Pages for the grading jobs people actually do.
Find the workflow by what students submit, the rubric being enforced, the human review role, the report audience, and the deployment stakes.
The three workflows that deserve their own depth.
Each page is written as a real scenario: what goes wrong in human grading, what criteria matter, what the AI should inspect, and what output the teacher or scoring leader should trust.
Source-based writing, arguments, DBQs, and literary analysis.
Score claim, evidence, reasoning, organization, conventions, multilingual learner patterns, and rubric-specific criteria.
Proofs, diagrams, partial credit, and mathematical reasoning.
Read notation and diagrams, check proof structure, isolate invalid steps, and explain where partial credit should land.
Messy workbook pages become next-day instruction.
Grade pages at class scale, build error books, identify recurring misconceptions, and generate parent-ready feedback.
Assessment context changes the controls.
The use case is not only the subject. It is also the institution, the stakes, the review policy, and the deployment environment.
Tutoring school
Fast cloud onboarding, teacher review queues, error books, mastery dashboards, and parent-friendly feedback.
High-stakes exam
Anchor approval, human-in-loop calibration, condition-code routing, audit logs, subgroup reports, and on-prem deployment.
Institution or district
Private cloud, SSO, rubric libraries, role permissions, LMS integration, and controlled data boundaries.
Curriculum publisher
Item-bank analytics, feedback templates, multi-subject coverage, batch scoring, and content improvement loops.
When a workflow needs custom handling.
A workflow needs custom handling when the submission format, rubric criteria, human review protocol, or reporting expectation changes. Otherwise it belongs as a subject variant.
Submission, decision, human, stakes.
Implementation stays concrete: what the learner submitted, what decision must be made, who reviews it, and which controls are required before scale.
Start with the submission
What does the student submit: typed text, handwriting, images, PDFs, speech, diagrams, tables, or mixed-language work?
Name the scoring decision
Is the output a score, feedback, routing flag, misconception cluster, item statistic, or technical-report table?
Choose the human role
Teacher, SME, scoring leader, psychometrician, table leader, parent, or student. The report changes by audience.
Set deployment stakes
Cloud for speed, private cloud for isolation, on-prem for high-stakes exams and customer-controlled custody.
A broad subject map, starting with 72 concrete labels.
Most subjects do not need their own page on day one. They need a clear rubric, representative examples, and the right reporting view. This map shows where workflow variants can expand.