// PopArt-reset consumer kernel reads 1.0, resets PopArt slot 63
|
||||
// EMA to identity normalization, and writes 0.0 back, making
|
||||
// it one-shot per fold.
|
||||
"sp18_popart_reset_flag" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
POPART_RESET_FLAG_INDEX, SENTINEL_POPART_RESET_FLAG,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
POPART_RESET_FLAG_INDEX,
|
||||
SENTINEL_POPART_RESET_FLAG,
|
||||
);
|
||||
}
|
||||
}
|
||||
// TDB_* Welford accumulators — B-leg, mirrors HRC_* layout.
|
||||
// All 6 sentinels = 0.0 (SENTINEL_WELFORD_ZERO).
|
||||
"sp18_tdb_target_mean" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_TARGET_MEAN_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_TARGET_MEAN_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_tdb_target_m2" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_TARGET_M2_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_TARGET_M2_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_tdb_diff_mean" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_DIFF_MEAN_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_DIFF_MEAN_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_tdb_diff_m2" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_DIFF_M2_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_DIFF_M2_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_tdb_prev_target" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_PREV_TARGET_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_PREV_TARGET_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_tdb_sample_count" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::{
|
||||
TDB_SAMPLE_COUNT_INDEX, SENTINEL_WELFORD_ZERO,
|
||||
};
|
||||
fused.trainer().write_isv_signal_at(
|
||||
TDB_SAMPLE_COUNT_INDEX,
|
||||
SENTINEL_WELFORD_ZERO,
|
||||
);
|
||||
}
|
||||
}
|
||||
"sp18_b_leg_reserved" => {
|
||||
if let Some(ref fused) = self.fused_ctx {
|
||||
use crate::cuda_pipeline::sp14_isv_slots::SP18_B_LEG_RESERVED_INDEX;
|
||||
fused.trainer().write_isv_signal_at(
|
||||
SP18_B_LEG_RESERVED_INDEX,
|
||||
0.0,
|
||||
);
|
||||
}
|
||||
}
|
||||
// SP15 Phase 1.2 (2026-05-06): cost-net sharpe slots.
|
||||
// OFI_IMPACT_LAMBDA_INDEX=407 is an Invariant-1 anchor (NOT
|
||||
// a stateful EMA) — rewrite the constructor's value at fold
|
||||
|
||||
@@ -2,6 +2,36 @@
|
||||
|
||||
**Status:** Populated during Plan 1 Task 6 (A.5 orphan audit). Updated on every commit per Invariant 7.
|
||||
|
||||
## 2026-05-08 — SP18 v2 Pre-Phase Task PP.3: 22 fold-reset registry entries + dispatch arms
|
||||
|
||||
Pre-Phase scaffolding completing the FoldReset wire-up for the 22 SP18 ISV slots from PP.2:
|
||||
|
||||
- `crates/ml/src/trainers/dqn/state_reset_registry.rs` — 22 new `RegistryEntry` records (10 D-leg + 12 B-leg) with FoldReset category. Each entry's `description` includes the slot index, sentinel value, and the prefix marker "SP18 D-leg" or "SP18 B-leg" (asserted by the new `sp18_fold_reset_entries_present` lock test).
|
||||
- `crates/ml/src/trainers/dqn/trainer/training_loop.rs::reset_named_state` — 22 new match arms, each rewriting the slot to its sentinel via `fused.trainer().write_isv_signal_at(<INDEX>, <SENTINEL>)`. Without these arms the existing `every_fold_and_soft_reset_entry_has_dispatch_arm` contract test fires (catches the "add registry entry, forget dispatch arm → runtime panic at fold boundary" bug pattern from SP5 #281 + SP7 T7).
|
||||
|
||||
**Sentinel mapping** (matches `crates/ml/src/cuda_pipeline/sp14_isv_slots.rs` — single source of truth):
|
||||
|
||||
| Slot | Name | Sentinel const | Value |
|
||||
|---|---|---|---|
|
||||
| 483 | sp18_hold_reward_pos_cap | SENTINEL_HOLD_REWARD_POS_CAP | 5.0 |
|
||||
| 484 | sp18_hold_reward_neg_cap | SENTINEL_HOLD_REWARD_NEG_CAP | -10.0 |
|
||||
| 485 | sp18_hold_reward_decomp_diag | SENTINEL_HOLD_REWARD_DIAG | 0.0 |
|
||||
| 486 | sp18_hold_opp_cost_fire_rate_ema | SENTINEL_HOLD_OPP_FIRE_RATE | 0.0 |
|
||||
| 487..492 | sp18_hrc_* (6 Welford slots) | SENTINEL_WELFORD_ZERO | 0.0 |
|
||||
| 493 | sp18_td_error_mag_ema | SENTINEL_TD_ERROR_MAG_EMA | 0.0 |
|
||||
| 494 | sp18_q_next_target_p99 | SENTINEL_Q_NEXT_TARGET_P99 | 0.0 |
|
||||
| 495 | sp18_q_next_minus_reward_p99 | SENTINEL_Q_NEXT_MINUS_REWARD_P99 | 0.0 |
|
||||
| 496 | sp18_v_share_trend_diag | SENTINEL_V_SHARE_TREND_DIAG | 0.0 |
|
||||
| 497 | sp18_popart_reset_flag | SENTINEL_POPART_RESET_FLAG | **1.0** |
|
||||
| 498..503 | sp18_tdb_* (6 Welford slots) | SENTINEL_WELFORD_ZERO | 0.0 |
|
||||
| 504 | sp18_b_leg_reserved | (literal 0.0) | 0.0 |
|
||||
|
||||
**Slot 497 POPART_RESET_FLAG semantics** (B-DD11): Sentinel = 1.0 (the only non-zero diagnostic sentinel in the 22-slot block). At each fold start, FoldReset writes 1.0; on the first epoch of that fold the Phase 5 PopArt-reset consumer kernel reads 1.0, resets PopArt slot 63 EMA back to identity normalization (μ=0, σ=1), and writes 0.0 back to slot 497 — making the reset one-shot per fold. Switching `q_next` source from rewards-distributed to Q-distributed (Phase 5 atomic refactor) shifts the PopArt input distribution; the per-fold reset is cheap insurance against the 1+ epoch adaptation lag.
|
||||
|
||||
**Atomicity envelope**: Pre-Phase scaffolding only — registry + dispatch arms wired in lockstep per the `every_fold_and_soft_reset_entry_has_dispatch_arm` contract; no producer kernel, no HEALTH_DIAG emit, no consumer migration. Phase 0–7 lands the actual wiring.
|
||||
|
||||
**Plan**: `docs/superpowers/plans/2026-05-08-sp18-reward-shape-hold-attractor.md` § Pre-Phase Task PP.3.
|
||||
|
||||
## 2026-05-08 — SP18 v2 Pre-Phase Task PP.2: 22 ISV slots [483..505) D-leg + B-leg
|
||||
|
||||
Pre-Phase scaffolding for the combined SP18 v2 plan: allocates 22 ISV slots split across two atomic legs per `feedback_no_partial_refactor`.
|
||||
|
||||
Reference in New Issue
Block a user