Between 24 June and 14 July 2026 we combined three registries of people missing after the La Guaira earthquake and deduplicated them, free of charge. The search on this page is closed. We were never the origin of these records — they came from several public and partner registries, and they were never an official source. If you are looking for someone, go directly to the registries themselves:
Three registries,
one record per person
After the La Guaira earthquake, several independent efforts collected reports of missing people and lists of hospital patients. The same person was often reported many times — sometimes in more than one registry. Tilores combined those sources and deduplicated them into one record per real person — without losing any data — and left every decision in human hands.
We combined three independent registries — venezuelareporta.org, Desaparecidos Terremoto Venezuela, and buscatupaciente.com — with Tilores entity resolution, and published the results free to organisations and families searching in Venezuela. This is what that produced.
How the project ran
Earthquake in La Guaira. Independent registries begin collecting reports of missing people.
First import: venezuelareporta.org, 53,693 records. First deduplication run.
buscatupaciente.com asks us to screen ~9,053 hospital patients against the registry.
A third registry — Desaparecidos Terremoto Venezuela — is added. Public search opens.
Three-source run: 140,831 records resolved into 107,075 distinct people.
Final run after five more days of source checks: 141,080 records resolved into 107,229 distinct people.
Three registries, combined and deduplicated
Final entity-resolution run · 14 July 2026, 16:55 VET (22:55 CEST)
From three independent sources
After entity resolution (preliminary)
≈24% of all records were duplicates
Sources: venezuelareporta.org (59,425 records), Desaparecidos Terremoto Venezuela (70,215), and buscatupaciente.com (11,440, hospital patient lists).
The distinct-people total is a preliminary estimate: with more aggressive matching rules, we did not re-verify the over-merge rate before the project closed.
Still missing, located, and yet to match
Only had "missing" reports
Found, safe, or matched to a hospital patient
A searchable pool for families still looking
"Still missing" counted open cases in the combined registry (people with only missing reports) — not confirmed casualties. It is a slight overstatement: the daily job caught new reports, but a registry editing an existing report from "missing" to "found" in place was not always picked up.
The same person in more than one registry
venezuelareporta.org and Desaparecidos Terremoto Venezuela were independent registries of the same people missing after the earthquake, so they overlapped enormously. Entity resolution cross-linked them into one record per person.
The same person recorded in more than one place
Appeared in two of the sources
Appeared in all three sources at once
Connecting missing-person reports with hospital patients
Hospital patient lists — buscatupaciente.com ↗ plus Desaparecidos Terremoto Venezuela's hospital lists — were screened against the missing-persons registry by national ID (cédula) and name, to locate people who were alive and in care.
People connected to a hospital patient record (1,880 records linked)
Only had a "missing" report and appeared as hospital patients
A "missing" report plus a "found / safe / hospital" record
Read this carefully. The 84 leads were candidates for human review, not confirmed reunifications. And 24,296 hospital patients were never matched: a searchable pool for families still looking.
What families saw when they searched
From 3 July, anyone could search the combined registry by national ID (cédula) or first + last name. One search returned one card per person — not per report — collapsing every duplicate across all three registries into a single answer.
Illustration only. Invented name and details, showing the shape of a result. We never published names on this page, and the search is now closed.
In a crisis, duplicate reports cost time
When many people report at once — and across several different registries — duplicates are not a cosmetic problem. They inflate the apparent number of missing people, scatter one person's information across many reports, and bury the fact that matters most: that someone has already been found.
Someone comparing several lists by hand cannot reliably tell that "María C. González" and "Maria Gonzalez" are the same person, or that a missing report already has a hospital patient waiting elsewhere in the data.
Entity resolution, in plain language
Entity resolution is the technology that recognises when several records refer to the same real person, even when the details are written differently. We combined the three sources — 141,080 records — and resolved them into distinct people. Matching was anchored on the national ID (cédula), plus name (with phonetic matching), age (within a ±4-year window), and approximate location.
On a subject this sensitive, a false positive is worse than no match. So the configuration was deliberately cautious:
We combined 141,080 records from three sources. A job re-checked all three sources every day and fed new reports back in; entity resolution refreshed incrementally, re-processing only what had changed. That is why a registry which grew by 87,000 records in under three weeks never needed a full rebuild — the last five days alone added 249 reports and 154 more distinct people.
The matching rules
Before anything was compared, every field was normalised: accents and capitals removed, spacing standardised, and connector words ("de", "la", "los"…) stripped. Names also went through a Spanish phonetic normalisation — b/v, g/j, c/s/z, ll/y, silent h, qu/k — so spelling variants lined up. The national ID was reduced to digits, and masked IDs (with • or *) were discarded.
Two reports were merged into the same person when they satisfied one of these rules:
Same cédula (digits only) and the first name agreed — the name check stopped a typo in the ID from merging different people.
"V-12,345,678 · José González" + "12345678 · Jose Gonzalez" → same person
The reporter's phone or email matched + first and last name agreed + age equal or empty; compound records were excluded.
Two reports from +58 412-555-1234: "María González, 30" and "Maria Gonzalez" (no age) → same person (re-submission)
Name (phonetic) + age + a shared distinctive location. Linked the same person reported by different people, even across different registries.
"Jesús Giménez, 24, Playa Grande, Naiguatá" (venezuelareporta) + "Jesus Jimenez, 25, Naiguatá" (Desaparecidos Terremoto Venezuela) → same person
Humans stayed in control
Every match was a suggestion for human review. A person confirmed before any case was closed, because a wrong match could stop the search for someone still missing.
The figures on this page come from the final resolution run of 14 July 2026 and do not change. They reflect the de-duplicated registry; the raw per-source totals are larger because the same people appeared repeatedly.
We worked from the registries' public data, in email contact with the teams behind them. If you run a registry or a relief effort and this would help you, we'd still like to hear from you: hello@tilores.io.
Three deliverables
One record per real person, merging reports from all three sources, with a "found / safe" status if any of their reports indicated one.
33,851 duplicate records merged: which reports were the same person, keeping one primary, using each source’s original identifiers.
Hospital patients linked to their missing-person report by national ID and name, to help locate people who were alive and in care.
Three things this project taught us
Roughly a quarter of all records were duplicates. Read raw, the registries suggested far more missing people than there were — and 4,931 people who had already been found were still counted as missing somewhere in the data.
Deduplicating one list is straightforward. Linking three lists built by different people, with different fields and different spellings, is where entity resolution earns its keep: 23,395 people appeared in more than one of them.
We tuned deliberately cautious and accepted the cost: a preliminary distinct-people count we never re-verified, rather than a confident number built on merges we could not defend.
None of this is specific to a disaster. The same duplicates sit in customer databases, patient records, and supplier lists — the stakes are just lower and the deadline less obvious.
This is what entity resolution does under pressure.
141,080 records from three sources, resolved into 107,229 people. If you run a registry, a relief effort, or a hospital's data, we'd be glad to talk.