# Knowledge Attribution Causal Ablation on Total (Combined ECLeKTic, MultiLoKo, G-MMLU) (test)

Source: [Knowledge Attribution Causal Ablation on Total (Combined ECLeKTic, MultiLoKo, G-MMLU) (test)](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-19T08:04:30.198Z

Evaluation of Knowledge Attribution Causal Ablation performance on the combined test set comprising ECLeKTic, MultiLoKo, and G-MMLU datasets.

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 | Knowledge Attribution Causal Ablation |
| Dataset (including any stated split) | Total (Combined ECLeKTic, MultiLoKo, G-MMLU) (test) |
| Metric | Ablation Success Rate |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [Ablation Success Rate](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Ablation%20Success%20Rate>): Higher is better; unit: PERCENT.

- [Spurious Gain Rate](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Spurious%20Gain%20Rate>): Lower is better; unit: PERCENT.

- [Rate Difference](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Rate%20Difference>): Lower is better; unit: PERCENT.

- [Num Q's w/ Inconsistency](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Num%20Q%27s%20w%2F%20Inconsistency>): Lower is better; unit: SCALAR.

- [Num Q's w/ Experts ID'd](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Num%20Q%27s%20w%2F%20Experts%20ID%27d>): Higher is better; unit: SCALAR.

- [Avg Experts ID'd per Q](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Avg%20Experts%20ID%27d%20per%20Q>): Higher is better; unit: SCALAR.

- [Original Correct Count](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Original%20Correct%20Count>): Higher is better; unit: SCALAR.

- [Num Q's All Incorrect](<https://www.sota2.com/research/sota/knowledge-attribution-causal-ablation-on-total-combined-eclektic-multiloko-g-mmlu-test.md?metric=Num%20Q%27s%20All%20Incorrect>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Ablation Success Rate | Result date | Report paper |
| --- | --- | --- | --- |
| XICI | 48 | Not recorded | 69bb82ed0636e5bf4afc18be |
| XICI | 35.5 | Not recorded | 69bb82ed0636e5bf4afc18be |
| Random Question-Shuffling (baseline) | 7.4 | Not recorded | 69bb82ed0636e5bf4afc18be |
| Random Question-Shuffling (baseline) | 7.3 | Not recorded | 69bb82ed0636e5bf4afc18be |
| Random Expert Set of Same Size (baseline) | 5 | Not recorded | 69bb82ed0636e5bf4afc18be |
| Random Expert Set of Same Size (baseline) | 4.4 | Not recorded | 69bb82ed0636e5bf4afc18be |

### XICI

Ablation Success Rate: 48. Result ID: 69bbadf047f34ce924a55e0a.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"GLM-4.5-Air","Details":"45 MoE layers, 5760 total routed experts","Configuration":"max 25, τ = 0.01, layers 6-40"}. This record does not assert independent reproduction.

### XICI

Ablation Success Rate: 35.5. Result ID: 69bbadf047f34ce924a55e0d.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"Qwen3-30B-A3B-Instruct-2507","Details":"48 MoE layers, 6144 total experts","Configuration":"max 25, τ = 0.005, layers 6-42"}. This record does not assert independent reproduction.

### Random Question-Shuffling (baseline)

Ablation Success Rate: 7.4. Result ID: 69bbadf047f34ce924a55e0e.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"Qwen3-30B-A3B-Instruct-2507","Details":"48 MoE layers, 6144 total experts","Configuration":"max 25, τ = 0.005, layers 6-42"}. This record does not assert independent reproduction.

### Random Question-Shuffling (baseline)

Ablation Success Rate: 7.3. Result ID: 69bbadf047f34ce924a55e0b.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"GLM-4.5-Air","Details":"45 MoE layers, 5760 total routed experts","Configuration":"max 25, τ = 0.01, layers 6-40"}. This record does not assert independent reproduction.

### Random Expert Set of Same Size (baseline)

Ablation Success Rate: 5. Result ID: 69bbadf047f34ce924a55e0c.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"GLM-4.5-Air","Details":"45 MoE layers, 5760 total routed experts","Configuration":"max 25, τ = 0.01, layers 6-40"}. This record does not assert independent reproduction.

### Random Expert Set of Same Size (baseline)

Ablation Success Rate: 4.4. Result ID: 69bbadf047f34ce924a55e0f.

[Reporting paper](<https://www.sota2.com/research/paper/69bb82ed0636e5bf4afc18be.md>)

Recorded evaluation configuration: {"Model":"Qwen3-30B-A3B-Instruct-2507","Details":"48 MoE layers, 6144 total experts","Configuration":"max 25, τ = 0.005, layers 6-42"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
