# The raw sources stay LOCAL and are never committed.
#
# This repository is public and Apache-2.0. The measurement fixtures are real
# third-party course material, which this project holds no right to relicense
# -- committing it would make LICENSE assert something false about those files.
#
# What IS committed is the part that carries the judgment: `corpus.py`, this
# README, and every hand-written file under `ground-truth/`. That analysis is
# original work, and it is also the reusable half -- another contributor
# reproduces the METHOD by pointing `corpus.py survey` at their own material,
# not by re-reading these bytes.
#
# The cost, stated plainly: nobody else can reproduce the exact numbers, only
# the procedure. For a manual harness that never runs in CI, that is the right
# trade.
# `sources/*`, not `sources/`: git does not descend into an excluded
# DIRECTORY, so a negation for a file inside one can never re-include it.
# Excluding the contents instead leaves the exception below reachable.
sources/*

# EXCEPTION: `medium-10-reunion-plataforma.md` is committed.
#
# The rule above exists because the other sources are third-party material this
# project cannot relicense. That reason does not apply to this one: it was
# WRITTEN for this repository as #694's oracle fixture, so it is original work
# under the repository's own licence.
#
# Committing it is the whole point. #694's complaint is that "extraction
# improved" and "extraction regressed" are both unfalsifiable claims; an oracle
# whose source cannot be checked out reproduces that problem one level up.
# With this file tracked, anyone cloning can run the measurement and get the
# same numbers, even with no third-party corpus present at all.
!sources/medium-10-reunion-plataforma.md
