# Detecting early-warning signals for conversational breakdown on CGA-Wiki ConvoKit release (test)

Source: [Detecting early-warning signals for conversational breakdown on CGA-Wiki ConvoKit release (test)](<https://www.sota2.com/research/sota/detecting-early-warning-signals-for-conversational-breakdown-on-cga-wiki-convokit-release-test>)

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

Data updated: 2026-05-12T10:31:15.502Z

Evaluation of models on the test split of the CGA-Wiki ConvoKit release dataset, specifically targeting the detection of early warning signals that precede conversational breakdown. Metrics include tau\_derail, tau\_civil, p, and d.

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 | Detecting early-warning signals for conversational breakdown |
| Dataset (including any stated split) | CGA-Wiki ConvoKit release (test) |
| Metric | Tau Derail |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 8 of 8 |
| 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

- [Tau Derail](<https://www.sota2.com/research/sota/detecting-early-warning-signals-for-conversational-breakdown-on-cga-wiki-convokit-release-test.md?metric=Tau%20Derail>): Higher is better; unit: SCALAR.

- [Tau Civil](<https://www.sota2.com/research/sota/detecting-early-warning-signals-for-conversational-breakdown-on-cga-wiki-convokit-release-test.md?metric=Tau%20Civil>): Higher is better; unit: SCALAR.

- [P-Value](<https://www.sota2.com/research/sota/detecting-early-warning-signals-for-conversational-breakdown-on-cga-wiki-convokit-release-test.md?metric=P-Value>): Lower is better; unit: SCALAR.

- [Effect Size d](<https://www.sota2.com/research/sota/detecting-early-warning-signals-for-conversational-breakdown-on-cga-wiki-convokit-release-test.md?metric=Effect%20Size%20d>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Tau Derail | Result date | Report paper |
| --- | --- | --- | --- |
| VADER VAR | 0.086 | Not recorded | 6a02b411dbc836b87d8246bb |
| Lex. Div. AC1 | 0.081 | Not recorded | 6a02b411dbc836b87d8246bb |
| q-hat CE VAR | 0.055 | Not recorded | 6a02b411dbc836b87d8246bb |
| Toxicity VAR | 0.055 | Not recorded | 6a02b411dbc836b87d8246bb |
| q-hat DA VAR | -0.129 | Not recorded | 6a02b411dbc836b87d8246bb |
| q-hat RM VAR | -0.133 | Not recorded | 6a02b411dbc836b87d8246bb |
| AMD Var | -0.2 | Not recorded | 6a02b411dbc836b87d8246bb |
| Lex. Div. VAR | -0.327 | Not recorded | 6a02b411dbc836b87d8246bb |

### VADER VAR

Tau Derail: 0.086. Result ID: 6a03016547f34ce924e67751.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"Baseline (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### Lex. Div. AC1

Tau Derail: 0.081. Result ID: 6a03016547f34ce924e67753.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"Baseline (Lag-1 autocorrelation trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### q-hat CE VAR

Tau Derail: 0.055. Result ID: 6a03016547f34ce924e6774e.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"CSD indicator (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### Toxicity VAR

Tau Derail: 0.055. Result ID: 6a03016547f34ce924e67750.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"Baseline (Variance trend)","Window size (W)":"5","Base model":"toxic-bert"}. This record does not assert independent reproduction.

### q-hat DA VAR

Tau Derail: -0.129. Result ID: 6a03016547f34ce924e6774c.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"CSD indicator (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### q-hat RM VAR

Tau Derail: -0.133. Result ID: 6a03016547f34ce924e6774d.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"CSD indicator (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### AMD Var

Tau Derail: -0.2. Result ID: 6a03016547f34ce924e6774f.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"CSD indicator (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

### Lex. Div. VAR

Tau Derail: -0.327. Result ID: 6a03016547f34ce924e67752.

[Reporting paper](<https://www.sota2.com/research/paper/6a02b411dbc836b87d8246bb.md>)

Recorded evaluation configuration: {"Indicator Type":"Baseline (Variance trend)","Window size (W)":"5"}. This record does not assert independent reproduction.

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