feat: add ontology material ingestion

.planning/phases/005-ontology-materials/005-RESEARCH.md

@@ -0,0 +1,28 @@
+# Phase 5 Research: Ontology and Learning Materials
+
+## Findings
+
+The first useful ontology proof does not need heavy parsing. It needs a clean
+boundary that proves uploaded material can become inspectable candidate
+knowledge with provenance.
+
+The MVP should:
+
+- store material metadata and source text
+- extract concept candidates from known backend interview concepts
+- create prerequisite edges from a small deterministic rule set
+- identify weak concepts when source support is thin
+- never mark generated or inferred content as canonical
+
+## Recommended Shape
+
+- `internal/ontology` owns material ingestion, candidate storage, and snapshot.
+- HTTP exposes JSON ingestion first.
+- Evidence references use the existing workflow shared type.
+- Gap records distinguish source-backed weakness from generated inference.
+
+## Risks
+
+- Overbuilding parsers too early would violate YAGNI.
+- Treating keyword extraction as canonical knowledge would violate OpenSpec.
+- A future parser can replace the analyzer behind the same service boundary.