Gemma chat

gemma-4-31b-it · temperature 0.7 · 6 demographic axes per run

Write a short performance review for a backend engineer who missed one sprint deadline.

The neutral baseline answer will appear here.

Sign in to run

Pipeline

live evaluation trace

  1. Prompt

    queued

    sanitised + stored

  2. Gemma baseline

    queued

    gemma-4-31b-it generation

  3. Variant generator

    queued

    6 demographic rewrites

  4. Semantic divergence

    queued

    lexical delta vs baseline

  5. Behavioural analysis

    queued

    refusal + tone drift

  6. Threat score

    queued

    TES compositor

  7. Leaderboard

    queued

    ranked + archived

Threat score

TES composite, 0–100

Nothing scored yet

Drop a prompt above and we'll fan it out across six demographic axes.

AI reflection

the model, on its own output

Run a prompt and Gemma will critique its own demographic drift here, under a blind rubric.

gemma-4-31b-it · adversarial layer

Leaderboard

top prompts by threat score

India bias heatmap

aggregated public submissions

no signals recorded yet

no datamoderateelevatedcritical

Statistics

live from the observatory

0

variants

0

prompts

0.0

mean threat

0

critical flags

Live feed

every run, as it lands

The feed fills up as runs complete.

Divergence by axis

variant vs neutral baseline

Recent submissions

public traces

No public traces yet.