buying-signalslisted

# Intent Signals ## Pipeline contract **Stage 9 of 20.** Reads `08-leads.csv`. Writes `09-signals.csv` (`account_domain, signal_type, signal_detail, source_url, event_date, collected_date, confidence`). - One row per signal, not per account. No signal found = no row. Never write a placeholder signal. - `event_date` is when it happened, `collected_date` is today. A signal without a `source_url` does not get written. - **Verify every signal before it is written — research sub-agents get this wrong more often than they get it right.** For each one: fetch `source_url` and confirm it resolves (not 404, not an unrelated page); confirm the page is about THIS `account_domain` and not a same-named company (Array the fintech vs Array Technologies the solar firm; hellotilt.com vs tilt.app); confirm `event_date` is inside the window. A live job post with no posting date is recorded with `date_precision=observed`. A source that blocks fetching is kept out until a human opens it. Log every drop with its reason. - Cheap sources first: GetLeads MCP `list_funding_signals` / `list_acquisition_signals` (1 credit per record), then job boards and the company site. - Per-signal collection playbooks: `references/` (hiring surge, fundraising, new-in-role, tech on website, job-posting language, champion moves). Signal-to-message mapping: `references/trigger-mapping.md`. Files live in `runs/<run>/`. Column definitions: [`/prospecting-system`](../prospecting-system/SKILL.md). Standalone use: ignore