How QuoteRunner is measured

50 simulated small-business owners, each with known true facts (and 1–2 facts they don't know), talk to the same agent this site runs. Every field it records is checked against the truth.

Showing: run 1 · 2026-11-08 · 3 conversations · agent gpt-5-nano (reasoning low) · commit cd5506f-dirty · “smoke test”

100.0% field accuracy
0 invented values, in 0 of 3 conversations
100% conversations finished
5.0 questions on average, vs a 25-field form
$0.0019 agent cost per conversation (+$0.0005 for the simulated customer)

What the numbers mean

Sample conversations

Accuracy by field

FieldCorrectAccuracy
Annual revenue3/3 100%
Business name3/3 100%
Deep fryers1/1 100%
Pool on premises1/1 100%
Industry3/3 100%
Licensed capacity (children)1/1 100%
Employees3/3 100%
Business vehicles3/3 100%
Offers delivery1/1 100%
Claims in last 3 years3/3 100%
Seats1/1 100%
Serves alcohol1/1 100%
State3/3 100%
State licensed1/1 100%
Transports children1/1 100%
Years in business3/3 100%

By industry

daycare100.0%
other100.0%
restaurant100.0%

By speaking style

many facts100.0%
rambling100.0%
typos100.0%

Every conversation in run 1

#IndustryStyleAccuracy QuestionsFinishedProblems
1 daycare rambling 100% 4 yes
2 restaurant typos 100% 8 yes
3 other many facts 100% 3 yes

Run history

RunDateNotePersonasAccuracy InventedFinishedAvg questions
4 2026-11-08 repeat of run 3, same code, to measure run-to-run variance 50 99.3% 0 90% 7.5
3 2026-11-08 after summary fixes (state names, hide industry other, classification hint) 50 99.8% 1 98% 6.8
2 2026-11-08 first full run 50 98.9% 0 78% 10.8
1 2026-11-08 smoke test 3 100.0% 0 100% 5.0