# Changelog of AI Sales Research Pack Generator (`zerrouki-samir/ai-sales-research-pack-generator`) Actor

- **URL**: https://apify.com/zerrouki-samir/ai-sales-research-pack-generator/changelog.md
- **Full Actor documentation**: https://apify.com/zerrouki-samir/ai-sales-research-pack-generator.md

## Changelog

### \[0.2.3] - 2026-09-05

BYO xAI (Grok) via the OpenAI-compatible API. Default model is `grok-build-0.1`.

### \[0.2.0] - 2026-09-05

Public-ready heuristic packs. No LLM key required for a usable row.

- Browser-render pricing / careers / contact when static HTML lacks prices or roles.
- Classify by the first path segment; ignore leftover nav CTAs.
- Extract `businessSignals.pricing` (plan amounts or `isCustomOrSalesLed`).
- Clean company names, drop mascot/security emails, HQ phrases, broader funding lines.
- Heuristic summary is body text plus those signals, not the meta description.

### \[0.1.0] - 2026-09-05

Initial design spec: input schema, output/dataset schema, key-value store schema, extraction
strategy, error-handling matrix, pricing proposal, and worked example input/output. No `src/`
implementation yet.

### \[0.1.1] - 2026-09-05

Deployed: private GitHub repo + private Apify Actor, connected via a push webhook for
automatic builds.

### \[0.1.8] - 2026-09-05

Cap pages per category (2 products, 1 blog, 1 other) so Pricing / Careers / Contact
still get fetched after a `/solutions` flood.

### \[0.1.7] - 2026-09-05

Stay on the marketing host. `www` and the bare domain are the same site; `developer.`,
`community.`, and off-site redirects (Loom → Atlassian support) are not. Duplicate
redirect targets are stored once.

### \[0.1.6] - 2026-09-05

Quality pass after the first platform run (`6G1kNGYOKL8E27qX2`, 10 companies, 7 ok /
3 false `blocked`).

- Do not treat the word "captcha" on a 200 page as a bot block (Intercom/Zapier/Webflow).
- Fetch pages in category-priority order so a sitemap of blog posts cannot starve
  About / Pricing / Careers (HubSpot, ClickUp).
- Classify by path segment, not substring (`/company-news` is no longer `about`).
- Canonicalize URLs so `/` and \`\` are one page; skip PDFs and tracking params.
- Ignore empty `mailto:`, image-filename "emails", share buttons, and personal
  LinkedIn `/in/` profiles. Read `sameAs` from JSON-LD.
- 4xx/5xx HTML is a failed fetch, not a research page. 403/429 get one browser retry.
- Playwright waits for `domcontentloaded`, not `networkidle`. Plain Chrome UA.
- Input cap of 100 companies, default concurrency 5, safer prefill, LICENSE, COST.md.