# Changelog of OLX.pl Car Scraper (`rastriq/olx-pl-scraper`) Actor

- **URL**: https://apify.com/rastriq/olx-pl-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/rastriq/olx-pl-scraper.md

## Changelog

### 0.1 (build 0.1.2, 2026-06-22)

- Imported verbatim from the deployed `latest` build on 2026-11-14.

### 0.2 (beta, 2026-11-14)

- Root cause of 0 items: CloudFront WAF returns `403 Request blocked` by egress IP reputation
  (whole site, including `robots.txt`). The endpoint `/api/v1/offers/` is unchanged and
  explicitly allowed by robots.txt.
- A 403/429 now retries on a fresh proxy session (3 attempts). A run that extracts nothing
  now FAILS with a clear status message instead of SUCCEEDED with 0 items.
- `SUMMARY` record in the default key-value store (items, requests, blocked responses).
- Parser fixes: `make` was always null (OLX has no `make` param; it is the leaf category,
  resolved once per brand through the same endpoint); `model` uses the label (`Passat`, not
  `passat`) and is null for "other" models; `hp` now filled from `enginepower`.
- **Blocked, not promotable.** Diagnostic run `M65WgkcJNCWrXZW4C` (build 0.2.1, `_diagnose`
  input, RESIDENTIAL PL): 3 proxy sessions x {browser UA, honest UA} x {`/robots.txt`,
  `/api/v1/offers/`} = 12/12 `403 Error from cloudfront` (PoPs WAW51, ATL58). Because
  `robots.txt` is blocked too, it is not the URL or the headers: CloudFront/AWS WAF rejects the
  client itself. Apify datacenter and GCP egress are blocked as well. The same API returns
  normal JSON from a non-blocked egress. Getting through would need TLS impersonation or a
  real browser, which is out of scope for this fix.

### 0.3 (beta, 2026-11-14)

- The 403 is a client-fingerprint block, not IP reputation: with a Chrome TLS/HTTP2 fingerprint
  (`curl_cffi`, `impersonate="chrome"`) both `robots.txt` and `/api/v1/offers/` return 200 from
  the same egress that got 403 with httpx. Use of the browser fingerprint was approved by the
  owner on 2026-11-14.
- HTTP client switched from httpx to `curl_cffi.requests.AsyncSession`; the User-Agent now comes
  from the impersonation profile. Session rotation on 403/429 is unchanged.
- `httpx` dropped from runtime dependencies; `curl_cffi==0.17.0` pinned.

### 0.4 (beta, 2026-11-15)

- Store page only; extraction code and HTTP client (curl_cffi, Chrome fingerprint) unchanged.
- README rewritten for the Store (~1,400 words): data table, completeness figures from run
  `aU5x2eMdSMx0kqCp4`, input/output examples, pricing, limits and FAQ.
- `dataset_schema.json`: was a bare `fields` map with no views; now a draft-07 schema with typed,
  nullable fields, descriptions, examples, `required`, and two views (`overview`, `all_fields`).
  `actor.json` wires it via `storages.dataset`.
- `output_schema.json` was a dataset view under `actorSpecification`; it is now a real output schema
  (`actorOutputSchemaVersion: 1`) pointing at the dataset's overview.
- Input schema: `max_pages` prefill 1 (~40-50 cars), proxy prefill/default datacenter
  (`{"useApifyProxy": true}`, no residential), sections and clearer descriptions. Field names unchanged.
- `defaultRunOptions.memoryMbytes` 512 -> 256 (max measured: 59-61 MB; beta smoke `uqaPipefCZocvlKCV` 59 MB).
- Tests: `tests/test_schemas.py` validates a real run's items against the dataset schema.
