# Automated failure mode detection alignment on RIFT (test) Source: [Automated failure mode detection alignment on RIFT (test)]() Collection index: [llms.txt]() Data updated: 2026-04-03T06:23:52.604Z Evaluation of model alignment robustness on the RIFT test set, specifically measuring performance across various failure modes such as subjectivity, ungroundedness, and hackability. 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 | Automated failure mode detection alignment | | Dataset (including any stated split) | RIFT (test) | | Metric | Subjectivity Score | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 4 of 4 | | 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 - [Subjectivity Score](): Higher is better; unit: PERCENT. - [Non-Atomic Score](): Higher is better; unit: PERCENT. - [Ungroundedness Score](): Higher is better; unit: PERCENT. - [Misaligned/Rigid Score](): Higher is better; unit: PERCENT. - [Missing Criteria Score](): Higher is better; unit: PERCENT. - [Hackability Score](): Higher is better; unit: PERCENT. - [Low Signal Score](): Higher is better; unit: PERCENT. - [Redundant Criteria Score](): Higher is better; unit: PERCENT. ## Recorded results | Method | Subjectivity Score | Result date | Report paper | | --- | --- | --- | --- | | LLMaJ | 86.1 | Not recorded | 69cf49680636e5bf4afcd8bb | | Alignment | 85.4 | Not recorded | 69cf49680636e5bf4afcd8bb | | IRR | 84.3 | Not recorded | 69cf49680636e5bf4afcd8bb | | Reward variance | 84.3 | Not recorded | 69cf49680636e5bf4afcd8bb | ### LLMaJ Subjectivity Score: 86.1. Result ID: 69cf5c7b47f34ce924b88aeb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluator":"LLMaJ","Model":"GPT-5.2"}. This record does not assert independent reproduction. ### Alignment Subjectivity Score: 85.4. Result ID: 69cf5c7b47f34ce924b88aed. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluator":"Alignment"}. This record does not assert independent reproduction. ### IRR Subjectivity Score: 84.3. Result ID: 69cf5c7b47f34ce924b88aec. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluator":"IRR"}. This record does not assert independent reproduction. ### Reward variance Subjectivity Score: 84.3. Result ID: 69cf5c7b47f34ce924b88aee. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Evaluator":"Reward variance","Conditioned Model":"GPT-5.2","Generator Model":"GPT-5 mini","Number of Responses":"4"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.