Rubric
Used to score the performance task (both tracks) and the school final project. Designed to be usable by a teacher, a peer reviewer, or a scoring agent — every level describes observable behavior.
Last reviewed: 2026-10-25
Criterion 1 — Accuracy checking
Does the learner treat claims and numbers as unverified until checked?
| Level | Observable behavior |
|---|---|
| 4 — Routine | Finds all planted errors and near-misses; verifies each against an appropriate source; distinguishes “wrong”, “unverifiable” and “true but unsourced”; the corrected text contains no claim they cannot stand behind. |
| 3 — Solid | Finds the main factual errors and the fabricated source; verifies most claims; corrected text is safe, though one minor claim remains unchecked. |
| 2 — Partial | Finds obvious errors (wrong numbers) but accepts fluent claims and plausible citations; some corrections, others copied through. |
| 1 — Missing | Treats fluency as truth; corrects style only; no verification performed. |
Criterion 2 — Data judgment
Does the learner recognize what may and may not be pasted/shared, and act on exposure?
| Level | Observable behavior |
|---|---|
| 4 — Routine | Identifies the data-privacy problem and its identifiability logic (name + address + order number = person); rewrites the instruction correctly; states the containment/reporting path for existing exposure; notes tool-approval distinctions (approved vs. personal accounts). |
| 3 — Solid | Identifies the personal-data paste and fixes it; tool distinctions mostly right; containment path named but not sequenced. |
| 2 — Partial | Notices the paste “seems private” but mis-classifies (e.g., calls customer data “internal”); no containment path. |
| 1 — Missing | No data awareness shown; keeps the pasting instruction. |
(School adaptation — criterion 2 becomes Source judgment: does the learner cite where their own facts came from and flag claims that need sources? Levels mirror 4→1: every claim sourced → main claims sourced → “I checked” without a source → no sourcing.)
Criterion 3 — Bias recognition
Does the learner spot unfairness and connect it to a cause?
| Level | Observable behavior |
|---|---|
| 4 — Routine | Identifies the CV-ranking bias problem and the subtle ones (vendor-claims conflict of interest, “fully compliant” assertion); explains the cause (pattern learned from uneven data/authority claims); proposes the human-in-the-loop alternative. |
| 3 — Solid | Identifies the CV-ranking problem and connects it to bias; proposes human screening. |
| 2 — Partial | Says “the AI might be biased” without connecting cause or consequence; misses the subtle authority claims. |
| 1 — Missing | No bias recognition; accepts ranking people by machine score. |
(School adaptation: learners identify one unfair output pattern and a matching training-data cause — the Lesson 3 objective.)
Criterion 4 — Justification
Does the learner explain *why — the log is reasoning, not a changelist?*
| Level | Observable behavior |
|---|---|
| 4 — Routine | Every change in the log names the principle (the affected-person test, identifiability, source verification); escalations name the right owner; the response could teach someone else. |
| 3 — Solid | Most changes carry a reason tied to a module principle; one or two are asserted without a why. |
| 2 — Partial | Changes listed with thin labels (“wrong”, “fixed”); principles absent. |
| 1 — Missing | No log, or a log of opinions without reference to any check or principle. |
Using this rubric
- Passing bar (suggested): Level 3 or above on all four criteria for the workplace task; Level 2 minimum per criterion for the school project, with growth expected between lessons.
- Scorers: a teacher, a peer (using the peer protocol), or a labeled AI scorer. Any AI-assisted scoring is labeled as such in results and never presented as human judgment.
- The rubric deliberately avoids weighing writing quality: the object is judgment.