Domain Intelligence Playbook

Turn hundreds of thousands of expiring-domain rows into a ranked drop-day list of zero to five defensible local-service assets. The normal target is two or three registrations per day.

Start with the machine architecture Review the scoring model


Operating target

Stage Expected daily volume
Expiring-feed rows 100,000–500,000
Early-filter survivors 5,000–25,000
Lifecycle watchlist 250–1,500
Confirmed pendingDelete candidates 20–100
Ranked drop-day queue 5–20
Registered domains 0–5

The correct result can be zero. The limit is a ceiling, not a quota.

Core rules

Evidence before opinion
Store raw evidence separately from scores and AI explanations.
Cheap decisions first
Do not pay for SEO data until deterministic rules have removed most candidates.
Expiration is not the drop
Treat the published expiration date as a discovery signal. Monitor registry status and calculate the actionable drop window only after pendingDelete is confirmed.
Standard registration only
Exclude auctions, backorders, premium aftermarket listings, and domains that require anything beyond normal registration price.
Human approval, automated timing
Approve the ranked queue before drop day, then let the registrar integration submit registration attempts at the predicted drop time.
Business outcomes win
Optimize for qualified lead profit and asset value, not third-party SEO metrics alone.
Layer Recommendation
Scheduling and orchestration Temporal Cloud or cron for the MVP
Feed ingestion Python + PostgreSQL COPY
Data processing Python + Polars
System of record PostgreSQL
SEO and keyword evidence DataForSEO
AI review and generation OpenAI Responses API
Review console Next.js
Site factory WordPress + WP-CLI + REST API
Build automation GitHub Actions
QA Playwright + Lighthouse CI + Screaming Frog
DNS and edge Cloudflare
Measurement Search Console + GA4 + CallRail
Reporting Metabase

No S3 layer is required for the first version. Download the provider file to temporary local storage, load it into an unlogged PostgreSQL staging table, merge it into canonical tables, record its checksum and import result, then delete the temporary file.