Deep Prospecting

Deep prospecting is the process for identifying and verifying the heirs/decision-makers behind a distress lead, and determining who must actually sign for the property to sell. Core philosophy: "When everyone hits a wall, we bring a shovel."

Version 5 (2026-07-29). The heir engine changed. v4 resolved relatives through the Enformion/Endato Person Search. v5 replaces that with SmartSkip, measured head to head on live Knox and Blount County records. Enformion's person search is retired from this skill. Enformion BusinessV2 is retained, for entity owners only, where nothing else works.

Why the change, with the numbers that drove it:

• Coverage. Enformion returned zero relatives on 6 of 12 owners tested. SmartSkip returned relatives on 12 of 12.
• Phones. Enformion's relativesSummary carries names but no phone numbers. Every relative you actually want to call costs another billed search. SmartSkip returns the relatives and their phones in one batch row.
• Cost. For a 100-owner batch with 682 relatives: SmartSkip plus Tracerfy = $15.90. The Enformion equivalent = $78.20. That is 4.9x.
• Precision. On the validation record, SmartSkip returned exactly 3 relatives and all 3 appeared in the published obituary. Enformion returned 50 (a capped blob) plus out-of-state numbers that looked like wrong-person bleed.

What did NOT change: the research layer. SmartSkip is wrong about death (Gotcha 1). The obituary and web pass is now mandatory, not optional.

The v5 stack

Typical single record: about $0.24 all in.

Grounding rule (unchanged, and it is the whole point): every name, address, phone, and relationship must come from a source you actually retrieved. Never infer or fabricate an heir. Surface conflicts; do not resolve them by guessing.

THE GOTCHAS. Read these before running anything.

1. SmartSkip's Deceased flag is unreliable, and there is NO date-of-death column

Live proof: SmartSkip returned Deceased = false for a man who died 12/06/2025 and whose obituary was published by a funeral home. SmartSkip carries a Deceased boolean and no DOD field at all. Never let SmartSkip decide who is alive. Date of death comes from the obituary and web pass, always.

2. THE SPOUSE-OBITUARY TRAP. An obituary on the record does NOT mean the owner died.

This is the highest-value check in the skill and it is easy to miss.

A live CRM record sat on the Obituary list with a recent obituary date and a status of Deep Prospecting, which reads as "owner is dead, go find the heirs." The obituary was for the owner's husband, not the owner. She was alive and she owned the property. The correct action was not heir research at all; it was to call the owner, gently, eight months after she was widowed. A caller working the un-researched record would have opened by asking for a man who had died.

Always confirm WHOSE obituary it is before treating a record as an heir case. Match the decedent's name against the owner of record. If they differ, the record is a living owner with a recent death in the household, which is a different and often better lead. Correct the record, do not just work it.

3. SmartSkip's relationship labels are coarse. Verify them.

The column is literally called "Possible Type." In a 100-record batch, 63% of relationships came back generic ("Relative" or "In-Law"). On the validation record it labeled the husband of 62 years as a plain "Relative." The obituary fixed it. Treat SmartSkip relationships as a starting hypothesis and let the research layer overwrite them.

4. The wallet does not pay for bulk skip

Topping up the SmartSkip wallet does not fund a bulk run. Verified live: with $25.00 in the wallet, a paid batch charged the saved Stripe card and left the wallet untouched. Bulk skip always bills the card through payment-intent. Budget on the card, not the wallet.

5. Unpaid orders are invisible. Keep the bulkSkipId.

An order that has not been paid does not appear in GET /bulk-skip. If you lose the bulkSkipId returned at upload, you cannot find the order again. Persist it before paying.

6. Entity owners cannot be name-traced at all

SmartSkip requires a first and last name. LLCs, trusts and estates return nothing, and Tracerfy is consumer-only. In a real vacant pull, 35 of 321 owners (11%) were entities. Those route to Enformion BusinessV2 (see references/entity-owners-enformion-business.md). Filter entities out of the batch up front rather than discovering this per record.

7. The owner rule wins on a shared household line

When the owner and a relative share a number, that number carries source and tier only, no relationship tag. Otherwise the dial sheet labels the owner's own landline "Husband" and the caller misreads who is picking up.

8. The DOD sanity check still applies, and still anchors on the publication date

Reject an obituary match whose date of death is more than 3 years before the notice's publication date (date_published, not the date you ingested the record). This still matters in v5: a stale-index date of 2004 surfaced on a live record during validation.

9. Never put a person's name in a phone tag

Names live on the message board using the last-4 method. Phone tags carry source, dial tier and relationship only.

Workflow

Step A. SmartSkip the owner (the heir engine)

Build a CSV with the required columns (First Name, Last Name, Mailing Address) plus the property address, then run scripts/smartskip_trace.py. Full API contract in references/smartskip-api.md.

export SMARTSKIP_EMAIL="..." SMARTSKIP_PASSWORD="..."
export SKIPTRACE_RUN_DIR="$HOME/dp_run"

# FREE: upload + auto-map columns + calculate. Prints the billable row count.
python3 scripts/smartskip_trace.py submit --csv owners.csv

# CONFIRM THE COST WITH THE OPERATOR, then bill the card:
python3 scripts/smartskip_trace.py pay --id <bulkSkipId>

# poll, then pull the export
python3 scripts/smartskip_trace.py status   --id <bulkSkipId>
python3 scripts/smartskip_trace.py download --id <bulkSkipId> --out export.csv

submit, status and download are free. Only pay charges. Never chain straight to payment on someone else's account without confirming the row count first.

Parse with scripts/parse_smartskip.py, which handles both export layouts (vertical "Campaign" and horizontal "CRM"), keeps Subject plus Relatives, drops Associates, drops absurd ages, and infers a canonical relationship tag.

Expected yield (Knox and Blount, vacant and distressed owners): about 89% hit rate, roughly 6.8 relatives per record, about 16.6 unique numbers per record, and 93% of relatives arrive with at least one phone.

Step B. Tracerfy gap-fill (small, cheap, often unnecessary)

Only for relatives SmartSkip named but returned no phone for. That is roughly 7% of relatives, and in practice all of them carry a mailing address for Tracerfy to anchor on. On a 100-record batch this was 45 people and 90 cents. Skip the step entirely when the gap is zero, which is common on single records.

Step C. Obituary and web research (MANDATORY for any suspected death)

This is what replaced Enformion, and it does the job Enformion was originally brought in to do: grounding. The difference from the v3 hallucination problem matters. Back then an LLM was asked to invent the heir set from obituary prose. Now SmartSkip hands over a grounded relative list and the research layer only has to confirm relationships and supply the date of death. That is a far safer job.

Do all of these:

1. Search the decedent's name plus city plus "obituary" plus the year. Funeral-home pages are the best source.
2. Confirm whose obituary it is against the owner of record (Gotcha 2).
3. Extract the verbatim survivors paragraph. Reconcile it against the SmartSkip relative list in both directions: relatives SmartSkip found that the obituary does not name, and survivors the obituary names that SmartSkip missed (married-out daughters are the classic miss).
4. Overwrite SmartSkip's coarse relationship labels with the obituary's real ones.
5. Record the date of death and run the 3-year sanity check.
6. If pages sit behind Cloudflare or a JS wall, use the Scrapfly fetcher. County, records and genealogy portals are the sweet spot; hardened people-search aggregators frequently IP-ban it.

references/google-dorking.md has the query patterns. references/deed-analysis.md covers title.

Step D. TrestleIQ scoring

Score the deduped union of every number: existing CRM numbers, SmartSkip numbers, Tracerfy numbers. Dedupe globally first, since a number's tier is the same wherever it appears. Skip numbers that already carry a recent tier.

Dial tiers (identical across every skill in this library): 81-100 Dial First, 61-80 Dial Second, 41-60 Dial Third, 21-40 Dial Fourth, 0-20 Drop.

Step E. Signing analysis

Determine who must sign, ranking decision makers by signing authority under state intestacy rules, from verified living heirs only. Never assert an heir interpretation; flag it. For probate, label relationships relative to whoever was skip traced ("Sons of [PR]").

Reconcile the signer set against the published obituary's survivor list, not the relative graph alone.

Step F. Deliverable

Produce the research pack (templates/research-pack-template.md) and render it to PDF for upload as a record attachment. Lead the pack with the headline correction whenever research changed what the record means, as in the validation example, because that is what actually stops a bad call.

Write back to the CRM: phones with source, tier and relationship tags, one combined message-board post, and property tags.

When to Use This Skill

• Skip trace returned no usable phone numbers
• Called 3 or more attempts with no contact
• Vacant mailing address discovered
• Return mail (bad address)
• Probate cases (often only a docket number)
• Entity or LLC ownership (needs the decision maker, see Gotcha 6)
• Conflicting owner or address information in public records

Cost model

Trestle is usually the largest line item on a big batch because it prices per unique number. Dedupe globally before scoring and skip already-tiered numbers.

What v5 removed

The Enformion person search. scripts/enformion_person_search.py is retained only as a deprecated reference for anyone still running v4. Its failure modes, for the record: zero relatives half the time, no phones on the relatives graph, a roughly 50-relative cap that silently truncates, a surname gate that drops married-out daughters, isDeceased flags that lag reality, and wrong-person matches when anchored on city and zip alone rather than the full street line.