Your key never reaches our servers. · it stays in your browser tab, in memory · every tool is open source — check it yourself
Deterministic LLM evaluation

AI Crash Test

Three tools for one question: can you trust this system tomorrow?

All three grade with fixed predicates — no LLM-as-judge — so a score that moves means the system moved, not the test. Bring your own key; it never leaves your tab.

2 live services no LLM-as-judge BYOK, browser-direct source on GitHub
01 / The lenses

Three questions, three tools

Same discipline pointed at different failure modes.

Model Drift live
Is it changing under me?

A frozen 35-task suite runs against 16 live models every day. Every run is kept, and a drop against the previous run is flagged automatically — including the false alarms, which get published too.

dailyno key neededlongitudinal
AI Crash Test live
Does it break under attack?

Point your own model at adversarial batteries — injection, tool-abuse, hallucination-bait, refusal calibration, retrieval grounding, agent trajectories — and get a severity-weighted vulnerability report with a fail card for every miss.

adversarialyour key, your browserseverity-weighted
Harness Builder beta
Which configuration should I run?

A harness config is data: roles, models, scaffolding, topology. Draw the shape or paste the JSON, declare the axes to vary, and it runs the whole matrix — against the built-in coding suite or your own tasks. Then it does the part a leaderboard skips: a per-task paired comparison that says whether one shape actually beat another, or whether the suite is too small to tell.

visual graph editorpaired sign testbring your own suite9 providersscoring needs the local backend
02 / The engine

One grading engine, not three that agree

A verdict is a function of the output, so it reproduces exactly. Rerun it and argue with it.

gradecore fixed predicates · pure, no I/O
AI Crash Test
imports it — 8 modules
Harness Builder
imports it — server.py
Model Drift
does not — it predates the extraction and still carries its own copy

The drift board is the one that hasn't been folded in yet.

It was written before gradecore was extracted and still grades with its own copy of the predicates. That is a duplication to pay down, not a design — and it is named here because a page claiming "one engine" while one of its three tools quietly runs a second copy would be exactly the kind of unchecked claim these tools exist to catch.

ONE HONEST EXCEPTION Running generated code is I/O, not a pure predicate. The harness tool's python check executes locally and labels itself as such in every result.
ONE KEY RULE The browser calls the provider directly with your key. No proxy, no server-side storage, no "remember me". The ingest endpoint actively refuses any payload that looks key-shaped, so even a buggy client can't leak one.
ONE WIRE FORMAT Every run serializes to the same eval_run.json, so a crash-test result and a drift run land in the same history and can be compared over time.
FAILURES GET PUBLISHED The drift board's field notes include the week four regression alerts turned out to be rate limits and single-question noise. A tool that never reports its own false positives hasn't looked.