84de278dfee5b1d12f480a2eaabc42e9e6213c19
980 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
jgrusewski
|
84de278dfe |
feat(sp14): B.2 — register 11 SP14 ISV slots for fold-boundary reset
Each EGF pearl EMA / state slot resets to its Pearl-A sentinel at fold
boundary, mirroring sp13_aux_dir_acc_short_ema / long_ema entries.
Atomic refactor (feedback_no_partial_refactor): both halves land
together — registry entry + reset_named_state dispatch arm.
Reset slots (11 total, sentinel in parens):
- Q_DISAGREEMENT_SHORT/LONG_EMA (slots 383, 384) → 0.5
- K_AUX_ADAPTIVE (385) → K_BASE_AUX = 20.0
- K_Q_ADAPTIVE (386) → K_BASE_Q = 15.0
- BETA_RATE_LIMITER_ADAPTIVE (387) → BETA_BASE = 0.5
- AUX_DIR_ACC_VARIANCE_EMA, Q_DISAGREEMENT_VARIANCE_EMA,
ALPHA_GRAD_RAW_VARIANCE_EMA (388, 389, 390) → 0.0
(initial k = k_base, β = β_base via ISV-driven controllers)
- GATE1_OPEN_STATE (391) → 0.0 (closed)
- ALPHA_GRAD_SMOOTHED (393) → 0.0
- AUX_DIR_ACC_POST_OPEN_MIN (394) → 1.0 (no min observed)
ALPHA_GRAD_RAW (slot 392, recomputed every step from variance EMAs)
and GRADIENT_HACK_LOCKOUT_REMAINING (slot 395, decays at epoch
boundary) are NOT in the fold-reset registry; both naturally
re-initialise without explicit reset.
Also corrects the isv-slots.md SP14 table: slots 392 and 395 were
incorrectly marked FoldReset in the B.1 entry; corrected to reflect
their actual reset semantics (NOT reset / epoch-boundary decay).
Producer + consumer wiring lands in subsequent tasks (B.3-B.12);
this commit is additive infrastructure only — no behavior change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
d63cb7992e |
feat(sp14): B.1 — sp14_isv_slots.rs with 13 new ISV slot constants
Allocates ISV slots [383..396) for the Aux→Q Wire + Earned Gradient Flow pearl (Layer B of SP14). Mirrors sp13_isv_slots.rs pattern. The plan originally documented [381..394), but Phase 0 verification found SP13 closeout added HOLD_RATE_TARGET_INDEX=381 and HOLD_RATE_OBSERVED_EMA_INDEX=382 after the plan was written, so the range shifts by +2. Slots fall into 4 functional groups: - Q-disagreement EMAs (short, long; K=4↔K=2 mapping with Hold/Flat masked) - Adaptive controllers (k_aux, k_q, β; variance-driven) - Welford variance EMAs (3, one per adaptive scalar) - Schmitt state + α_grad outputs + circuit breaker Plus 14 structural constants for numerical-stability anchors: K_BASE_*, K_MIN, VARIANCE_REF_*, BETA_BASE, BETA_MAX, SCHMITT_BAND, WARMUP_STEPS_FALLBACK, LOCKOUT_*, Q_DISAGREEMENT_BASELINE. Per feedback_isv_for_adaptive_bounds: adaptive bounds (k_*, β, post_open_min, lockout) live in ISV; numerical anchors live as structural constants. Per pearl_first_observation_bootstrap: all EMAs reset to sentinels and Pearl-A bootstraps on first observation. Producer + consumer wiring lands in subsequent tasks (B.2-B.12); this commit is additive infrastructure only — no behavior change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
f75786fc5a |
fix(sp14): A.1 — C51 inv_a_std floor lift (1e-6 → 1e-3)
c51_grad_kernel.cu line 275: lift floor from 1e-6 to 1e-3 in \`inv_a_std = 1.0f / (a_std + 1e-3f)\`, capping the magnitude-branch gradient amplifier at 1000 instead of ~1e6 in the degenerate case. Why: Smoke A produced 1109 GRAD_CLIP_OUTLIER events with C51 grad reaching 9.5e6 — the SP7 budget controller saturated at the EPS_DIV floor instead of rebalancing proportionally. Phase-0 verification against the actual kernel found the amplifier is NOT the spec's claimed −log(p)/p divide (which does not exist; the kernel uses the CE-stable expf(lp) - proj form at line 81). The actual amplifier is inv_a_std = 1/(a_std + 1e-6) at line 275, gated by \`if (d == 1) grad_val *= inv_a_std\` at line 282. When magnitude advantage logits collapse near-uniform (Smoke A: var_q=9e-10), a_std → ~1e-9, so inv_a_std → ~1e6. Per feedback_isv_for_adaptive_bounds, this is a numerical-stability anchor (Invariant 1: prevent division-by-near-zero amplification), not a behavioural bound. ISV-driven bounds govern behavior; the existing 1e-12f floor on a_std at line 274 is also a structural anchor — same class of fix. Validation gate: Smoke A2-A GRAD_CLIP_OUTLIER count <100 in fold 2 (was 1109 pre-fix). The 3-order-of-magnitude reduction in worst- case amplification should bring C51 grad spikes back under SP7 budget controller authority. Audit doc: Fix 40 added (parallel to Fix 39 for A.2 and Fix 41 for A.3); the stale "A.1 deferred" note was removed in the A.3 commit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
1420383212 |
fix(sp14): A.3 — stagnation warmup gate at fold boundary
compute_aux_w_p0b's stagnation term was firing inappropriately at fold reset because Pearl-A first-observation bootstrap forces both EMAs equal: short_ema = sentinel 0.5 → first observation X long_ema = sentinel 0.5 → first observation X (same update) improvement = max(0, short - long) = 0 stagnation = (1 - 0/max(deficit, 0.005)) = 1.0 aux_w *= (1 - 0.7) = 0.3 → spurious decay on a non-stagnation Fix: add epochs_in_fold parameter; skip stagnation when < 1. Wait one full epoch for the α=0.3 vs α=0.05 short/long EMA timescale split to produce real improvement signal. Atomic refactor (feedback_no_partial_refactor): all 5 callers migrated — 1 production site (training_loop.rs:4257) plus 4 existing unit tests at trainers/dqn/trainer/tests.rs. Test: aux_w_stagnation_warmup_gate_epoch_0 verifies: - Epoch 0: stagnation = 0, aux_w = base × deficit_amp = 0.625 - Epoch 1: stagnation = 1.0, aux_w = floor 0.15 Combined with A.2 (clamp lift), Fold 2's aux_w should now hit 0.625 in epoch 0 (the controller's intended post-deficit-amp value) instead of being collapsed to 0.164. Audit doc: Fix 41 added; stale "A.1 deferred" note from a prior killed agent was also removed (A.1 lands in the next commit, not deferred per user direction). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
731cae4c80 |
fix(sp14): A.2 — lift set_aux_weight clamp to SP13 P0b range [0.15, 1.5]
The pre-fix clamp at gpu_dqn_trainer.rs:14722 was [0.05, 0.3] — the SP11-era cap. P0b's controller computes aux_w in [0.15, 1.5] (base 0.5 × [0.3, 3.0]) but the setter silently chopped everything above 0.3, masking the deficit-amplification term entirely. Smoke A trace confirmed: Fold 0/1: raw aux_w = 0.66-0.80 → clamped to 0.30 (deficit invisible) Fold 2: raw aux_w = 0.164 (stagnation; below clamp) → 45% deficit Post-fix: deficit-amp term `(1 + 5 × deficit)` actually expresses through to the trainer. Fold 2 stagnation will get the designed floor 0.15 instead of being capped at 0.3 — but the upper range 1.5 also opens, so deficit-amp can pull aux_w up when accuracy is below target. Constants imported from sp13_isv_slots.rs (AUX_W_BASE=0.5, AUX_W_HARD_FLOOR_RATIO=0.3, AUX_W_HARD_CEIL_RATIO=3.0). No new slots; existing constants exposed as the clamp bounds. Test: set_aux_weight_clamp_range verifies constants resolve correctly. A.1 (C51 atom-probability floor) deferred — Phase 0 verification found the spec's stated `−log(p)/p · ∂p/∂z` divide does NOT exist in c51_grad_kernel.cu at HEAD 037c24116; actual kernel uses the numerically-stable `expf(lp) - proj`. The 1109 GRAD_CLIP_OUTLIER events in Smoke A are real but their mechanism is different. Will be re-spec'd as a separate task post-SP14. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
037c24116d |
plan(sp14): implementation plan for spec v3
Two-sub-project plan: - Layer A (4 tasks, ~50 LOC): C51 atom-floor, set_aux_weight clamp lift, stagnation warmup gate, smoke validation - Layer B (15 tasks, ~1180 LOC): 13 ISV slots, 4 kernels (q_disagreement, alpha_grad, gradient_hack_detect, dir_concat_qaux), forward wire, backward gradient gating, orchestrator wire-up, HEALTH_DIAG, tests, smoke Each task has bite-sized TDD steps (write failing test, run fail, implement, run pass, commit) per the writing-plans skill conventions. Phase 0 verification tasks (A.0, B.0) anchor against current code at HEAD 40e737a18+ before edits. 19 total tasks across 2 layers. Each layer commits independently as an atomic feature; intermediate per-task commits during Layer B keep the wire safety-protected (forward concat lands before backward gating in B.9 → B.10 sequence). 8 explicit kill criteria for Smoke A2-B per spec B.7. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
40e737a181 |
docs(sp14): spec v3 — K=4 direction + ISV-adaptive β rate limiter
Two further user-flagged corrections from second-critical review:
1. Direction Q-head emits K=4 actions, NOT K=3.
Authoritative source: state_layout.cuh:123-126
#define DIR_SHORT 0 // open/maintain short
#define DIR_HOLD 1 // keep current position (no-op)
#define DIR_LONG 2 // open/maintain long
#define DIR_FLAT 3 // close all to zero
The "default: 3 — Short/Flat/Long" comment at gpu_dqn_trainer.rs:2438
is stale pre-SP13. Production callsites all set branch_0_size: 4.
SP13 added DIR_HOLD as a separate fourth direction action (Hold-pricing).
The config default comment was never updated when DIR_HOLD landed.
This is exactly the feedback_trust_code_not_docs failure mode — a
single stale comment would have silently corrupted the q_disagreement
signal (Hold/Flat being indices 1/3 instead of just Flat=1).
Updates:
- B.1 action-space context: 4 actions with Hold AND Flat both
non-committal (Hold = keep position, Flat = exit to zero)
- B.2.3 q_disagreement mapping: K=4↔K=2 with both Hold and Flat
masked from disagreement signal (no new directional commitment
to evaluate); only Short and Long picks contribute to disagreement
- Edge case handling for all-Hold/all-Flat batches
2. Adaptive β rate limiter (was structural β=0.9).
Per feedback_isv_for_adaptive_bounds, β should be signal-driven
not hardcoded. v3 derives β from variance of α_grad_raw, mirroring
the k_aux/k_q variance-driven steepness pattern in B.2.5.
Formula:
β = clip(β_base + variance_alpha_raw / variance_ref_alpha,
[β_base, β_max])
β_base = 0.5 (light smoothing baseline; ~2-step half-life)
β_max = 0.95 (heavy smoothing; ~20-step half-life)
Stable α_grad_raw → β = β_base (preserves directional intent)
Volatile α_grad_raw → β → β_max (dampens jitter)
Adds 2 ISV slots:
ALPHA_GRAD_RAW_VARIANCE_EMA_INDEX (Welford variance)
BETA_RATE_LIMITER_ADAPTIVE_INDEX (current β value)
ISV slot count: 11 → 13 (net +2 for variance + adaptive β).
LOC estimate: ~1150 → ~1180 (negligible delta; 3 Welford
variances now in alpha_grad_compute_kernel instead of 2).
HEALTH_DIAG pearl_egf_diag emit updated to expose all three
adaptive scalars (α, β, k_aux/k_q) plus all three driving
variances (var_alpha, var_aux, var_q) for full observability.
Verified against current code at HEAD
|
||
|
jgrusewski
|
d243a6f080 |
docs(sp14): spec v2 — critical review corrections
8 corrections from code-anchored critical review at HEAD
|
||
|
jgrusewski
|
eaf4adcb98 |
docs(sp14): spec — aux→Q wire + Earned Gradient Flow pearl
Designs the SP14 chain on top of SP13 Layer B (HEAD
|
||
|
jgrusewski
|
6657e56265 |
feat(sp13): B1.1b — producer kernel + replay direct path + experience collector
Final piece of the SP13 Layer B chain. B1.1a flipped the aux head from K=1
MSE regression to K=2 softmax CE classification but aux_nb_label_buf was
zero-init — model was training on "all bars are class 0 (down)". B1.1b
lands the producer kernel that fills i32 -1/0/1 labels from the 30-bar
price trajectory, the replay direct-path 8th gather that carries those
labels into the trainer, and the experience collector hoist that ensures
bar_indices_pinned is always populated (producer + hindsight relabel both
consume it). Aux head finally trains on real classification signal.
Recovery commit: this completes a B1.1b agent dispatch that crashed
mid-edit. The implementer had landed ~95% of the cascade (kernel file,
build.rs, replay buffer signature + direct path, fused_training getter,
trainer accessor, both training_loop.rs callers, kernel field + cubin
loader on the experience collector) before being killed. The missing
pieces (experience collector launch + bar_indices_pinned hoist + 6
producer tests + audit doc) were completed manually post-crash and
verified end-to-end.
Three contracts (atomic single commit per feedback_no_partial_refactor):
1. NEW aux_sign_label_kernel.cu producer — pure per-thread O(1) map
reading targets[bar*6+2] (raw_close column) at bar and bar+lookahead,
writing -1 (skip if bar+lookahead >= total_bars), 0 (down/flat under
strict greater-than tie-break), or 1 (up). Replaces B0 alloc_zeros.
2. Replay direct-path 8th gather — set_trainer_buffers gains 8th arg
trainer_aux_sign_labels_ptr; direct branch in sample_proportional
adds gather_i32_scalar into the trainer ptr; fallback gather wrapped
in if !direct_to_trainer (avoids wasted DtoD). Direct-mode
GpuBatchPtrs return points aux_sign_labels_ptr at trainer ptr.
3. Experience collector bar_indices_pinned cpu-fill hoist — moved out
of if hindsight_fraction > 0.0 so producer + hindsight share it.
Files (9 total):
- crates/ml/src/cuda_pipeline/aux_sign_label_kernel.cu (NEW)
- crates/ml/build.rs (cubin registration)
- crates/ml/src/cuda_pipeline/gpu_experience_collector.rs (kernel
field + cubin loader + struct init + hoist + producer launch)
- crates/ml-dqn/src/gpu_replay_buffer.rs (8th arg + direct gather +
fallback skip + GpuBatchPtrs return)
- crates/ml/src/cuda_pipeline/gpu_dqn_trainer.rs (aux_nb_label_buf_ptr
accessor mirrors 6 existing trainer-buf accessors)
- crates/ml/src/trainers/dqn/fused_training.rs
(trainer_aux_sign_labels_buf_ptr getter)
- crates/ml/src/trainers/dqn/trainer/training_loop.rs (both
set_trainer_buffers callers updated)
- crates/ml/tests/sp13_layer_b_oracle_tests.rs (6 NEW producer tests)
- docs/dqn-wire-up-audit.md (B1.1b section)
Hard rules upheld:
- feedback_no_partial_refactor: every consumer of the 3 contracts
migrates atomically
- feedback_no_atomicadd: producer is pure map; no reductions
- feedback_cpu_is_read_only: producer GPU-only; only host work is
pre-existing bar_indices_pinned cpu-fill (hoisted unchanged)
- feedback_no_stubs: kernel output flows through real chain — ring
buffer → direct gather → aux_nb_label_buf → CE consumer
- feedback_no_legacy_aliases: 8-arg setter gets
#[allow(clippy::too_many_arguments)] not an alias shim
- feedback_no_htod_htoh_only_mapped_pinned: targets_buf and
bar_indices_pinned both pre-existing mapped-pinned
Build + test:
- cargo check --workspace clean (only pre-existing warnings)
- cargo check --workspace --tests clean
- 17 tests in sp13_layer_b_oracle_tests.rs:
- 2 CPU-only (fingerprint bump + HEALTH_DIAG snap stable) pass
- 15 GPU on RTX 3050 Ti pass (9 B1.1a + 6 new B1.1b producer):
aux_sign_label_monotone_up_all_ones
aux_sign_label_monotone_down_all_zeros
aux_sign_label_flat_all_zeros_strict_gt
aux_sign_label_last_30_bars_skip
aux_sign_label_boundary_first_valid_last_skip
aux_sign_label_multi_episode_per_episode_skip
Producer tests cover every edge case in the kernel:
- Monotone trajectories (label=1 / label=0 across all valid bars)
- Flat tie-break (strict greater-than means flat → 0)
- Skip sentinel for last lookahead bars
- First/last bar boundary (bar=0 valid, bar=L-1 skip)
- Multi-episode global skip semantics
Next: Smoke A — L40S 5-epoch validation of full SP13 stack
(P0a + P0b + B0 + B0.1 + B1.0 + B1.1a + B1.1b). Expected: aux head
trains on real K=2 softmax CE labels; aux_dir_acc_short_ema rises above
0.5 within first epoch (vs B1.1a degraded baseline at 0.5);
HEALTH_DIAG aux_b1_diag emits per-epoch with n_down/n_up/n_skip/mask_frac.
If aux_dir_acc_short_ema > 0.55 by epoch 5, B1.1b is validated and the
chain merges to main.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
7d10ea8b3e |
feat(sp13): B1.1a — K=1→2 + softmax CE kernel rewrites + struct flips
Flips the aux next-bar head from K=1 MSE regression to K=2 softmax
cross-entropy classification. Kernel ABIs, struct fields, partial-buf
shapes, and per-step ISV producers all migrate atomically; the producer
that fills `aux_sign_labels` with real -1/0/1 from the price trajectory
lands separately in B1.1b.
Why split B1.1a from B1.1b: the original B1 brief was decomposed (B1.0
+ B1.1) after six full-B1 dispatches confirmed agent-session capacity
is the bottleneck, not cascade understanding. B1.1 itself is now further
split into B1.1a (kernel/struct/test cascade — this commit) and B1.1b
(producer kernel + replay direct path + experience collector hoist +
remaining tests + Smoke A). B1.1a is contract-consistent atomic
kernel-side; B1.1b lands the producer + smoke.
Why labels stay zero-init: B1.1a is producer-less by design. The
`aux_nb_label_buf: CudaSlice<i32>` is `alloc_zeros<i32>` so every
sample receives label 0 ("down"). The model converges on "predict
class 0 (down) everywhere" until B1.1b lands the producer kernel that
fills real -1/0/1 from the 30-bar price trajectory. This degraded
training behavior is intentional and known — the cascade is internally
consistent (every consumer of K-flip / softmax tile / CE / i32 label
migrates atomically per `feedback_no_partial_refactor`); the labels are
placeholder. Local unit tests (CE correctness, dir_acc correctness,
isv_tanh correctness, fingerprint bump) validate B1.1a in isolation;
no L40S smoke runs between B1.1a and B1.1b.
Four contracts (atomic in this commit):
1. K_NB flip 1 → 2: AUX_NEXT_BAR_K constant, compute_param_sizes
([121]/[122] grow), fingerprint seed rename
(PARAM_AUX_NB_W2/B2 → PARAM_AUX_NB_W2_K2/B2_K2 — bumps the hash),
forward + backward kernels, partial-buf allocs (nb_w2 [B,H]→[B,K,H],
nb_b2 [B]→[B,K]), saxpy spec table, max_aux_tensor_len,
aux_nb_pred_buf renamed to aux_nb_logits_buf per
feedback_no_legacy_aliases.
2. Softmax tile: aux_next_bar_forward writes [B, K] softmax via
in-kernel stable softmax (max-shift form, K=2 single-thread fanout
mirrors regime kernel); 4 consumers read the tile (loss, backward,
dir-acc, isv-tanh). NEW field aux_nb_softmax_buf [B, K].
3. MSE → CE: aux_next_bar_loss_reduce reads softmax + i32 labels,
masks -1, divides by B_valid (mean-over-valid-rows), writes loss +
B_valid scalar. aux_next_bar_backward reads B_valid so loss + grad
share the same divisor — derivatives of the same scalar function.
NEW field aux_nb_valid_count_buf [1]. All-skip batch produces
loss = 0 (no NaN; fmaxf(valid, 1) divisor) and zero gradients.
Numerical floor 1e-30 prevents -log(0) = +inf in extreme-logit path.
4. i32 label dtype: aux_nb_label_buf flipped f32 → i32; -1 mask
sentinel handled across loss + backward + dir-acc + isv-tanh.
The strided_gather of next_states[:, 0] retired entirely.
Cascade (atomic per feedback_no_partial_refactor):
- aux_heads_kernel.cu: aux_next_bar_forward gains K + softmax tile
output (via in-kernel stable softmax); aux_next_bar_loss_reduce
ABI flipped (softmax + i32 labels, mean-over-valid CE,
valid_count_out); aux_next_bar_backward ABI flipped (softmax +
i32 labels + valid_count, K-fanout d_logits, masked rows zero
across the K-vector)
- aux_dir_acc_reduce_kernel.cu: read softmax + i32 labels, argmax
over K, output grew 3 → 6 floats (added n_down/n_up/n_skip);
shmem 4 → 6 int arrays
- aux_pred_to_isv_tanh_kernel.cu: read softmax tile, compute
mean(softmax[:, 1] - softmax[:, 0]); tanh transcend retired
(structural [-1, +1] bound via softmax components per
pearl_bounded_modifier_outputs_require_structural_activation)
- gpu_aux_heads.rs: AUX_NEXT_BAR_K 1 → 2; forward_next_bar gains K +
logits_out + softmax_out args; next_bar_loss_reduce gains K +
valid_count_out; backward_next_bar gains softmax_in + labels_i32_in
+ valid_count_in + K
- gpu_dqn_trainer.rs: compute_param_sizes ([121]/[122]); fingerprint
seed rename (W2/B2 → W2_K2/B2_K2); struct fields (logits, softmax,
i32 label, valid_count); aux_dir_acc_buf 3 → 6 floats; partial-buf
allocs grow; max_aux_tensor_len extended; saxpy spec table updated;
orchestrator launchers (launch_aux_dir_acc_reduce,
launch_aux_pred_to_isv_tanh, launch_sp13_aux_dir_metrics) gain K
arg; strided_gather block deleted entirely
- training_loop.rs: aux_b1_diag HEALTH_DIAG line reads
aux_dir_acc_buf [3..6] for n_down / n_up / n_skip + mask_frac;
doc comment update for the per-step aux dir-metrics block
- tests/sp13_phase0_oracle_tests.rs: 6 dir_acc + 3 isv_tanh tests
rewritten in-place to new ABI (no shadow tests, per
feedback_no_legacy_aliases)
- tests/sp13_layer_b_oracle_tests.rs (NEW): 11 B1.1a tests — 5 CE
loss/backward (single-row, batch-mixed, all-skip-NaN, backward
single-row, backward batch-mixed); 2 dir_acc (handcrafted argmax,
all-skip NaN-safe); 2 isv_tanh (bounded fuzz, mean handcrafted);
2 layout regression (fingerprint bump, HEALTH_DIAG snap stable)
- docs/dqn-wire-up-audit.md: B1.1a section
Hard rules upheld:
- feedback_no_partial_refactor: every consumer of K-flip / softmax tile
/ CE / i32 labels migrates atomically — kernels + orchestrators +
struct + diag + existing oracle tests
- feedback_no_atomicadd: block tree-reduce only; CE loss reduce uses
2 parallel partial-reduction strips; CE backward uses existing
per-sample partial → final aux_param_grad_reduce pattern
- feedback_cpu_is_read_only: aux_nb_label_buf is GPU-resident
CudaSlice<i32>; HEALTH_DIAG aux_b1_diag reads via mapped-pinned
aux_dir_acc_buf (no DtoH)
- feedback_no_stubs: every new buffer + kernel arg wired through to
a real consumer; CE forward / loss / backward chain executes
end-to-end against placeholder labels (degraded behavior, not stub)
- feedback_no_legacy_aliases: aux_nb_pred_buf renamed in-place
(no shim); PARAM_AUX_NB_W2/B2 renamed to _W2_K2/_B2_K2 in seed
(no _DEPRECATED alias); 9 existing oracle tests rewritten in-place
- feedback_no_cpu_test_fallbacks: 9 GPU tests gated #[ignore]; 2
layout-regression tests are CPU-only (pub const + size_of)
- feedback_no_htod_htoh_only_mapped_pinned: every CPU↔GPU buffer
in tests + production is MappedF32Buffer / MappedI32Buffer
- feedback_isv_for_adaptive_bounds: no hardcoded thresholds added
(1e-30 numerical floor on log is stability epsilon, not tunable)
- feedback_trust_code_not_docs: 8/8 Phase 0 anchors verified at
HEAD
|
||
|
jgrusewski
|
75e94858c5 |
feat(sp13): B1.0 — ISV[117] retirement + scale-free MSE bridge
Retires ISV[117]=AUX_LABEL_SCALE_EMA_INDEX together with its producer
kernel (aux_label_scale_ema_update), launch site, backward pass-through,
StateResetRegistry entry, HEALTH_DIAG snapshot field, and unit test.
Why: labels at the data layer are z-normalised, so the
mean(|label|) EMA tracked by ISV[117] sits at ~1.0 empirically.
Dividing by max(scale, 1e-6) before the residual `(pred - label)`
reduces to `(pred - label)` within rounding. The divisor was a
defensive scaffold from when the data layer carried mixed-scale
labels (1e-3 log returns vs 5000 raw prices); z-normalisation made
that scaffold redundant.
This is a numerical bridge, NOT the final fix. B1.1 lands on top:
- Aux head 1→2 dim (next-bar regression → 2-class direction logit)
- MSE → CE loss flip
- aux_dir_acc reads softmax over the 2 logits
- aux_pred_to_isv_tanh rewrite as logit-diff
- Producer kernel that fills aux_sign_labels with real -1/0/1 from
the 30-bar price trajectory (B0 plumbing currently zero-init)
- dqn_param_layout fingerprint bump (head dim changes)
- aux_b1_diag HEALTH_DIAG metric
- 17+ GPU oracle unit tests
Cascade (atomic per feedback_no_partial_refactor):
- aux_heads_kernel.cu: aux_next_bar_loss_reduce + aux_next_bar_backward
drop `isv` + `isv_label_scale_index` params; residual is (pred - label)
- aux_heads_loss_ema_kernel.cu: aux_label_scale_ema_update kernel deleted
- gpu_aux_heads.rs: kernel field/loader + launch_label_scale_ema +
isv_* args from next_bar_loss_reduce / backward_next_bar all dropped
- gpu_dqn_trainer.rs: Step 2b producer launch + ISV slot uses dropped;
AUX_LABEL_SCALE_EMA=117 line retained in fingerprint seed
(no fingerprint bump in B1.0; B1.1 will bump on head-dim flip)
- gpu_health_diag.rs + health_diag.rs: aux_label_scale snapshot field
dropped; aux block 4→3 floats, downstream offsets shift down by 1,
WORD_TOTAL 150→149, snapshot_size_is_stable test 150*4 → 149*4
- health_diag_kernel.cu: WORD_AUX_LABEL_SCALE removed, downstream
offsets shift, static_assert(WORD_TOTAL == 149)
- state_reset_registry.rs: isv_aux_label_scale_ema FoldReset dropped
- training_loop.rs: reset_named_state arm + HEALTH_DIAG read +
aux line label_scale field all dropped
- sp4_producer_unit_tests.rs: load_aux_label_scale_ema_kernel helper +
sp4_aux_label_scale_ema_writes_step_obs_via_pearl_a_then_converges_pearl_d
test dropped
Hard rules upheld:
- feedback_no_partial_refactor: every consumer of ISV[117] migrates
atomically — kernel + Rust orchestrator + producer launch + backward
+ HEALTH_DIAG + reset registry + unit test all in this commit
- feedback_no_stubs: not a stub — divisor is removed at every site,
not aliased through a 1.0_const shim
- feedback_no_legacy_aliases: no legacy AUX_LABEL_SCALE_EMA_INDEX → 1.0
alias function
- feedback_no_hiding: doc comments forward to B1.1 explicitly; no
underscore suppression or #[allow(dead_code)]
Build: cargo check --workspace --tests clean.
Tests: snapshot_size_is_stable passes at 149*4=596 bytes.
cargo test -p ml --lib + cargo test -p ml-dqn --lib compile.
Net delta: 10 files, −288 LOC.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
6a869ad366 |
fix(sp13): B0 cascade gap — 5 unaudited insert_batch call sites
The B0 audit (commit
|
||
|
jgrusewski
|
62ab8ed850 |
feat(sp13): B0 — replay buffer i32 ring + GpuBatchPtrs plumbing
Pure plumbing: threads a new i32 column (aux_sign_labels) through every
layer of the replay path so B1 can wire the aux head's CrossEntropy
classification target without touching any aggregator or batch-shape
contract on its own. No consumer reads the column yet — all labels are
zero-initialized; smoke between B0 and B1 should be bit-identical to
parent
|
||
|
jgrusewski
|
0ad5b6fa42 |
chore(sp13): script bug fix + Layer B implementation notes
While P0b smoke (train-sw4ws on
|
||
|
jgrusewski
|
bdc5cb8bb2 |
feat(sp13): P0b — aux_w deficit+stagnation controller + Hold cost lift
P0a smoke (train-67gqb on
|
||
|
jgrusewski
|
f934ea1719 |
feat(sp13): P0a atomic — Hold-pricing + dir_acc instrumentation (additive)
Tests user's hypothesis (Hold being FREE is the bug, not Hold itself) by
pricing Hold via ISV-driven adaptive controller targeting 20% Hold-rate.
11 new SP13 ISV slots [372..383). 5 new GPU kernels:
- aux_dir_acc_reduce_kernel.cu (correct/pos_pred/pos_label/valid → 3 scalars)
- hold_rate_observer_kernel.cu (packed batch_actions decode, count(Hold)/B)
- apply_fixed_alpha_ema_kernel.cu (preserves short/long timescale split that
Wiener-optimal apply_pearls_ad_kernel would collapse)
- aux_pred_to_isv_tanh_kernel.cu (mean(tanh(aux_pred)) → ISV[375])
- 3 reward-composition sites in experience_kernels.cu subtract isv[HOLD_COST]
on Hold actions (segment_complete pre-asymmetric-cap, positioned-non-event
per-bar, flat per-bar)
Host-side controller in training_loop.rs:
excess = max(0, observed - target)
hold_cost = HOLD_COST_BASE × (1 + 5 × excess), clamped [0.5×, 5.0×base]
Per-step observer + EMA chain in gpu_experience_collector.rs after
experience_action_select. Per-epoch HEALTH_DIAG emit:
aux_dir_acc target/short/long/pred_tanh
hold_pricing observed_rate/target/cost
4-way action space stays (ExposureLevel::Hold preserved). Replay buffer /
fxcache compatibility preserved. SP11 (11/11) + SP12 (14/14) tests no
regression. SP13 P0a oracle tests: 14/14 on RTX 3050 Ti.
Spec/plan: docs/superpowers/{specs,plans}/2026-05-04-sp13-redefine-success-for-predictive-skill.md (v3)
Audit: docs/dqn-wire-up-audit.md (SP13 P0a section appended)
v2 → v3 reframe: P0a.T3 v2 implementer's audit found DirectionAction enum
doesn't exist (codebase uses 8-variant fused ExposureLevel cascading through
77 files). v3 reframes from "eliminate Hold" (250 LOC + 32-test cascade) to
"price Hold" (additive, no contract change, no cross-crate cascade).
Tension with pearl_event_driven_reward_density_alignment acknowledged in spec
— per-bar Hold cost is exposure-NEGATIVE (pulls policy AWAY from Hold-default,
inverse of the pearl's failure mode), models real economic carry, ISV-bounded
by controller. Faithful reward modeling, not artificial shaping.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
ba83fcd1f5 |
docs(sp13): v3 spec + plan — Hold-pricing replaces Hold-elimination
P0a.T3 v2 implementer's audit revealed `DirectionAction` enum doesn't exist; the codebase uses an 8-variant fused `ExposureLevel` (ShortSmall/Half/Full, Hold, LongSmall/Half/Full, Flat) with cross-crate consumers across 77 files and 32+ test files pinning the 8-variant invariant. Atomic Hold elimination would cascade massively. User insight (2026-05-04): Hold being FREE is the bug, not Hold itself. MFT trading legitimately needs multi-bar holds; we want the model to use them deliberately, not as a CQL-bias lazy default. Holding isn't free in the real world — broker fees, margin interest, opportunity cost. v3 reframes as Hold-pricing: - 4-way action space stays; ExposureLevel::Hold stays; no cross-crate cascade - 3 new ISV slots (380-382): HOLD_COST_INDEX, HOLD_RATE_TARGET_INDEX, HOLD_RATE_OBSERVED_EMA_INDEX - Hold-rate observer: small GPU kernel + Pearls A+D smoothing - Hold-cost controller: 5-line deficit-driven formula (excess > target → cost rises 1×→5× base; observed ≤ target → relax) - Per-bar reward subtraction at action == DIR_HOLD site - 2 GPU oracle tests for the controller P0a.T3 cuts from ~250 LOC + 32-test cascade → ~120 LOC additive. T1+T2 already-staged work unchanged. T4/T5/Layer B/C/D structure preserved. Tension with pearl_event_driven_reward_density_alignment acknowledged in spec — per-bar Hold cost is exposure-NEGATIVE (away from Hold), models real economic carry, ISV-bounded by controller. Inverse of the pearl's failure mode. Faithful reward modeling, not artificial shaping. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
0ca45ef61d |
docs(sp13): v2 spec + plan after critical review
Spec v2 supersedes v1 with five major fixes from the critical review: - Drop alpha-vs-benchmark reward (mathematically tautological — Long trades had alpha = -costs always against always-long benchmark) - Split Phase 0 into 0a (Hold-only, clean test of user hypothesis) + 0b (aux amplification probe, only if 0a partial), avoiding the v1 confounded experiment - Bound direction-skill bonus relative to |alpha| (cap_ratio × |alpha|) to prevent reward gaming on small-alpha correct-direction losers; spec adds 8-quadrant worked-example matrix verifying the no-negative-EV invariant - Replace 5-epoch absolute-threshold gate with 10-epoch trajectory criterion to avoid false-negatives from aux head underconvergence - Aux_w controller adds dual-EMA stagnation detector (decays toward base when no improvement) — prevents permanent destabilization of Q-head in data-limited case Plan v2 mirrors spec changes: - Phase 0a (Hold elimination + dir_acc instrumentation, ~300 LOC, 1 atomic commit) - Phase 0b (aux_w controller replacement, conditional, ~80 LOC) - Layer B (aux head regression -> binary classification, ~120 LOC) - Layer C (skill bonus + luck discount with calibrated bounds, ~150 LOC) - Layer D (30-epoch validation + 3 new pearls) ISV slot allocation [372..380): drops slot 371 (BENCHMARK_PNL_CUMULATIVE), adds 374 (AUX_DIR_ACC_LONG_EMA — stagnation), 379 (SKILL_BONUS_CAP_RATIO). Plan integrates Explore-agent touch-list (50-70 sites) with corrected enum ordering: Short=0, Long=1, Flat=2 (preserves codebase Short-first convention, avoids 30+ stale-comment churn). Adds direction-bias signal handling at experience_kernels.cu:5159 (Hold's 0.5 softening gate vanishes -> [1, 1, 0]). Three new pearls planned for Layer D close-out: - pearl_redefine_success_for_predictive_skill - pearl_skill_bonus_must_be_alpha_bounded (calibration lesson) - pearl_reward_quadrant_audit_required (meta-pearl) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
a1681abc46 |
Revert "exp(sp13): aux_w=1.0 override + directional accuracy metric for data investigation"
This reverts commit
|
||
|
jgrusewski
|
d2a27a0042 |
exp(sp13): aux_w=1.0 override + directional accuracy metric for data investigation
One-shot diagnostic to answer the SP13 root question: does the data have
predictable directional signal at the bar level?
Wires the existing `aux_next_bar_loss_reduce` kernel (which already had a
4-strip shmem reduction emitting `[dir_acc, pos_pred_frac, pos_label_frac]`
into a 3-float output) end-to-end:
- `gpu_aux_heads.rs`: launcher takes `dir_acc_out_ptr`, allocates 4×AUX_BLOCK
shmem to back the four parallel reductions.
- `gpu_dqn_trainer.rs`: adds `aux_nb_dir_acc_buf` (3 f32 device buffer),
threads it through the loss-reduce launch, exposes `read_aux_dir_acc()`
accessor for once-per-epoch DtoH readback.
- `training_loop.rs`: pins `aux_w = 1.0` (instead of the ISV-driven 0.05–0.3
clamp) so the supervised aux head dominates the loss; emits a new
`HEALTH_DIAG[ep]: aux_dir_acc accuracy=… pos_pred_frac=… pos_label_frac=…`
line per epoch.
Verdict thresholds:
* dir_acc > 55% by ep 5 ⇒ data has signal, DQN failing to use it
* dir_acc ≈ 50% throughout ⇒ data lacks signal at this timescale
* dir_acc 60–70% ⇒ strong signal we're not using
EXPERIMENT BRANCH — revert this commit after the investigation reads back the
5-epoch table from smoke logs. All five touch points are tagged
"SP13 data-investigation" / "EXPERIMENT" for clean revert.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
1c645264e6 |
test(sp12): GPU oracle tests for the 3 reward math changes
Adds the GPU oracle test scaffold deferred from commit
|
||
|
jgrusewski
|
17cfbb2503 |
fix(sp12): per-trade event-driven reward composition
Three architectural changes in one atomic commit per spec |
||
|
jgrusewski
|
ecab584c32 |
spec(sp12): per-trade event-driven reward composition (v3)
Three architectural changes in one unified design to push the model from
HFT noise extraction (62% trade rate, sharpe-gaming) toward MFT alpha-hunting:
1. Asymmetric bounded cap (-10/+5) — restores loss aversion erased by
SP11 symmetric cap. 2:1 ratio matches Kahneman/Tversky prospect theory.
Anchor: pearl_audit_unboundedness_for_implicit_asymmetry.
2. Min-hold soft penalty with temperature curriculum — patience requirement
at exit. Soft factor = deficit/(deficit+T), T anneals 50→5 over 50 epochs.
Forces commitment without paralysis in early training.
3. Zero per-bar shaping (gate micro/opp_cost on events) — eliminates
continuous-reward gradient that pulls toward continuous exposure.
Anchor: pearl_event_driven_reward_density_alignment.
Combined: reward fires only on trade events with prospect-theory loss
aversion + commitment requirement. Pure per-trade event-driven Q-learning
properly aligned with per-trade P&L objective.
~50 LOC across 3-4 files. No new ISV slots in Phase 1 (constants only).
Cost ~€1.30 (€0.30 smoke + €1.00 30-epoch validation).
Empirical motivation: train-multi-seed-pmbwn 50-epoch on commit
|
||
|
jgrusewski
|
6a259942e9 |
docs(sp11): close-out — pearl_symmetric_clamp_audit + audit doc
Documents the SP11 sharpe-degradation root cause investigation: - |
||
|
jgrusewski
|
b92dcc3dfc |
chore(sp11): remove reward-chain diagnostic instrumentation
Instrumentation from |
||
|
jgrusewski
|
348f6078b8 |
fix(sp11): plan_isv symmetric clamp — 4 sites mirror reward-cap bug
Implementer of
|
||
|
jgrusewski
|
35db310893 |
fix(sp11): symmetric reward cap — losses were unbounded
experience_kernels.cu:2788:
float capped_pnl = fminf(base_reward, 10.0f);
^^^^^^^^^^^^^ caps profits, NOT losses
Diagnostic instrumentation in smoke-test-k9drh on commit
|
||
|
jgrusewski
|
774d7552a0 |
diag(sp11): instrument reward chain to find the 5000x inflater
Smoke smoke-test-gwfn8 on
|
||
|
jgrusewski
|
7a19c51522 |
docs(sp11): B1b smoke-recovery close-out — audit + 3 memory pearls
Layer C close-out for SP11 B1b smoke-recovery work. Two atomic commits land the fix: - |
||
|
jgrusewski
|
fd24b53833 |
fix(sp11): B1b launch-order — reward_component_ema before mag-ratio canary
smoke-test-4rbv9 on
|
||
|
jgrusewski
|
b3b4d02789 |
fix(sp11): B1b smoke-recovery — z-score normalization for mag-ratio canary
Linear magnitude ratios in reward_component_mag_ratio_compute_kernel amplified popart's intrinsic O(100) magnitude over the other 5 components' O(0.1-2) magnitudes, causing controller to saturate w_pop toward MAX_WEIGHT regardless of actual signal quality. Replaced with z-score: z[c] = mag[c] / max(sqrt(var[c]), EPS_DIV). 6 new ISV slots [361..367) for per-component variance EMAs computed via Welford's online algorithm in extended popart_component_ema_kernel and reward_component_ema_kernel. Atomic per feedback_no_partial_refactor: slot allocation + state-reset registry + 2 producer kernels + canary signature + launcher Pearls A+D + tests. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
85069ba75e |
spec(sp11): amend B1b smoke-recovery — z-score normalization for mag-ratio canary
smoke-test-6wd2c on commit |
||
|
jgrusewski
|
61b2fa962b |
fix(sp11): B1b bug 3 — cf-component feedback loop in mag-ratio canary
Deep audit on |
||
|
jgrusewski
|
b435d25bec |
fix(sp11): B1b bug-hunt fix-up — stale rc[] init + cf_flip ordering
Bug-hunt review on
|
||
|
jgrusewski
|
5e16b67ca6 |
fix(sp11): B1b follow-up — add slot 360 for popart-component mag EMA
Per spec §4 amendment at
|
||
|
jgrusewski
|
034ba16801 |
feat(sp11): B1b — structural reward composition refactor (production flip)
Per spec §3.5.3 amended at
|
||
|
jgrusewski
|
d5e1214f25 |
fix(sp11): B1a — saboteur GPU multiplication + SimHash state_stride
Per feedback_cpu_is_read_only, saboteur effective scale now computed on-device: - saboteur_generate_params kernel signature gains isv ptr + saboteur_intensity_mult_slot parameter - Kernel reads `mult = fmaxf(isv[saboteur_intensity_mult_slot], SABOTEUR_MIN)` (sentinel-0 defense for cold-start before A2's controller first runs) and applies `effective_scale = base × mult` to perturbation generation - gpu_experience_collector.rs launcher updated; only one call site SimHash state_stride parameter added to lookup + update kernels — prepares for B1c replay-time curiosity wiring against trainer. states_buf which is STATE_DIM_PADDED=128-strided. Kernel inner loop reads `state[i × state_stride + d]` (was `i × 42 + d`). Proj-init kernel unchanged (writes projection, doesn't read states). Pre-requisite for B1c (curiosity wiring) and a small atomic step toward full SP11 production behavior. cargo check + build clean; 6/6 SP11 GPU tests + 14/14 contract tests still pass. |
||
|
jgrusewski
|
302992f63a |
fix(sp11): B0 — controller renorm Σ=1 → mean=1 (post-A2 spec amendment)
Per spec §3.4.3 amended at
|
||
|
jgrusewski
|
44fb4531a8 |
fix(sp11): A2 follow-up — delete dead launchers + expand XOR-fold rationale
Code-quality review on
|
||
|
jgrusewski
|
25eba79ad5 |
feat(sp11): A2 — controller kernel + SimHash novelty buffer
reward_subsystem_controller_kernel: 5 canaries → 10 outputs, true Z-score (delta_ema/sqrt(var_ema)), sigmoid blending, weight renormalization to Σ=1, saboteur post-clamp, curiosity permanent floor (0.2 × bound). Pearls A+D chained on outputs per spec §3.4.1. novelty_simhash_kernel: 42×16 random projection → 16-bit SimHash code, 1M-slot bucket count table for novelty signal `1/sqrt(1+count)`. Race- tolerated update per feedback_no_atomicadd (under-counts bias novelty UPWARD — safe direction). novelty_simhash_proj_init_kernel: Philox-seeded GPU init for the projection matrix (CPU is read-only per feedback_no_cpu_forwards). HEALTH_DIAG `sp11_reward` line emits 10 outputs + improvement_z each epoch. Reset registry: novelty hash table reset arm wired (closes the A0 deferral); projection matrix is frozen at trainer init for run lifetime, not reset. All 20 SP11 slots populate every step. No consumer reads them yet — training behavior unchanged from A1. 3 new GPU oracle tests pass on RTX 3050 Ti (controller midpoint, weight renorm, saboteur clamp). Spec: docs/superpowers/specs/2026-05-04-sp11-reward-as-controlled-subsystem.md §3.4 §3.5.2 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
jgrusewski
|
91b48bc7a5 |
feat(sp11): A1 — three canary producer kernels (no behavior change)
Adds val_sharpe_delta + saboteur_engagement + reward_component_mag_ratio
GPU producers for the SP11 reward-as-controlled-subsystem chain. Each is
a single-block producer chained with apply_pearls_ad_kernel for Pearls
A+D smoothing per pearl_first_observation_bootstrap.md +
pearl_wiener_optimal_adaptive_alpha.md. All three write to slots in
[350..360) which no consumer reads yet — Layer A is additive; consumer
migration lands atomically in Layer B.
A1.1 — val_sharpe_delta_compute_kernel.cu
Two-pass: writes raw delta + (delta - prev_delta_ema)^2 to scratch.
Chained Pearls A+D (n_slots=2) → ISV[VAL_SHARPE_DELTA_EMA_INDEX=350,
VAL_SHARPE_VAR_EMA_INDEX=351]. Host writes val_sharpe to mapped-pinned
history[1]; rotation handled in training_loop.rs at val emit boundary
(a literal already-computed value — no host-side compute, no htod_copy).
A1.2 — saboteur_engagement_compute_kernel.cu
Per-bar |Δreward| > 0.01 × ISV[PNL_REWARD_MAGNITUDE_EMA_INDEX] check
with block tree-reduce (no atomicAdd per feedback_no_atomicadd). The
per-bar Δreward signal is produced by experience_env_step's saboteur
perturbation site as `traded × |reward| × max(|eff_spread − 1|,
|eff_slip − 1|)` — a structural proxy for the cost-differential the
saboteur imposed on bars where the model traded. Single kernel-side
emit (no parallel reward computation), per spec §3.3.1.
Chained Pearls A+D → ISV[SABOTEUR_ENGAGEMENT_RATE_INDEX=358].
A1.3 — reward_component_mag_ratio_compute_kernel.cu
Reads ISV[REWARD_POPART_EMA_INDEX..+6) (the SP4 reward-component
magnitude EMAs), normalises to ratios, and mirrors popart magnitude
into scratch[6] as a side-output. ONE non-pointer parameter
(popart_ema_base_slot) — no _unused param per feedback_no_stubs.
Two chained Pearls A+D launches:
n_slots=6 → ISV[REWARD_COMPONENT_MAG_RATIO_BASE..+6)
n_slots=1 → ISV[PNL_REWARD_MAGNITUDE_EMA_INDEX=359]
(slots non-contiguous: 352..358 then 359.)
Wire-up (per feedback_wire_everything_up):
- 3 cubin entries appended to crates/ml/build.rs
- 3 kernel handles + val_sharpe_history_pinned (MappedF32Buffer[2]) +
saboteur_delta_reward dev-ptr cache fields on GpuDqnTrainer
- 3 launchers (launch_sp11_*) + 1 setter (set_sp11_saboteur_delta_reward_buf)
- saboteur_delta_reward_per_sample buffer field on GpuExperienceCollector
- experience_env_step kernel signature extended with the new buffer arg;
every call site in the same commit per feedback_no_partial_refactor
- training_loop.rs init wires collector→trainer setter; val emit boundary
invokes launch_sp11_val_sharpe_delta_compute; per-epoch metrics block
invokes launch_sp11_mag_ratio_compute then
launch_sp11_saboteur_engagement_compute (mag_ratio first so the
signal-relative threshold base is populated before the saboteur reader)
- SP5_SCRATCH_TOTAL grown 266 → 276 (10 new scratch slots: 2+1+7)
- docs/isv-slots.md SP11 section updated to reflect A1 producers
3 GPU oracle tests in crates/ml/tests/sp11_producer_unit_tests.rs
pass on RTX 3050 Ti via MappedF32Buffer fixtures (zero htod_copy /
dtoh_sync_copy / alloc_zeros — feedback_no_htod_htoh_only_mapped_pinned
compliant).
Note on Step 8a path: the plan offered two routes for the saboteur
Δreward producer — in-kernel diff emission OR a small dedicated
reader-of-existing-buffers. The existing reward path emits ONE reward
(not both with/without), so the dedicated-reader alternative was
infeasible. The in-kernel emission landed as a small write site at the
END of experience_env_step (after total_reward_per_sample is finalised),
threading saboteur_eff_spread/saboteur_eff_slip from the perturbation
site forward to the END via stack vars. Single new kernel parameter,
single new GPU-only buffer, single existing call site updated.
Spec: docs/superpowers/specs/2026-05-04-sp11-reward-as-controlled-subsystem.md §5
Plan: docs/superpowers/plans/2026-05-04-sp11-reward-as-controlled-subsystem.md (Task A1)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
jgrusewski
|
201b59dfbc |
fix(sp11): A0 sweep — eliminate remaining stale wiener-buffer refs
A0 follow-up (
|
||
|
jgrusewski
|
1e5a65912c |
fix(sp11): A0 follow-up — update stale wiener-buffer + isv-slots header
Code-quality review on
|
||
|
jgrusewski
|
bf3a32d63a |
feat(sp11): A0 — allocate 20 ISV slots [340..360) + 20 reset entries
Pure infrastructure. No producer kernels, no consumer reads. Existing training paths trace identically because no consumer reads slots [340..360) yet. Layout-fingerprint bumped to ISV_TOTAL_DIM=360. Spec: docs/superpowers/specs/2026-05-04-sp11-reward-as-controlled-subsystem.md |
||
|
jgrusewski
|
e0d3abd9d2 |
plan(sp11): patch all 5 violations + 2 gaps from review
feedback_no_htod_htoh_only_mapped_pinned (tests not exempt):
- All test fixtures converted from htod_copy/dtoh_sync_copy to
MappedF32Buffer with host_slice / host_slice_mut access.
- Novelty hash buffer + projection matrix changed from
cudarc::CudaSlice<f32> + alloc_zeros to MappedF32Buffer.
feedback_no_cpu_forwards (CPU is read-only):
- Projection matrix initialization changed from host-side StdRng +
host_slice_mut writes to a one-shot GPU init kernel
(novelty_simhash_proj_init_kernel) using Philox seeded from
config.seed. No host RNG, no host writes.
feedback_no_cpu_compute_strict (saboteur multiplication):
- B1 step 5 reverted from Rust-side `read_isv_slot * scale` to
GPU-side: pass base scale + ISV pointer + slot index to the
saboteur perturbation kernel; multiplication happens on-device.
feedback_trust_code_not_docs (grad-ratio terminology):
- Spec §3.3.1 "per-component grad EMA" was wrong — SP4 grad-balancer
is per-branch (4 slots), not per-reward-component. Renamed:
reward_component_grad_ratio_compute_kernel
→ reward_component_mag_ratio_compute_kernel
REWARD_COMPONENT_GRAD_RATIO_BASE
→ REWARD_COMPONENT_MAG_RATIO_BASE
Source: existing REWARD_POPART_EMA_INDEX..+6 (per-component reward
magnitude EMAs from SP4 reward_component_ema_kernel). Semantic
equivalent for the controller's exploit/diversify blend.
feedback_no_stubs (dead parameter):
- Removed `eps_div_idx_unused` from mag_ratio kernel signature.
Saboteur engagement: missing producer specified
- Spec §3.3.1's two-reward-arrays formulation replaced with single
`saboteur_delta_reward_buf` produced by the saboteur perturbation
kernel itself (single reward computation, diff emitted as side
output). Engagement kernel signature simplified to one input array.
PNL_REWARD_MAGNITUDE_EMA_INDEX (slot 359): producer wired
- mag-ratio kernel mirrors `isv[REWARD_POPART_EMA_INDEX]` to
scratch_out[6]; chained apply_pearls_ad targets slot 359.
Replay sample kernel location specified
- graph_utility_kernels.cu:71 (gather_f32_scalar). New sibling kernel
`gather_replay_reward_with_curiosity` defined; replaces the existing
scalar gather (no legacy alias per feedback_no_legacy_aliases).
novelty_simhash_lookup runs before, novelty_simhash_update after.
A2 controller test placeholders → full GPU oracle assertions
- Three controller tests (z=0 midpoint, weight renorm, saboteur clamp)
have full mapped-pinned fixtures with assertions on weight sum,
individual values, post-clamp bounds.
Plan now passes:
- feedback_no_htod_htoh_only_mapped_pinned (tests + production)
- feedback_no_cpu_forwards (CPU never writes/computes for GPU)
- feedback_no_cpu_compute_strict (all multiplications GPU-side)
- feedback_no_atomicadd (race-tolerated non-atomic, safety documented)
- feedback_no_partial_refactor (Layer B atomic; saboteur kernel sig
change touches all callers in the same commit)
- feedback_no_stubs (no dead parameters)
- feedback_trust_code_not_docs (corrected spec terminology)
- feedback_wire_everything_up (every new field has init + producer)
- feedback_no_legacy_aliases (old gather_f32_scalar replaced, deleted)
1447 lines, +268 from previous version.
|
||
|
jgrusewski
|
d8b44e0829 |
plan(sp11): implementation plan for reward-as-controlled-subsystem
Task-by-task plan for the SP11 spec at HEAD
|
||
|
jgrusewski
|
9395b983cc |
spec(sp11): fix all 13 review issues — straight-up implementation
Critical bugs fixed:
1. Z-score formula: was mean(Δ)/mean(|Δ|), bounded to [-1,+1] — sigmoid
never saturated, controller stuck in [0.27, 0.73]. Now true Z-score:
delta_ema / sqrt(delta_var_ema) with two-pass var. Renamed canary
slot 351 from VAL_SHARPE_STD_EMA to VAL_SHARPE_VAR_EMA.
2. Component weight renormalization added — Σweights = 1 enforced
post-floor so PopArt's reward-magnitude EMA stays uncontaminated.
3. SABOTEUR_MIN bound violation — engagement-floor (0.1) was silently
pulling output below 0.5 stated minimum. Added post-multiplication
clamp to [SABOTEUR_MIN, SABOTEUR_MAX].
4. ratios[] undeclared — now __shared__ float ratios[6] block-loaded
once from ISV.
5. §3.6 slot count off-by-five (15 → 20). All 20 slots get FoldReset
entries; cold-start window behavior specified explicitly (no NaN path).
Architectural fixes:
6. Q-overconfidence concern addressed in scope — controller's
REWARD_CF_WEIGHT_INDEX covers CQL conservatism. No SP11′ deferral;
if T10 fails, extend controller outputs (e.g., target-update τ).
7. Curiosity recompute-at-replay specified (§3.5.1) — replay buffer
stores base reward only; curiosity added per-tuple at replay time
against current visit-count and current ISV. Eliminates stale-signal
replay contamination that would worsen ep1-overfitting.
Smaller fixes:
8. Novelty signal specified — SimHash 42×16 projection → 16-bit code,
1M-slot per-bucket count table, 1/sqrt(1+count). Hash table lives
in state-reset registry (cleared at fold boundary).
9. Saboteur engagement operationally defined — per-bar
|reward_with_saboteur − reward_without_saboteur| > EPS_ENGAGEMENT,
block tree-reduce, EMA'd. EPS_ENGAGEMENT = 0.01 × PnL_EMA.
10. Duplicate sentence in §7 component-starvation paragraph removed.
11. Validation criteria strengthened (§9): aggregation across 9
trajectories specified; primary metric = median peak-epoch ≥ 10
(baseline median peak-epoch = 1); secondary = drop-from-peak ≤ 5%
by ep20; tertiary = ep20 mean ≥ baseline ep1 mean. §9.3 fix-forward
response codified — no rollback.
12. Phases split per project pattern — Layer A additive (3 commits:
slots, canaries, controller), Layer B atomic consumer migration
(1 commit, including replay-time curiosity), Layer C validation +
close-out. No falsification gate; smoke is validation only.
13. Pearls A+D applied to controller outputs (§3.4.1) — chained
apply_pearls_ad_kernel after controller writes scratch, smoothed
values land in ISV. Consumers read smoothed slots.
Constants surviving (Invariant-1 anchors only, all rate-not-regime):
EPS_DIV, WEIGHT_HARD_FLOOR, SABOTEUR_MIN/MAX, CURIOSITY_PERMANENT_FRACTION,
CURIOSITY_BOUND_FRACTION, WEIGHT_FLOOR_FRACTION, ENGAGEMENT_FLOOR.
Spec is now full straight-up implementation; smoke validates Layer A
infrastructure and Layer B consumers, T10 validates SP11 success metric.
~1550 LOC across 3 commits in Layer A + 1 atomic commit in Layer B +
close-out in Layer C.
|
||
|
jgrusewski
|
d49fbbe4e2 |
spec(sp11): reframe curiosity as feature + add permanent floor
User correction: curiosity is the *fix* for the ep1-peak overfitting pathology, not a hazard to defend against. Reframed §7 from "Risks" to "Design notes" — curiosity bound is a signal-relative scale, not a defensive cap. Fix the contradiction this exposed in the formula: previous `curiosity_pressure = stagnant_or_worse * curiosity_bound` went to zero when improving, which would cancel the always-on exploration the §7 narrative now relies on. Replace with permanent-floor pattern per pearl_blend_formulas_must_have_permanent_floor: curiosity_floor = 0.2 * curiosity_bound (CURIOSITY_PERMANENT_FRACTION) curiosity_dynamic = stagnant_or_worse * curiosity_bound curiosity_pressure = max(curiosity_dynamic, curiosity_floor) Now curiosity is always ≥ 20% of bound (anti-overfitting baseline) and rises toward the bound when stagnant (stagnation breaker). Updated unit-test guidance to assert pressure > 0 even at z=+10. CURIOSITY_PERMANENT_FRACTION=0.2 added to Invariant-1 fraction list. Saboteur-rising-with-improvement reframed as adversarial-load feature rather than over-stress risk. |
||
|
jgrusewski
|
a72a23e6b1 |
spec(sp11): reward as controlled subsystem — fully ISV-driven
Brainstorm spec for SP11. Resolves the policy-stagnation pathology
surfaced in T10 train-multi-seed-xkjkb seed-0 ep0-14: model finds a
stable fixed point at ep1 (peak val sharpe 80.61), then OVERFITS to
it across remaining epochs (decline 80.61 → 70.58). Q-values grow but
val performance declines because reward function has no improvement
pressure.
Architecture (every input ISV-driven):
- Z-score-driven adaptation (no hardcoded "improving" threshold):
improvement_z = val_sharpe_delta_ema / max(val_sharpe_std_ema, EPS)
- 10 ISV outputs: 6 component weights + curiosity_pressure +
saboteur_intensity_mult + adaptive weight_floor + curiosity_bound
- 5 ISV canaries: val_sharpe_delta + val_sharpe_std (Z-score noise
estimate) + 6 per-component grad ratios + saboteur engagement +
PnL magnitude EMA (signal-relative curiosity bound)
- 4 new producer kernels (controller + 3 canary computers)
- Audit + migrate hardcoded cf_weight=0.3 in mse_loss_kernel.cu:318
and c51_loss_kernel.cu:789, plus other shaping multipliers
- NEW reward dimension: curiosity bonus, bounded by PnL magnitude
Per pearl_controller_anchors_isv_driven: every threshold replaced with
sigmoid(z) — no constants encode "what counts as improving". Per
pearl_blend_formulas_must_have_permanent_floor: every weight has
adaptive floor preventing zero-out. Per pearl_engagement_rate_self_
correction: saboteur intensity self-corrects via engagement rate canary.
Per pearl_cold_start_exit_signal_or: improvement signal OR'd from
multiple canaries so single-signal-failure doesn't stall controller.
New pearl authored alongside spec: pearl_reward_as_controlled_subsystem
— meta-principle that every reward path degree of freedom is a unified
controller output. Subsumes controller-anchor pearl at the reward layer.
Scope: 20 ISV slots, 4 producer kernels, audit + migration of
hardcoded reward shaping constants, 1 atomic commit.
~1300-1700 LOC; ~2.5-3 hours subagent work.
Success metric: val_sharpe[ep20] > val_sharpe[ep1] (Fix 33-38 baseline
peaked at ep1; SP11 should shift peak later as model continues
learning).
|