purchase-advisor

Version 1.3.0.

Someone wants to buy a specific kind of product (a TV, a stand mixer, a computer) and does not yet know which one. The run moves through scan, interview, constraint cost, hunt, verify, recommend, and ends with one recommendation they can act on today. Choosing is the job: a run that hands back a neutral pile of candidates has pushed the hardest part onto the user.

Storage goes through the effort-store skill. Load it before writing anything. Declares: root purchases; notes Requirements, Buying guide, Shortlist. The effort name is the product in a few words: Stand mixer.

0. Resume or start

find the product in the effort store before anything else. If an effort exists, offer to resume: the requirements and buying guide stand. Ask what has changed since; a moved budget or requirement reopens the interview for that field only, and update writes the new answer into Requirements as a dated section before the hunt launches. Location is re-asked on every resume regardless of what else changed; it is never carried forward from the saved note. When the Buying guide has no tester roll-call (an effort from before version 1.3.0) or the location moved to another country, build the roll-call as section 1 describes and update it into the Buying guide before the hunt. The run then re-enters at the hunt with the criteria as updated, and update adds the new results to the Shortlist note as a dated section above the old ones. Second-hand hunts especially span weeks; the saved effort is what makes waiting cheap.

1. Scan

In-session: a few searches and reads to learn the category, plus one search per tester for the roll-call, before asking the user anything, so the interview asks the questions this category deserves rather than generic ones. No subagent; the expensive search comes later, once it can be fully constrained.

Come out holding five findings: