# Changelog of Amazon Seller Scraper (`codingfrontend/amazon-seller-scraper`) Actor

- **URL**: https://apify.com/codingfrontend/amazon-seller-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/codingfrontend/amazon-seller-scraper.md

### Captured-source parser correction — 2026-10-02

- Accept the observed India-to-amazon.com canonical only with the same seller ID and explicit India seller-profile state; source-request host checks are unchanged.
- Read complete anchor tags, retain same-seller `/storefront` links, and normalize only exact bare-host marketplace aliases for public profile links.
- Omit a zero feedback-total placeholder only when it conflicts with complete visible feedback; preserve consistent explicit zeros.
- Regression-test identity mismatches, external/credentialed canonical links, marketplace provenance, source link shapes, rating/count distinctions, and dataset schema.

### 1.2.0 — 2026-10-02

- Add opt-in exact-byte response evidence with SHA256, source/status metadata, parser settings, and OUTPUT evidence keys, including complete denied and redirect bodies.
- Add offline source replay and local evidence validation. Evidence persistence failure stops further source requests and cannot report full success.
- Keep a single sticky proxy session per run and stop at access barriers; no proxy/identity rotation.
- Preserve public seller/profile URL modes, optional feedback and storefront products. Reject non-2xx terminal responses rather than parsing them as successful pages.
- Make npm run check execute syntax checks plus all runnable tests. Version 1.2 is prepared for bounded cloud validation; no live-source success is asserted by fixture tests.

## Changelog

### Local reconstruction, 2026-10-01

- Fixed source/redirect identity validation and removed arbitrary page-title seller names.
- Separated positive-feedback percentages from star ratings, removed 12-month lifetime-count fallback, and retained explicit zero without null/boolean numeric coercion.
- Validated supported proxy settings and credential-/port-free source URLs, normalized optional arrays, and reported accurate transport flags.
- Scoped proxy flags to Actor configuration and omitted optional direct-egress claims when managed runtime routing is unknown, without changing the gateway.
- Redacted bounded diagnostics, added strict `PARTIAL`/`errorUrl` schema support, and counted rows only after successful storage.
- Added adversarial parser/normalization/schema tests, recursive test discovery, local dataset/OUTPUT validation, and isolated real-SDK entrypoint fixture checks. These are current local checks, not live target validation or cloud deployment evidence.

### 1.1.0

- Replaced the legacy browser-oriented implementation with bounded public HTML requests.
- Added exact-host URL and redirect validation, response-size and redirect limits, challenge detection, and explicit proxy setup.
- Separated run diagnostics into `OUTPUT` so failed requests cannot become paid dataset rows.
- Added deterministic seller identities, null-free optional fields, strict schemas, dataset/output validation, and source-backed parser tests.
- Reduced maximum run, product, and feedback bounds and removed the ignored `headless` input.
- Switched to a deterministic Node 22 `npm ci` image and refreshed production dependencies.
- Accepted headerless public HTML only when bounded content sniffing finds standard document markers.
- Added bounded content-type and response-byte evidence for empty or non-HTML target responses.