evil_MoE/docs/results.md

# Results, organized by the question each run answers

Deploy-eval is the headline metric: knob-off forward on the recency-clean held-out
TEST set (ids>=3243, base solve ~0.1, n=119), single-mode `run_tests` env, Qwen3-4B.
Regenerate the table with `just results-deploy` (scripts/results_deploy.py, auto-discovers
every `out/runs/*/deploy_test.json`); `just results` gives the live training-hack table.

OLD eval1 results (training-hack metric, the Q1-Q13 mechanism/basis/refresh studies on the
pre-recency-clean eval) are archived in [results_eval1_archive.md](results_eval1_archive.md).

## How to read this

- **Headline = solve_deploy - hack_deploy** (both alone are gameable; the gap is the honest
  objective: solve the task without learning the cheat). Higher is better.
- **hack_deploy / solve_deploy** = knob-off deploy eval (quarantine deleted for routeV; the
  trained model for vanilla) on test n=119. **hack_train / solve_train** = last-5-step student
  rates during training (converged regime).
- **All rows n=1** (single seed 43) unless noted; treat gaps <=0.03 as noise.
- Watch the **pool/pairs confound** across rows (see `argv` / `train_set`); only same-axis
  pairs are A/B-comparable (called out in the answer).

---

## Q14. 🥇 routeV deploy on the recency-clean eval2 test set (the current headline)

<!-- METRIC: deploy_test.json, knob-off forward (quarantine deleted for routeV; trained model
     for vanilla), eval_set=test = recency-clean held-out ids>=3243 (base solve ~0.1), n=119,
     60-step fast preset, Qwen3-4B, single-mode run_tests env, seed 43. NOT comparable to Q12
     (old n=64 eval, pre the 2026-05-23 grader-bug / recency-clean fix that moved base solve
     0.94->0.1). REGENERATE: `just results-deploy` (scripts/results_deploy.py) auto-discovers
     every out/runs/*/deploy_test.json -- this table is a curated copy of that output.
     Smoke runs (seed 41, steps 30, tiny-random, hack=0) are excluded.
     completed src: _dir6_routeV_s43 (job 8) / _dir6_routeV_pertoken_s43 (job 9) /
       _dir6_routeV_random_s43 (job 10) / _dir8_routeV_authored_perroll_s43 (job 15).
     pending: _dir8_vanilla_s43 (16 RUNNING) / _dir8_routeV_actvote_authored_s43 (19) /
       _dir8_lora_routeV_authored_s43 (20) / _dir8_routeV_randomV_authored_s43 (21). commit e26f5fe. -->

Everything above (Q1-Q13) is on the OLD eval. Q12's route2 numbers used n=64 prompts before the
recency-clean fix; the env is now single-mode `run_tests` and the held-out test set is
recency-clean (ids>=3243, base solve ~0.1). This is the corrected substrate. All rows: seed 43,
60 steps, deploy = knob-off forward on test n=119. Headline = solve_deploy - hack_deploy.
Note the pool/pairs confound across rows (see `argv`); the only single-axis A/Bs are called out
in the answer.

| arm | pairs | gran | hack ↓ | solve ↑ | headline |
| :-- | :-- | :-- | --: | --: | --: |
| routeV per-token   | prog_wide | per-token   | 0.042 | 0.143 | +0.101 |
| routeV authored    | authored  | per-rollout | 0.076 | 0.118 | +0.042 |
| routeV prog_wide   | prog_wide | per-rollout | 0.101 | 0.126 | +0.025 |
| routeV random-V    | prog_wide (Haar dir) | per-rollout | 0.101 | 0.109 | +0.008 |
| vanilla GRPO       | -- | -- | running (job 16) | | |
| routeV act_vote    | authored | per-rollout (global vote) | queued (19) | | |
| routeV LoRA-B      | authored | per-rollout | queued (20) | | |
| routeV random-V    | authored (Haar dir) | per-rollout | queued (21) | | |

**Answer: three single-axis reads already hold; the suppression magnitude waits on vanilla.**
- *Direction doesn't matter at per-rollout (H2 absorption, on eval2):* real-V (prog_wide, 0.101)
  == random-V (prog_wide, 0.101). Replicates the old-eval H2 result on the clean set. The open
  question is whether AUTHORED direction matters (job 21 random-V-authored vs job 15) -- queued.
- *Pairs matter:* authored per-rollout 0.076 < prog_wide per-rollout 0.101 (clean A/B -- same
  granularity, same dense pool, differ only in pairs). Authored helps even though real-vs-random
  didn't, so the gain is the pair CONTENT, not direction sharpness.
- *Granularity matters most:* per-token 0.042 < per-rollout 0.101 (both prog_wide). The arm the
  per-token ablation -- lowest deploy hack AND highest solve (0.143) of any completed run.

All single seed (n=1), so treat <=0.03 gaps as noise. NOT yet interpretable as suppression until
vanilla (job 16, running): if vanilla deploy hack is ~0.10 then even per-token's 0.042 is only a
modest cut over base; if vanilla is high (>0.3 as on the old eval), all routeV arms suppress
strongly. Pairs separability (orthogonal, job 17): authored_all p@10=0.70 beats prog_wide 0.20
(`out/diag/pairs_compare.csv`).

Training-`rout` note (not deploy): grad-cosine routing cliffs (0.63@step6 -> 0.09@step20, GRPO
advantage flattening); act_vote sustains late (0.88@step17) by gating on activations -- see
RESEARCH_JOURNAL 2026-06-08. Whether that converts to deploy suppression is what job 19 tests.

## Dynamics note (sizing the convergence test)

Per-step trajectories (mix=0.125 g8, seed 41): `hack_s` rises 0→~0.6-0.75 and
**plateaus by step ~13-16**; `gt_s` (solve) stays **noisy-flat at ~0.1-0.5 the
whole run, it never climbs**. The attractor in this surrogate regime is full
*hack*, not full solve — so "run until full solve" has no target. The
convergence question is therefore: once vanilla hack plateaus (~step 15), does
projected stay below it or catch up? A 60-step run (~2.2h at g8) sees 3x past
the plateau; a 1000-step run (~36h) is wasteful.

## Open / queued (no result yet)

- **convergence at ≥3 seeds (#121)**: the n=1 seed-42 run (Q11) shows the gap
  closing by step 60, but that could be a seed-42 high-hack draw. Need 2+ more
  seeds before concluding the suppression erodes vs survives.
- **pairset content at ≥3 seeds (#122)**: Q10's mechanism>framing>placebo
  ordering is n=1 per row; replicate `prog_wide` and the placebo on 2+ seeds.
- **route arm at scale (#182)**: running; validates routing's ablated-eval
  hack<kept on Qwen3-4B before the 3-way none/erase/route cells (#130).
- **k-slice (k=1/2/5)**: only smoke-tested, no 4B results.
- **Stage 2/3 cross-*mechanism* generalisation**: the load-bearing test --
  extract v_hack from hack A, check it stops the *unknown* hack B the student
  would otherwise learn. Q10 (held-out *framing*) is a weaker cousin.