Grading

know-code uses agent-proposed grading with human review and attest. The human does not self-assign a score.

Flow

1. Human completes browser quiz (know-code ask) → answers.json
2. Agent reads answers + quiz + diff context → writes grade-proposal.json
3. Human runs know-code grade --review (TUI shows per-question scores + feedback)
4. Human attests passphrase → sealed grade.json
5. Human runs know-code pass → sealed gate.json

# Agent (after ask)
know-code grade propose --json    # optional context helper
# Agent writes .know-code/grade-proposal.json

# Human
know-code grade --review          # or --accept to skip score adjustment
know-code pass

Pass bar

Overall score ≥ 0.8. Below the bar, grade --review exits 2 — re-teach and re-quiz.

grade-proposal.json

Written by the agent (unsigned). Binds diffHash and answersDigest. See the skill reference grading-rubric.md in the repo.

Emergency self-score

Disabled by default (allowSelfScore: false). Enable only for emergencies:

know-code config set allowSelfScore true
know-code grade --score 0.85

Config

Field	Default	Meaning
requireGradeProposal	true	Require agent proposal before human grade
allowSelfScore	false	Permit legacy grade --score