Evidence beats demos

The TYR-X proof log documents how AI assistants perform against real forwarding and commercial work. One entry a week. Every entry answers five questions: What was the task? What was the assistant expected to do? What actually happened? Where was a person required? Would I run it again?

Successes show what is becoming possible. Failures show what is not ready. Improvements matter more than a perfect demo. If a week is missed, it says so.

Categories

Commercial research
can an assistant find and structure information a forwarding salesperson can use?
Sales preparation
can it turn research into account preparation rather than generic AI copy?
Follow-up
can it prepare the right next action without losing context?
RFQ and quoting experiments
can it extract, structure and prepare parts of a freight RFQ while preserving the required human controls? (Experiments only. Not a service.)
Assistant reliability
what happens when information is incomplete, contradictory or missing?

Proof log

Week 1 — no proof published.