feat: Milestone 9 — exposure filtering layer (student_public / research_instructor)

[shaypal5]

Summary

• exposure/filters.py: BundleFilter frozen dataclass defining which metadata artifacts to include; FILTERS dict keyed by ExposureMode; get_filter() accessor.
• exposure/redaction.py: write_metadata_dir() — writes metadata/ containing graph.json, graph.graphml, world_spec.json, latent_registry.json, and mechanism_summary.json.
• exposure/modes.py: apply_exposure(bundle, root, mode) — dispatches filtering; skips entire metadata/ directory for student_public mode, writes full truth for research_instructor.
• Wired into api/bundle.py between dataset-card and manifest steps so exposure filtering is part of the standard bundle-write pipeline.
• 24 new tests covering both exposure modes, filter correctness, metadata file presence/absence, and round-trip integrity.
• Total: 545 tests passing.

Test plan

• pytest tests/exposure/ — 24 tests for filter lookup, metadata writing, and mode dispatch
• pytest — full suite (545 passing)
• ruff check . — clean
• Verify student_public bundles contain no metadata/ directory
• Verify research_instructor bundles contain full metadata/ with all 5 artifacts

🤖 Generated with Claude Code

[github-actions]

pr-agent-context report:

No unresolved review comments, failing checks, or actionable patch coverage gaps were found on PR
#14. Treat this PR as all clear unless new signals appear.

Run metadata:

Tool ref: v4
Tool version: 4.0.20
Trigger: pull request opened
Workflow run: 25092190678 attempt 1
Comment timestamp: 2026-04-29T05:16:59.547794+00:00
PR head commit: a4cb963d4b60b377aa934e4ab184df90618e9085

[Copilot]

Pull request overview

Adds an exposure-mode layer to the bundle write pipeline so that “student_public” bundles omit hidden-truth metadata while “research_instructor” bundles include a metadata/ directory with additional truth artifacts.

Changes:

• Introduces leadforge.exposure modules for per-mode filter lookup and mode dispatch, plus metadata writing for research_instructor.
• Wires exposure handling into leadforge.api.bundle.write_bundle() before manifest generation.
• Adds a new tests/exposure/ suite validating both modes and metadata content.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
leadforge/exposure/filters.py	Defines BundleFilter, FILTERS, and get_filter() for ExposureMode → filter resolution.
leadforge/exposure/modes.py	Adds apply_exposure() entry point used during bundle writing.
leadforge/exposure/redaction.py	Implements write_metadata_dir() to write 5 hidden-truth metadata artifacts in instructor mode.
leadforge/api/bundle.py	Integrates exposure handling into the standard write_bundle() pipeline before manifest generation.
tests/exposure/test_exposure.py	Adds unit + integration tests for filter correctness, metadata presence/absence, and determinism.
tests/exposure/__init__.py	Creates the exposure test package.
.agent-plan.md	Updates milestone tracking/status text.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.