Open assessment reference
A common scale for AI fluency.
Twelve sub-competencies, four levels, and a rule for what counts as proof. Free to use and to challenge.
The framework says what to look for. This says how far along you are.
The AI Fluency Framework of Rick Dakan and Joseph Feller, developed with Anthropic, defines four competencies and twelve sub-competencies. It is a competency model, not an assessment: it names the skills without saying what each looks like at a beginner or an expert level, or what evidence would settle the question. This reference adds those three things, and nothing else.
Forty-eight descriptors in total, one per sub-competency per level. Each is a single observable action, falsifiable on real work, with no product named and no adverb of degree.
The rule of evidence
Three ways to see the work, and the limits of each.
The framework's own teaching course names three complementary approaches: what someone produces, how they work over time, and what they understand about their own practice. Combining them gives the fullest picture. Each family below serves one of the three, and declares what it cannot see.
Two instruments, so far.
The scenario test
Twelve workplace situations, one per sub-competency, in your browser. Nothing is uploaded and nothing is stored: your result lives in the link, and the link is yours to keep or discard.
The trace instrument
An MCP server that reads your own conversation export on your own machine, from any assistant, and turns it into evidence. It detects the eleven behaviours Anthropic measured at population scale, with an excerpt for every detection so you can contest it.
Traces alone cap every placement at level 2. That limit is deliberate.
What it refuses to be.
Fifteen minutes for an estimate. Longer for a placement.
Start with the twelve situations. The result tells you where you stand and, just as plainly, what it could not see.