A professional brief for how this system runs, and how to deploy it per client.

Not a lead list. A five-stage automaton that turns real buying signals into qualified, tracked revenue opportunities — with every record carrying evidence and every stage handing a human its next step.

HUNT → REFINE → CAPTURE → ACT → LEARN → (feeds HUNT)

Layer	Where	What it does automatically
HUNT	cron/hunt.ts, email/lead-intake.ts, IG webhook	Watches a real source (sheet/JSON feed every 15 min), ingests forwarded email, and receives IG DMs/comments via Meta webhook. Real signals only.
REFINE	backend/outbound.ts	Scores any real public signal (0–100), writes the signal + recommended action, and auto-drafts a personalized first-touch quoting the actual message.
CAPTURE	lib/store.ts	Promotes a prospect/signal to a verified lead the moment a real email/phone is confirmed. Budget seekers are rejected; nothing is invented.
ACT	cron/automate.ts, pipeline status API	Every 15 min: flags hot leads idle >24h, nudges them to the work queue, and (only with ENABLE_AUTOSEND=1) sends that auto-drafted first touch to verified contacts. Pipeline moves new → … → won/lost with full history.
LEARN	/api/analytics	Reports conversion by stage, source, and signal so budget flows to what actually produces revenue.

Cron	Frequency	Job
cron/hunt.ts	every 15 min	Pull the configured real source into the pipeline
cron/automate.ts	every 15 min	Nudge stale hot leads + auto-send first touches (guarded)
email/lead-intake.ts	on email	Qualified every forwarded inquiry instantly

• No invented information. A lead needs a real contact; a signal without an email is helpfully drafted, never fabricated.
• Auto-send is OFF by default. Set ENABLE_AUTOSEND=1 to let the engine email. Otherwise it dry-runs for human review — responsible, not spam.
• All sends and stage changes are logged so the operator sees what the machine did and why.

This engine is whitelabeled via env config, so the exact same code powers a new client without edits:

1. Remix this val into the client's org (own store, env, crons, dashboard — full isolation per client).
2. Set DLX_BRAND_NAME to the client's brand (used in capture copy + drafts).
3. The client connects their own channels: their website form to /api/ingest, their Google Sheet to DLX_HUNT_URL, their Meta app to the IG webhook, and/or forwards DMs to their deployment's email intake address.
4. You operate: monitor logs, tune disqualifiers/scoring in qualifier.ts, enable ENABLE_AUTOSEND when the client approves, and review /api/analytics.

Because each deployment is isolated, you can run many brands safely from one Codebase, and a client's data, channels, and secrets never mix.

• Connect a real source: Google Sheet CSV → set DLX_HUNT_URL (+ DLX_HUNT_TOKEN if private).
• Point any site form at POST /api/ingest and/or embed /capture.
• Publish dlxdesigns-…@valtown.email as an intake address for DMs/leads.
• (Optional) Connect the Instagram/Meta webhook per the README guide.
• (Optional) Add a free-tier GEMINI_API_KEY for AI-refined signals.
• (Optional) Run database/schema.sql in Supabase to enable the mirror.
• Review /api/analytics weekly; tune scoring in qualifier.ts.