Observed arrival · 2026-11-04

Rekount’s repeated-answer audit

rekount.net Visit website

Editorial interest 83/100 Selection signal · not a rating of the site

Rekount audits enterprise data agents for whether their answers stay correct across repeated asks and rewordings.

Landing page captured for the 2026-11-04 issue.

For: Teams evaluating enterprise data agents
Worth noticing: In one 72-wording example, it distinguishes 40 flaky wordings from four that were never right.

Field notes

One live demonstration asks Microsoft Fabric, Databricks Genie, and a GPT model writing SQL the same taxi-trip question; Rekount gives the data-derived count as 3,848,642 and shows differing model answers. The homepage also says its nine-industry census found 34% of answers agreed on by all three agents were still wrong when checked against the data. These are site-reported results, not independently validated in the supplied extract.

Observed signals

Read the marks

Editorial observations of this landing page, not a rating.

○OpenPublic substance visible

$PaidCommerce or pricing visible

✦PrettyNotable craft visible

●ProPolished or operationally mature

◎NicheUnusually specific use

One card from the complete issue

Tattoo Roulette Enters the Museum

323,487 arrived 1,000 judged 976 catalogued Enter the complete issue