foxhunt

Files

jgrusewski 9c0aaacdfb docs(sp22): smoke scope — α ACTIVE β DEAD (B6 not landed)

Auditing the β reward shaping path revealed a producer gap:
- experience_kernels.cu:3614-3628 correctly computes r_aux_align from
  scale_β = isv_signals[537]
- But slot 537 has NO PRODUCER — SP11 controller still emits 6 components
  (N_COMPONENTS=6.0f), Phase B6 extension to 7 components hasn't landed
- → scale_β stays at alloc_zeros 0 → r_aux_align = 0 → β no-op

Current smoke validates α atom-shift mechanism ONLY. β remains dormant.

Updated verdict outcome table: WR-shift implies α works alone; WR-stable
with dist-shift implies α active but β needed (land B6 next, ~1-2hr).

Phase B6 scope documented (single-kernel edit to SP11 controller +
launcher update for N_COMPONENTS).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-13 08:26:46 +02:00

infra

feat(infra): GPU training launcher with local/cloud routing

2026-02-23 11:19:58 +01:00

plans

docs(sp22): H6 Phase 3 runbook — atom-shift math derivation for Steps 7+8

2026-05-13 01:45:40 +02:00

superpowers

docs(claude): foxhunt agents & skills rollout — phase 1 close-out

2026-05-10 22:19:28 +02:00

dqn-backward-nan-audit.md

fix(dqn): SP3 Mech 10 — ISV-driven h_s2 activation clamp + slot 49 diag