v0.1.0 — first release as a 2scraper-family scraper

Replaces the April 2026 prototype entirely. The two share no data
contract: prices are numbers, spinny_id is sku, kilometers is
km_driven, there are exit codes and a run sidecar, and --mode api is
gone. If anything consumed the old output it needs rewriting rather than
adjusting — see the CHANGELOG's "Replaced, not extended".

Scrapes Spinny used-car city listings, filtered listings and car detail pages
through Playwright, Selenium or pyppeteer, with JSON/CSV output, a
run-metadata sidecar and this family's exit-code contract. 627 offline
checks, two CI workflows, and every number in the README measured and
dated.

The one thing to understand: --pages scrolls

A Spinny listing has exactly one page. ?page=N is silently ignored —
pages 1, 2 and 3 of one URL return byte-identical first cards under HTTP
200 — and there is no rel=next, no numbered anchor and no "load more"
anywhere. So --pages N means N batches of 20, reached by scrolling, and
--concurrency above 1 is refused with that reason in every engine.

A planner that built ?page=N would have fetched page one N times, found no
new car id, concluded the listing was exhausted, and reported a complete
run holding a twentieth of the catalogue.

You need nothing paid

Measured 2026-09-11: an ordinary local Chromium on a home connection, no API
key and no proxy, returned 62 cars at 100% price coverage. No refusal of
any kind was observed on this site — from that connection, an Amsterdam
datacentre exit, a Chennai residential exit, a bare GitHub runner or the
Scraping Browser API. The canary therefore runs daily with no credential,
unlike every sibling repo's; its first dispatch returned 82 cars.

Live-verified

• All three engines return the same 62 car ids with zero differing
  columns.
• The luxury listing returns 40 of the 40 cars the site advertises.
• --mode detail joins its listing row on 15 columns, and adds the exact
  odometer, the owner count, the colour, the seating capacity and the
  insurance details.
• Every paid path was run: Scraping Browser API (62 cars), residential proxy
  on an Indian exit (62 cars), Fingerprint API (applied live), Scraper API
  (200, 1.87 MB, 0 cars — by construction, the grid is not in the first
  response).

Findings worth knowing

• A listing page carries no structured data about its cars — five JSON-LD
  blocks and not one names a vehicle. A detail page does, and spells the
  price offers.Price, with a capital P.
• Four prices per car, on two bases. Mixing them computes a plausible
  negative discount, so every run reports the all-in-versus-displayed share.
• reCAPTCHA Enterprise v3 loads on every page and challenges nothing, so
  recaptcha and g-recaptcha were removed from the inherited marker set: a
  marker that matches every page of the site it guards is worse than no
  marker.
• A listing URL with no city renders nothing while advertising the
  national total — /used-suv-cars/s/ says "2549 Used SUV cars in India" and
  serves none of them.

Defects the live runs found, each now pinned by a check

fingerprint_client could not read .env; a --proxy the run was about to
ignore could still end the run; the readiness wait reported a timeout on a
page that had painted (off by one, all three engines); pages_completed
disagreed between engines on identical runs; and a stalled lazy load reported
a complete run holding 22 of an advertised 1546 — now scroll_stalled,
status partial, exit 6.

Full detail in CHANGELOG.md.