966 B

Raw Blame History

Multi-Source Ingestion

The multi-source ingestion layer lets Didactopus build one draft domain pack from several heterogeneous inputs describing the same course or topic.

Why this matters

Real course material is often scattered across:

syllabus files
lesson notes
transcripts
assignment sheets
HTML pages
supplemental markdown

A single-source parser is too narrow for serious curriculum distillation.

Pipeline

detect adapter by file extension or naming convention
normalize each source into a NormalizedSourceRecord
merge sources into a NormalizedCourse
extract concept candidates
run rule-policy passes
emit merged draft pack
emit conflict report and attribution manifest

Conflict report categories

duplicate lesson titles across sources
repeated key terms with different local contexts
modules with no explicit exercises
project-like content needing manual review
lessons with thin mastery signals