feat(data-quality): product_type classification + form_factor recovery + lead_time docs

sql/119: product_type enum (module|dac|aoc|aec|switch|nic|accessory) computed by
trigger trg_transceivers_biu from part_number/category/reach on every insert/update.
Backfilled 41008 rows (36284 module, 2791 dac, 1835 aoc, 29 aec, 27 accessory,
26 switch, 16 nic). Cables and switches/NICs no longer counted as modules;
BASE-T/RJ45 copper modules protected from cable demotion; AOC/DAC keyword beats
the optical-token guard. Consumers filter product_type=module.

sql/120: conservative form_factor recovery. Corrects only provably-wrong rows
(form_factors.max_speed < speed_gbps) via an explicit title token; 87 fixed
(e.g. 400G cables mislabelled SFP+ to QSFP-DD), 947 opaque rows left untouched
(never guessed). Scraped original preserved in new column form_factor_raw.

sql/121: document lead_time_days as not-collected (0 percent populated on all
time-series tables) so no consumer fabricates a default.

scraper/utils/db.ts: note that product_type/form_factor are trigger-maintained.
This commit is contained in:
root 2026-07-04 16:04:45 +00:00
parent 7c9bd10941
commit dd2c917f05
5 changed files with 313 additions and 0 deletions

View File

@ -320,3 +320,7 @@ Types: FEAT · FIX · UI · DATA · AI · INFRA
{"d":"2026-06-11","t":"DATA","m":"Flexoptix-Datenblatt-Audit: alle 805 katalog-bestätigten Teile programmatisch gegen gespeicherte fx_specifications (Datenblatt) geprüft. Nach Terabit-Parser-Fix + Re-Derive aus Specs: Form Factor 802/802 (100%), Speed 801/801 (100%), 0 Abweichungen. Jeder Wert kommt aus dem Datenblatt, nicht aus dem Namen."} {"d":"2026-06-11","t":"DATA","m":"Flexoptix-Datenblatt-Audit: alle 805 katalog-bestätigten Teile programmatisch gegen gespeicherte fx_specifications (Datenblatt) geprüft. Nach Terabit-Parser-Fix + Re-Derive aus Specs: Form Factor 802/802 (100%), Speed 801/801 (100%), 0 Abweichungen. Jeder Wert kommt aus dem Datenblatt, nicht aus dem Namen."}
{"d":"2026-06-11","t":"DATA","m":"Switch-Daten-Audit: 681 Switches, nur 267 (39%) mit Port-Config, 414 ohne (überwiegend Cisco-Bulk-Import). Vorhandene Port-Configs 99.3% plausibel, 248/267 mit Hersteller-URL, nur 35 mit datasheet_url. Demo-Switches (Aruba CX10000, Arista, Juniper) vollständig+korrekt. LÜCKE: autoritatives Port-Sourcing per Hersteller-Datenblatt für die 414 fehlt = separates Scraper-Projekt pro Vendor."} {"d":"2026-06-11","t":"DATA","m":"Switch-Daten-Audit: 681 Switches, nur 267 (39%) mit Port-Config, 414 ohne (überwiegend Cisco-Bulk-Import). Vorhandene Port-Configs 99.3% plausibel, 248/267 mit Hersteller-URL, nur 35 mit datasheet_url. Demo-Switches (Aruba CX10000, Arista, Juniper) vollständig+korrekt. LÜCKE: autoritatives Port-Sourcing per Hersteller-Datenblatt für die 414 fehlt = separates Scraper-Projekt pro Vendor."}
{"d":"2026-06-11","t":"FIX","m":"Switch-Kompatibilität = 'läuft wirklich': Matcher verlangt jetzt zusätzlich, dass Flexoptix den Transceiver für den Switch-VENDOR codieren kann (fx_compatibilities). Physisch-passend reicht nicht. CX 10000: 629 physisch -> 283 echt-Aruba-codierbar (283/283 verifiziert). Cisco 8818: 55 (alle Cisco-codierbar). Plus Chassis-Port-Format (max_per_slot_NNNG_FF) geparst. Jede Empfehlung jetzt: katalog-bestätigt + datenblatt-genau + passt physisch + codierbar = garantiert lauffähig."} {"d":"2026-06-11","t":"FIX","m":"Switch-Kompatibilität = 'läuft wirklich': Matcher verlangt jetzt zusätzlich, dass Flexoptix den Transceiver für den Switch-VENDOR codieren kann (fx_compatibilities). Physisch-passend reicht nicht. CX 10000: 629 physisch -> 283 echt-Aruba-codierbar (283/283 verifiziert). Cisco 8818: 55 (alle Cisco-codierbar). Plus Chassis-Port-Format (max_per_slot_NNNG_FF) geparst. Jede Empfehlung jetzt: katalog-bestätigt + datenblatt-genau + passt physisch + codierbar = garantiert lauffähig."}
{"d":"2026-07-04","t":"DATA","m":"product_type-Klassifikation (sql/119): neues Enum module|dac|aoc|aec|switch|nic|accessory, deterministisch per Trigger trg_transceivers_biu aus part_number/category/reach gesetzt (self-healing bei jedem Insert/Update, Preis-Updates unberührt). Backfill 41008 Zeilen: 36284 module, 2791 dac, 1835 aoc, 29 aec, 27 accessory, 26 switch, 16 nic. Kabel (DAC/AOC/AEC) und Switches/NICs (ConnectX/Quantum/N-Port) nicht mehr als Modul gezählt. BASE-T/RJ45-Kupfermodule vor Kabel-Demotion geschützt (Guard vor copper-Regel); AOC/DAC-Keyword schlägt Optik-Token-Guard (AOC ist optisch, trägt legitim nm/MMF/SR). Konsumenten filtern product_type='module'. Missed-cable=0 verifiziert. Per-Vendor 100G+: NADDOD 179 Module/423 Kabel/28 Switch-NIC, ATGBICS 1331/297, QSFPTEK 22/98, ProLabs 1690/818."}
{"d":"2026-07-04","t":"DATA","m":"form_factor-Recovery (sql/120): korrigiert NUR beweisbar-falsche Zeilen (form_factors.max_speed < speed_gbps) mit explizitem Formfaktor-Token im Titel. 87 gefixt (u.a. 400G-Kabel als SFP+ getaggt -> QSFP-DD; QSFP-DD@800G -> QSFP-DD800; Q56DD -> QSFP-DD), 947 opake Zeilen (z.B. EOLS-1303-40-D, MGB-2GSR) unangetastet gelassen, nichts geraten. Gescraptes Original in neuer Spalte form_factor_raw bewahrt. Recovery auch im Ingest-Trigger (Neuzugänge self-healing)."}
{"d":"2026-07-04","t":"DATA","m":"lead_time_days als NICHT ERFASST dokumentiert (sql/121, COMMENT auf price_observations/stock_observations/stock_snapshots): 0% befüllt über 1.378.984 + 85.406 + 0 Zeilen. Plumbing existiert (crawler-llm stock-schema erfasst es, upsertPriceObservation nimmt Param), aber aktiver Scraper-Pfad liefert nie einen Wert und LLM-Crawler-Tabelle stock_snapshots ist leer. Kein Default fabrizieren; NULL = unbekannt. Befüllen ist Scraper-Aufgabe."}
{"d":"2026-07-04","t":"DATA","m":"Befund FS.COM-Untererfassung bestätigt (kein Fix, Scraper-Thema): 392 Zeilen/378 distinct SKUs vs ATGBICS 8420 (22x-Lücke). 391/392 sind bereits product_type='module' -> Klassifikation ist NICHT die Ursache, reine Crawl-Tiefe. Ursache in fs-com.ts + Seed-Scope (Kategorie-Abdeckung) + evtl. zu aggressive Quarantäne von fs.com/c/-URLs. Empfehlung: FS.com-Crawl-Seed erweitern (Scraper-Pipeline, nicht raten)."}

View File

@ -480,6 +480,12 @@ export async function upsertStockObservation(params: {
return true; return true;
} }
// NOTE: product_type (module|dac|aoc|aec|switch|nic|accessory) and form_factor
// corrections are set automatically by the DB trigger trg_transceivers_biu
// (see sql/119 + sql/120). Callers do NOT pass or maintain them — the trigger
// classifies from part_number/category/reach on every insert and update, so
// every scraper stays consistent without touching this call site. Consumers
// filter product_type='module' for a true module count.
export async function findOrCreateScrapedTransceiver(params: { export async function findOrCreateScrapedTransceiver(params: {
partNumber: string; partNumber: string;
vendorId: string; vendorId: string;

View File

@ -0,0 +1,144 @@
-- Migration 119: product_type classification (module vs cable vs switch/NIC/accessory)
-- ─────────────────────────────────────────────────────────────────────────────
-- Problem: TIP ingests DAC/AOC/AEC cables, switches and NICs as `transceivers`
-- rows (one per cable length). `form_factor` is unreliable (a 400G DAC can be
-- tagged 'SFP+'). Every "assortment breadth" metric is therefore cable-inflated,
-- and consumers (Lupo) had to re-derive a text heuristic on every query.
--
-- Fix: a deterministic `product_type` enum, computed at ingest by a BEFORE
-- INSERT/UPDATE trigger from the free-text part_number + category + reach.
-- Consumers filter `product_type = 'module'` for a true module count.
--
-- Taxonomy: module | dac | aoc | aec | switch | nic | accessory
-- Classification is keyword-first (an explicit 'AOC'/'DAC' token beats the
-- optical-token guard, because an Active Optical Cable legitimately carries
-- nm/MMF/SR tokens). BASE-T / RJ45 electrical transceivers are protected as
-- modules before the copper/DAC rule so genuine RJ45 optics are never demoted.
--
-- Idempotent. Applied manually via `docker exec -i tip-postgres psql` on Erik
-- (the _migrations tracker is only used by scripts/migrate.ts; recent migrations
-- are applied by hand — we still record this file there for correctness).
-- ─────────────────────────────────────────────────────────────────────────────
-- 1. Column + CHECK constraint --------------------------------------------------
ALTER TABLE transceivers ADD COLUMN IF NOT EXISTS product_type TEXT;
DO $$
BEGIN
IF NOT EXISTS (
SELECT 1 FROM pg_constraint WHERE conname = 'transceivers_product_type_check'
) THEN
ALTER TABLE transceivers ADD CONSTRAINT transceivers_product_type_check
CHECK (product_type IS NULL OR product_type IN
('module','dac','aoc','aec','switch','nic','accessory'));
END IF;
END;
$$;
COMMENT ON COLUMN transceivers.product_type IS
'Deterministic product class set by trigger tip_transceivers_biu(): '
'module=optical/electrical pluggable, dac/aoc/aec=cables, switch/nic=infrastructure, '
'accessory=loopback/attenuator/amplifier/mux/etc. Consumers filter product_type=''module''.';
-- 2. Deterministic classifier ---------------------------------------------------
-- IMMUTABLE: depends only on its inputs, so it is safe in triggers and indexes.
CREATE OR REPLACE FUNCTION tip_classify_product_type(
p_part_number TEXT,
p_category TEXT,
p_reach_label TEXT
) RETURNS TEXT
LANGUAGE sql IMMUTABLE PARALLEL SAFE AS $$
SELECT CASE
-- 1. NIC / adapter cards
WHEN COALESCE(p_part_number,'') ~* '(connectx|supernic|bluefield|smartnic|adapter card|\mmcx[0-9])'
THEN 'nic'
-- 2. Switches / network infrastructure (N-port matches "64-Port", not "Twin-Port")
WHEN COALESCE(p_part_number,'') ~* '(\mswitch\M|switches|\mquantum\M|\mspectrum\M|tomahawk|\mqm9[0-9]{3}|q3[0-9]{3}-ra|[0-9]+-port|\mdgx\M|\mgpu\M)'
THEN 'switch'
-- 3. Hard accessory keywords (never a module even if the title says "transceiver"/"copper")
WHEN COALESCE(p_part_number,'') ~* '(loopback|\mlb\M|attenuator|amplifier|\medfa\M|\moadm\M|demux|multiplexer|\mmux\M|transponder|transport[ -]system|microscope|cleaning|cassette|\mtester\M|media converter)'
THEN 'accessory'
-- 4. AEC — active electrical cable (Mellanox MCA copper-active family)
WHEN COALESCE(p_part_number,'') ~* '(\maec\M|active electrical|\mmca[0-9])'
THEN 'aec'
-- 5. AOC — active optical cable (Mellanox MFA/MFS active fibre families)
WHEN COALESCE(p_part_number,'') ~* '(aoc|active optical|\mmfa|\mmfs[0-9])'
THEN 'aoc'
-- 6. BASE-T / RJ45 electrical transceiver = MODULE (guard before the copper/DAC rule)
WHEN COALESCE(p_part_number,'') ~* '(base-?t|[0-9]+gbase-?t|rj-?45|8p8c)'
THEN 'module'
-- 7. DAC / passive copper / twinax (Mellanox MCP passive-copper family)
WHEN COALESCE(p_part_number,'') ~* '(dac|twinax|direct attach|copper|passive|splitter|breakout|\mmcp|\mcab-)'
THEN 'dac'
-- 8. Category-driven cable signals (scraper set a cable category but a bland title)
WHEN COALESCE(p_category,'') ~* 'AOC' THEN 'aoc'
WHEN COALESCE(p_category,'') ~* '(DAC|Copper|Cable)' THEN 'dac'
-- 9. Bare "<n>m (<n>ft) ..." length title — a cable, unless it carries an optical token
WHEN COALESCE(p_part_number,'') ~* '^[0-9.]+ *m( *\([0-9.]+ ?ft\))? '
AND COALESCE(p_part_number,'') !~* '(sr[0-9]|dr[0-9]|fr[0-9]|lr[0-9]|er[0-9]|zr|base|[0-9]+nm|mmf|smf|km|cwdm|dwdm|bidi|lambda)'
THEN 'dac'
-- 10. Short reach <=10 m with no optical token — a copper DAC
WHEN COALESCE(p_reach_label,'') ~* '^[0-9.]+ *m$'
AND COALESCE(NULLIF(regexp_replace(p_reach_label,'[^0-9.].*$',''),'')::float, 999) <= 10
AND COALESCE(p_part_number,'') !~* '(sr[0-9]|dr[0-9]|fr[0-9]|lr[0-9]|er[0-9]|zr|base|[0-9]+nm|mmf|smf|km|cwdm|dwdm|bidi|lambda)'
THEN 'dac'
-- 11. Category-driven accessory residue
WHEN COALESCE(p_category,'') IN ('Accessory','Adapter / Converter','Switch / Media Converter',
'Switch / Network Infrastructure','NIC / Adapter','Mux / Passive Optical',
'Product Family','Loopback / Test Module')
THEN 'accessory'
-- 12. Default: a genuine optical/electrical module
ELSE 'module'
END;
$$;
-- 3. Trigger — classify every row on write ------------------------------------
-- Recomputes only when a classification input changed (or product_type is NULL),
-- so ordinary price/verification updates are untouched. Deterministic + self-healing.
CREATE OR REPLACE FUNCTION tip_transceivers_biu() RETURNS TRIGGER
LANGUAGE plpgsql AS $$
BEGIN
IF TG_OP = 'INSERT'
OR NEW.product_type IS NULL
OR NEW.part_number IS DISTINCT FROM OLD.part_number
OR NEW.category IS DISTINCT FROM OLD.category
OR NEW.reach_label IS DISTINCT FROM OLD.reach_label
THEN
NEW.product_type := tip_classify_product_type(NEW.part_number, NEW.category, NEW.reach_label);
END IF;
RETURN NEW;
END;
$$;
DROP TRIGGER IF EXISTS trg_transceivers_biu ON transceivers;
CREATE TRIGGER trg_transceivers_biu
BEFORE INSERT OR UPDATE ON transceivers
FOR EACH ROW EXECUTE FUNCTION tip_transceivers_biu();
-- 4. Backfill every existing row ------------------------------------------------
UPDATE transceivers
SET product_type = tip_classify_product_type(part_number, category, reach_label)
WHERE product_type IS DISTINCT FROM
tip_classify_product_type(part_number, category, reach_label);
-- 5. Indexes for the common consumer filter ------------------------------------
CREATE INDEX IF NOT EXISTS idx_transceivers_product_type
ON transceivers (product_type);
CREATE INDEX IF NOT EXISTS idx_transceivers_ptype_speed
ON transceivers (product_type, speed_gbps);
-- 6. Record in the migration tracker (idempotent) ------------------------------
INSERT INTO _migrations (filename)
VALUES ('119-product-type-classification.sql')
ON CONFLICT (filename) DO NOTHING;
-- 7. Report --------------------------------------------------------------------
DO $$
DECLARE rec RECORD;
BEGIN
RAISE NOTICE '=== product_type distribution ===';
FOR rec IN SELECT product_type, count(*) c FROM transceivers GROUP BY 1 ORDER BY 2 DESC LOOP
RAISE NOTICE ' %: %', rpad(COALESCE(rec.product_type,'(null)'), 10), rec.c;
END LOOP;
END;
$$;

View File

@ -0,0 +1,124 @@
-- Migration 120: conservative form_factor recovery (provably-wrong rows only)
-- ─────────────────────────────────────────────────────────────────────────────
-- Problem: some scrapers default form_factor to 'SFP'/'SFP+', so a 400G QSFP-DD
-- DAC ends up tagged 'SFP+'. We only ever correct a form_factor that is
-- PROVABLY wrong — i.e. its max electrical speed (form_factors.max_speed_gbps)
-- is below the row's actual speed_gbps — and only when the part_number carries
-- an explicit form-factor token that CAN carry that speed. Opaque part numbers
-- (e.g. 'EOLS-1303-40-D') are left untouched; we never guess a form factor.
--
-- The scraped original is preserved in form_factor_raw for audit/rollback.
-- Ingest-time self-healing is added to the tip_transceivers_biu() trigger.
-- Idempotent.
-- ─────────────────────────────────────────────────────────────────────────────
-- 1. Preserve the scraped original ---------------------------------------------
ALTER TABLE transceivers ADD COLUMN IF NOT EXISTS form_factor_raw TEXT;
COMMENT ON COLUMN transceivers.form_factor_raw IS
'Form factor exactly as scraped, before tip_recover_form_factor() correction. Audit/rollback only.';
UPDATE transceivers SET form_factor_raw = form_factor
WHERE form_factor_raw IS NULL;
-- 2. Explicit form-factor token extraction (most-specific first) ----------------
CREATE OR REPLACE FUNCTION tip_extract_ff_token(p_part_number TEXT) RETURNS TEXT
LANGUAGE sql IMMUTABLE PARALLEL SAFE AS $$
SELECT CASE
WHEN p_part_number ~* 'qsfp-?dd-?800|qdd-?800' THEN 'QSFP-DD800'
-- Q56DD / QSFP56-DD is the 400G double-density QSFP connector (Dell naming) → QSFP-DD
WHEN p_part_number ~* 'q56dd|qsfp56-?dd' THEN 'QSFP-DD'
WHEN p_part_number ~* 'osfp-?xd' THEN 'OSFP-XD'
WHEN p_part_number ~* 'osfp224' THEN 'OSFP224'
WHEN p_part_number ~* 'osfp112' THEN 'OSFP112'
WHEN p_part_number ~* '\mosfp\M' THEN 'OSFP'
WHEN p_part_number ~* 'qsfp112' THEN 'QSFP112'
WHEN p_part_number ~* 'qsfp-?dd|\mqdd\M|qsfpdd' THEN 'QSFP-DD'
WHEN p_part_number ~* 'qsfp56|\mq56\M' THEN 'QSFP56'
WHEN p_part_number ~* 'qsfp28|\mq28\M' THEN 'QSFP28'
WHEN p_part_number ~* 'qsfp\+|qsfp-?plus' THEN 'QSFP+'
WHEN p_part_number ~* 'sfp-?dd|sfpdd' THEN 'SFP-DD'
WHEN p_part_number ~* 'sfp112' THEN 'SFP112'
WHEN p_part_number ~* 'sfp56' THEN 'SFP56'
WHEN p_part_number ~* 'sfp28' THEN 'SFP28'
WHEN p_part_number ~* 'sfp\+|sfp-?plus' THEN 'SFP+'
WHEN p_part_number ~* 'cfp2-?dco' THEN 'CFP2-DCO'
WHEN p_part_number ~* 'cfp2' THEN 'CFP2'
WHEN p_part_number ~* 'cfp8' THEN 'CFP8'
WHEN p_part_number ~* 'cfp4' THEN 'CFP4'
WHEN p_part_number ~* '\mcfp\M' THEN 'CFP'
WHEN p_part_number ~* '\mxfp\M' THEN 'XFP'
ELSE NULL
END;
$$;
-- 3. Recovery: return corrected FF only when the current one is PROVABLY wrong ---
CREATE OR REPLACE FUNCTION tip_recover_form_factor(
p_part_number TEXT, p_form_factor TEXT, p_speed_gbps NUMERIC
) RETURNS TEXT
LANGUAGE plpgsql STABLE PARALLEL SAFE AS $$
DECLARE
v_token TEXT := tip_extract_ff_token(p_part_number);
v_cur_max NUMERIC;
v_tok_max NUMERIC;
BEGIN
IF v_token IS NULL OR p_form_factor IS NULL OR p_speed_gbps IS NULL THEN
RETURN p_form_factor;
END IF;
IF upper(v_token) = upper(p_form_factor) THEN
RETURN p_form_factor; -- already correct
END IF;
SELECT max_speed_gbps INTO v_cur_max FROM form_factors WHERE upper(name) = upper(p_form_factor);
SELECT max_speed_gbps INTO v_tok_max FROM form_factors WHERE name = v_token;
-- current provably wrong for the speed, and the token can actually carry it
IF v_cur_max IS NOT NULL AND v_tok_max IS NOT NULL
AND p_speed_gbps > v_cur_max AND v_tok_max >= p_speed_gbps THEN
RETURN v_token;
END IF;
RETURN p_form_factor; -- not confident → leave as-is
END;
$$;
-- 4. Backfill the confidently-recoverable rows ---------------------------------
UPDATE transceivers
SET form_factor = tip_recover_form_factor(part_number, form_factor, speed_gbps)
WHERE form_factor IS DISTINCT FROM tip_recover_form_factor(part_number, form_factor, speed_gbps);
-- 5. Extend the ingest trigger with FF self-healing ----------------------------
CREATE OR REPLACE FUNCTION tip_transceivers_biu() RETURNS TRIGGER
LANGUAGE plpgsql AS $$
BEGIN
-- product_type: recompute only when a classification input changed
IF TG_OP = 'INSERT'
OR NEW.product_type IS NULL
OR NEW.part_number IS DISTINCT FROM OLD.part_number
OR NEW.category IS DISTINCT FROM OLD.category
OR NEW.reach_label IS DISTINCT FROM OLD.reach_label
THEN
NEW.product_type := tip_classify_product_type(NEW.part_number, NEW.category, NEW.reach_label);
END IF;
-- form_factor: preserve the scraped original, then correct only provably-wrong values
NEW.form_factor_raw := COALESCE(NEW.form_factor_raw, NEW.form_factor);
NEW.form_factor := tip_recover_form_factor(NEW.part_number, NEW.form_factor, NEW.speed_gbps);
RETURN NEW;
END;
$$;
-- (trigger trg_transceivers_biu already bound in migration 119 — replacing the
-- function body above is sufficient.)
INSERT INTO _migrations (filename)
VALUES ('120-form-factor-recovery.sql')
ON CONFLICT (filename) DO NOTHING;
-- 6. Report --------------------------------------------------------------------
DO $$
DECLARE v_fixed INT; v_left INT;
BEGIN
SELECT count(*) INTO v_fixed FROM transceivers WHERE form_factor IS DISTINCT FROM form_factor_raw;
SELECT count(*) INTO v_left FROM transceivers t JOIN form_factors f ON upper(f.name)=upper(t.form_factor)
WHERE t.speed_gbps > f.max_speed_gbps;
RAISE NOTICE 'form_factor corrected on % rows; % provably-wrong rows left untouched (opaque, not guessed).', v_fixed, v_left;
END;
$$;

View File

@ -0,0 +1,35 @@
-- Migration 121: document lead_time_days as NOT COLLECTED
-- ─────────────────────────────────────────────────────────────────────────────
-- Verified 2026-07-04: lead_time_days is 0 % populated on every time-series table
-- price_observations : 1,378,984 rows → 0 filled
-- stock_observations : 85,406 rows → 0 filled
-- stock_snapshots : 0 rows (LLM-crawler path is dormant)
--
-- The plumbing exists (crawler-llm stock-schema captures lead_time_days and
-- writes it to stock_snapshots; upsertPriceObservation accepts a leadTimeDays
-- param) but the ACTIVE ingestion path never supplies a value, and there is no
-- evidence any tracked vendor page exposes a machine-readable lead time.
--
-- Rather than fabricate a default, this migration documents the column as
-- not-collected so no consumer invents one. Lupo already renders "k.A." / "n/a"
-- and blends only the two signals it actually measures (availability, breadth).
-- Populating this field is a SCRAPER-side task (extend the active crawlers to
-- read a delivery/lead-time badge) — out of scope for a schema migration.
-- ─────────────────────────────────────────────────────────────────────────────
COMMENT ON COLUMN price_observations.lead_time_days IS
'NOT COLLECTED (verified 2026-07-04): 0 % populated. The active scraper path never '
'supplies a lead time. Do NOT default this — treat NULL as "unknown / not captured". '
'Populating it requires scraper work (read a delivery-time badge on the vendor page).';
COMMENT ON COLUMN stock_observations.lead_time_days IS
'NOT COLLECTED (verified 2026-07-04): 0 % populated. Treat NULL as "unknown". See '
'price_observations.lead_time_days.';
COMMENT ON COLUMN stock_snapshots.lead_time_days IS
'Captured by the crawler-llm path (stock-schema.lead_time_days) but that path is '
'currently dormant (0 rows). Treat NULL as "unknown / not captured".';
INSERT INTO _migrations (filename)
VALUES ('121-lead-time-not-collected-doc.sql')
ON CONFLICT (filename) DO NOTHING;