A rubric for portfolio reviews that mentors actually use

Portfolios fail when reviewers improvise criteria on the spot. We publish a rubric tied to the artifacts each course promises: automation graphs, SQL packs, runbooks, or token RFCs. Each dimension uses plain language anchors. Instead of vague "polish" scores, we ask whether a reader could execute the rollback path without pinging the author. That keeps debate grounded. We also leave room for mild disagreement. Two mentors may score adjacent bands; the learner sees both comments and chooses which adjustment to prioritize. The goal is judgment with traceability, not false precision. If you are hiring reviewers internally, borrow the rubric structure even if you change weights. Consistency beats charisma when learners compare notes after graduation.