RavnLab tools · free · interactive · version 1.0

Set the review gate. Watch what still reaches a user.

A confidence gate is the simplest safety net an AI system has: ship answers the model is sure about, send the rest to a human. It only works if the model is sure for the right reasons. Move the gate below and watch the wrong answers that slip through - then switch the model and watch the net fail.

a RavnLab demonstration · pairs with the Confident-Wrong Checklist and the Plausible-Wrong Benchmark

how to read it

Each dot is one answer the system produced

Left to right is how confident the model was. Blue dots were right; red dots were wrong. The gate is the vertical line: everything to its right gets shipped automatically, everything to its left goes to a human to check. You are tuning one number - where to put the gate - and trading two costs against each other: wrong answers that reach a user, and the human effort of review.

the simulation

Move the gate

Its confidence tracks its accuracy: when it says 90%, it is right about 90% of the time. Most wrong answers sit on the left, at low confidence.

Answers plotted by the model's confidence, coloured by whether they were right or wrong, with a movable review gate. The live counts below the chart report the same information in text. ← sent to a human shipped automatically → 0% confidence 100% confidence how sure the model was
right answer wrong answer 48 answers · illustrative
the gate50% confidence
0
wrong answers that reach a user
0
wrong answers caught for review
0
answers a human must check
0
right answers shipped with no wait

Illustrative data - the shape of the problem, not one model's scores. A demonstration we produced, no client involved. The pattern is the human-review gate from our reliable-agent-patterns work.

what it shows

You cannot threshold your way out of miscalibration

With the calibrated model, the confidence number is a real dial: raise the gate and you reliably ship fewer wrong answers, because the model is rarely confident about a wrong one. With the overconfident model, the wrong answers hide among the confident ones - at the same gate it ships far more of them, and no setting sorts them out unless you send almost everything to a human. The gate is only as good as the confidence behind it. That is what calibration buys, and it is measurable.

Is your model sure for the right reasons?

The gate above is a toy. The real work is measuring whether your system's confidence tracks its accuracy on the cases that matter, then closing the gap. That is the conversation we want to have.

Book a call