The past-paper corpus stays outside the repository

See overview and v2.

Context

The import command needs a directory of past-paper source files to read. The repository holding v2 is meant to be public.

Decision

The corpus lives in a local directory named by an environment variable (CORPUS_DIR), never committed. Nothing in the tree points a reader at where the data comes from; that is written down in the docs instead. Parser fixtures are hand-written in the source format rather than copied from real papers.

Alternatives rejected

Committing a sample of the real corpus for fixtures, so the parser and its tests would have something to run against out of the box.

Why

The papers are Al-Karim University material. Committing any of it would have landed it in git history, which cannot be undone cheaply once pushed. Hand-written fixtures let the parser and its tests be read and understood by someone with no access to AKU content at all.

Consequences

A fresh clone of the repository cannot run the import command until CORPUS_DIR is pointed at a local copy of the corpus, and the parser's tests run against hand-written fixtures rather than real paper text.

Status

Current.