# Steering Validation on Real Agents LLMAgent, Claude Haiku 4.5 N-scaling Phase 4 (val)

Source: [Steering Validation on Real Agents LLMAgent, Claude Haiku 4.5 N-scaling Phase 4 (val)](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val>)

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

Data updated: 2026-04-10T08:20:55.742Z

Evaluation of steering validation performance for the Real Agents LLMAgent setup using the Claude Haiku 4.5 N-scaling Phase 4 validation set. Metrics include judge accuracy, contamination rate, and context token usage.

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 | Steering Validation |
| Dataset (including any stated split) | Real Agents LLMAgent, Claude Haiku 4.5 N-scaling Phase 4 (val) |
| Metric | Haiku Judge Accuracy |
| 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

- [Haiku Judge Accuracy](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md?metric=Haiku%20Judge%20Accuracy>): Higher is better; unit: PERCENT.

- [GPT-4o-mini Judge Accuracy](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md?metric=GPT-4o-mini%20Judge%20Accuracy>): Higher is better; unit: PERCENT.

- [Contamination Rate](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md?metric=Contamination%20Rate>): Lower is better; unit: PERCENT.

- [Avg Context Tokens](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md?metric=Avg%20Context%20Tokens>): Lower is better; unit: SIZE.

- [P-value](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md?metric=P-value>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Haiku Judge Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| DACS | 83.7 | Not recorded | 69d883fb7d0172140ec77999 |
| DACS | 79.8 | Not recorded | 69d883fb7d0172140ec77999 |
| Base | 63.3 | Not recorded | 69d883fb7d0172140ec77999 |
| Base | 62.6 | Not recorded | 69d883fb7d0172140ec77999 |

### DACS

Haiku Judge Accuracy: 83.7. Result ID: 69d8b2e547f34ce924c17884.

[Reporting paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

Recorded evaluation configuration: {"N":"5","Scenario":"ra2\_n5"}. This record does not assert independent reproduction.

### DACS

Haiku Judge Accuracy: 79.8. Result ID: 69d8b2e547f34ce924c17882.

[Reporting paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

Recorded evaluation configuration: {"N":"3","Scenario":"ra1\_n3"}. This record does not assert independent reproduction.

### Base

Haiku Judge Accuracy: 63.3. Result ID: 69d8b2e547f34ce924c17885.

[Reporting paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

Recorded evaluation configuration: {"N":"5","Scenario":"ra2\_n5"}. This record does not assert independent reproduction.

### Base

Haiku Judge Accuracy: 62.6. Result ID: 69d8b2e547f34ce924c17883.

[Reporting paper](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999.md>)

Recorded evaluation configuration: {"N":"3","Scenario":"ra1\_n3"}. This record does not assert independent reproduction.

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