Commit Graph
4 Commits
Author SHA1 Message Date
wassnameandClaudypoo adca442253 feat(#41): routeA activation gate replaces routeV grad gate
Gate now scores each rollout by dot(pooled bottleneck act, v_act) captured on
the no-grad logpi_old forward (quarantine-ablated, matching the sampling
policy); masks are pinned BEFORE the single grad-carrying forward, so the
grad-gate's pass-1 backward is gone. Thresholds: rolling 256-act buffer,
z-normalized, two-threshold Otsu (winsorized 1/99); warmup pins absorb until
128 scores. Buffer stores pooled acts and re-scores against the current v_act,
so the forward-only refresh (every 5 steps) needs no flush. No bimodality
guard: calibration showed Otsu tail separation ~2.4-2.8 buffer-sd on every
condition including pure Gaussians, so no shape statistic discriminates.

Deleted with the arm wiring (rename-on-logic-change: routeA never conflates
with routeV runs): extract_vhack_grad.py, _build_v_grad, route_band_edges,
_pair_cos, the pass-1 autograd.grad block, grad_probe training wiring,
v_grad_k/route_std_*/routeV_random_v_seed config, smoke-topk recipe.
c-probe stays in lora2r.py for scripts/diag_pinning.py only.

verify_science_invariants: all-in-one count 27 -> 42 (stale since c33b810
added the wave-2 behavior2 pairs) + assert the 8-pair routeA training subset.

Smoke: routeA/vanilla/absorb/solvemix all pass (gate exercises warmup, Otsu
zones, refresh, deploy ablation) -- /tmp/claude-1000/smoke_routeA.log.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-06-11 12:38:19 +00:00
wassname 270c4f5a27 misc 2026-06-11 11:07:28 +00:00
wassname 55937a86fb rename python package projected_grpo -> vgrout
git mv src/projected_grpo -> src/vgrout and find-replace the module name in
all imports (.py), `-m projected_grpo.*` invocations (justfile), and the
[project] name (pyproject; setuptools auto-discovers via where=["src"]).

Left RESEARCH_JOURNAL.md untouched: its commands/paths are dated lab notes
tied to past commits, so rewriting them would falsify provenance. Repo dir,
git remote, and absolute paths unchanged.

Verified: `import vgrout` and `python -m vgrout.train --help` load the full
graph; verify_rewards.py + verify_gate_anchor.py (both import vgrout) pass.
Full `just smoke` is blocked upstream by missing gitignored data artifacts
(out/pools/{substrate,teacher_pool}, out/vhack/*smoke*), unrelated to the rename.
2026-06-05 14:51:48 +08:00
wassnameandClaudypoo 4c9071cca0 A5: build held-out-mode (hack,clean) pairs from student rollouts
scripts/pairs_from_rollouts.py mirrors pairs_from_pool but sources the
student's own rollouts.jsonl and splits hack/clean by env_mode+exploited
(the per-mode weak detector). Same-prompt pairing, asserts prompt equality.
Smoke-validated: parse + classify + loud-fail paths green on smoke rollouts
(0 hacks -> 0 pairs, as expected). Unblocks A5 once job 95 harvest lands.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
2026-06-03 00:59:07 +00:00