Accreditation Mapper — Outcomes-to-Standards Documentation Team

Handles the compliance layer above the course: mapping course outcomes to program outcomes to an accreditation standard's criteria, assembling the evidence behind each mapping, finding the gaps, and drafting self-study prose that a review panel can trust. This skill extends the suite's alignment idea one level up — the Alignment Gate checks outcome ↔ assessment ↔ schedule within a course; this skill checks course ↔ program ↔ standard, with the same discipline: claims are verified against artifacts, not accepted because they look complete.

Prime rule: the mapping is the professor's claim; the evidence check is this skill's job. A matrix cell saying "LO2 supports Program Outcome 3" is asserted by the professor — the skill then verifies whether assessment evidence actually exists behind it and labels the cell honestly. A beautiful matrix with hollow cells is documentation theater, and review panels can smell it.

Quick Start

Map my CS 201 outcomes to our ABET student outcomes
帮我把课程目标对应到毕业要求，做工程教育认证的课程矩阵
Assemble the evidence package for our program assessment report
Which of our program outcomes have weak assessment coverage?
Draft the continuous-improvement section of our self-study from last year's changes

Modes

Mode dispatch rule: evidence, gap, and self-study all consume a confirmed matrix. If none exists, run map first — drafting compliance prose from an unverified mapping is exactly the failure mode this skill exists to prevent. Detect intent in any language.

Does NOT trigger

Agent Team (4)

Workflow (map mode)

Phase 0  INTAKE      — professor supplies the program outcomes AND the standard's
                       criteria as actual current text (verbatim, with version/year).
                       The generic structures in references/accreditation_frameworks.md
                       are scaffolding for orientation only — never a substitute.
                       Missing standard text = [NEEDS PROFESSOR INPUT], full stop.
         🧑 checkpoint: standard version + text confirmed
Phase 1  REGISTER    — standards_analyst normalizes criteria into a register: ids,
                       verbatim text, evidence type demanded (direct / indirect /
                       process), measurability notes, vague-criterion flags
         🧑 checkpoint: register confirmed (incl. professor's reading of vague criteria)
Phase 2  MATRIX      — matrix_builder builds course-LO × program-outcome × criterion
                       matrix; professor claims each cell's strength (Introduce /
                       Reinforce / Master, or the institution's own scale)
Phase 3  VERIFY      — per-cell evidence status from the passport: LO.assessed_by →
                       assessment_plan → artifacts[] — does the evidence chain exist?
                       CLAIMED / EVIDENCED / HOLLOW per cell
         🧑 checkpoint: matrix + hollow-cell flags + over-mapping counts presented;
            professor confirms, revises claims, or accepts gaps knowingly
Phase 4  RECORD      — confirmed matrix saved from templates/outcome_matrix_template.md,
                       pinned to the standard version; this is the artifact the
                       `evidence`, `gap`, and `self-study` modes consume

evidence mode runs evidence_assembler over the confirmed matrix → templates/evidence_index_template.md. gap mode reads the matrix column-wise (criteria with no EVIDENCED cells across the mapped courses) and hands remediation candidates to course-designer redesign. self-study mode requires both the matrix and the evidence index and runs selfstudy_writer.

Iron rules

1. The standard's text comes from the professor — current and verbatim. Bodies revise criteria; the reference file's generic structures are labeled possibly-outdated scaffolding. No current text from the professor → the mapping target is [NEEDS PROFESSOR INPUT], and no matrix is built against a guess.
2. Claims and evidence are separated. The professor claims mappings; the skill verifies evidence existence. Every cell carries CLAIMED (asserted, chain not yet checked), EVIDENCED (assessment evidence chain exists in the passport), or HOLLOW (claimed, no evidence chain). The skill never upgrades a label as a courtesy.
3. No compliance inflation. Self-study prose strength is capped by evidence status: "students demonstrate X" requires an artifact showing it. HOLLOW cells cannot generate "students demonstrate" sentences — they generate gap statements with remediation plans, which is what honest continuous improvement looks like.
4. Read-only on the course design. Gaps route to course-designer; this skill never edits outcomes to fit a standard. An auditor that rewrites what it audits stops being an audit — same rule as Gate 1.5, one level up.
5. Version pinning. The matrix records which version/year of the standard it maps against. A standard revision invalidates the matrix loudly — the skill refuses to reuse a matrix pinned to a superseded version without re-confirmation — never silently.

Outputs

• outcome_matrix.md — from templates/outcome_matrix_template.md; the mapping of record, version-pinned to the standard
• evidence_index.md — from templates/evidence_index_template.md (evidence mode)
• gap_analysis.md — prioritized coverage gaps with remediation routing (gap mode)
• self_study_section.md — footnoted draft, every factual sentence traceable (self-study mode)

References

• references/accreditation_frameworks.md — generic shapes of ABET, AACSB, institutional review, and 工程教育认证; evidence-type taxonomy; mapping anti-patterns. Orientation only — carries its own version warning.
• templates/outcome_matrix_template.md
• templates/evidence_index_template.md
• Shared: shared/course_passport_schema.md, shared/alignment_gate_protocol.md, shared/checkpoint_protocol.md