# Causal Faithfulness Evaluation on HaluEval Adversarial Source: [Causal Faithfulness Evaluation on HaluEval Adversarial]() Collection index: [llms.txt]() Data updated: 2026-05-22T07:41:28.589Z Evaluation of model causal faithfulness performance using the HaluEval Adversarial dataset, measured by nAOPC, nComp, and \|∆CE\| scores. 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 | Causal Faithfulness Evaluation | | Dataset (including any stated split) | HaluEval Adversarial | | Metric | nAOPC | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 28 of 28 | | 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 - [nAOPC](): Higher is better; unit: PERCENT. - [nComp](): Higher is better; unit: SCALAR. - [\|∆CE\|](): Lower is better; unit: SCALAR. ## Recorded results | Method | nAOPC | Result date | Report paper | | --- | --- | --- | --- | | Retrain | 100 | Not recorded | 6a0fe320f984739ed6739e05 | | SAEBoost | 100 | Not recorded | 6a0fe320f984739ed6739e05 | | FaithfulSAE | 100 | Not recorded | 6a0fe320f984739ed6739e05 | | GAE | 100 | Not recorded | 6a0fe320f984739ed6739e05 | | Finetune | 98.8 | Not recorded | 6a0fe320f984739ed6739e05 | | Fixed | 98.5 | Not recorded | 6a0fe320f984739ed6739e05 | | TERM | 98.4 | Not recorded | 6a0fe320f984739ed6739e05 | | GAE | 96.8 | Not recorded | 6a0fe320f984739ed6739e05 | | SAEBoost | 96.5 | Not recorded | 6a0fe320f984739ed6739e05 | | GAE | 95.3 | Not recorded | 6a0fe320f984739ed6739e05 | | Finetune | 93.2 | Not recorded | 6a0fe320f984739ed6739e05 | | Fixed | 93 | Not recorded | 6a0fe320f984739ed6739e05 | | Retrain | 93 | Not recorded | 6a0fe320f984739ed6739e05 | | Finetune | 92.5 | Not recorded | 6a0fe320f984739ed6739e05 | | SAEBoost | 90.9 | Not recorded | 6a0fe320f984739ed6739e05 | | FaithfulSAE | 90.8 | Not recorded | 6a0fe320f984739ed6739e05 | | Fixed | 89.9 | Not recorded | 6a0fe320f984739ed6739e05 | | FaithfulSAE | 89.9 | Not recorded | 6a0fe320f984739ed6739e05 | | TERM | 89.8 | Not recorded | 6a0fe320f984739ed6739e05 | | TERM | 89.6 | Not recorded | 6a0fe320f984739ed6739e05 | | Retrain | 89.4 | Not recorded | 6a0fe320f984739ed6739e05 | | GAE | 87.1 | Not recorded | 6a0fe320f984739ed6739e05 | | SAEBoost | 84 | Not recorded | 6a0fe320f984739ed6739e05 | | FaithfulSAE | 73.8 | Not recorded | 6a0fe320f984739ed6739e05 | | Fixed | 73.5 | Not recorded | 6a0fe320f984739ed6739e05 | | TERM | 73 | Not recorded | 6a0fe320f984739ed6739e05 | | Finetune | 57.9 | Not recorded | 6a0fe320f984739ed6739e05 | | Retrain | 52.1 | Not recorded | 6a0fe320f984739ed6739e05 | ### Retrain nAOPC: 100. Result ID: 6a1008a4fbc373592c08afe7. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### SAEBoost nAOPC: 100. Result ID: 6a1008a4fbc373592c08afe8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### FaithfulSAE nAOPC: 100. Result ID: 6a1008a4fbc373592c08afe9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### GAE nAOPC: 100. Result ID: 6a1008a4fbc373592c08afea. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### Finetune nAOPC: 98.8. Result ID: 6a1008a4fbc373592c08afe6. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### Fixed nAOPC: 98.5. Result ID: 6a1008a4fbc373592c08afe4. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### TERM nAOPC: 98.4. Result ID: 6a1008a4fbc373592c08afe5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### GAE nAOPC: 96.8. Result ID: 6a1008a4fbc373592c08afc0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### SAEBoost nAOPC: 96.5. Result ID: 6a1008a4fbc373592c08afbe. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### GAE nAOPC: 95.3. Result ID: 6a1008a4fbc373592c08afe3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Finetune nAOPC: 93.2. Result ID: 6a1008a4fbc373592c08afdf. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Fixed nAOPC: 93. Result ID: 6a1008a4fbc373592c08afdd. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Retrain nAOPC: 93. Result ID: 6a1008a4fbc373592c08afe0. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Finetune nAOPC: 92.5. Result ID: 6a1008a4fbc373592c08afbc. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### SAEBoost nAOPC: 90.9. Result ID: 6a1008a4fbc373592c08afe1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### FaithfulSAE nAOPC: 90.8. Result ID: 6a1008a4fbc373592c08afe2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Fixed nAOPC: 89.9. Result ID: 6a1008a4fbc373592c08afba. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### FaithfulSAE nAOPC: 89.9. Result ID: 6a1008a4fbc373592c08afbf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### TERM nAOPC: 89.8. Result ID: 6a1008a4fbc373592c08afde. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### TERM nAOPC: 89.6. Result ID: 6a1008a4fbc373592c08afbb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### Retrain nAOPC: 89.4. Result ID: 6a1008a4fbc373592c08afbd. [Reporting paper]() Recorded evaluation configuration: {"Model":"Pythia-1.4B"}. This record does not assert independent reproduction. ### GAE nAOPC: 87.1. Result ID: 6a1008a4fbc373592c08afb9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### SAEBoost nAOPC: 84. Result ID: 6a1008a4fbc373592c08afb7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### FaithfulSAE nAOPC: 73.8. Result ID: 6a1008a4fbc373592c08afb8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Fixed nAOPC: 73.5. Result ID: 6a1008a4fbc373592c08afb3. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### TERM nAOPC: 73. Result ID: 6a1008a4fbc373592c08afb4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Finetune nAOPC: 57.9. Result ID: 6a1008a4fbc373592c08afb5. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. ### Retrain nAOPC: 52.1. Result ID: 6a1008a4fbc373592c08afb6. [Reporting paper]() Recorded evaluation configuration: {"Model":"GPT-2 Small"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.