Web research • Fact-checking • AI evaluation

TMK Research Lab

I investigate difficult open-web questions, verify them against primary sources, and document the reasoning so anyone can audit it.

Open-web investigation

Locating primary documents across government portals, institutional databases, PDFs and archives.

Fact-checking and verification

Separating evidence from assumption, and recording the exact locator behind every claim.

Structured documentation

Auditable evidence chains, search logs and clean JSON deliverables another person can reproduce.

Projects

Ten completed benchmarks across history, space, law, politics, technology, culture, science, economics and public records — each with a fixed answer and a full evidence trail.

Independent Research Demonstration — Not Client Work

AIBRC-001 · Public records and administrative history

Identifying a Ghanaian Local Authority

A short, verifiable answer hidden behind legislative, administrative, geographic and census clues, with the two search shortcuts found in validation disclosed.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-001 · History of technology

AI Browsing Benchmark Research Challenge

A completed, fully sourced browsing benchmark built around the 1858 transatlantic telegraph cable, with independently checkable clues and a reproducible evidence trail.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-002 · Space mission records

Voyager Launch-Order Benchmark

A second completed browsing benchmark on a different domain: a space-mission question answered entirely from NASA records, including a technical flight-data PDF.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-003 · Legal case records

Landmark Judgment Benchmark

A third completed benchmark in a legal domain: a court-judgment date recoverable only from law reports, a judgment transcript and national archive catalogues.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-004 · US court records

Consolidated Appeal Benchmark

A legal-records benchmark where the famous date is the wrong answer: the task is the first oral argument date of a consolidated 1950s appeal, printed only in the official reporter.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-005 · Politics and treaties

International Politics Benchmark

A politics benchmark built on treaty records: the signing date is famous, the date the instrument actually took effect is not, and both are documented by the depositary.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-006 · Technology history

Computing History Benchmark

A technology benchmark where the widely cited month is wrong and the correct day depends on which time zone the log is read in.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-007 · Culture and museums

Cultural Heritage Benchmark

A culture benchmark on a famous museum theft, built so that the holding institution corroborates the actors while the day-level date must come from a second, checkable record.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-008 · Science and publishing

Scientific Record Benchmark

A science benchmark that asks for a journal issue date rather than a discovery date, so the famous year and the Nobel year both act as traps.

Read the case study
Independent Demonstration Project — Not Client Work

AIBRC-009 · Economics and central banking

Economic History Benchmark

An economics benchmark that separates the day a central-banking statute was signed from the day the twelve regional banks actually opened, using the system's own records and a 1914 Treasury announcement.

Read the case study
Read the research notes