Schema

The corpus is a graph of 13 entity kinds plus a Claim record type that layers sourced assertions over any subject.

Entities

Kind	Purpose
person	The subject of the corpus (exactly one in v0.1.0).
organization	Employer, client, school, or publishing venue.
role	One tenure at one organization.
project	An initiative within or across roles.
achievement	A discrete outcome, typically a CV-bullet atom.
skill	Competency, tiered foundational / applied / domain.
education	Formal credential or substantial training.
publication	Paper, article, talk, podcast.
artifact	Linkable deliverable: repo, demo, design doc.
testimonial	Third-party quote.
cover_letter_seed	Reusable narrative fragment (first person).
target	A specific job application being tailored for.
source_doc	Pointer into the content-addressable source store.

Design invariants

1. Claims over overwrites. Refined facts append; originals stay for audit. The generator scores and picks at render time.
2. Structured impact. Metrics are numeric with explicit units, not parsed from prose. Renderers format freely.
3. Three-layer skill taxonomy. Foundational, applied, domain. Flattening hides the signal recruiters key off.
4. Visibility tiers. Every entity and claim carries public / nda / private; redaction is a render-time filter, not a rewrite.
5. Content-addressable source store. Ingested binaries live outside the repo, keyed by sha256. The corpus references them via SourceDoc metadata.
6. Per-entity schema versioning. Each file carries its own schema_version, so migrations run file-by-file without a flag day.

Reference

Export JSON Schemas per kind:

cv-corpus schema --out schemas/