TMK

Research journal

How I investigate

An open account of the working method behind the benchmarks: how questions are framed, how sources are judged, what gets logged, and what gets dropped.

  1. Entry 01

    Framing a question that survives a search engine

    A question is only useful if the plain-language version of it does not answer itself. Every draft is typed into a search box first. If the answer appears in a snippet, the wording is changed — usually by removing the proper nouns and describing the situation instead. The search that broke the draft is kept in the validation log rather than deleted.

  2. Entry 02

    Separating the famous date from the correct one

    Most well-documented events carry several dates: the signing and the entry into force, the hardware install and the first transmission, the argument and the judgment. Asking for the less-quoted one keeps the answer short and objectively verifiable while defeating answers produced from memory rather than from a source.

  3. Entry 03

    How a source earns its place

    Each candidate is scored against five criteria before it is cited: who issued it, whether it is primary or secondary, whether the specific claim can be located inside it, whether it is publicly reachable, and whether it agrees with an independent record. An institutional record that only confirms a year is still recorded — its limits are part of the evidence.

  4. Entry 04

    Quoting only what was read

    Text is quoted only where the wording was opened and read on the page. Where a detail could not be confirmed directly it is described rather than quoted, and no page number, locator or excerpt is written from memory. Where the strongest available evidence is encyclopaedic rather than archival, the write-up says so.

  5. Entry 05

    Recording the failures

    The searches that returned nothing useful matter as much as the one that worked. They show which paths a capable researcher or browsing agent would try, and they are the reason a benchmark can be trusted to be difficult rather than merely obscure.

  6. Entry 06

    Knowing when a benchmark is finished

    A benchmark is complete when the answer is fixed, every clue points at a source another person can open, the decoys are documented, and the remaining limitations are written down. Anything still unresolved stays in the reflection section instead of being smoothed over.

The method in practice

Every case study below applies the same process end to end.

Published write-ups