-- Migration 122: FS.com delivery dates + lead_time now collected; purge units_sold garbage
-- ─────────────────────────────────────────────────────────────────────────────
-- Follow-up to migration 121, which documented lead_time_days as NOT COLLECTED and
-- flagged that populating it was a scraper-side task. That task is now done:
--
--   * fs-com.ts (via fs-com-parse.ts) parses the bare word-date FS.com renders on
--     each warehouse line ("… im DE-Lager, 8. Juli 2026 Lieferbar") into
--     warehouse_de_delivery_date / warehouse_global_delivery_date /
--     backorder_estimated_date, and derives lead_time_days = min(date − today).
--   * upsertStockObservation now writes lead_time_days.
--
-- Historical rows stay NULL until the next FS.com run re-scrapes each SKU; NULL
-- therefore still means "unknown / not yet captured", not "zero lead time".
--
-- Second fix: units_sold was ~10× inflated for decimal-abbreviated values
-- ("4.8K"→48000, "1.4M"→14000000) because the K/M parser stripped the decimal
-- point. The parser is fixed AND ingest now drops values above a sanity cap.
-- This migration removes the clearly-impossible historical values (the multi-
-- million "outliers") so no consumer ingests them before the re-scrape. Sub-cap
-- values that may still be mildly inflated self-correct on the next FS.com run
-- (a changed value writes a fresh observation; consumers read the latest).
-- ─────────────────────────────────────────────────────────────────────────────

-- 1. Purge impossible units_sold (matches the ingest gate default, FS_MAX_UNITS_SOLD=2,000,000).
UPDATE stock_observations
SET units_sold = NULL
WHERE units_sold IS NOT NULL
  AND units_sold > 2000000;

-- 2. Re-document the columns as collected-by-scraper (reversing migration 121's note).
COMMENT ON COLUMN stock_observations.warehouse_de_delivery_date IS
  'FS.com DE-Lager availability date (collected 2026-07-04+ by fs-com-parse). '
  'NULL = warehouse in stock now, vendor did not quote a date, or SKU not yet re-scraped.';

COMMENT ON COLUMN stock_observations.warehouse_global_delivery_date IS
  'FS.com Global-Lager availability date (collected 2026-07-04+ by fs-com-parse). '
  'NULL semantics as warehouse_de_delivery_date.';

COMMENT ON COLUMN stock_observations.backorder_estimated_date IS
  'FS.com Nachlieferung (backorder) estimated date (collected 2026-07-04+ by fs-com-parse). '
  'NULL = no backorder date quoted or SKU not yet re-scraped.';

COMMENT ON COLUMN stock_observations.lead_time_days IS
  'Soonest availability in days = min(delivery_date − today) across warehouses that quote a '
  'date. Collected 2026-07-04+ by the FS.com scraper; other vendors do not expose a lead time. '
  'Treat NULL as "unknown / not captured" (do NOT default to 0).';

COMMENT ON COLUMN stock_observations.units_sold IS
  'FS.com "verkauft" lifetime sell-through counter. Parser de-abbreviates K/M correctly and '
  'ingest drops values > 2,000,000 as implausible. FS.com only; NULL elsewhere.';

INSERT INTO _migrations (filename)
VALUES ('122-fs-com-delivery-dates-and-units-sold-gate.sql')
ON CONFLICT (filename) DO NOTHING;