Skip to content

Grading Rubrics for FCC Assignments

Six ready-to-use rubrics for common FCC-based assignments. Each rubric scores 4-5 dimensions across 4 levels (Emerging, Developing, Proficient, Exemplary). Use them directly or adapt to your course weighting.

These rubrics align with the 30+ quality gates defined in src/fcc/data/governance/quality_gates.yaml and the assessment rubrics in src/fcc/data/docs/assessment_rubrics.yaml.

Scoring convention

  • Emerging (1): significant gaps; revision required
  • Developing (2): functional but missing depth
  • Proficient (3): meets all stated requirements
  • Exemplary (4): exceeds requirements with insight or polish

Sum dimensions for a raw score; convert to letter grade per course policy.

Rubric 1: Custom Persona YAML

Use after weeks 3-4 of the 12-week curriculum.

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
R.I.S.C.E.A.R. completeness Missing 3+ of the 10 components Missing 1-2 components or one is shallow All 10 present and non-trivial All 10 present; each reflects domain expertise and cross-references literature
Role clarity Role overlaps existing persona or is vague Role is distinct but scope is fuzzy Clearly differentiated role with testable boundaries Role occupies a previously underserved niche with rationale
Quality-gate alignment No gates wired; schema fails Passes schema but no gates referenced All applicable quality gates referenced in responsibilities New quality-gate candidate proposed with justification
Collaborator graph No collaborators listed Flat list, no edge types Typed collaborators with reasonable edge types Collaborators form a coherent team; upstream/downstream roles justified
Adoption checklist Generic or missing Checklist present but vague Concrete, actionable checklist matching role Checklist is usable by a real team on day one

Common pitfalls

  • Missing archetype or role_adoption_checklist (see common mistakes M6)
  • Category collision with the 20 core categories
  • Responsibilities that are actually role_skills

Cross-links: quality_gates.yaml, custom-persona-design-guide.md.

Rubric 2: Scenario Authoring

Use after week 6.

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
Testability No pass/fail criteria Criteria stated but not checkable Every criterion maps to a quality gate or metric Criteria + expected ranges + tolerance
Coverage One persona, one phase All 3 phases but thin All 3 phases with >= 3 distinct personas Rich persona selection with justification per slot
Realism Toy/contrived inputs Plausible but generic Grounded in a real domain problem Real data + documented provenance
Reproducibility No seed, no provider config Seed set; provider unspecified Seed + provider + model + temperature all set Includes mock fallback for CI reproducibility
Workflow choice No workflow reference Default workflow only Workflow variant chosen and justified Custom workflow + explanation of node selection

Common pitfalls

  • Omitting setup.ai_config (causes silent provider drift across runs)
  • Quality-gate ID typos
  • Missing inputs: block

Rubric 3: Plugin Development

Use after week 10.

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
Contract correctness Doesn't load via entry point Loads but wrong plugin_type Correct type + all required methods Type + methods + contract tests + idempotency
Tests No tests Smoke test only Unit tests + >= 80% coverage Unit + integration + failure-mode tests
Docs No docstring, no README Module docstring only Full API docs + usage example Docs + tutorial + troubleshooting section
Dependency hygiene Adds unpinned core deps Pinned but broad Pinned, narrow, extras-only Works with zero extras; optional deps guarded by import
Observability No events, no traces Some debug prints Publishes relevant events Events + traces + metrics + constitution tier

Common pitfalls

  • Entry point in pyproject.toml references wrong module path (see M14)
  • Forgetting plugin_type class attribute (see M15)
  • Leaking credentials into logged events

Rubric 4: Workflow Visualization

Use after week 5 or week 11.

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
Architecture clarity Boxes with no labels Labeled but cluttered Clean, readable, legend present Layered diagram with phases clearly zoned
Graph correctness Wrong node types Correct nodes, wrong edges Valid per workflow schema Validates AND generates a diff against default variants
Node/edge fidelity Fewer nodes than required Required count met; generic Each node has justification Nodes tied to specific R.I.S.C.E.A.R. responsibilities
Phase zoning FIND/CREATE/CRITIQUE not visible Labeled phases Phases color-coded and counted Phases + sub-phases with transition rationale
Accessibility No alt text, low contrast Alt text present Alt text + high contrast + legend Full WCAG AA; text-only fallback included

Rubric 5: Compliance Audit Report (EU AI Act Mapping)

Use after week 11 or in a dedicated compliance module.

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
Risk classification Category not assigned Assigned but not justified Correct category + justification referencing Reg 2024/1689 articles Classification with counter-scenarios and sensitivity analysis
Requirement mapping < 50% of applicable requirements cited 50-80% cited All applicable cited with evidence All cited + NIST AI RMF crosswalk + gap analysis
Evidence quality Assertions without artifacts Some artifacts Every requirement has traceable evidence Evidence graph built with compliance.evidence_graph
Remediation plan None Vague roadmap Prioritized plan with owners Plan + acceptance criteria + re-audit trigger
Reproducibility Manual audit only Semi-scripted Uses CompliancePipeline Full automated audit integrated with CI

Cross-links: src/fcc/data/compliance/eu_ai_act_requirements.yaml, src/fcc/data/compliance/nist_ai_rmf_mapping.yaml.

Rubric 6: Capstone Project

Use at end of semester (week 12).

Dimension Emerging (1) Developing (2) Proficient (3) Exemplary (4)
Integration breadth 1-2 subsystems used 3-4 subsystems 5-6 subsystems (personas, workflows, plugin, event bus, KG, docs) All required 7 subsystems + at least one advanced (RAG/federation/compliance)
Custom personas Template-only 2 custom personas 3+ custom, well-specified 3+ with dimension profiles and discernment matrix
Custom plugin Non-functional Functional, single type Functional + tested + documented Cross-plugin orchestration (2+ plugin types interacting)
Event-bus evidence No events Events fire but unfiltered >= 5 event types published and consumed Full subscriber with filtering, serialization, and replay demo
Presentation Unclear demo Demo works, narrative weak Clear demo + architecture walkthrough Publication-quality narrative + design decisions + lessons learned

Scoring guidance: a Capstone should earn >= 3 on every dimension to be considered Proficient overall. Dimensions at 1 should block the grade until revised; this communicates that integration is the point of the capstone.

Common pitfalls (across all rubrics)

  1. Using mock mode for a live demo -- fine for CI, confusing on stage. Require AI mode with temperature=0 for capstone demos.
  2. No version pinning -- students should cite FCC version explicitly.
  3. Missing cross-references -- reward student work that links back to docs/for-beginners/, guidebook chapters, and ADRs.
  4. Uneven depth -- a beautiful persona plus a toy workflow is a 2, not a 3.