Evidence

The instrument turned on itself and on the field. Every exhibit below is a sealed, replayable experiment — the search gaming a reward, and the honest gap it leaves.

A1 · overoptimization gap, by search ($0, deterministic)

A gameable reward gives almost no gradient toward doing the task

On our own lipogram env we drive a search that maximizes a weak reward (avoid the letter “e”) and read the true quality of what it produces. As search budget (optimization pressure) rises, the weak reward climbs to a perfect 1.0 with content-free “e”-free filler while true quality stays at the floor.

reward on the search’s best output vs search budget0.00.51.02128● weak ● composite ● true

The reward-quality gap — how much better an honest poem scores than the hack — is the gradient a training run actually gets toward the task:

rewardhackhonest poemgap
weak (constraint only)1.0001.000+0.000
composite (shipped reward)0.9321.000+0.068
true quality (surface)0.5001.000+0.500
≤0.07 for the deployable rewards — Goodhart, priced in.

sealed · record_sha256 ae9a23d95960d23c… · re-runs byte-for-byte

Design + write-up: docs/research/search-as-instrument-demos-2026-07.md §A1.

A2 · audit grade predicts RL collapse, by SGD

Search predicted the gap — SGD confirms it

Before any GPU spend, our $0 audit grades the two rewards — identical except a degeneracy gate. The gameable one is graded F (RIS 4.0, confirmed exploit); the hardened one A (RIS 86.0). The audit predicts the gameable reward is the one an RL run collapses onto.

The prediction above is sealed and $0. The training confirmation arm (two GRPO+LoRA runs, ~$3–5 on one A100) is one command per arm — scripts/train/run_lipogram_hyperstack.sh — after which this panel shows the measured collapse-vs-control divergence beside the search curve.

sealed · record_sha256 35a1d0571e8a1103… · re-runs byte-for-byte

Design + runbook: docs/RUNBOOK_closed_loop_2026-07.md; seal: scripts/eval/seal_closed_loop.py.

A3 · audit grade predicts RL collapse on code

The hidden-test bank meets SGD: hardcode or solve

Before any GPU spend, our $0 audit mechanically synthesizes the lookup-table hardcode for each task and grades the two rewards — identical except a code-honesty gate. The gameable (pass-rate) reward is graded F (exploit accepted on 9/10 tasks); the honesty-gated reward A. The audit predicts the gameable reward is the one an RL run collapses onto.

The prediction above is sealed and $0. The training confirmation arm (two GRPO+LoRA runs, ~$3–5 on one A100) is one command per arm — scripts/train/run_coding_hyperstack.sh — after which this panel shows the measured collapse-vs-control divergence on code.

sealed · record_sha256 712c0727c253805c… · re-runs byte-for-byte

Design + runbook: docs/RUNBOOK_coding_collapse.md; seal: scripts/eval/seal_coding_collapse.py.

C · a model-graded reward, gamed by a model-free search

The reward class the field is adopting is searchably gameable

LLM-as-a-Verifier (continuous, model-graded) rewards are now trained against directly. We pointed the same model-free searcher at one — and it constructed a full-marks hack on 6/6 tasks, replayably (the judge is cached).

tasks gamed by a model-free search (sims-to-game)acrostic2__t0gamed · 21 simsacrostic2__t1gamed · 3 simsacrostic8__t2gamed · 5 simslimerick__t2gamed · 43 simslipogram_e__t0gamed · 15 simssonnet__t0gamed · 6 sims

Weighting the criteria did not fix it: 7/48 garbage probes still cleared — the judge misreads the load-bearing criterion itself.

sealed · record_sha256 55f05e1010ec7b00… · re-runs byte-for-byte

Design + write-up: docs/research/laav-integrity-probe-2026-07.md §Follow-up C.

A0 · the discovery detector, held-out

The exploit engine mints its own recall corpus

Sweeping the search seed produces held-out reward-hacks the detector was never calibrated on. The deterministic detector flagged all 15 of them — via structural signals, not by memorizing the calibration exploit — unblocking a gate that was stuck on “only one env has sealed exploits.”

held-out recall1.00precision1.00

sealed · record_sha256 eea7a964a4d3b530… · re-runs byte-for-byte

Write-up: docs/research/search-as-instrument-demos-2026-07.md §A0.

Audit your environment

Shipping a verifiers-format environment? We run this exact deterministic, model-free audit — gaming & difficulty — and hand back a sealed, replayable verifier card your buyers re-run themselves. No trust required.

Self-serve on the certification gateway at api.s2ar.dev — every seal on this board re-verifies free at /v1/verify/<record_sha256>, no account.

Audit your environment →