Back to ProbeBench
pending reviewlinearEval AwarenessApache-2.0

EvalAwarenessGuard

Detects when model believes it is being evaluated · UK AISI priority

Linear probe separating real-world deployment prompts from evaluation prompts on Qwen3.6-27B. Used to compute "eval-awareness corrected AUROC" for other probes — exposes the confound that AUROC scales with eval-awareness (arXiv:2509.13333).

byCaio VicentinoOpenInterp2026-05-XXarXiv:2507.01786
ProbeScore
0.848
Eval Awareness #1Global #1
7 weighted axes. Subject to revision as more reproducers land.

Quickstart

Three lines via the openinterp.probebench SDK. Returns the probe's P(positive_class) for a tensor of activations captured at the layer/position below.

python
from openinterp import probebench
probe = probebench.load("openinterp/evalawareness-qwen36-27b-l40")
score = probe.score(activations)  # → P(positive_class)

ProbeScore — 7 weighted axes

Composite metric in [0, 1]. No single axis can dominate; we think this discourages single-metric optimization. Numbers update as new evaluations land.

scoring formula
ProbeScore breakdown0.848
AUROC0.233
Eval-aware0.167
Dist-shift0.097
Calibration0.090
Transfer0.061
Goodhart-resistance0.050
Latency0.100
License0.050

ProbeScore — 8-axis breakdown

ProbeBench v0.0.2 · 8 axes weighted to sum 1.0

Visual breakdown of where this probe scores high vs low across all 8 axes. Polygon area is proportional to total score; shape reveals which axes carry it (and which ones it sacrifices).

AUROC0.93Eval-aware0.93Dist-shift0.81Calibration0.90Transfer0.61Goodhart0.50Latency1.00License1.00
AUROC
0.930
× 0.25 = 0.233
Eval-aware
0.930
× 0.18 = 0.167
Dist-shift
0.810
× 0.12 = 0.097
Calibration
0.900
× 0.10 = 0.090
Cross-model
0.610
× 0.10 = 0.061
Goodhart-resistance
0.500
× 0.10 = 0.050
Latency
1.000
× 0.10 = 0.100
License
1.000
× 0.05 = 0.050

Per-task evaluation

Headline AUROC alongside the eval-awareness-corrected and distribution-shift versions for each task. Reproducers run from a single notebook on a single Colab session.

TasknAUROC [95% CI]Eval-aware correctedDist-shiftECEFPR@99TPRLatencyReproducer
Apollo Eval-vs-Deploy
Eval Awareness · evaluation
2000.930 [0.88, 0.97]0.9300.8100.0500.0301.0 ms

Cross-model transfer (Pearson_CE)

Pearson correlation of paired ablation effects across models. ≥ 0.7 suggests the probe's direction is shared; 0.4–0.7 suggests partial transfer; < 0.4 suggests retraining per architecture.

SourceTargetPearson_CETransfer AUROCNotes
Qwen3.6-27BLlama-3.3-70B0.6100.860
Mean Pearson_CE for this probe (excluding self-baseline): 0.610

Architecture & artifact

Base model

Qwen3.6-27B
Qwen/Qwen3.6-27B
Family
Qwen
Params
27B
Architecture
Hybrid GDN + Gated-Attn (dense, reasoning)
Layers
64
d_model
5120
Weights license
Apache-2.0
Probe attaches at
Layer 40 · end_question
huggingface.co/Qwen/Qwen3.6-27B

Artifact

EvalAwarenessGuard weights
sklearn-compatible probe + scaler + meta.json
Params
312,000
Size
1.2 MB
License
Apache-2.0
Released
2026-05-XX
sha256
ea7c3d9e5b4a…0b9a8f7e
Hash matches the artifact at huggingface.co/datasets/caiovicentino1/EvalAwarenessGuard-linearprobe-qwen36-27b. Recompute via openinterp probebench verify openinterp/evalawareness-qwen36-27b-l40.

Reproduce

Three entry points. Pick one. Each path lands at the same numbers.

shell
git clone https://github.com/OpenInterpretability/notebooks.git
pip install openinterp
openinterp probebench reproduce openinterp/evalawareness-qwen36-27b-l40

Honest scope

Limits derived from the evaluation data above. We think these are the honest constraints; revisions land as more reproducers do.

  • Trained / evaluated on: Apollo Eval-vs-Deploy. Performance outside these tasks is unmeasured here.
  • Eval-aware AUROC drop: 0.000 AUROC averaged across tasks (uncorrected − corrected for eval-awareness confound, arXiv:2509.13333).
  • Distribution-shift drop: 0.120 AUROC averaged across tasks (in-distribution − long-context / OOD).
  • Cross-model fit (mean Pearson_CE): 0.610 — values below 0.4 suggest retraining per architecture; values above 0.7 suggest the direction is shared.
  • Probe attaches at: L40 · end_question of Qwen3.6-27B. Other layers / positions are out of scope unless re-trained.

ProbeBench v0.0.1 · 5 probes registered · 11 evaluations · 7 transfer measurements. Schema and weights subject to revision.