Teardowns
Every service, performed on a live system
No mock screenshots, no unnamed clients. Each teardown runs the service on our own production system and publishes the numbers, so you can judge the work before you ask for pricing.
AI Cost & Grounding Audit
All 8 published questions instrumented to the token, with replays of the recorded runs and their real receipts.
exhibit: recorded run replay with token receipt
Optimization Sprint
The real cost history told honestly, a per question no caching counterfactual table, and what each technique contributes.
exhibit: cache hit rate simulator on measured tokens
Model Watch
The same model with and without grounding, side by side, including the round where the ungrounded answer was fine.
exhibit: spot the hallucination
The cost anatomy
One production answer reverse engineered to the token: the per round waterfall, the cached prefix, the counterfactuals.
the original single answer teardown