GCC Account Desk

Evidence-based qualification of Indian GCC (global capability centre) accounts for specialist recruitment firms.

It answers: which GCCs show relevant hiring demand, which observed roles match your services, what evidence supports that, what is still unknown, and which accounts deserve investigation. Public hiring activity is never treated as a recruitment-agency mandate.

Requires zero AI API keys and zero environment variables. No database, no auth. Data is static verified JSON; preferences and decisions live in localStorage.

Screens

Screen	Route
Account Explorer (search, filter, sort)	/
Account Intelligence Detail	/accounts/[id]
Recruitment Service Profile	/profile
Qualification Workspace (gated decisions, audit trail)	/workspace
Account Research Brief (print/PDF + JSON export)	/brief/[id]

Three kinds of hiring evidence, kept apart

Category	What it is	Never means
Verified vacancy	A posting opened and read on the company's own job board (title, location, seniority, skills, link)	—
Capability signal	A skill or function named in an announcement	An open role
Hiring plan	An announced intention or headcount target, with a timeline only if sourced	Current openings

Service fit, hiring evidence and commercial intent (external agency mandate) are separate dimensions. High service fit implies neither vacancies nor a mandate; a missing mandate is Unknown, never "No".

Data-first design

1. src/data/gcc-dataset.json: 19 accounts, 21 sources, 21 evidence items, 80 role observations (4 verified vacancies). See docs/SOURCE_REGISTER.md and DATA_AUDIT.md.
2. src/lib/validate.ts: integrity gate, also enforced at app start-up. Every role's wording must appear in its evidence; weak sources can't back "verified"; vacancies must come from a company job board with a posting URL.
3. src/lib/engine.ts: one deterministic engine shared by every screen. Results carry human-readable reasons with evidence references; there is no single score.
4. src/lib/review.ts: reason required for Investigate / Reject / Qualified; Qualified is impossible without documented evidence, which you record, that the company uses or seeks external recruitment agencies. Decisions append to an audit trail with an assessment snapshot; profile changes never rewrite history.
5. src/lib/persist.ts: validated localStorage persistence. src/lib/brief.ts: briefs from structured templates; the JSON export is the same object the page renders.

The engine consumes a plain Dataset, so a future ingestion step (see AI_METHODOLOGY.md) could add evidence without changing qualification logic.

Run and test

npm install
npm run dev                          # http://localhost:3000
npm test                             # 67 unit tests (Vitest)
npm run typecheck && npm run lint
npx playwright install chromium      # first time only
npm run build && npm run test:e2e    # production build + 9 browser tests

Deploy to Vercel (zero environment variables)

Import the repo in Vercel, or npx vercel login && npx vercel --prod.

Known limitations

• Vacancies were verified for one account only (Smartsheet's public job board). For the other 18 accounts the careers page was not checked, so "Verified vacancies: Not verified" means unchecked, not none.
• The Smartsheet job board shows no posting dates, so posted dates are Unknown.
• 19 accounts (target 20–25). 4 accounts lack verified role information and show Insufficient evidence.
• Evidence text comes from automated page extraction and was checked by a second automated pass; a human should open the links before quoting figures externally.
• Role families and vacancy seniority labels are the curator's classification.
• Decisions are per browser (localStorage); no sync or multi-user support.
• Freshness is computed against the viewer's current date, so the dataset looks staler over time until refreshed.