2 minute readAPIAgent-ready
Every computed price now says what it means
compute_price answers now carry a price object beside the number: the expected acceptance at each price, the range the price was chosen in, what it was conditioned on, the evidence behind it, what it cost to compute, and whether it was an exploring draw or a negotiation round.
An agent acting on a price needs more than the number. Every compute_price answer, over the API and the MCP tool, now carries price_object:
- an acceptance curve, the chance a buyer says yes at each price with a band, when the workspace has learned one; empty when it has not, never guessed
- the range the price was chosen in, and whether that range came from your limits, the workspace band, or nothing
- what the price was conditioned on: what is sold, how it is charged, the unit, the buyer, the product, the channel, the function
- the evidence behind it: learned, assumed or cached, the learned elasticity and how many outcomes it rests on
- what the answer cost to compute, from the meter
- what the model layer did: a plain quote, an exploring draw with how likely it was, or a negotiation round's continue or stop
- beside it, what a decision layer would have done for the same quote (quote, explore, or stop paying to learn), judged by what one learning round is worth against what it costs on the meter; reported only, it never moves the price
Nothing already in the answer changes, and no price moves. Each part is defined in the vocabulary at /.well-known/lastprice-vocabulary.json.