Loading…Loading

Reading the sheet…

Telltale

Load-test an LLM's position before you ship it.

Deterministic, explainable and sealed. It measures whether a stated position survives scripted social pressure. It is not a safety certification.

Product

Estate
Load bench
Lineup
Method
Agent
Export

Open source

Star on GitHub
Issues
Documentation
Live deployment

MIT licensed. Built for alignment reviewers who need a number they can defend in a review.

Telltale — Telltale-Hold benchmark. Not affiliated with any model provider.

https://github.com/aniruddhaadak80/telltale

The estate

Load tests in this session

Ownership is an anonymous cookie in this browser. You can see everything you have graded here and nothing anyone else has.

New load test

Trials you graded

Authored examples are never counted here.

Mean hold grade

—

Out of 100, at a 1.00× service load.

With permanent set

Trials whose position did not return after the unload turn.

Weakest band

No trials yet

Nothing here yet. Open the load bench, pick a script, paste a transcript and grade it. The first test takes about a minute and nothing is saved until you save it.

Run your first load test

Band

Order

Filter and order are in the URL.

No trials match this filter.

Scores are produced by telltale-grade/1.0.0, a deterministic function with no model in it. Turn the service load on any trial to re-rate it against a more aggressive user base without touching the transcript.