Research · in the open
Studies
Planned tests of AI answers. Each study states its scope, its sample and its limits before its result.
A field note is one observation. A study is a planned test: its questions are fixed before the engines are asked, every answer is kept, and it says what the result can and cannot support. Where the analysis was not planned in advance, the study says that too.
0·001 study, 1 example
scope read 29 August 2026 · exploratory · Denes Csaszar · published on denescsaszar.com
Answer engines rarely word an answer the same way twice. This study asked eight answer engines eighteen questions about one company, in seven capture waves between 4 July and 25 August 2026, and compared how alike the answers were inside one wave with how alike they were across waves. In 85 of 103 question-by-engine cells, the answers were less alike across waves than inside one.
The study is exploratory, and its similarity measure compares wording, not facts. The result is net of engine and question, not of period: in this archive each wave is also one sitting, so it cannot say whether the difference is change over time.
Why it matters here: before anything on this site calls an answer changed, the answer has to be set against how much it moves on its own. This study measured that floor.
Read the study → Field note 001, the observation behind it →
Cite as Csaszar, D. (2026). The floor test, scope 08284347. https://denescsaszar.com/drift-2026-08
invented data · not a result
A complete reading with invented data: three questions about an invented product, seven anonymised engines, two languages, three draws, the locked ground truth, the scoreboard and its SHA-256 fingerprint. It shows what a real reading contains without publishing one.
Read the example reading →
The next study needs two research partners: companies whose own published facts get tested, free, results first to them. How it works →
More studies are in preparation. Each one appears here once its data and its limits are ready to publish, and a study that finds nothing is listed too.