Scout Multi-Use-Case Build Plan

Date: 2026-05-20

Required Use Cases

Scout must cover all of these, not choose one:

1. Company intelligence: about page, overview, leadership, socials, key URLs.
2. PRISM: company + exec + investor + careers + hiring + news bundle for Algolia prospect research.
3. Investor: IR pages, decks, reports, filings, events.
4. Careers: career pages, ATS detection, departments, hiring signals.
5. Jobs: discover companies, find jobs, score against a private profile, no auto-apply.
6. Products: crawl product/category pages and produce Algolia-ready records.
7. News: newsroom/blog/recent announcement signals.
8. Research: generic page/site/document extraction and opportunity discovery.

Test Scenario Families

Company / PRISM

Targets: Adobe, Salesforce, Nike, Atlassian, Intuit.

Assertions:

• official website resolved,
• about/leadership/social links captured,
• source URLs preserved,
• PRISM bundle emits company, career, investor, and news records.

Investor

Targets: Salesforce IR, Adobe IR, Nike IR, Intuit IR, Microsoft IR.

Assertions:

• investor relation page detected,
• assets classified as page/report/deck/filing/event,
• dates and URLs preserved where available.

Careers / Jobs

Targets: Adobe, Atlassian, Intuit, Salesforce, Greenhouse, Lever, Ashby, Workday.

Assertions:

• ATS platform detected,
• role pages extracted,
• candidate profile scoring works,
• salary/seniority filters are aggressive,
• private profile data does not enter public fixtures.

Products

Targets: Lacoste, Estée Lauder, Sephora, Nike, Patagonia.

Assertions:

• category/product links separated,
• stable objectID,
• brand/name/url/image/price/category/provenance emitted,
• hard sites produce blocked-page evidence and browser fallback results when available.

Research / News / Docs

Targets: company newsrooms, docs sites, blogs, PDFs, simple marketing sites.

Assertions:

• content converted to records,
• provenance preserved,
• source pages and validation artifacts are written.

Build Discipline

• Add tests before implementation for each slice.
• Keep Pydantic contracts at boundaries.
• Update repo docs and vault after each slice.
• Run unit, integration, pyright, ruff, CLI smoke, HTTP smoke, and browser verification before claiming completion.