Backfill Project Sidewalk contributors from the NSF annual reports

[jonfroehlich]

Why

Project Sidewalk now consumes /api/v1/projects/sidewalk/people/ as its single source of truth for who contributed to PS. That endpoint is driven by ProjectRole rows, so a contributor without one is invisible to Sidewalk — even if they have a member page.

Diffing the four NSF Crowd+AI annual report participant spreadsheets (award #2125087, Y1–Y4 = Oct 2021 → Sep 2025) against the live API found 115 participant rows = 74 unique people. 45 were already linked correctly; the other 29 split into two gaps:

Gap	Count	Who
Member page exists, but no Sidewalk ProjectRole	10	Y1 undergrads + faculty/collaborators (Heer, Hosseini, Zhihan Zhang, Kuang, Milandin, Lakshmanan, Duanhao Zhang, Alex Liu, Rapo, Buchanan)
No Person record at all	17	every one from the UIC subaward (Eisenberg / Labbé / Hammel's team)

What

Adds seed_sidewalk_participants, modelled on seed_project_aliases (#944): a baked-in seed table plus a create-only handler, wired into docker-entrypoint.sh as step 4.10d. There is no shell or DB access on the test/prod servers, so a startup one-shot is the only path for this.

Create-only is the load-bearing property, since this runs on every container start long after editors have tuned these records in /admin. It never modifies an existing Person, never adds a Position to anyone who already has one, and never touches an existing ProjectRole. A link-only entry whose Person can't be found is skipped with a warning rather than recreated, so a rename or merge can't spawn a duplicate.

Dates come from the NSF reporting periods (Oct 1 – Sep 30) — the finest granularity the spreadsheets offer, since they record which annual report someone appeared in, not the day they started. Consecutive appearances merge into a single role.

Also adds Title.PROJECT_COORDINATOR for the UIC project coordinator, whose role "Research Scientist" overstated. Worth knowing for future contributors: a new Title requires a TITLE_ORDER_MAPPING entry — admin/position_admin.py sorts the dropdown by it and would KeyError otherwise — so a test now pins every Title as mapped. No migration is committed (website/migrations/ is gitignored per-environment); the entrypoint's makemigrations/migrate generates the choices-only AlterField on deploy, leaving the column untouched.

Data judgment calls

• Lauren Frame gets two Positions — undergrad through May 2024, then UIC's Occupational Therapy Doctorate from Aug 2024. The NSF sheets had her Y2/Y3 status backwards (Graduate in Y2, Undergraduate in Y3).
• Xia Su and Lisa Orii are deliberately excluded — they appear in the reports but shouldn't be listed as PS contributors.
• Headshots are out of scope. Person.save() supplies the Star Wars fallback; real headshots get uploaded via /admin, which is what sets the Cropper.js crop box (Would be so nice to get an instant preview of an image that you add for cropping #1299). 16 are still being collected.

Testing

python manage.py test website --settings=makeabilitylab.settings_test → 620 passed, 8 skipped. 12 new tests cover idempotency, that admin edits survive a re-run, the held-back sentinel for unconfirmed titles, the mid-grant title change, the new Title, and that seeded people surface through the real API route.

Dry run against a local DB:

seed_sidewalk_participants (dry run): 17 people, 18 positions, 18 project roles created; 0 held back pending a title.

18 positions because Lauren has two. On prod expect 17 people, 18 positions, 27 roles — the extra 9 roles are the link-only people the local dev DB doesn't have.

No UI changes, so no Pa11y run or screenshots (data-only).

Follow-ups (not in this PR)

• 4 people whose existing ProjectRole dates don't cover the years they were reported — Michael Duan, Devon Snyder, Dylan Bunarto, Manjiri Dehankar. /admin edits; they already appear in the API, only the dates are wrong.
• Headshot uploads for the 16 outstanding UIC folks.

🤖 Generated with Claude Code