A chosen offer still has to deliver.

The commercial question

Can a purpose-built, outcome-shaped endpoint win against credible alternatives at comparable job economics, and can it fulfil the promise?

The underlying programme compared offer cards in controlled buyer-agent tasks and checked selected endpoints against working infrastructure. The public account is anonymised.

What the records show

The October 9–10 wave report records 88 valid buyer-agent decisions across several task-specific offers. Rival offers were modelled from published terms; their vendor credentials were not connected.

Some selections were followed by successful execution, while others exposed failures, including incomplete outputs and timing problems. The record kept those failures. One earlier pass with an incomplete rival set was excluded from evidence.

The useful finding was not that every offer won. The test made the gap between a convincing card and a working product visible, and that gap determined the next product change.

What this does not establish

These are bounded sandbox decisions, not 88 independent customers, live marketplace share or realised revenue. Presentation, task fixtures, model behavior and competing-card construction affect the result.

Selection does not establish contribution margin. That needs delivery cost, retry exposure and repeat-use economics. Claims about a market need separate live validation.

How it informs client work

Test the offer against credible alternatives. Execute the chosen product when access permits. Keep the exceptions and failure cases. Then decide whether the result is strong enough for a live pilot.