Research
Research reports
When AI is asked to organize an account of a problem, what does it add, and what does it drop? Does it put a year on "last October"? Does it keep conflicting memories as they are? Does it keep evidence the writer said does not exist? Does it give advice nobody asked for?
We report this one measurement at a time. Each report has a question, the conditions, one results table, a side-by-side of the source and the output, and the limitations. We state conclusions only within what we measured. When an earlier conclusion turns out to be wrong, the correction stays on record.
The tests use fictional first-person accounts written in Japanese.
Reports
- September 27, 2026Report 001: Does an AI add a year to "last October"?
- September 28, 2026Report 002: Does a wavering memory survive after an AI organizes it?
- September 28, 2026Report 003: Do evidence items known not to exist survive the organizing?
- September 28, 2026Report 004: When asked only to organize, does an AI also write advice?
- September 28, 2026Report 005: Does splitting the organizing work into stages make it better?
- September 28, 2026Report 006: If you tell the AI "do not add years," does it stop inventing them?
- September 29, 2026Report 008: Does adding a rule to unify names change how evidence is picked up?
- September 29, 2026Report 009: What do intake forms and recording practices separate?
Test cases (fictional accounts)
- chat_log.txt (Japanese)
- harras_long.txt (Japanese)
- harras_year.txt (Japanese)
- mumble_accident.txt (Japanese)