.gitignore

/target
.worktrees/
data/
!case-studies/heritage-aerial/data/
!case-studies/heritage-aerial/data/**
!case-studies/mcp-ies-rag-demo/data/
!case-studies/mcp-ies-rag-demo/data/**
!case-studies/modip-plastics-kg/data/
!case-studies/modip-plastics-kg/data/**
!case-studies/robot-safety-security-crosswalk/data/
!case-studies/robot-safety-security-crosswalk/data/**
!case-studies/foundry-owl-crosswalk/data/
!case-studies/foundry-owl-crosswalk/data/**
# OntoAxiom ground truth + scored results — the benchmark claims are unverifiable without them
!benchmark/ontoaxiom/data/
!benchmark/ontoaxiom/data/**
# Ignore generated OWL files in root, but keep benchmark references
/*.owl

# Reasoner benchmark artifacts (downloaded corpora, jars, generated files)
benchmark/reasoner/corpora/
benchmark/reasoner/lib/
benchmark/reasoner/*.class
benchmark/reasoner/results/
__pycache__/
python/oo_lite_test.duckdb
.identifiers-guard
.superpowers/

# Pipeline output demo/README.md documents as regenerated on every run and
# not checked in (see its "Derived artifacts" section). Was untracked but
# not ignored, so it showed up as permanent noise in `git status`.
demo/derived/
demo/corpus_extracted/
demo/_review.jsonl

# Confidentiality guards, promoted from one case study's .gitignore to the root.
# The pattern that protected `case-studies/foundry-owl-crosswalk/` worked, and it
# protected exactly one directory out of eighteen. A naming discipline defeated
# by one misnamed file is not a guard.
#
# Outreach drafts name real recipients. Never publish them.
OUTREACH.md
**/OUTREACH.md
**/OUTREACH*.md
**/outreach/
**/CAMPAIGN*.md
**/PROSPECTS*.md
**/PRIVATE*.md
**/BIDNOBID*.md
**/*-internal.md

# Environment files. `studio/.env` was committed once; it held a port number and
# no harm was done, which is the only reason this is a near miss and not a story.
.env
.env.*
!.env.example
**/.env
!**/.env.example

# In-progress research. This was protected only by .git/info/exclude, which is
# local-only: it does not travel with the repo, it is gone on a fresh clone, and
# it is invisible to a collaborator or to an agent working in one.
benchmark/handshake_gate/

# The native Isabelle checker, linked from the committed generated ML by
# isabelle/build_native.sh. A build artefact, not a source: it is a machine-specific
# binary linked against the Poly/ML runtime in whichever Isabelle installation was
# found, so committing it would ship a file that runs on one machine.
isabelle/build/

# Claude Code agent worktrees: machine-local, never part of the repo.
.claude/worktrees/

# rocq/ build artefacts. Nothing here is a source.
#
# theories/FixtureData.v is GENERATED from tests/fixtures/horn/ by
# gen_fixture_data.sh on every build, deliberately rather than committed: a
# committed copy of a fixture is a second file that can drift from the first, and
# a kernel gate over a drifted copy is a gate over nothing.
#
# driver/rocq_horn_core.ml* is the extraction output and is NOT committed, which
# is the opposite of what isabelle/driver/oo_horn_generated.ML does. The reason
# the Isabelle export is committed is that regenerating it needs several
# gigabytes of Isabelle; regenerating this one needs a container image CI already
# pulls, so committing it would buy nothing and add an artefact no job rebuilds.
rocq/build/
rocq/theories/FixtureData.v
rocq/theories/*.vo
rocq/theories/*.vok
rocq/theories/*.vos
rocq/theories/*.glob
rocq/theories/.*.aux
rocq/Makefile.rocq
rocq/Makefile.rocq.conf
rocq/.Makefile.rocq.d
rocq/driver/rocq_horn_core.ml
rocq/driver/rocq_horn_core.mli
rocq/driver/*.cmi
rocq/driver/*.cmx
rocq/driver/*.o