Make projects private by default with a single is_visible flag (#1300)

[jonfroehlich]

Closes #1300.

Summary

Public project visibility was a hodgepodge: an implicit "has a thumbnail AND a publication" check duplicated across views and templates. This replaces it with a single editor-controlled Project.is_visible flag so we can:

• create a project entry the moment a project begins,
• add people to it while we're working (keeping the people-tracking accurate for NSF reporting, etc.),
• keep it hidden from the public until it's ready to go live.

What changed

• Model (models/project.py): new is_visible field. Project.save() defaults new projects to private (False); can_show_online() now returns bool(is_visible). The field is intentionally nullable with no DB default so the migration adds existing rows as NULL rather than silently hiding every current project on first deploy.
• One-shot backfill (backfill_project_visibility, wired into docker-entrypoint.sh): resolves legacy NULL rows to their previous visibility (thumbnail + publication). Keyed on is_visible__isnull=True, so it's idempotent and never clobbers a later manual admin choice. Same pattern as backfill_num_pages.
• Replaced the scattered checks with is_visible=True in the landing page (index), project gallery (project_listing, active + completed), and member views; dropped the now-redundant per-item guards in project_listing.html. Ordering uses nulls_last so a visible project with no publication yet doesn't sort to the top.
• Individual project page: private projects return 404 for the public but remain previewable by logged-in staff (so you can build a project before going live).
• Admin: is_visible added to fieldsets / list_display / list_filter, rendered as a clean checkbox ("Visible on website").
• Supporting: data-health check surfaces is_visible; seed_demo_projects marks demo projects visible. Version bumped to 2.6.0.

Testing

• 130 tests pass (python manage.py test website --settings=makeabilitylab.settings_test), including 16 new ones in test_project_visibility.py covering the private-by-default behavior, backfill idempotency, gallery/landing filtering, and the staff-preview gate. Added a make_project fixture helper.
• Backfill verified in the real entrypoint on local dev: "Resolved 18 legacy project(s) — 12 visible, 6 private" — existing visibility preserved.
• Gate verified live: private project → 404 for anonymous, visible → 200.

UI / accessibility note

No markup changed — the project gallery renders the same unchanged snippet; only which projects appear is different (driven by the queryset). The one visible UI addition is a checkbox on the admin Project form. Pa11y shows no new violations (only the site's pre-existing global footer/color-contrast baseline).

Deploy plan

Merging to master deploys to -test only; production is intentionally not tagged yet so we can validate on makeabilitylab-test.cs.washington.edu first. The entrypoint backfill will preserve existing projects' visibility there.

🤖 Generated with Claude Code

[jonfroehlich]

Follow-up: full visibility coverage (commit 281fe1b)

After an audit of every place a project can be mentioned, found six surfaces beyond the gallery/landing/member/chip set that still referenced a private project. All now route through is_visible:

Surface	Before	Now
Awards page	showed project name as plain text	private project not listed at all (new Award.get_visible_projects)
"Related projects" on a project page	filtered on gallery_image only	also requires is_visible
People listing chips	get_projects_sorted_by_contrib filtered by pub count	also filters is_visible=True
News item related projects	news.project.all()	display filtered to visible (related-news lookup unchanged)
Project-listing umbrella filter	counts/names from project_set.all()	visible-only (no count inflation / name leak)
Landing carousel	private project's banner showed name + 404 link	banners on a private project excluded

Intentionally not changed: /view-project-people/ — per discussion this is an internal people-tracking tool and is left exempt.

Each surface has a regression test in test_project_visibility.py (21 visibility tests; 135 total, all green).