1. ID comes first
If the ID matches exactly, the case moves close to the maximum. If the IDs conflict, the score drops even when the name looks similar.
Labs
Matching lab
This space separates experimental work from the main directory. We will test the zero-cost pipeline here for cross-matching public missing-person reports with public located-person lists.
Visual review of possible matches
This page shows possible matches for human review. The percentage is heuristic: an exact ID match carries much more weight than name similarity.
How the score is calculated
The current formula does not use machine learning. It is a review heuristic: it combines strong and weak signals to suggest human priority.
2. Then the name
Without an ID, the main weight comes from normalized name similarity: exact, very close, strong, or moderate.
3. Age and location adjust
Age and location only adjust the result. An exact or nearby age adds a little; a large difference subtracts. Location adds a small boost and never replaces identity.
How to read the percentages
93–99%: high review priority
80–92%: possible match
0–79%: weak or ambiguous signal
The percentage does not confirm identity on its own. Review the evidence shown on the card and, when available, verify the ID, age, hospital, and original source.
…Loading labs summary
Loading candidate matches
Initial version
We start without a paid backend: GitHub Actions for source syncs, static JSON files for publishing results, and manual review before confirming any match.
Expected outputs
- Normalized missing-people dataset.
- Normalized located-people dataset.
- Candidate-match dataset with score and explanation.
Zero-dollar infrastructure
- Scheduled GitHub Actions to fetch Kobo and Localizados.
- Repository scripts for parsing, normalization, and scoring.
- Generated datasets under docs/ to serve them through GitHub Pages.
- Manual review of suggested matches before anything is published.
Pipeline we will build
- Ingest: fetch Kobo and Localizados on each run.
- Parsing: extract name, ID, age, contact, and location from free text.
- Normalization: clean accents, casing, whitespace, and name variants.
- Matching: generate candidates by ID, name, age, and location.
- Review: publish suggested matches separately for human verification.
Confirmed public sources
KoboToolbox · TerremotoVE
Public JSON feed with reports of missing people, missing families, rescued people, and related events.
Localizados Venezuela API
Public read-only API with people located in hospitals and other facilities.
Later this can move to a subdomain like labs.directorioterremotovenezuela.org, but starting with /labs/ is the simplest and cheapest option.