We did not repeat the same demo three times. Each experiment was designed to expose a different failure mode in AI-assisted prospect research.
Experiment 01 · Japan Outdoor Market
Can the system keep business roles accurate across a long search?
A Canadian outdoor-accessories brand explored Japan. The test focused on importer, distributor, retailer, and manufacturer classification, deduplication, skeptical audit, and contact-route separation.
What happens when the original market-entry assumption is too narrow?
The mission began with “find distributors.” Repeated market evidence showed that integrators and engineering firms also controlled access to deployment, so the mission was formally revised.
42 universe20 researched12 initial qualified7 final
If a strong LLM is already good, what does the framework add?
FlyPig used itself to find Canadian early adopters, then a fresh high-capability LLM solved the same business problem without reading the FlyPig method.
13 FlyPig final15 control final1 major process change