Cloud build SUCCEEDED (buildId: KM8KY7ND5M8ioyuAN, actorId: wIv3c5VLteB2wvfOr).
Cloud QA PASS: runId=cEwa4b15p9SkIHtoG, 3/3 rows returned with make,
model, year, price (EUR), mileage, fuel, transmission, engine power,
color, registration, seller, region, photos, description. RESIDENTIAL
proxy resolved and consumed (0.0017 GB). PPE charges fired
(actor-start=1, result-row=3). Dataset preview URLs (JSON + CSV) both
200. Maintenance badge cleared.
Root-cause fix for the 30-day 13% customer success rate.
polovniautomobili.com rewrote its frontend to Next.js; the old
server-rendered <article data-classifiedid> markup, per-ad @type=Car
LD-JSON, dataLayer.push({...}) and the uk-grid spec table are all
gone. search_cars()/search_prices() returned zero rows on every
run, which the scraper correctly treated as fail-loud
(RuntimeError: No listings parsed) — hence the near-total run
failure rate.
Rewrote src/parser.py and src/scraper.py to read the site's
<script id="__NEXT_DATA__"> JSON blob instead: searchResults.results
on the list page (make/model/year/price/mileage/fuel/gearbox/power in
one shot) and productData + contactDetails on the detail page
(full spec sheet, seller, description, photo gallery).
Photo CDN also moved gcdn.polovniautomobili.com ->
cdn.polovniautomobili.com; the new parser builds full-size URLs
from productData.images[].fileName instead of a hostname-pinned
regex, so it no longer breaks on CDN subdomain drift.
Dropped the now-unused selectolax dependency (no more HTML-tree
parsing — everything comes from structured JSON).
Adds tests/fixtures/{search,detail}.html (captured live 2026-07-10)
and tests/test_parser.py — locks in the new __NEXT_DATA__ parsing
against real markup so a future regression fails loud in CI, not in
production.
Adds tests/fixtures/input.qa.json — mandatory cloud-QA fixture,
missing from the previous release.
Bumps to 0.2.0 (pyproject.toml + actor.json) to mark the breaking
parser rewrite.
