Playbook
Try this versionAsk for a product and watch this version research it, step by step. It's a real check (it costs what a user's check costs).
Tests
Models and reasoning effort for this run (overrides on top of the version; the config hash records them)
Test runs
| # | Version | Config | Mode | Status | Passed | Score | Judge | People | Cost | Time | Started |
|---|
Test cases
Only cases a person has confirmed are run. "Add to test set" on a check in Research pre-fills one.
| Request | Where | Good answer | Confirmed | In tests |
|---|
Bench
Which playbook, models and effort get closest to the reference answer, at what cost and speed.
Agreement with the reference vs cost per case
Where the offers came from (leads → kept → verified), selected config
Configs on this snapshot
Mean over the config's runs, with the range when it ran more than once. Agreement: product, featured shop, featured price within 2%, price-at-home shop, Match verdicts, against the reference.
| Config | Runs | Agreement | Product | Best shop | Best price | Home | Match P / R | Score | Judge | Cost / case | Time p50 / p95 | Verified | Replay |
|---|
Research
Every price check, stage by stage, and what people did with the answer.
Checks
| When | Asked | Product | Outcome | Best | Offers | Time | Cost | Taps |
|---|
Orders
Buy For You orders on the central Buy&Ship account: what each needs next.
| Order | Date | Customer | Items | Total | Payment | Status | Next step |
|---|
Buy&Ship inbox
Emails from Buy&Ship in the central mailbox, linked to orders.
Users
Click a person to read their conversation with lowtide.
| Name | Phone | Role | Orders | Joined |
|---|
People who asked on WhatsApp. They never sign up, so they are not in the list above; chats are kept for 30 days.
| Phone | Country | Via | Asked | Last asked | First seen | Last active |
|---|