Skip to content

Release Notes

Versioned changelog for the Juddges App platform. Versions track annotated git tags prod-vX.Y.Z and correspond to manually built/pushed Docker images.

v1.3.0 — 2026-05-13

Commits: 10 commits since prod-v1.2.1

Highlights

  • Meilisearch engine upgraded to v1.43 (from v1.13) across both dev and prod compose. Includes a documented dump-based migration runbook (docs/how-to/upgrade-meilisearch.md) exercised end-to-end against a v1.13.3 snapshot before release.
  • Search topics are now durable in Supabase. Meilisearch can be rebuilt from a Supabase snapshot — Supabase is the new ground truth for the autocomplete topics index.
  • Search analytics attribute to the logged-in user, with a new GET /api/search/analytics/history endpoint so users can retrieve their own search history.
  • Three new CLI scripts for topic operations: import, sync to Meili, and validate.
  • Highlight fix on document cards — query matches stay visible even under the 3-line clamp.
  • Service-worker hardening (v3) — RSC payloads no longer cached; fixes the stale auth-redirect issue from v1.2.1.

Frontend Changes (frontend/)

New Features

Search proxies forward Supabase session for analytics attribution
  • Technical: frontend/app/api/search/{autocomplete,documents,topic-click}/route.ts now forward the user's Supabase Authorization: Bearer … token to the backend, so analytics rows can be attributed to the authenticated user. Anonymous traffic is unchanged (NULL user_id).
  • Why: Enables per-user search history and meaningful "your top searches" analytics — previously all traffic was anonymous in search_analytics.
  • Impact: Foundation for personalized search history and account-level insights. (Commit: 2c8722c)

Bug Fixes

Highlight stays visible under line-clamped document cards
  • Technical: New recenterHighlightSnippet helper in frontend/lib/highlight.ts plus ensureMarkVisible prop on QueryHighlight shifts a snippet's mark close to the start before clamp truncation cuts it off. DocumentCard passes ensureMarkVisible={!showExtended}.
  • Why: With 3-line clamping, the matched word often fell off the bottom — users couldn't see why a result was returned.
  • Impact: Every result card now visibly proves its match. (Commit: 819c084)
Service worker skips RSC payloads and narrows static cache (v3)
  • Technical: frontend/public/sw.js bumped to CACHE_VERSION = v3. Never caches ?_rsc= requests; static-asset cache limited to immutable build hashes; install/activate flow tightened.
  • Why: Previous SW occasionally replayed middleware auth-redirects to logged-in users, sending them to the marketing home page after sign-in.
  • Impact: Auth redirects behave correctly across reloads. (Commit: b4c6cab)

UI / Test Maintenance

  • "Preview Plan" subtext removed from user-card popover (frontend/lib/styles/components/user-card.tsx). (Commit: e4559c3)
  • Drifted unit tests repaired for landing page, schema editor, i18n, api, query-highlight, document-card, and collection-batch. (Commit: 6837fd9)

Backend Changes (backend/)

New Features

Persistent search_topics table — Supabase is the ground truth
  • Technical: New migration supabase/migrations/20260513000001_create_search_topics_table.sql adds a search_topics table keyed by run_id. New backend/app/services/search_topics_store.py provides persist_search_topics_run and load_latest_search_topics_run. scripts/generate_search_topics.py now persists topic runs to Supabase, then rebuilds Meilisearch from that snapshot via the new push_topics_run_to_meilisearch helper.
  • Why: Previously, Meili was the only home of generated topics — losing the Meili volume meant regenerating topics from scratch. This makes topics durable and auditable, and decouples regeneration from rebuild.
  • Usage: Run scripts/generate_search_topics.py as before — it auto-persists. To rebuild Meili from the existing snapshot, run scripts/sync_meilisearch_topics.py.
  • Impact: Faster recovery from Meili volume loss; topic runs are versioned and inspectable. (Commit: e42a525)
Search analytics attributed to user + history endpoint
  • Technical: Migration 20260513000002_add_user_id_to_search_analytics.sql adds nullable user_id UUID + partial index. record_search_query / record_topic_click accept user_id. autocomplete, documents_search, and topic_click endpoints capture it via get_optional_user. New migration 20260513000003_create_get_user_search_history.sql exposes RPC get_user_search_history (auth-required, RLS-safe) backing GET /api/search/analytics/history.
  • Why: Sets up a foundation for per-user history UX and meaningful engagement metrics.
  • Usage: Logged-in users: GET /api/search/analytics/history returns their recent searches; anonymous calls return rows with user_id = NULL for aggregate analytics only.
  • Impact: Personalized history is now possible; admin analytics dashboards can split traffic by anonymous vs authenticated. (Commit: 2c8722c)

API Changes

  • New endpoint: GET /api/search/analytics/history — returns the authenticated user's recent searches (RLS-enforced).
  • Analytics endpoints (/autocomplete, /documents, /topic-click) now optionally accept Bearer auth and attribute rows to the authenticated user.

Scripts & Tooling

Three new search-topics CLI tools
  • scripts/sync_meilisearch_topics.py — re-publish the latest persisted Supabase snapshot to Meilisearch (no regeneration).
  • scripts/import_search_topics_snapshot.py — load a snapshot from JSON into the search_topics table.
  • scripts/validate_search_topics.py — validate a topic run for completeness, doc counts, and schema conformance.

Usage:

docker compose run --rm backend python scripts/sync_meilisearch_topics.py
docker compose run --rm backend python scripts/import_search_topics_snapshot.py <path.json>
docker compose run --rm backend python scripts/validate_search_topics.py

(Commit: d9a7be7)

Infrastructure & DevOps

Meilisearch bumped to v1.43 (dev + prod)
  • Technical: docker-compose.dev.yml and docker-compose.yml both pinned to getmeili/meilisearch:v1.43.0. The prod service image is parameterized via ${MEILISEARCH_IMAGE_TAG:-v1.43.0} so rollback to v1.13 is a one-line .env edit + recreate. Memory ceiling and nofile ulimits unchanged.
  • Why: Older v1.13 line is no longer current; the v1.43 line includes performance and stability improvements relevant to hybrid vector indexes.
  • How to migrate prod: follow the dump-based runbook at docs/how-to/upgrade-meilisearch.md. The procedure was exercised on 2026-05-13 against a local v1.13.3 snapshot (12,307 judgments + 149 topics) and validated end-to-end before this release. Acceptance checks include /version, document counts, embedder config preservation, direct keyword search, and backend autocomplete with <mark> highlighting.
  • Impact: Engine upgrade is decoupled from any schema or product change. Phase 2 (judgments_pl / judgments_en index split) and Phase 3 (multi-search query path) are tracked separately. (Commits: 3f0fdfb, prod compose update in this release.)

Documentation

  • Dataset citation cleanup: README.md, docs/architecture/*, docs/getting-started/*, docs/open-science/*, docs/how-to/data-ingestion.md plus new ecosystem and home-page references — cite JuDDGES HF datasets with DOIs and drop the HFforLegal/case-law reference. (Commit: 9f57061)
  • populate-search-topics runbook updated to document the Supabase ground-truth flow and the sync_meilisearch_topics.py rebuild path. (Commit: e42a525)

Breaking Changes

None at the public API level. A few internal notes for operators:

  • Meilisearch on-disk format is not backward-compatible v1.13 → v1.43. A direct image tag bump without a dump-based migration will fail at boot. The migration runbook is mandatory; see Migration Notes below.
  • search_analytics rows written before this release have user_id = NULL permanently — the column is nullable on purpose.
  • Topic regeneration (generate_search_topics.py) now writes to Supabase first; if SUPABASE_SERVICE_ROLE_KEY is missing in the run environment, the script will fail before touching Meili. Old behavior (Meili-only) is no longer supported.

Migration Notes

  1. Upgrade Meilisearch from v1.13 → v1.43 (do this first; it requires a dump-based migration). Follow docs/how-to/upgrade-meilisearch.md step-by-step. The pre-existing prod data volume cannot be reused as-is across the v1.13 → v1.43 gap; the new container will refuse to boot.
  2. Run new Supabase migrations (auto-applied by your normal deploy):
  3. 20260513000001_create_search_topics_table.sql
  4. 20260513000002_add_user_id_to_search_analytics.sql
  5. 20260513000003_create_get_user_search_history.sql
  6. Bootstrap topic snapshot (one time, after deploy):
docker compose run --rm backend python scripts/generate_search_topics.py

This populates search_topics and rebuilds Meili in one pass. 4. Verify with scripts/validate_search_topics.py and check /api/v1/search/autocomplete?q=fraud.

Known Issues

  • Meili settings atomic-apply at prod startup still occasionally drifts on the bge-m3 embedders block. Verify filterableAttributes / synonyms post-deploy.

Full Commit Log

6837fd9 test(frontend): repair drifted unit tests for landing, schema editor, i18n, api, and collection-batch
819c084 fix(search): recenter snippet so highlight survives line-clamp in document cards
3f0fdfb chore(infra): bump dev Meilisearch to v1.43
d9a7be7 feat(search-topics): add CLI tools to sync, import, and validate topic snapshots
e42a525 feat(search-topics): persist topic runs in Supabase and rebuild Meili from snapshot
9f57061 docs(datasets): cite JuDDGES HF datasets with DOIs, drop HFforLegal/case-law
2c8722c feat(search): attribute analytics to user + add history endpoint
b4c6cab fix(sw): skip RSC payloads, narrow static cache, bump CACHE_VERSION v3
e4559c3 chore(navbar): drop "Preview Plan" subtext from user-card popover