Draft specification
Status: Draft (v1.0) · Not yet in effect — thresholds below are proposed defaults pending finalization
Today, learners build portfolio artifacts they can show. Human-reviewed credentials are not yet available. This draft describes the proposed assessment and review process for a future credential.
1 · What we verify
UofAi does not certify course completion. We verify demonstrated capability: evidence that a person can use AI tools to produce quality work on real tasks, with sound judgment.
The unit of evidence is an artifact — a work product created by the learner during an evaluated practice session (“Rep”) on their own real work or a realistic scenario from their function. An artifact includes the task context, the learner's process (prompts, iterations, tool decisions), the output, and the learner's own verification of that output.
A portfolio is the set of artifacts a learner submits toward a credential — a minimum of 5 artifacts spanning the skill areas the credential covers. A credential may require more artifacts when its published scope calls for broader evidence.
2 · Three tiers of assessment
Tier
What happens
Who scores
Where it applies
Tier 1
Every Rep is scored against the published rubric.
AI, against calibrated rubrics
All plans, unlimited
Tier 2
A trained human reviewer scores the full portfolio against this Standard before any credential is issued.
Human reviewer
Every issued credential, no exceptions
Tier 3
A random sample of AI-scored work is independently re-reviewed by a human to measure and publish scoring reliability.
Human reviewer, blind
10% of AI-scored work, ongoing
No UofAi Verified Capability Credential is ever issued on AI scoring alone.
3 · The rubric
1
Did the learner decompose the problem and choose an appropriate approach?
2
Was AI used where it adds leverage, with effective prompting and iteration?
3
Did the learner check, correct, and take ownership of AI output? (Weighted highest. This is the dimension employers pay for.)
4
Does the final artifact meet a professional bar for its function?
5
Does the process shown generalize beyond this single task?
Credential threshold
Portfolio average ≥ 3.0 with no dimension averaging below 2.5. These thresholds are being validated during the pilot phase; changes require a version increment.
4 · Reviewers and calibration
Human review only means something if the humans are calibrated. Every reviewer proves agreement with reference scores before touching a live portfolio.
5 · Audit and published reliability
10% of Tier-1 AI-scored artifacts are randomly selected for blind human re-review. We track divergence between AI and human scores and publish the aggregate reliability figure on this page, updated quarterly. If divergence exceeds tolerance on any dimension, that dimension's AI scoring is recalibrated and affected in-flight portfolios re-reviewed.
6 · Integrity
7 · What the credential asserts — and what it doesn't
It asserts
This person demonstrated, on real work products, capability meeting the thresholds above under Standard v1.0 (draft), verified by a calibrated human reviewer.
It does not assert
Mastery of a specific vendor tool, performance in any specific job, or anything about work performed outside evaluated sessions.
8 · Versioning
This Standard is versioned. Material changes (rubric, thresholds, audit rate, reviewer requirements) increment the version and are recorded in the change log. Credentials always reference the version they were issued under.
Questions or recognition inquiries: hello@uofai.com. Employers can verify any credential through its public verification record at uofai.com/verify/… — every issued credential carries a unique verification link.