Datasetpublic

G-reen/fastdetector-val-stat-val

Auto-Generated FastDetector Dataset Dataset: G-reen/fastdetector-val-stat-val Globals Config: config/globals_val.toml Analysis Config: config/analysis.toml Rows: 12,271 Evaluation Results Prompt Subsets: 4 (direct_reference, indirect_reference, revise, rewrite) Generator Configs: 7 (Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6), Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7), claude-sonnet-4-5-20250929 (Temp: Unknown), claude-sonnet-5 (Temp: Unknown), gpt-3.5-turbo… See the full description on the dataset page: https://huggingface.co/datasets/G-reen/fastdetector-val-stat-val.

sourceHugging Faceupdated 3h agoView on Hugging Face

0likes574downloads

Dataset Card

Auto-Generated FastDetector Dataset

—Dataset: G-reen/fastdetector-val-stat-val
—Globals Config: config/globals_val.toml
—Analysis Config: config/analysis.toml
—Rows: 12,271

Evaluation Results

—Prompt Subsets: 4 (directreference, indirectreference, revise, rewrite)
—Generator Configs: 7 (Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6), Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7), claude-sonnet-4-5-20250929 (Temp: Unknown), claude-sonnet-5 (Temp: Unknown), gpt-3.5-turbo (Temp: Unknown), gpt-4o (Temp: Unknown), hy3 (Temp: Unknown))
—Classifiers: 19 (EditLens Roberta-Large Score, EditLens Roberta-Large Bucket, EditLens Llama-3.2-3B Score, EditLens Llama-3.2-3B Bucket, EditLens Giga RoBERTa-Large Score, EditLens Giga RoBERTa-Large Bucket, EditLens Giga Llama-3.2-3B Score, EditLens Giga Llama-3.2-3B Bucket, Perplexity (Llama-3.2-3B-Instruct), Perplexity (Llama-3.2-3B), Entropy (Llama-3.2-3B-Instruct), Entropy (Llama-3.2-3B), Top-p Outliers (Llama-3.2-3B-Instruct), Top-p Outliers (Llama-3.2-3B), Top-k Outliers (Llama-3.2-3B-Instruct), Top-k Outliers (Llama-3.2-3B), FastDetectGPT (Llama-3.2-3B-Instruct), FastDetectGPT (Llama-3.2-3B), Binoculars)
—Filter Conditions: cosdist >= 0.03 OR softngram >= 0.06
—Evaluation / Validation Rows: 11,043 / 1,228 (validation_size = 0.1)
—Base Columns: original (Human), final_response (AI)

The best classifier was EditLens Giga Llama-3.2-3B Score with an AUROC of 0.9521. The hardest prompt subset was rewrite with a TPR of 0.1685, and the hardest generator config was hy3 (Temp: Unknown) with a TPR of 0.2105.

Classifier	Threshold	AUROC	TPR	FPR	Accuracy	F1
✔️ EditLens Giga Llama-3.2-3B Score	0.3188	0.9521	0.6328	0.0028	0.8150	0.7738
EditLens Llama-3.2-3B Score	0.3526	0.9352	0.6510	0.0056	0.8227	0.7859
EditLens Giga RoBERTa-Large Score	0.6381	0.9268	0.5297	0.0075	0.7611	0.6891
EditLens Roberta-Large Score	0.5167	0.9039	0.4454	0.0080	0.7187	0.6130
EditLens Llama-3.2-3B Bucket	0.0000	0.8702	0.7540	0.0235	0.8653	0.8484
EditLens Giga RoBERTa-Large Bucket	0.0000	0.8609	0.7409	0.0281	0.8564	0.8377
EditLens Giga Llama-3.2-3B Bucket	0.0000	0.8523	0.7079	0.0052	0.8514	0.8265
EditLens Roberta-Large Bucket	0.0000	0.8193	0.6775	0.0576	0.8100	0.7810
Perplexity (Llama-3.2-3B-Instruct)	4.4416	0.6590	0.0721	0.0024	0.5348	0.1342
Entropy (Llama-3.2-3B-Instruct)	1.4199	0.6561	0.0867	0.0033	0.5417	0.1590
Top-k Outliers (Llama-3.2-3B-Instruct)	0.0212	0.6316	0.0635	0.0026	0.5304	0.1191
Perplexity (Llama-3.2-3B)	1.2539	0.6053	0.0003	0.0012	0.4995	0.0005
Binoculars	1.0377	0.6047	0.0139	0.0043	0.5048	0.0272
Top-p Outliers (Llama-3.2-3B)	0.0026	0.6046	0.0071	0.0025	0.5023	0.0140
Entropy (Llama-3.2-3B)	0.3161	0.5986	0.0004	0.0017	0.4993	0.0007
Top-k Outliers (Llama-3.2-3B)	0.0045	0.5871	0.0054	0.0020	0.5017	0.0108
Top-p Outliers (Llama-3.2-3B-Instruct)	0.0244	0.5437	0.0243	0.0060	0.5091	0.0471
FastDetectGPT (Llama-3.2-3B-Instruct)	3.2075	0.5106	0.0050	0.0054	0.4998	0.0099
❗ FastDetectGPT (Llama-3.2-3B)	-3.0452	0.4278	0.0115	0.0052	0.5032	0.0226

Classifier metrics averaged within each prompt and generator subset:

Subset	Average AUROC	Average TPR	Average FPR	Average Accuracy	Average F1
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	0.7756	0.3731	0.0096	0.6818	0.4244
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	0.7427	0.2729	0.0093	0.6318	0.3590
Prompt: revise	0.7341	0.3286	0.0088	0.6599	0.3695
Prompt: direct_reference	0.7329	0.3292	0.0094	0.6599	0.3939
Prompt: indirect_reference	0.7303	0.2819	0.0081	0.6369	0.3578
Model: gpt-3.5-turbo (Temp: Unknown)	0.7281	0.3027	0.0083	0.6472	0.3734
Model: gpt-4o (Temp: Unknown)	0.7254	0.3201	0.0087	0.6557	0.3658
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	0.7077	0.2659	0.0090	0.6284	0.3272
Model: claude-sonnet-5 (Temp: Unknown)	0.6947	0.2400	0.0098	0.6151	0.3028
Prompt: rewrite	0.6329	0.1685	0.0108	0.5788	0.2378
❗ Model: hy3 (Temp: Unknown)	0.6092	0.2105	0.0097	0.6004	0.2780

✔️ marks the best AUROC, ❗ the worst.

Statistics of Interest

[image] [image] [image] [image]

Appendix

Table of contents 1. Univariate Analysis 2. Correlation Heatmap 3. Distance Histograms 4. Distance Histograms per Prompt Subset 5. Distance Histograms per Generator Config Subset 6. Classifier: EditLens Roberta-Large Score - Performance: - Thresholding: - Classification Histograms: 7. Classifier: EditLens Roberta-Large Bucket - Performance: - Thresholding: - Classification Histograms: 8. Classifier: EditLens Llama-3.2-3B Score - Performance: - Thresholding: - Classification Histograms: 9. Classifier: EditLens Llama-3.2-3B Bucket - Performance: - Thresholding: - Classification Histograms: 10. Classifier: EditLens Giga RoBERTa-Large Score - Performance: - Thresholding: - Classification Histograms: 11. Classifier: EditLens Giga RoBERTa-Large Bucket - Performance: - Thresholding: - Classification Histograms: 12. Classifier: EditLens Giga Llama-3.2-3B Score - Performance: - Thresholding: - Classification Histograms: 13. Classifier: EditLens Giga Llama-3.2-3B Bucket - Performance: - Thresholding: - Classification Histograms: 14. Classifier: Perplexity (Llama-3.2-3B-Instruct) - Performance: - Thresholding: - Classification Histograms: 15. Classifier: Perplexity (Llama-3.2-3B) - Performance: - Thresholding: - Classification Histograms: 16. Classifier: Entropy (Llama-3.2-3B-Instruct) - Performance: - Thresholding: - Classification Histograms: 17. Classifier: Entropy (Llama-3.2-3B) - Performance: - Thresholding: - Classification Histograms: 18. Classifier: Top-p Outliers (Llama-3.2-3B-Instruct) - Performance: - Thresholding: - Classification Histograms: 19. Classifier: Top-p Outliers (Llama-3.2-3B) - Performance: - Thresholding: - Classification Histograms: 20. Classifier: Top-k Outliers (Llama-3.2-3B-Instruct) - Performance: - Thresholding: - Classification Histograms: 21. Classifier: Top-k Outliers (Llama-3.2-3B) - Performance: - Thresholding: - Classification Histograms: 22. Classifier: FastDetectGPT (Llama-3.2-3B-Instruct) - Performance: - Thresholding: - Classification Histograms: 23. Classifier: FastDetectGPT (Llama-3.2-3B) - Performance: - Thresholding: - Classification Histograms: 24. Classifier: Binoculars - Performance: - Thresholding: - Classification Histograms:

Univariate Analysis

Every statistic the report does arithmetic on, over the 11,043-row evaluation split. Invalid counts rows whose value is missing or non-finite; those rows are excluded from the other columns.

Statistic	N	Mean	Median	Std	Min	Max
jaccard_1	11,043	0.6774	0.7622	0.2268	0.0000	1.0000
jaccard_2	11,043	0.8003	0.9099	0.2345	0.0216	1.0000
levenshtein	11,043	1990.5932	1355.0000	2350.3864	11.0000	65876.0000
softngram	11,043	0.6141	0.6751	0.3128	0.0000	1.0000
cosdist	11,043	0.2197	0.1545	0.2016	-0.0047	1.0315
bertscore	11,043	0.1463	0.1529	0.0689	0.0041	0.4338
bertscore_precision	11,043	0.1439	0.1483	0.0679	0.0036	0.5106
bertscore_recall	11,043	0.1475	0.1534	0.0760	0.0014	0.4319
moverscore	11,043	0.5620	0.6027	0.1646	0.1011	1.0810
reranker	11,043	-2.8248	-4.6875	5.8203	-11.1875	19.0000
EditLens Roberta-Large Score (Human)	11,043	0.0582	0.0205	0.0989	0.0064	0.9993
EditLens Roberta-Large Score (AI)	11,043	0.5103	0.4360	0.3677	0.0065	0.9996
EditLens Roberta-Large Bucket (Human)	11,043	0.0697	0.0000	0.3182	0.0000	3.0000
EditLens Roberta-Large Bucket (AI)	11,043	1.4634	1.0000	1.2944	0.0000	3.0000
EditLens Llama-3.2-3B Score (Human)	11,043	0.0455	0.0225	0.0627	0.0003	0.8558
EditLens Llama-3.2-3B Score (AI)	11,043	0.5326	0.5065	0.3411	0.0011	1.0000
EditLens Llama-3.2-3B Bucket (Human)	11,043	0.0245	0.0000	0.1638	0.0000	3.0000
EditLens Llama-3.2-3B Bucket (AI)	11,043	1.5378	1.0000	1.1914	0.0000	3.0000
EditLens Giga RoBERTa-Large Score (Human)	11,043	0.0258	0.0049	0.0962	0.0013	0.9999
EditLens Giga RoBERTa-Large Score (AI)	11,043	0.5949	0.7005	0.3972	0.0014	1.0000
EditLens Giga RoBERTa-Large Bucket (Human)	11,043	0.0448	0.0000	0.3034	0.0000	3.0000
EditLens Giga RoBERTa-Large Bucket (AI)	11,043	1.7717	2.0000	1.2732	0.0000	3.0000
EditLens Giga Llama-3.2-3B Score (Human)	11,043	0.0118	0.0039	0.0359	0.0004	0.9339
EditLens Giga Llama-3.2-3B Score (AI)	11,043	0.5171	0.4856	0.3743	0.0007	1.0000
EditLens Giga Llama-3.2-3B Bucket (Human)	11,043	0.0064	0.0000	0.1001	0.0000	3.0000
EditLens Giga Llama-3.2-3B Bucket (AI)	11,043	1.5328	1.0000	1.2510	0.0000	3.0000
Perplexity (Llama-3.2-3B-Instruct) (Human)	11,043	19.5039	16.0945	16.2568	1.6941	757.4117
Perplexity (Llama-3.2-3B-Instruct) (AI)	11,043	15.3748	12.1771	18.1414	1.0586	990.4898
Perplexity (Llama-3.2-3B) (Human)	11,043	14.4906	12.4128	10.3796	1.0484	485.7397
Perplexity (Llama-3.2-3B) (AI)	11,043	12.8463	10.4403	12.8358	1.0502	551.9839
Entropy (Llama-3.2-3B-Instruct) (Human)	11,043	2.6259	2.5882	0.4948	0.4899	7.6310
Entropy (Llama-3.2-3B-Instruct) (AI)	11,043	2.3039	2.3097	0.6394	0.0945	6.7115
Entropy (Llama-3.2-3B) (Human)	11,043	2.5256	2.5162	0.4527	0.0784	5.6794
Entropy (Llama-3.2-3B) (AI)	11,043	2.3627	2.3732	0.5247	0.0607	6.0152
Top-p Outliers (Llama-3.2-3B-Instruct) (Human)	11,043	0.0585	0.0556	0.0171	0.0000	0.1746
Top-p Outliers (Llama-3.2-3B-Instruct) (AI)	11,043	0.0558	0.0538	0.0184	0.0000	0.2000
Top-p Outliers (Llama-3.2-3B) (Human)	11,043	0.0431	0.0426	0.0129	0.0000	0.1304
Top-p Outliers (Llama-3.2-3B) (AI)	11,043	0.0385	0.0381	0.0155	0.0000	0.1667
Top-k Outliers (Llama-3.2-3B-Instruct) (Human)	11,043	0.1076	0.1010	0.0450	0.0000	0.5435
Top-k Outliers (Llama-3.2-3B-Instruct) (AI)	11,043	0.0881	0.0809	0.0522	0.0000	0.5481
Top-k Outliers (Llama-3.2-3B) (Human)	11,043	0.0882	0.0823	0.0405	0.0000	0.4565
Top-k Outliers (Llama-3.2-3B) (AI)	11,043	0.0782	0.0703	0.0479	0.0000	0.5294
FastDetectGPT (Llama-3.2-3B-Instruct) (Human)	11,043	-1.6809	-1.6693	1.6419	-15.3432	25.5512
FastDetectGPT (Llama-3.2-3B-Instruct) (AI)	11,043	-1.6320	-1.6166	1.6742	-15.5276	10.1506
FastDetectGPT (Llama-3.2-3B) (Human)	11,043	-0.1344	-0.1042	1.0496	-6.4372	5.0980
FastDetectGPT (Llama-3.2-3B) (AI)	11,043	0.2291	0.1899	1.5314	-15.9713	10.6377
Binoculars (Human)	11,043	0.8558	0.8638	0.0974	0.0074	1.1340
Binoculars (AI)	11,043	0.8842	0.8839	0.0715	0.0480	1.2209

Correlation Heatmap

Pearson correlation between every statistic of interest, computed over the rows where both statistics are present.

[image]

Distance Histograms

[image] [image] [image] [image] [image] [image] [image] [image] [image] [image]

Distance Histograms per Prompt Subset

[image] [image] [image] [image] [image] [image] [image] [image] [image] [image]

Distance Histograms per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Roberta-Large Score

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.9039	0.4454	0.0080	0.7187	0.6130
Prompt: direct_reference	6,166	0.9029	0.5420	0.0071	0.7674	0.6997
Prompt: indirect_reference	5,584	0.9101	0.4628	0.0057	0.7285	0.6302
Prompt: revise	6,032	0.9654	0.5020	0.0090	0.7465	0.6645
❗ Prompt: rewrite	4,304	0.8080	0.2054	0.0107	0.5974	0.3378
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.8669	0.3823	0.0072	0.6875	0.5502
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9699	0.5792	0.0074	0.7859	0.7301
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.8969	0.4632	0.0074	0.7279	0.6300
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.8855	0.3462	0.0090	0.6686	0.5109
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.9006	0.4756	0.0073	0.7342	0.6415
Model: gpt-4o (Temp: Unknown)	3,308	0.9484	0.5677	0.0097	0.7790	0.7198
Model: hy3 (Temp: Unknown)	3,060	0.8471	0.2771	0.0078	0.6346	0.4313

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.5167.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Roberta-Large Bucket

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.8193	0.6775	0.0576	0.8100	0.7810
Prompt: direct_reference	6,166	0.8347	0.7035	0.0597	0.8219	0.7980
Prompt: indirect_reference	5,584	0.7998	0.6329	0.0527	0.7901	0.7510
✔️ Prompt: revise	6,032	0.9047	0.8481	0.0570	0.8956	0.8904
❗ Prompt: rewrite	4,304	0.7023	0.4591	0.0618	0.6987	0.6037
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7522	0.5427	0.0548	0.7440	0.6795
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9013	0.8375	0.0610	0.8883	0.8823
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.8171	0.6705	0.0548	0.8078	0.7772
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.7867	0.6173	0.0558	0.7808	0.7379
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.8150	0.6669	0.0607	0.8031	0.7721
Model: gpt-4o (Temp: Unknown)	3,308	0.8922	0.8174	0.0532	0.8821	0.8739
Model: hy3 (Temp: Unknown)	3,060	0.7526	0.5569	0.0627	0.7471	0.6877

Thresholding:

—Direction: higher_is_ai
—Swept for f1 with a found threshold of 0.0000.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Llama-3.2-3B Score

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.9352	0.6510	0.0056	0.8227	0.7859
Prompt: direct_reference	6,166	0.9467	0.7506	0.0029	0.8738	0.8561
Prompt: indirect_reference	5,584	0.9277	0.6089	0.0061	0.8014	0.7540
Prompt: revise	6,032	0.9823	0.7891	0.0063	0.8914	0.8790
❗ Prompt: rewrite	4,304	0.8595	0.3694	0.0079	0.6808	0.5364
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.9130	0.5682	0.0039	0.7821	0.7228
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9845	0.8444	0.0057	0.9193	0.9128
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.9187	0.6163	0.0053	0.8055	0.7601
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.9184	0.5564	0.0103	0.7731	0.7103
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.9382	0.6788	0.0013	0.8387	0.8080
Model: gpt-4o (Temp: Unknown)	3,308	0.9670	0.7618	0.0036	0.8791	0.8630
Model: hy3 (Temp: Unknown)	3,060	0.9000	0.4954	0.0092	0.7431	0.6586

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.3526.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Llama-3.2-3B Bucket

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.8702	0.7540	0.0235	0.8653	0.8484
Prompt: direct_reference	6,166	0.8993	0.8067	0.0195	0.8936	0.8835
Prompt: indirect_reference	5,584	0.8477	0.7081	0.0233	0.8424	0.8180
Prompt: revise	6,032	0.9456	0.9038	0.0225	0.9406	0.9384
❗ Prompt: rewrite	4,304	0.7510	0.5279	0.0307	0.7486	0.6774
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.8344	0.6830	0.0235	0.8297	0.8005
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9537	0.9162	0.0251	0.9456	0.9439
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.8476	0.7099	0.0234	0.8432	0.8191
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.8378	0.6968	0.0282	0.8343	0.8079
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.8803	0.7698	0.0185	0.8757	0.8609
Model: gpt-4o (Temp: Unknown)	3,308	0.9179	0.8452	0.0212	0.9120	0.9057
Model: hy3 (Temp: Unknown)	3,060	0.8037	0.6261	0.0242	0.8010	0.7588

Thresholding:

—Direction: higher_is_ai
—Swept for f1 with a found threshold of 0.0000.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Giga RoBERTa-Large Score

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.9268	0.5297	0.0075	0.7611	0.6891
Prompt: direct_reference	6,166	0.9299	0.6656	0.0084	0.8286	0.7952
Prompt: indirect_reference	5,584	0.9233	0.5494	0.0079	0.7708	0.7056
Prompt: revise	6,032	0.9780	0.5676	0.0056	0.7810	0.7216
❗ Prompt: rewrite	4,304	0.8524	0.2560	0.0084	0.6238	0.4050
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.8938	0.4866	0.0039	0.7414	0.6530
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9817	0.7064	0.0051	0.8506	0.8254
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.9243	0.5114	0.0074	0.7520	0.6734
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.9095	0.4429	0.0128	0.7151	0.6085
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.9228	0.5838	0.0073	0.7883	0.7338
Model: gpt-4o (Temp: Unknown)	3,308	0.9593	0.6258	0.0060	0.8099	0.7670
Model: hy3 (Temp: Unknown)	3,060	0.8870	0.3190	0.0105	0.6542	0.4798

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.6381.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Giga RoBERTa-Large Bucket

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.8609	0.7409	0.0281	0.8564	0.8377
Prompt: direct_reference	6,166	0.8837	0.7843	0.0308	0.8767	0.8642
Prompt: indirect_reference	5,584	0.8411	0.6981	0.0229	0.8376	0.8112
Prompt: revise	6,032	0.9429	0.9025	0.0272	0.9377	0.9354
❗ Prompt: rewrite	4,304	0.7394	0.5079	0.0321	0.7379	0.6596
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.8251	0.6673	0.0267	0.8203	0.7878
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9540	0.9213	0.0268	0.9473	0.9459
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.8511	0.7206	0.0267	0.8469	0.8248
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.8161	0.6571	0.0301	0.8135	0.7789
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.8647	0.7480	0.0297	0.8592	0.8416
Model: gpt-4o (Temp: Unknown)	3,308	0.9086	0.8331	0.0278	0.9027	0.8954
Model: hy3 (Temp: Unknown)	3,060	0.7903	0.6065	0.0288	0.7889	0.7418

Thresholding:

—Direction: higher_is_ai
—Swept for f1 with a found threshold of 0.0000.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Giga Llama-3.2-3B Score

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.9521	0.6328	0.0028	0.8150	0.7738
Prompt: direct_reference	6,166	0.9532	0.7408	0.0039	0.8685	0.8492
Prompt: indirect_reference	5,584	0.9527	0.6010	0.0025	0.7992	0.7496
Prompt: revise	6,032	0.9871	0.7576	0.0017	0.8780	0.8613
❗ Prompt: rewrite	4,304	0.8957	0.3443	0.0033	0.6705	0.5110
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.9270	0.5629	0.0020	0.7805	0.7195
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9926	0.8261	0.0011	0.9125	0.9042
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.9435	0.5862	0.0033	0.7914	0.7376
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.9422	0.5506	0.0064	0.7721	0.7073
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.9546	0.6616	0.0013	0.8301	0.7957
Model: gpt-4o (Temp: Unknown)	3,308	0.9770	0.7376	0.0012	0.8682	0.8484
Model: hy3 (Temp: Unknown)	3,060	0.9210	0.4686	0.0046	0.7320	0.6362

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.3188.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: EditLens Giga Llama-3.2-3B Bucket

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.8523	0.7079	0.0052	0.8514	0.8265
Prompt: direct_reference	6,166	0.8873	0.7775	0.0062	0.8857	0.8718
Prompt: indirect_reference	5,584	0.8294	0.6615	0.0047	0.8284	0.7941
Prompt: revise	6,032	0.9334	0.8694	0.0040	0.9327	0.9281
❗ Prompt: rewrite	4,304	0.7182	0.4419	0.0060	0.7179	0.6104
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.8137	0.6308	0.0059	0.8125	0.7708
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.9477	0.8974	0.0040	0.9467	0.9439
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.8360	0.6765	0.0060	0.8352	0.8041
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.8129	0.6321	0.0083	0.8119	0.7706
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.8633	0.7282	0.0033	0.8625	0.8411
Model: gpt-4o (Temp: Unknown)	3,308	0.8969	0.7956	0.0030	0.8963	0.8847
Model: hy3 (Temp: Unknown)	3,060	0.7780	0.5608	0.0059	0.7775	0.7159

Thresholding:

—Direction: higher_is_ai
—Swept for f1 with a found threshold of 0.0000.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Perplexity (Llama-3.2-3B-Instruct)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6590	0.0721	0.0024	0.5348	0.1342
Prompt: direct_reference	6,166	0.6967	0.1278	0.0032	0.5623	0.2260
Prompt: indirect_reference	5,584	0.7303	0.1186	0.0021	0.5582	0.2116
Prompt: revise	6,032	0.6478	0.0153	0.0013	0.5070	0.0300
Prompt: rewrite	4,304	0.5306	0.0116	0.0033	0.5042	0.0229
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7846	0.1963	0.0046	0.5959	0.3270
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.7638	0.1585	0.0040	0.5773	0.2727
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6671	0.0127	0.0013	0.5057	0.0250
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6561	0.0115	0.0019	0.5048	0.0228
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.7187	0.1069	0.0007	0.5531	0.1930
Model: gpt-4o (Temp: Unknown)	3,308	0.6297	0.0085	0.0030	0.5027	0.0167
❗ Model: hy3 (Temp: Unknown)	3,060	0.3831	0.0026	0.0013	0.5007	0.0052

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 4.4416.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Perplexity (Llama-3.2-3B)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6053	0.0003	0.0012	0.4995	0.0005
Prompt: direct_reference	6,166	0.6369	0.0000	0.0010	0.4995	0.0000
Prompt: indirect_reference	5,584	0.6837	0.0004	0.0004	0.5000	0.0007
Prompt: revise	6,032	0.5877	0.0000	0.0017	0.4992	0.0000
Prompt: rewrite	4,304	0.4859	0.0009	0.0019	0.4995	0.0019
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7310	0.0000	0.0007	0.4997	0.0000
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.7075	0.0000	0.0011	0.4994	0.0000
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6116	0.0007	0.0033	0.4987	0.0013
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6048	0.0006	0.0013	0.4997	0.0013
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.6705	0.0000	0.0000	0.5000	0.0000
Model: gpt-4o (Temp: Unknown)	3,308	0.5564	0.0000	0.0012	0.4994	0.0000
❗ Model: hy3 (Temp: Unknown)	3,060	0.3468	0.0007	0.0007	0.5000	0.0013

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 1.2539.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Entropy (Llama-3.2-3B-Instruct)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6561	0.0867	0.0033	0.5417	0.1590
Prompt: direct_reference	6,166	0.6927	0.1512	0.0042	0.5735	0.2617
Prompt: indirect_reference	5,584	0.7222	0.1393	0.0018	0.5688	0.2442
Prompt: revise	6,032	0.6454	0.0252	0.0020	0.5116	0.0491
Prompt: rewrite	4,304	0.5341	0.0121	0.0056	0.5033	0.0237
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7538	0.1846	0.0072	0.5887	0.3098
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.7714	0.1990	0.0051	0.5969	0.3305
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6646	0.0307	0.0040	0.5134	0.0594
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6524	0.0167	0.0019	0.5074	0.0327
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.7225	0.1392	0.0007	0.5693	0.2442
Model: gpt-4o (Temp: Unknown)	3,308	0.6113	0.0218	0.0024	0.5097	0.0425
❗ Model: hy3 (Temp: Unknown)	3,060	0.4066	0.0039	0.0013	0.5013	0.0078

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 1.4199.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Entropy (Llama-3.2-3B)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.5986	0.0004	0.0017	0.4993	0.0007
Prompt: direct_reference	6,166	0.6216	0.0000	0.0013	0.4994	0.0000
Prompt: indirect_reference	5,584	0.6622	0.0007	0.0007	0.5000	0.0014
Prompt: revise	6,032	0.5880	0.0000	0.0023	0.4988	0.0000
Prompt: rewrite	4,304	0.5002	0.0009	0.0028	0.4991	0.0019
Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.6885	0.0000	0.0013	0.4993	0.0000
✔️ Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.7191	0.0000	0.0023	0.4989	0.0000
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.5963	0.0007	0.0033	0.4987	0.0013
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.5835	0.0006	0.0013	0.4997	0.0013
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.6368	0.0000	0.0000	0.5000	0.0000
Model: gpt-4o (Temp: Unknown)	3,308	0.5881	0.0006	0.0024	0.4991	0.0012
❗ Model: hy3 (Temp: Unknown)	3,060	0.3638	0.0007	0.0013	0.4997	0.0013

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.3161.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Top-p Outliers (Llama-3.2-3B-Instruct)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.5437	0.0243	0.0060	0.5091	0.0471
Prompt: direct_reference	6,166	0.5685	0.0389	0.0065	0.5162	0.0745
Prompt: indirect_reference	5,584	0.5653	0.0362	0.0047	0.5158	0.0695
Prompt: revise	6,032	0.5165	0.0070	0.0063	0.5003	0.0137
Prompt: rewrite	4,304	0.5179	0.0121	0.0065	0.5028	0.0237
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.6862	0.0920	0.0091	0.5414	0.1671
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.5510	0.0160	0.0074	0.5043	0.0312
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.5040	0.0020	0.0047	0.4987	0.0040
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.5050	0.0045	0.0026	0.5010	0.0089
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.5486	0.0422	0.0092	0.5165	0.0803
Model: gpt-4o (Temp: Unknown)	3,308	0.5369	0.0054	0.0048	0.5003	0.0108
❗ Model: hy3 (Temp: Unknown)	3,060	0.4724	0.0105	0.0039	0.5033	0.0206

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.0244.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Top-p Outliers (Llama-3.2-3B)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6046	0.0071	0.0025	0.5023	0.0140
Prompt: direct_reference	6,166	0.6575	0.0130	0.0039	0.5045	0.0255
Prompt: indirect_reference	5,584	0.6563	0.0079	0.0014	0.5032	0.0156
Prompt: revise	6,032	0.5772	0.0030	0.0017	0.5007	0.0059
Prompt: rewrite	4,304	0.5006	0.0033	0.0033	0.5000	0.0065
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7106	0.0065	0.0046	0.5010	0.0129
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.6405	0.0108	0.0011	0.5048	0.0214
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6080	0.0013	0.0020	0.4997	0.0027
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6215	0.0006	0.0019	0.4994	0.0013
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.6888	0.0218	0.0026	0.5096	0.0425
Model: gpt-4o (Temp: Unknown)	3,308	0.5114	0.0018	0.0018	0.5000	0.0036
❗ Model: hy3 (Temp: Unknown)	3,060	0.4563	0.0065	0.0039	0.5013	0.0129

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.0026.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Top-k Outliers (Llama-3.2-3B-Instruct)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6316	0.0635	0.0026	0.5304	0.1191
Prompt: direct_reference	6,166	0.6812	0.1174	0.0036	0.5569	0.2095
Prompt: indirect_reference	5,584	0.7073	0.1003	0.0018	0.5492	0.1820
Prompt: revise	6,032	0.6056	0.0153	0.0020	0.5066	0.0300
Prompt: rewrite	4,304	0.5006	0.0060	0.0033	0.5014	0.0120
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7581	0.1559	0.0039	0.5760	0.2688
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.7095	0.1311	0.0034	0.5639	0.2312
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6676	0.0241	0.0027	0.5107	0.0469
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6404	0.0122	0.0013	0.5054	0.0240
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.6728	0.0963	0.0013	0.5475	0.1755
Model: gpt-4o (Temp: Unknown)	3,308	0.5958	0.0157	0.0036	0.5060	0.0308
❗ Model: hy3 (Temp: Unknown)	3,060	0.3729	0.0033	0.0020	0.5007	0.0065

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.0212.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Top-k Outliers (Llama-3.2-3B)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.5871	0.0054	0.0020	0.5017	0.0108
Prompt: direct_reference	6,166	0.6325	0.0107	0.0016	0.5045	0.0211
Prompt: indirect_reference	5,584	0.6660	0.0075	0.0007	0.5034	0.0149
Prompt: revise	6,032	0.5569	0.0010	0.0030	0.4990	0.0020
Prompt: rewrite	4,304	0.4637	0.0014	0.0028	0.4993	0.0028
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.7119	0.0091	0.0026	0.5033	0.0181
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.6553	0.0125	0.0023	0.5051	0.0247
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6254	0.0007	0.0040	0.4983	0.0013
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6028	0.0006	0.0006	0.5000	0.0013
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.6291	0.0132	0.0013	0.5059	0.0260
Model: gpt-4o (Temp: Unknown)	3,308	0.5353	0.0006	0.0018	0.4994	0.0012
❗ Model: hy3 (Temp: Unknown)	3,060	0.3482	0.0007	0.0013	0.4997	0.0013

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of 0.0045.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: FastDetectGPT (Llama-3.2-3B-Instruct)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.5106	0.0050	0.0054	0.4998	0.0099
Prompt: direct_reference	6,166	0.5104	0.0055	0.0036	0.5010	0.0109
Prompt: indirect_reference	5,584	0.5135	0.0036	0.0054	0.4991	0.0071
Prompt: revise	6,032	0.4970	0.0053	0.0056	0.4998	0.0105
Prompt: rewrite	4,304	0.5258	0.0056	0.0079	0.4988	0.0110
✔️ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.6825	0.0072	0.0085	0.4993	0.0141
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.4567	0.0011	0.0068	0.4971	0.0023
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.4484	0.0060	0.0040	0.5010	0.0119
❗ Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.4456	0.0032	0.0038	0.4997	0.0064
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.5230	0.0026	0.0040	0.4993	0.0052
Model: gpt-4o (Temp: Unknown)	3,308	0.5283	0.0067	0.0060	0.5003	0.0131
Model: hy3 (Temp: Unknown)	3,060	0.4957	0.0085	0.0046	0.5020	0.0168

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 3.2075.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: FastDetectGPT (Llama-3.2-3B)

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.4278	0.0115	0.0052	0.5032	0.0226
Prompt: direct_reference	6,166	0.3820	0.0088	0.0045	0.5021	0.0173
Prompt: indirect_reference	5,584	0.3534	0.0093	0.0061	0.5016	0.0183
Prompt: revise	6,032	0.4643	0.0153	0.0053	0.5050	0.0299
Prompt: rewrite	4,304	0.5381	0.0130	0.0046	0.5042	0.0256
❗ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.2378	0.0026	0.0026	0.5000	0.0052
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.4385	0.0103	0.0086	0.5009	0.0201
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.4055	0.0094	0.0027	0.5033	0.0185
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.3715	0.0051	0.0064	0.4994	0.0101
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.3181	0.0059	0.0046	0.5007	0.0117
✔️ Model: gpt-4o (Temp: Unknown)	3,308	0.6038	0.0236	0.0060	0.5088	0.0458
Model: hy3 (Temp: Unknown)	3,060	0.6018	0.0229	0.0046	0.5092	0.0445

Thresholding:

—Direction: lower_is_ai
—Swept for fpr_0_5pct with a found threshold of -3.0452.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]

Classifier: Binoculars

Performance:

Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	22,086	0.6047	0.0139	0.0043	0.5048	0.0272
Prompt: direct_reference	6,166	0.6083	0.0104	0.0065	0.5019	0.0204
Prompt: indirect_reference	5,584	0.5833	0.0093	0.0036	0.5029	0.0184
Prompt: revise	6,032	0.6230	0.0153	0.0036	0.5058	0.0299
Prompt: rewrite	4,304	0.6018	0.0228	0.0028	0.5100	0.0444
❗ Model: Llama-4-Scout-17B-16E-Instruct-NVFP4 (Temp: 0.6)	3,066	0.5393	0.0072	0.0039	0.5016	0.0142
Model: Mixtral-8x7B-Instruct-v0.1 (Temp: 0.7)	3,508	0.6377	0.0211	0.0034	0.5088	0.0412
Model: claude-sonnet-4-5-20250929 (Temp: Unknown)	2,992	0.6117	0.0094	0.0053	0.5020	0.0184
Model: claude-sonnet-5 (Temp: Unknown)	3,120	0.6071	0.0058	0.0019	0.5019	0.0115
Model: gpt-3.5-turbo (Temp: Unknown)	3,032	0.5660	0.0112	0.0040	0.5036	0.0221
Model: gpt-4o (Temp: Unknown)	3,308	0.6185	0.0133	0.0060	0.5036	0.0261
✔️ Model: hy3 (Temp: Unknown)	3,060	0.6469	0.0281	0.0052	0.5114	0.0544

Thresholding:

—Direction: higher_is_ai
—Swept for fpr_0_5pct with a found threshold of 1.0377.

[image]

Classification Histograms:

[image]

Per Prompt Subset

[image] [image] [image] [image]

Per Generator Config Subset

[image] [image] [image] [image] [image] [image] [image]