`RESUME-DIG.md`

37 lines
# RENTV lead-recovery loop — resume procedure

Goal: raise the reachable-email coverage of the 5,136 unique RENTV leads from 84%
toward ~90% by recovering emails for named contacts whose company domain is missing
from the sheet. All work is LOCAL/$0 and STAGED — nothing is written to the Google
Sheet (that write is gated; run `infer_emails.py --apply` on Steve's go).

## Lanes (in priority order)
1. ✅ Deterministic pattern inference — MAXED (0 new; live `infer_emails.py` already ran).
2. ✅ Company-name normalization recovery — 72 clean emails, ambiguity-guarded
   (franchise brand-collision like Cushman&Wakefield→picor is BLOCKED). See
   `rentv_recovered_emails.csv`.
3. 🔄 Web-domain lookup — 238 companies (326 contacts) in `rentv_company_domain_todo.csv`
   need a company website→domain via the FREE stack (host WebSearch). IN PROGRESS.

## Loop step (do this each wake-up)
1. Read `rentv_company_domain_todo.csv` (ranked by appearance-weight, hottest first).
2. Read `data/company_domains_found.csv` — skip companies already resolved.
3. Take the next ~10 unresolved companies. For each, WebSearch
   `"<Company> commercial real estate official website"`, extract the primary domain
   (and a mainline phone if shown). Reject aggregator domains (linkedin, zoominfo,
   crunchbase, clutch, leadiq, bldup, facebook, etc.) — take the firm's OWN site.
4. Append rows to `data/company_domains_found.csv` (Company,Domain,Phone,Source).
5. Regenerate `rentv_websearch_recovered.csv` (first.last@domain for each unreachable
   contact at a resolved company; carry the mainline phone).
6. Commit. Re-schedule the next wake-up until the todo list is exhausted.

## Guards / honesty
- Every recovered email is a GUESS (first.last pattern) until verified — keep them in
  the "Likely/inferred" lane, never the verified Email column.
- MX-verify domains before trusting (see the MX pass; all 39 batch-2 domains passed).
- Ambiguous franchise brands (one name, many local domains) stay BLOCKED, not guessed.

## Done when
`rentv_company_domain_todo.csv` is fully covered by `data/company_domains_found.csv`,
then draft ONE gated memo proposing the sheet write-back for Steve's approval.