Grading at scale is two jobs. One is reading the work against the rubric. The other is recording why the score landed where it did. An education agent can draft the first. The instructor still owns the second. Dock holds the rationale, the rubric criteria the agent applied, the instructor's edits, and the attestation, so the grade that posts back to the SIS has a defensible record behind it.
Canvas, PowerSchool, and Google Classroom stay the system of record for the raw data. Dock is the system of record for what the agent interprets. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Dock surface: grading-rationale queue
| submission_ref | rubric_criterion | agent_score | agent_rationale | instructor_decision | reviewer | timestamp |
|---|---|---|---|---|---|---|
| canvas://courses/214/assignments/881/sub/4471 | Evidence use (4 levels) | 3 of 4 | Cites two primary sources but misattributes the second quote to Douglass instead of Garrison. | Approved at 3 of 4, comment added | jpark@school.edu | 2026-05-29T14:11Z |
| gclass://c/HS-Bio-3/coursework/22/sub/s-9921 | Lab procedure clarity | 2 of 4 | Steps 4 and 6 omit reagent volume. Hypothesis stated but not tied to method. | Overridden to 3 of 4: volumes implied in lab notebook | mvega@school.edu | 2026-05-29T15:02Z |
| powerschool://terms/S2/sections/4501/assignments/77/student/11820 | Argument structure | 4 of 4 | Thesis stated in paragraph 1, defended across three body paragraphs, restated with qualification. | Approved | jpark@school.edu | 2026-05-29T15:34Z |
The row is the unit of audit. The instructor reads the agent's rationale, accepts or overrides, and the score posts to the gradebook with the rationale attached.
The workflow
A student submits an essay in Canvas. The agent pulls the submission text and the rubric, scores each criterion, and writes a one-sentence rationale per criterion into a Dock row. The row points back to the Canvas submission ID. The instructor opens the queue, reads the rationales, edits where the agent misread, and clicks approve. Dock writes the approved score to Canvas via API and stores the full trail. For a PowerSchool section, the agent runs the same loop against the district rubric. For Google Classroom, it pulls the submission and the attached rubric and drafts against both. The instructor's attestation is what the registrar would see in an audit, not the agent's draft.
Why this matters
AAC&U's VALUE rubrics were built so faculty from different institutions could score the same authentic student work and agree on outcomes (AAC&U). The constraint is time. Carnegie Mellon's Eberly Center notes that rubrics reduce grading uncertainty and protect consistency across multiple graders, but only when each criterion gets read carefully (Eberly Center). The agent reads carefully on the first pass. The instructor reads where it matters. The grade that posts has a named human attached to it, which is the standard a parent or accreditor will ask about.
This is the same pattern as the research literature-review workflow: agent drafts, human attests, the row records both. It is also why we treat agent identity as a first-class credential. A rationale signed by "the AI" is not auditable. A rationale signed by a named agent, approved by a named instructor, with a timestamp, is.
Start a grading-rationale queue with Dock for education.
FAQ
Does the agent post grades directly to Canvas or PowerSchool? No. The agent drafts. The instructor approves. Only the approved score posts back via API. Dock keeps the draft, the edit, and the attestation as separate fields on the row. See agent audit and compliance.
What happens if the rubric changes mid-term? The agent re-fetches the rubric on each new submission. Existing rows keep the rubric version they were scored against. New rows show the new version. The instructor can re-open a closed row and re-attest under the new rubric if the district requires it.
How do we prove a specific instructor approved a specific grade? Each row stores the reviewer's authenticated identity, the timestamp, and the agent identity that drafted the rationale. The full lineage is queryable. The same model we use for agent identity lifecycle applies: every action has a named principal.
Can the agent grade without instructor approval for low-stakes work? The district sets the policy per assignment type. Formative quizzes can auto-post with agent scores. Summative work requires attestation. The policy lives on the assignment, and the Dock row records which policy applied. See agent audit and compliance for the audit trail.
