Measure Your AI Fluency

Score AI fluency across six skills — foundations, verification, prompting, workflow, safety, and role use — from your own 0–5 marks. The weakest skill is the headline (MIN, not the flattering average), and an essential gate holds anyone back to AWARE who can't verify a fact or handle data safely (Verification or Safety < 3/5) no matter how strong the rest. Tiers FLUENT / CAPABLE / AWARE, an overconfidence flag where self-rating outruns measured skill, and a per-person route to the exact training next — plus a team roll-up (weakest-learner tier + most-common gap). Runnable Python engine + a workbook that reproduces it + individual & team playbooks. For individual development and training planning — not a hiring, promotion, performance-review, or certification tool, and not legal advice.

Test whether you catch errors in AI output instead of trusting it. Review a piece of AI text deliberately seeded with six planted problems — a confidently-stated fabricated statistic, plausible-but-false claims, and biased framing — and mark which you caught; it scores your weighted catch-rate and returns INDEPENDENT / ASSISTED / RELIANT, with a hard gate on the fabricated statistic (miss the fake number and you can't be INDEPENDENT, capped to ASSISTED) and an over-confidence flag where you were highly confident on an item you missed — the over-reliance signal. A team roll-up reports the weakest-learner tier and the most-common missed error type. Deterministic and offline — your own marks drive the score, no AI grades you. Runnable Python engine + a workbook that reproduces it. The deep dive on the verification skill the AI Fluency Diagnostic flags. For individual development and training only — not a hiring, promotion, performance-review, or ranking tool, and not legal advice.

Run six AI safe-use drills past your team and find out if they catch the everyday mistakes — pasting client PII into a public tool, leaking a secret or credential, shadow AI, a hallucinated fact, prompt injection from pasted content, an over-permissioned connector. Mark recognition and correct action 0–5 per drill; the weaker mark caps the drill (you can't respond to a threat you never spotted) into a SAFE / RISKY / UNSAFE verdict, and the team band is your single weakest drill — DRILLED / UNEVEN / RAW — never the flattering average. A hard regulated-data gate forces the whole team to RAW if any regulated drill lands UNSAFE (the shipped sample averages 73 and is still RAW on one missed drill). Runnable Python scorer + a workbook that reproduces it + two playbooks (running the drills, the six failure modes) + a worked sample. It scores the drills, not people — not a score of individuals, not monitoring, not a security audit. Educational; not legal advice.

Practice prompts like a skill — graded drills with spaced repetition. Each prompt you write is scored 0–5 on the six RCTFCV criteria (Role, Context, Task, Format, Constraints, Verification-ask) on weights summing to 100, banded STRONG / WORKABLE / WEAK — with one hard gate: if the Verification-ask is below floor (you never ask the model to check its own work), the drill caps at WORKABLE no matter how high the total. A great prompt that doesn't ask for verification can't be STRONG. Then a Leitner spaced-repetition schedule (boxes 1–5 → review in 1/2/4/8/16 sessions, no calendar dates) puts the weak drills back in front of you until they stick. Grades prompt craft, not the model's output; deterministic and offline. Runnable Python engine + a workbook that reproduces it + two playbooks (the RCTFCV rubric + a deliberate-practice guide). The skill-building companion to the AI Fluency Diagnostic. Not legal, medical, or professional advice.