Decisions you can
measure.

Bring your own data or sample the Bluesky firehose. Test typed decisions across Clef models and inspect real inference performance.

01Input state

02Clef decision

03Typed answers

Configure the run

One schema, bounded samples, comparable results.

Decision template

CR Content router Intent, conversation potential, feed value choice · yes/no · score MT Moderation triage Policy area, review need, severity choice · yes/no · score SS Synthetic signals Text-only authorship signals and uncertainty choice · yes/no · score FC Feed curator Topic, shelf life, broad interest choice · yes/no · score + Custom classifier Define your own typed decision schema up to 8 questions

Input source

One input per line, or a JSON array

8 inputs availableText · one post per line

Run settings

Sample sizeMaximum 25 per run

125

Clef modelSelect for this run

Clef FlashFast · 9B · 24K context $0.038/M ClefPrecise · 27B · 64K context $0.240/M Clef OmniMultimodal · 64K context $0.150/M

Compare all modelsRuns identical inputs across all three

◇Private by designInputs are processed transiently and never persisted.

Performance

Measured at the Worker, around every inference.

Model measurements

Lower latency is better

Model	Success	Average	p50	p95	Range

Decisions you canmeasure.

Configure the run

Performance

Model measurements

Typed decisions

Decisions you can
measure.