Experiment registry
Useful claims should be testable.
Until a study has real outputs, it stays marked upcoming. This page is intentionally noindex for now.
TEST 001Upcoming
Minimal context vs detailed context
- Question
- Does adding relevant recipient detail produce more varied and useful ideas?
- Controlled
- Same people, budgets, deadlines and base prompt structure.
- Evaluation
- Specificity, category variety, constraint fit and obvious repetition.
TEST 002Upcoming
Which detail changes the answer most?
- Question
- Relationship, budget, timing, interests, avoid list or past gifts?
- Controlled
- One variable changed at a time.
- Evaluation
- How much the direction set actually moves.
TEST 003Upcoming
Same people, different gift finders
- Question
- How do current general AI tools differ on the same recipient briefs?
- Controlled
- 10 fixed recipient profiles and the same prompt structure.
- Evaluation
- Outputs anonymized before rating where practical.
What we will publish with a finished test
01Exact recipient profiles and prompts
02Raw or anonymized outputs
03Scoring criteria decided before results
04Limitations and anything that changed during the run
When the first study has real data, this page can be indexed and the result gets its own dated report.