LUMACODE PORTFOLIO · TEAM EXPERIENCE / P5
SIGNAL INTELLIGENCE
A qualification engine where every answer comes with its evidence.
- 10,451
- SIGNALS, EACH WITH EVIDENCE
- 9,779
- DOMAINS MATCHED
- 1,420
- COMPANIES OPENED FROM PUBLIC DATA
THE CHALLENGE
A US consultancy needed to find the utilities and manufacturers that actually run operational technology, and show why.
Public data about critical infrastructure is messy. Names repeat, summaries overcount, and a passing mention can look like proof. A plausible name is worse than no name, so the engine was built to prefer no answer to a wrong one.
WHAT MADE IT HARD
Summaries that overcount: registry figures are 346 and 179 entities, not the 369 and 238 the source workbook suggests.
Mentions that look like proof: only a register naming the organisation as its subject can open an account.
Real volume: the main page loads in 123.6 KB and queries run in 3 to 44 milliseconds on real data.
WHAT WE BUILT
One adapter per source, all with the same shape, so a new source is a file, not a project. Every source is built and tested against real public data.
Evidence modelled as tiers A, B and C instead of a blended score. An account qualifies on one verified Tier A record, or verified Tier B evidence from two independent sources. Verifying is a person's decision, so nothing qualifies itself.
A fuzzy name match never writes a domain: ten separate public power districts sharing the same phrase stay ten separate companies.
WHAT CHANGED
10,451 signals, each tied to the record it came from, 9,779 domains matched and 1,420 companies opened from public data, with 44 checks against a real database plus 18 mapping tests.
Six sources are ingesting.
"Every signal is tied to the record it came from, and the final call stays with a person."LUMACODE PORTFOLIO · TEAM EXPERIENCE
Want this kind of system in your business?