# Misalignment Detection on WorkBench (n=326) Source: [Misalignment Detection on WorkBench (n=326)]() Collection index: [llms.txt]() Data updated: 2026-07-03T06:22:16.421Z Evaluation of Misalignment Detection performance on the WorkBench dataset, specifically using the n=326 configuration, reporting Avg. IR, mean difference (Delta), and p-value. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Misalignment Detection | | Dataset (including any stated split) | WorkBench (n=326) | | Metric | Average IR | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 30 of 30 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Average IR](): Higher is better; unit: PERCENT. - [Mean Difference (Delta)](): Lower is better; unit: SCALAR. - [P-Value](): Lower is better; unit: SCALAR. ## Recorded results | Method | Average IR | Result date | Report paper | | --- | --- | --- | --- | | ProvePrompt | 52 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 49.5 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 45.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 44.4 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 43.1 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 41.4 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 40.7 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 39.4 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 38.3 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 37.5 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 37.3 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 29.4 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 27.5 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 26.9 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 26.7 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 24.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 19.8 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 18.1 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 17.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 16.9 | Not recorded | 6a4741ee143ebdcacacc5a79 | | DirectPrompt | 15.3 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 15.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 14.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvePrompt | 13.9 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 13.2 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 12.5 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 11.9 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 11 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 10 | Not recorded | 6a4741ee143ebdcacacc5a79 | | ProvenanceGuard | 4.8 | Not recorded | 6a4741ee143ebdcacacc5a79 | ### ProvePrompt Average IR: 52. Result ID: 6a4754fa60b88bb33991b2be. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemini 3 Flash"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 49.5. Result ID: 6a4754fa60b88bb33991b2b2. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-5"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 45.2. Result ID: 6a4754fa60b88bb33991b2b3. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"GPT-5"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 44.4. Result ID: 6a4754fa60b88bb33991b2b5. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-5 Mini"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 43.1. Result ID: 6a4754fa60b88bb33991b2ca. [Reporting paper]() Recorded evaluation configuration: {"Model":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 41.4. Result ID: 6a4754fa60b88bb33991b2bf. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemini 3 Flash"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 40.7. Result ID: 6a4754fa60b88bb33991b2cd. [Reporting paper]() Recorded evaluation configuration: {"Model":"Claude Haiku 4.5"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 39.4. Result ID: 6a4754fa60b88bb33991b2cb. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 38.3. Result ID: 6a4754fa60b88bb33991b2ce. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Claude Haiku 4.5"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 37.5. Result ID: 6a4754fa60b88bb33991b2c4. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 37.3. Result ID: 6a4754fa60b88bb33991b2b6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"GPT-5 Mini"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 29.4. Result ID: 6a4754fa60b88bb33991b2c1. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemini 3.1 Flash-Lite"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 27.5. Result ID: 6a4754fa60b88bb33991b2bb. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-4.1 Mini"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 26.9. Result ID: 6a4754fa60b88bb33991b2b8. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-4.1"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 26.7. Result ID: 6a4754fa60b88bb33991b2bc. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"GPT-4.1 Mini"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 24.2. Result ID: 6a4754fa60b88bb33991b2c5. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 19.8. Result ID: 6a4754fa60b88bb33991b2c2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemini 3.1 Flash-Lite"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 18.1. Result ID: 6a4754fa60b88bb33991b2cc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Claude Haiku 4.5"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 17.2. Result ID: 6a4754fa60b88bb33991b2bd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemini 3 Flash"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 16.9. Result ID: 6a4754fa60b88bb33991b2c8. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash-Lite"}. This record does not assert independent reproduction. ### DirectPrompt Average IR: 15.3. Result ID: 6a4754fa60b88bb33991b2b9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"GPT-4.1"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 15.2. Result ID: 6a4754fa60b88bb33991b2b7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-4.1"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 14.2. Result ID: 6a4754fa60b88bb33991b2b4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-5 Mini"}. This record does not assert independent reproduction. ### ProvePrompt Average IR: 13.9. Result ID: 6a4754fa60b88bb33991b2c7. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash-Lite"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 13.2. Result ID: 6a4754fa60b88bb33991b2b1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-5"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 12.5. Result ID: 6a4754fa60b88bb33991b2c0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemini 3.1 Flash-Lite"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 11.9. Result ID: 6a4754fa60b88bb33991b2ba. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-4.1 Mini"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 11. Result ID: 6a4754fa60b88bb33991b2c3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 10. Result ID: 6a4754fa60b88bb33991b2c6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemini 2.5 Flash-Lite"}. This record does not assert independent reproduction. ### ProvenanceGuard Average IR: 4.8. Result ID: 6a4754fa60b88bb33991b2c9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.