# Safety Gate Capacity Scaling on Safety Scaling Testbed N=100-10,000

Source: [Safety Gate Capacity Scaling on Safety Scaling Testbed N=100-10,000](<https://www.sota2.com/research/sota/safety-gate-capacity-scaling-on-safety-scaling-testbed-n-100-10-000>)

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

Data updated: 2026-04-02T08:33:02.227Z

Evaluation of model safety performance across varying capacities (N=100 to 10,000) using scaling-specific metrics like Classifier ceiling, Ball advantage, and Advantage 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 | Safety Gate Capacity Scaling |
| Dataset (including any stated split) | Safety Scaling Testbed N=100-10,000 |
| Metric | Classifier Ceiling (B=0.01) |
| Unit | SCALAR |
| 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

- [Classifier Ceiling (B=0.01)](<https://www.sota2.com/research/sota/safety-gate-capacity-scaling-on-safety-scaling-testbed-n-100-10-000.md?metric=Classifier%20Ceiling%20(B%3D0.01)>): Higher is better; unit: SCALAR.

- [Ball (σ=σ\*)](<https://www.sota2.com/research/sota/safety-gate-capacity-scaling-on-safety-scaling-testbed-n-100-10-000.md?metric=Ball%20(%CF%83%3D%CF%83*)>): Higher is better; unit: SCALAR.

- [Advantage (×)](<https://www.sota2.com/research/sota/safety-gate-capacity-scaling-on-safety-scaling-testbed-n-100-10-000.md?metric=Advantage%20(%C3%97)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Classifier Ceiling (B=0.01) | Result date | Report paper |
| --- | --- | --- | --- |
| Classifier-based safety gate | 12.4 | Not recorded | 69cdf7e80636e5bf4afcc91b |
| Classifier-based safety gate | 3.9 | Not recorded | 69cdf7e80636e5bf4afcc91b |
| Classifier-based safety gate | 1.2 | Not recorded | 69cdf7e80636e5bf4afcc91b |

### Classifier-based safety gate

Classifier Ceiling (B=0.01): 12.4. Result ID: 69ce299347f34ce924b74e6e.

[Reporting paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb55870636e5bf4afca49c.md>)

Recorded evaluation configuration: {"N":"10,000"}. This record does not assert independent reproduction.

### Classifier-based safety gate

Classifier Ceiling (B=0.01): 3.9. Result ID: 69ce299347f34ce924b74e6c.

[Reporting paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb55870636e5bf4afca49c.md>)

Recorded evaluation configuration: {"N":"1,000"}. This record does not assert independent reproduction.

### Classifier-based safety gate

Classifier Ceiling (B=0.01): 1.2. Result ID: 69ce299347f34ce924b74e6a.

[Reporting paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb55870636e5bf4afca49c.md>)

Recorded evaluation configuration: {"N":"100"}. This record does not assert independent reproduction.

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