seldon-course-pipeline

name	seldon-course-pipeline
description	End-to-end university course production — research findings to published multilingual course material with review gates

Seldon Course Pipeline — Automated Course Production

Chains research findings into published course material through a governed pipeline with quality gates.

Usage

/seldon course-pipeline [department] — run full pipeline for department (triggers research-cycle first) /seldon course-pipeline --from-cycle [cycle_id] — produce course from an existing research cycle /seldon course-pipeline batch [department] — produce courses for all confirmed findings lacking courses

Pipeline

RESEARCH ──→ OUTLINE ──→ CONTENT ──→ TRANSLATE ──→ REVIEW ──→ PUBLISH
   │            │           │            │            │          │
   └─ /seldon   └─ structure └─ write    └─ ES/PT/FR  └─ quality  └─ commit
     research-     sections     full       cultural      gate       + log
     cycle         + duration   module     adaptation    (>0.7)

Phase 1: RESEARCH

If --from-cycle is provided, load the existing cycle from state/streeling/research-cycles/{cycle_id}.json.

Otherwise, invoke /seldon research-cycle [department] to run a fresh research cycle.

Gate: Research must produce confirm or discover_question conclusion with confidence >= 0.5. If not, pipeline stops with status rejected and logs the reason.

Phase 2: OUTLINE

Generate a course outline from research findings:

1. Read department curriculum (state/streeling/departments/{dept}.department.json)
2. Check existing courses to avoid duplication (state/streeling/courses/{dept}/en/)
3. Determine level (beginner/intermediate/advanced) based on:
   • If finding extends a beginner topic → intermediate
   • If finding is foundational → beginner
   • If finding requires prior courses → advanced
4. Structure into sections (3-6 sections, 15-45 min total)
5. Plan practice exercises (at least 1 per main section)

Output: Outline object with section titles, estimated durations, exercise sketches.

Phase 3: CONTENT

Write the full course module in English following the standard format:

---
module_id: {dept-code}-{sequence}-{slug}
department: {department}
course: "{curriculum area}"
level: {level}
prerequisites: [{list}]
estimated_duration: "{N} minutes"
produced_by: seldon-course-pipeline
research_cycle: {cycle_id}
pipeline_id: {pipeline_id}
version: "1.0.0"
---

# {Title}

> **{Department Full Name}** | Level: {level} | Duration: {duration}

## Objectives
[3-5 learning objectives]

---

## {Sections with content}

### Practice Exercise
[Hands-on exercises]

---

## Key Takeaways

## Further Reading

---
*Produced by Seldon Course Pipeline {pipeline_id}.*
*Research: {cycle_id} | Belief: {value} ({confidence})*

Content quality requirements:

• Factual claims must trace to research evidence
• Examples must be concrete, not abstract
• Practice exercises must be doable without special tools (for music: just a guitar)
• Tone matches department persona's voice settings
• No unexplained jargon — define terms on first use

Save to: state/streeling/courses/{department}/en/{module_id}.md

Phase 4: TRANSLATE

Check policies/multilingual-policy.yaml for translation requirements.

For each required language (currently: es, pt, fr):

1. Translate content preserving structure and frontmatter
2. Apply cultural adaptation:
   • Spanish (es): Use Latin American music examples where relevant, formal "usted" for technical content
   • Portuguese (pt): Brazilian Portuguese with bossa nova / MPB examples for music
   • French (fr): Metropolitan French, reference chanson tradition for music
3. Preserve all Markdown formatting, code blocks, and frontmatter fields
4. Add language suffix to module_id in frontmatter

Save to: state/streeling/courses/{department}/{lang}/{module_id}.{lang}.md

Gate: Translation must preserve all sections, exercises, and key terms. Quick verification: section count must match English version.

Phase 5: REVIEW

Quality gate before publishing:

1. Accuracy check: Do claims match research evidence? (belief_confidence >= 0.7 to auto-approve)
2. Completeness check: All sections present? At least 1 exercise?
3. Curriculum alignment: Does module fit the department's curriculum progression?
4. Duplication check: No significant overlap with existing courses?
5. Format check: Frontmatter valid? Module ID follows convention?

Verdicts:

• approved — proceed to publish
• revise — return to CONTENT phase with specific feedback (max 2 revisions)
• reject — pipeline ends, log reason, flag for human review

If confidence < 0.7 but >= 0.5: flag for human review instead of auto-approving. If confidence < 0.5: auto-reject (should have been caught at Phase 1 gate).

Phase 6: PUBLISH

1. Verify all files are written to correct paths
2. Log pipeline run to state/streeling/course-productions/{pipeline_id}.json conforming to schemas/course-production.schema.json
3. Update department weights if research cycle was fresh (delegate to research-cycle compound step)
4. Commit all course files with conventional commit:

   feat: {DEPT-CODE}-{SEQ} — {course title}
   

5. Print summary:

Course Published: {module_id}
Department: {department}
Title: {title}
Level: {level} | Duration: {duration}
Languages: en, es, pt, fr
Research: {cycle_id} (belief: {value}, confidence: {confidence})
Files: {count} files across {language_count} languages

Batch Mode

When invoked with batch, scan state/streeling/research-cycles/ for cycles with:

• conclusion = confirm or discover_question
• belief_confidence >= 0.5
• No corresponding course in state/streeling/courses/{dept}/en/

Run the pipeline (phases 2-6) for each qualifying cycle. Print batch summary.

Pipeline State

Pipeline runs are logged to state/streeling/course-productions/{pipeline_id}.json.

This allows:

• Resuming interrupted pipelines
• Tracking production metrics (duration, token usage, revision count)
• Identifying bottleneck phases
• Measuring course production velocity per department

Governance

• Article 2 (Transparency): All production steps are logged with evidence trails
• Article 9 (Bounded Autonomy): Course scope bounded by department curriculum
• Article 11 (Ethical Stewardship): Content reviewed before publication
• Multilingual Policy: Translation required for departments in policy scope
• Streeling Policy: Knowledge transfer quality standards
• Confidence thresholds: Auto-publish >= 0.7, human review 0.5-0.7, reject < 0.5