-- ============================================================================
-- One-off backfill: job_applications.worker_profile_id
--
-- NOT A MIGRATION, and deliberately not one. `worker_profile_id` has existed
-- since migration 000001 and the column is unchanged; what was missing was the
-- frontend setting it when it filed an application from the talent pool. That
-- is fixed going forward. This connects the rows written before the fix.
--
-- A migration would make this part of the schema's history and run once per
-- environment whether or not it was wanted. A script runs when somebody decides
-- to run it, against the environment they name, after reading what it will do.
--
-- WHAT IT MATCHES
--
--   org_id  AND  case-insensitive email
--
-- and nothing else. No fuzzy matching, no name comparison, no "probably the
-- same person". Email is the identity `worker_profiles` itself already asserts:
--
--   CREATE UNIQUE INDEX worker_profiles_org_email_key
--     ON worker_profiles (org_id, email);
--
-- That constraint is why an ambiguous match is not merely unlikely but
-- IMPOSSIBLE: at most one worker profile can exist for any (org_id, email), so
-- the join below can never find two. The HAVING clause asserts it anyway —
-- cheap, and it fails loudly rather than silently picking one if that
-- constraint is ever relaxed.
--
-- `email` is `citext`, so the comparison is already case-insensitive; it is
-- written with lower() so the intent survives a column type change.
--
-- WHAT IT TOUCHES
--
--   job_applications.worker_profile_id   ONLY, and only where it is NULL.
--
-- No status, no score, no timestamp, no other column, no other table. It never
-- creates a worker_profiles row — an application with nobody behind it stays
-- unlinked, which is the honest answer.
--
-- SAFE TO RE-RUN. Rows already linked are excluded, so a second run changes
-- nothing.
--
-- USAGE
--   psql -d <database> -f scripts/backfill_worker_profile_id.sql
--
-- Wrapped in a transaction: the report and the update see the same rows, and a
-- failure leaves nothing half-applied.
-- ============================================================================

BEGIN;

\echo ''
\echo '── Before ─────────────────────────────────────────────────────────────'

WITH m AS (
  SELECT a.worker_profile_id AS current_link,
         (SELECT count(*) FROM worker_profiles w
           WHERE w.org_id = a.org_id
             AND lower(w.email::text) = lower(a.email::text)) AS candidates
  FROM job_applications a
)
SELECT count(*)                                                        AS applications,
       count(*) FILTER (WHERE current_link IS NOT NULL)                AS already_linked,
       count(*) FILTER (WHERE current_link IS NULL AND candidates = 1) AS will_link,
       count(*) FILTER (WHERE current_link IS NULL AND candidates = 0) AS no_worker,
       count(*) FILTER (WHERE current_link IS NULL AND candidates > 1) AS ambiguous
FROM m;

-- The guard. `worker_profiles_org_email_key` should make this impossible; if it
-- ever returns a row the backfill must not run, because picking one of two
-- people is exactly the kind of quiet wrong answer this script exists to avoid.
\echo ''
\echo '── Ambiguous matches (must be empty) ──────────────────────────────────'

SELECT a.id AS application_id, a.email, count(w.id) AS matching_workers
  FROM job_applications a
  JOIN worker_profiles w
    ON w.org_id = a.org_id
   AND lower(w.email::text) = lower(a.email::text)
 WHERE a.worker_profile_id IS NULL
 GROUP BY a.id, a.email
HAVING count(w.id) > 1;

\echo ''
\echo '── Linking ────────────────────────────────────────────────────────────'

UPDATE job_applications a
   SET worker_profile_id = w.id
  FROM worker_profiles w
 WHERE a.worker_profile_id IS NULL
   AND w.org_id = a.org_id
   AND lower(w.email::text) = lower(a.email::text)
   -- Belt and braces: only where exactly one profile matches.
   AND (SELECT count(*) FROM worker_profiles w2
         WHERE w2.org_id = a.org_id
           AND lower(w2.email::text) = lower(a.email::text)) = 1;

\echo ''
\echo '── After ──────────────────────────────────────────────────────────────'

WITH m AS (
  SELECT a.worker_profile_id AS current_link,
         (SELECT count(*) FROM worker_profiles w
           WHERE w.org_id = a.org_id
             AND lower(w.email::text) = lower(a.email::text)) AS candidates
  FROM job_applications a
)
SELECT count(*)                                                        AS applications,
       count(*) FILTER (WHERE current_link IS NOT NULL)                AS linked,
       count(*) FILTER (WHERE current_link IS NULL AND candidates = 1) AS still_matchable,
       count(*) FILTER (WHERE current_link IS NULL AND candidates = 0) AS no_worker
FROM m;

COMMIT;