route2 is an internal run-tag, not something a reader cares about.
Rename to route in the WIP banner, the routing-arm paragraph, and two
figure captions; describe the earlier relu-gate/shared-basis sketch as
'an early version' rather than v1.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
User settled it: prog_wide pairs were AI-authored (Claude), so the
synthetic/AI-written framing in contribution 2 is honest. Rather than
argue label-free, show one run_tests pair verbatim (app:pairs) and let
the reader judge the supervision.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Huang related-work bullet now states the actual differences (SVD of clean
update trajectory + warmup vs our contrastive pair-gradients in delta_S coords;
they project onto trusted, we project out hack; we quarantine+delete at deploy,
they only constrain training). Renamed docs/papers/grad_routing/paper_deng_*
-> paper_huang_* (untracked note; correct attribution is Huang et al. 2026).
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
- title -> 'Quarantining Reward-Hacking Gradients with a Hacking Representation'
- contributions: (1) adapt SGTM parameter-gradient masking from supervised
unlearning to RL reward hacking, route+ablate framing from gradient routing
but NOT Cloud's activation .detach(); (2) replace the data-label mask with a
RepE-extracted hack direction from ~10-21 pairs (live rollouts unlabeled).
- method 'Arms': call route SGTM-style post-backward parameter masking in SVD
basis, routed into a deletable subspace.
- related work: Cloud = localize-then-ablate idea only; SGTM = closest
mechanistic relative, their TPR/FPR knob = our weak-detector axis.
- title comment flags the OPEN synthetic-pairs question (headline v_hack is
hand-authored prog_wide, not AI-prompted).
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
- route2 -> route in all prose/captions/tables (route2 stays in % provenance
comments as the run-tag). A reader does not care about the version number.
- title: steering-vector framing; recorded naming reasoning as a comment (do NOT
claim label-free -- our pairs ARE labels; the backable scoped claim is held-out
hacks suppressed with zero labels of their own, earnable by A5).
- FLAG at contribution 1: our mechanism is SGTM-style post-backward parameter-
gradient masking, NOT Cloud's activation-level gradient routing. Author-verbatim
claim left intact but flagged inline; see docs/papers/grad_routing/sgtm_vs_ours.md.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Longrun caption: drop 'Pre-empts the "you stopped at 60 steps" critique: durable
not delayed' (answers an offstage referee objection) -> state the positive (gap
opens by step 60, persists to 200). Alignment bullet: apply the user's own flagged
humanizer note -- drop the agent-added 'not an enumeration ... nor a monitor'
X-not-Y-nor-Z clause, state 'needs only the hack subspace', remove the resolved note.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
train.py: allow mix_ratio=0 with a teacher pool set -> G_t=0, student-only GRPO
(guard the teacher-mixing branch on G_t>0, relax the (0,1) assertion to [0,1),
drop G_t==0 from the degenerate check). The pool stays loaded for the 4-mode
partition and route2 v_grad extraction; only the teacher-rollout MIX is removed.
Smoke (mix=0 + normal mix=0.5 + vanilla) all green.
Also: fill A4 long-run figure (fig:longrun) in main.tex, update writeup spec A4
status (route2 durable to 200; vanilla collapses ~88, not clean saturation).
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
- deploy hack/solve is now the headline metric for all arms, so turn the
mid-train deploy-eval on by default (smoke now covers the deploy path too);
200-step runs pass a sparser cadence explicitly.
- docs/spec/20260602_writeup_spec.md: durable A1-A7 paper-artifact tracker
(keynote fig+table, ablation table, long-run fig, generalisation, appendix).
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
- blog: mark as erase-n=2 draft, note route2/exploration-floor/deploy-eval are the
current direction; embed dyn_sub4_hack_overlay.png (force-added); ASCII em-dashes;
de-bold the arm list (#15 tell)
- README: add route2 arm + apples-to-apples deploy-eval to 'What we compare'; stale
banner on the n=1 mix=0.5 findings
- plot_dynamics: remove _mark_if_sparse (asymmetric sparse-only dots); EMA-held line
for all arms
- train.py: fix 'held-out greedy' -> 'held-out eval subset, T=0.7' (deploy eval is
sampled, not greedy)
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
The plotter picked hk_abl (dense proxy) whenever the COLUMN existed, but no-floor
runs (rollout_ablate_frac=0) emit hk_abl as 0/0 -> all-nan, so the deploy panel
came up empty. Test for finite data (_has_data) not column presence; fall back to
the sparse-but-real hk_dep (every eval_ablate_every steps). _ema carries values
across the nan gaps -> a held step-line.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
The distinct-basis A_q/B_q LoRA (~33M params at rank-16) gave the quarantine a
~100x capacity edge over delta_S, so routing-everything-there was the low-
resistance path: qE pinned ~0.97 (energy into the thrown-away knob) while the
deployed delta_S learned nothing (job 54). The cause was capacity imbalance, not
the routing gate (calibrated-tau already separated hack/clean, hkgap>0).
Consolidate to one adapter type: the quarantine is now delta_S_hack, the second
diagonal in the same frozen SVD basis, shape [r], capacity-matched to delta_S,
zeroed at deploy. route2's calibrated-tau gate parks the flagged rollouts' grad
into delta_S_hack.grad (like proj.py's route parks its subspace projection);
delta_S keeps the unflagged. Both diagonals train at one shared lr.
Removed: A_q/B_q params, v_act + extract_v_act, the act-mask arm (a shared
diagonal can't be per-token gated), route2_mask / route2_quarantine_rank /
route2_quar_lr_scale knobs, the separate quar optimizer group. Arm name
routing2_{act,grad} -> routing2. v_grad refresh extracts from delta_S (main)
with the quarantine ablated.
SGTM check: their gradient routing uses a hard detach on capacity-matched
reserved dims, no soft/tanh/sigmoid gate -- balance is the fix, not gating.
Smoked clean: tau/hkgap/qE render, ||delta_S_hack||>0 assert passes, exit 0.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Routing stays vector-based (cos>tau, not the detector flag) but tau is the
per-step EMA midpoint of the hack vs clean cos clouds (teacher+flagged-student
anchor hack; not-flagged anchor clean). Rides the cin drift; force-routes known
hacks; tau-routes unknown B. Logs tau + hkgap. No-cheat: detector only
calibrates, gt_pass never gates.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
DeepSeek/GPT-5.5/Gemini converge: (1) UNANIMOUS top concern -- prove the v_hack
DIRECTION is causal, not the detector flag/capacity (random-V + flag-only triad);
(2) route2-grad over-routes too (cos>0 = ~50% coin-flip by concentration, not a
granularity fix); (3) improvement B != erase only via on-policy generation, which
ablate-during-gen would remove.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
The hook gate is necessarily per-token ([G*s, r], nn.Linear flattens the
batch). _route2_grad_filter now sums each rollout's token gate-grads before
the cos(g_b, v_grad) flag, so routing is per-rollout (the preregistered GRPO
unit) and the sign is denoised. Per-token a clean rollout scatters ~50% of
tokens over cos>0 by noise, spuriously routing half its gradient mass.
Verified by deepseek-v4-pro review: gate identity, divide-out, eps-guard,
Arm B detach-route, R5 no-cheat all correct; this was the one finding.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Arm A (route2_mask=grad): per-rollout gate splice (identity at c=1) recovers
the per-sample delta_S grad after backward (c.grad = delta_S * g_b); train.py
divides it out (eps-guard |delta_S|>1e-6), flags rollouts by cos(g_b, v_grad)>0,
and SUBTRACTS them from delta_S.grad. Single-pass, no forward detach, no second
backward -- the cross-step mismatch that made the spec's A1 stale-mask awkward
never arises (routing is post-backward within the step). v_grad = unit-mean
gradient diff from extract_v_hack raw grads (gradient-space analogue of v_act).
route2 forces the combined (non-split) backward since cos_pre is NaN for it
anyway, which also gives the gate a single clean grad to read.
Drop route2_tau: never tuned; the mask is cos>0 (the natural hack-ward boundary)
and the load-time noise floor already filters axes.
v_hack path now auto-derives from --vhack-pairs-path (out/vhack/v_hack_pairset_
<stem>.safetensors): pass the pairset, the hack file auto-loads/extracts -- no
need to also pass --v-hack-path. run-substrate drops the redundant flag.
smoke: smoke-route2 (act) and new smoke-route2-grad both pass (||B_q||=0.109,
exit 0); erase shared-basis path unchanged (cout->0, fired~0.9).
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Routing-v2 spec (distinct-basis quarantine, two arms, proofs); related-work
no-cheat scorecard for TDGA/Cloud/SGTM/Confessions; full-text fetches of the
Deng and SGTM papers; journal entry for the run-31 confound + T1/T2 landing.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
External review (3 subagents) caught:
- blog: noise-floor drop is GLOBAL across modules, not per-Linear (proj.py:187)
- blog: route pseudocode used full c; route actually uses the same one-sided
gate as erase and quarantines the identical 'removed' vector (proj.py:124,199)
- spec: 'never seen by detector' -> clarify student trains on all 4 modes, the
detector just never labels C/D for v_hack extraction; cross-ref G3/task #107
Dismissed: reviewer claim that only exit_code survived (stale spec; live log
columns hk_rt/hk_so/hk_se/hk_fm confirm 4 modes) and a hallucinated 'Furthermore'.
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
Streaming table (StepLogger) redesign per user review:
- drop sprd/N/refr from the streaming view (constant / in argv / always '-')
- short names: cos_pre->cin, cos_pre_s/t->cin_s/t, cos_post->cout, gradn->gn
- 2 sig figs on loss; 1 on gn/lr
- cin/cin_s/cin_t/cout/fired only on projecting arms (no vanilla cos_post_cf)
- ADD per-mode cumulative student-hack columns hk_<rt|eq|xc|so|se|fm> on
multi-mode (substrate) runs -> shows WHICH loophole classes are learnt
- self-decoding legend() (only the columns this arm/mode-set shows)
- end-dump auto-renders any (n,d) tuple as n/d; drops sprd/N too
derisk_loopholes (#139): replace the 6 spoonfed exploit recipes with ONE
generic elicit (the faithful hint already discloses the mechanism; the model
must connect loophole+permission -> exploit = honest discoverability test) +
an exit-interview '### Notes' section, surfaced in the log (too-vague/too-blatant
hint signal).
Rename ship->deploy (Gradient Routing): the route arm's quarantine-deleted eval
is the DEPLOYED model; 'ablate' collided with the erase arm's gradient ablation.
train.py columns + row dict + plot_dynamics + plot_route_evidence + results.py;
dropped the dual-name back-compat reads.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Redesign the sentinel loophole from a module-global flag (GRADER_RESULT=PASS)
to a Solution-object .valid boolean the grader reads -- a concrete mechanism
the faithful hint can name (parallel to run_tests() naming the function),
fixing the discoverability gap (secret-token problem). Wire the stdout/file
detectors onto their natural pass-report families (_text_signals_pass /
result-named file with pass content) so the hinted exploit actually trips them.
verify_rewards: replace the 7 ad-hoc cross-mode cases with an EXHAUSTIVE 6x6
hack-x-mode matrix -- each hack must pay ONLY on its home subset (exploited iff
grade_mode==home_mode). 52/52 green; smoke green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CRIT (gpt-5.5): (1) _gt_correct keyed on a FIXED sentinel string the wrong solution
could print itself -> gt_correct=True. Now a fresh secrets.token_hex nonce per call,
unguessable, lives in outer globals not the solution namespace. (2) __is_safe used
isinstance(x,int), so an int SUBCLASS with __eq__=True passed as safe and __strict_eq
trusted it. Now exact types (type(x) in (...)). +2 verify cases (forge_sentinel,
int-subclass eq). 29/29.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CRIT (gpt-5.4 review): _gt_correct keyed correctness on exit-code-0, so a wrong
solution with os._exit(0) (uncatchable, bypasses the SystemExit guard) read
gt_correct=True in every mode -- breaking the strict oracle AND non-overlap
(a hard-exit hack looked genuinely correct everywhere). Verified the hole, then
fixed: correctness now requires REACHING a post-assert sentinel in stdout; any
early termination (sys.exit/os._exit/raise) or failing assert skips it. +3 verify
cases (os_exit @ exit_code/run_tests/sentinel), 25/25 pass.
IMPORTANT: build_substrate greedy round-robin could starve a mode when an even
assignment existed -> replaced with exact Kuhn bipartite matching, decrement
per_mode until all modes saturate, fail loud otherwise.
IMPORTANT: teacher rows stored foolable gt_pass (True on exit/eq exploits) ->
inflated teacher gt_t/PASS_RATE. Now store strict gt_correct.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
derisk #10: only exit_code is base-elicitable at scale (98%); sentinel 13.5%
(13 seeds), run_tests 2% (RL-emergent, pool-sourced), stdout/file/eq ~0. So the
teacher batch sources exit_code+sentinel from elicit files and run_tests from the
existing teacher pool. Scarcest-mode-first round-robin + pool_cap give an even
7/7/7 partition (21 problems, 40 rollouts). Spec records the elicitability finding.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Flags the non-overlap problem: gt_pass-based passed lets sys.exit/eq pay on every
subset -> must switch to passed_i = gt_correct OR channel_i with per-mode positive
detectors. Plus the per-problem env_mode gap.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
expose-K violates the paper's 3 criteria (no explicit prompting / ~0% base /
no leak); our T0 64.6% base rate is a red flag not a pass (criterion inverted).
New design: hack class = (grader flaw)+(factual hint); distinct mechanism = a
distinct GRADER mode, not a solution-side trick (C collapses into A/B). Candidate
menu M1/A/B/S/R/T + corrected de-risk bar (~0% base, emergent). expose-K code to
be ripped out.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Code writes+reads the new scheme; migrate_out_dirs.py moved 225 loose artifacts
(0 left at top level). Per-run checkpoints+rollouts now group under
runs/<ts>_<run_id>/ as train.safetensors/rollouts.jsonl. Figures land in
out/figs/ with a stable docs/figs/<name>.png symlink (figs.link_latest).
justfile also gains run-cell REFRESH param (online-erasure arm). Smoke +
smoke-vanilla + results all green on new paths. Requeue manifest preserves the
why/resolve labels that pueue reset wiped.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
out/ is 25GB/195 loose files. Target: one subdir per datatype, per-run
artifacts under runs/<ts>_<slug>/. NOT executed live: 11 queued jobs pass
out/ paths as literal args, so the data move + code-path edits run atomically
when the queue is idle. Archived the unreferenced *_OLD_step_format dirs now.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Q10: swap only pair-set content (all bases k=12/tau=0, trained k=5, seed-41
mix=0.125 frozen). prog_wide (mechanism) -0.226; semantic framings ~0; null_city
placebo +0.024. v_hack tracks the hack mechanism, not a generic honesty
direction. n=1 per row, baseline noise +/-0.06.
Q11: 60-step seed-42 mix=0.125, gap closes (vanilla 0.936, frozen 0.957,
refresh-2 0.907) -- projection delays but does not prevent hacking at this
horizon. n=1, confounded with mix/seed vs Q2.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
T5: eval_hack_solve helper + ablate_quarantine ctx; periodic ablated-eval
(hack_abl/solve_abl cols, appended so results.py indices unchanged) every
--eval-ablate-every steps; final kept-vs-ablated ROUTE EVAL BLUF. plot_dynamics
plots the ablated series for the routing arm (the coherence-gap fix: training
hack_s looks vanilla; routing only shows post-ablation).
External-review fixes (docs/spec/20260530_code_review.md):
- Critical: route now feeds delta_S the SAME g_proj as erase (was forcing
preserve_magnitude=False/overshoot=1, which diverged from erase before AdamW).
delta_S is its own AdamW param fed erase's grad, so route-ablated deployment
evolves identically to erase regardless of AdamW non-linearity. Only the
combined training forward over-moves (intended; never deployed). Corrected the
overclaiming docstrings (no "sum == g" / "reproduces vanilla" identity).
- Important: clip_grad_norm_ now covers delta_params + delta_hack_params
(no-op for none/erase; bounds the route update).
- Important: results.py paired-delta table includes routing (keyed on arm).
smoke route/erase/vanilla green: dsh route=0.0105 erase/none=0, span=2.9e-7,
ROUTE EVAL BLUF prints.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stage-1 (T3) of the routing spec. Adds a per-module quarantine knob
delta_S_hack (AntiPaSTO forward = delta_S + delta_S_hack, both 0 at init).
intervention=route parks the hack-ward grad component (g - cV to delta_S,
cV to delta_S_hack) instead of erasing it; eval ablates delta_S_hack.
- proj.py: route flag splits the grad (overshoot=1, no rescale -> the split
sums to g, so the training forward still moves hack-ward; route ⊇ erase).
- antipasto.py: second trainable knob, identity preserved at init.
- train.py: arm -> intervention {none,erase,route}; arm kept as a derived
display name so run-id/BLUF/results.py/plot classify are unchanged. opt
steps both knobs (hack knob grad=None under none/erase -> AdamW skips it,
so erase reproduces old `projected` bit-for-bit, R4). R3 span assert
(resid/||gh|| < 1e-4) + end-of-run ||delta_S_hack|| guard (route >0).
- results.py / plot_dynamics.py: read arm from the preset line (covers both
old --arm and new --intervention logs); plot classifies `routing`.
smoke: none ||dsh||=0, erase clean, route ||dsh||=0.0105 span=2.9e-7. 64
archived projected rows still parse.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
wassname