Removed the external Yelp Actor that served search mode first and rescued blocked direct URLs. Both modes now run the native Yelp browser first, then the public search index of Yelp business pages.
A run where the browser is blocked and the index returns nothing ends FAILED (UPSTREAM_FAILED) with no result charges.
RUN_SUMMARY.providerRequests counts public-index lookups; fallbackRunId was removed.
1.9 — 2026-08-12
Bounded automated quality-test prefill
Reduce only the active prefill to one business with review enrichment disabled so the five-minute quality test persists one truthful default-dataset row without unnecessary pagination.
Preserve customer defaults of 50 businesses, review enrichment enabled, and five reviews per business.
1.8 — 2026-08-10
Provider fallback integrity repair
Skip a provider row without a business name or independently useful business fields instead of billing an echoed direct URL as an unknown business.
Normalize fallback ratings outside Yelp's 1–5 range to null and empty strings to null before persistence.
1.7 — 2026-08-10
Publication-readiness surface
Added a bounded, non-public task SEO map of distinct business-listing workflows; launch tasks remain empty pending exact canary evidence and explicit task-publication approval.
Corrected README provenance guidance so data_source accurately distinguishes native extraction from the provider fallback.
1.6 — 2026-08-10
Data-integrity and billing repair
Reject Yelp error documents before they can be emitted as a business solely because an error heading matched the business-name selector.
Couple each persisted business row to the yelp-business PPE event through the SDK; a rejected complete record is never replaced with a smaller billable placeholder.
1.5 — 2026-06-11
Input schema UI cleanup
Removed duplicate sectionCaption keys from location, maxReviewsPerBusiness, and serpApiKey so the Apify form no longer renders repeated section headers (Apify only honors sectionCaption on the FIRST property in each section).
1.4 — 2026-05-30
SerpApi fallback for reliability under Yelp blocking
When the live scrape returns nothing in search mode (Yelp edge/IP block), the actor now
automatically falls back to SerpApi so it still returns data instead of failing
Fallback uses SerpApi's Yelp Search (business enumeration) + Yelp Reviews engines. It
populates business_name, yelp_url, yelp_business_id, categories, description, and reviews.
SerpApi's current Yelp parser does NOT expose aggregate rating/review_count/price/phone/
address, so those are null on fallback records
Every record now carries a data_source field ('scrape' = full live data, 'serpapi' = leaner fallback)
New inputs: useSerpApiFallback (default true) and serpApiKey (optional BYOK, secret).
The actor's own SerpApi key is stored as an encrypted Apify secret (env var), never in code
Honest-fail still applies: a run fails only if BOTH the scrape and the SerpApi fallback
return nothing. Direct-URL mode can't use SerpApi (no slug→place_id resolution)
1.3 — 2026-05-30
Reverted Camoufox — back to Chrome (kept all correctness fixes)
Testing proved Yelp's 403 is edge/IP-reputation level (blocked even direct /biz/ URLs
under Camoufox), not a browser-fingerprint problem — so Camoufox added a 713MB image
and slower Firefox for no anti-bot benefit. Reverted the engine to Chrome.
All 1.1 correctness/billing/schema fixes remain in place (soft-fail, honest-fail,
concurrency-safe billing, safePushData, cost-cap, JSON-LD numeric coercion).
Kept the Chrome path on Crawlee's fingerprint generator + retryOnBlocked + single-block
session retirement (strictly better than the original 1.0.x hand-rolled masking).
Reliability now depends on Yelp's IP block easing; durable fix tracked in CLAUDE.md
(Web Unlocker proxy or Yelp Fusion API).
1.2 — 2026-05-30
Anti-bot: switched to Camoufox (Firefox stealth browser)
Replaced Chrome + Crawlee fingerprints with Camoufox — a hardened Firefox build with
native anti-fingerprinting — to beat Yelp's aggressive 403 blocking
Per-launch Camoufox config binds spoofed geolocation/timezone/locale to the rotating
Apify residential proxy exit IP (coherent proxy↔fingerprint↔geo), WebRTC blocked,
humanized cursor
Dockerfile now uses the Firefox base image and fetches the Camoufox binary at build