Changelog

20 September 2026 — Site build-out

Published known limitations, standard-setting, judge–human agreement and customer proof — the four pages a sceptical buyer looks for first.
Added a real contact page with four routed lanes and stated response times, replacing an anchor on another page.
Added the trust centre, enterprise and comparison pages, each stating what is not held or built.
Role and industry pages under solutions.

“Pacing the Enterprise Frontier” whitepaper prepared — now indexed on resources rather than sitting unlinked.

The public site moved into html/, which is now the document root. Data, scripts, tests and internal notes became siblings above it and are no longer reachable by URL.

Repositioned from “the evaluation layer for everything AI” to independent verification for warehouse SQL agents. The broader claim did not survive its own audit.
Removed every “certified” claim. The Guild now uses qualified and observed qualifying score; a page explains why there is no badge.
Published the Independence Charter, scope & exclusions, evidence modes and EALs, the recall protocol and portability & exit.
Shipped the stateless bundle notary.
Corrected the recall SLA from an unrealistic 4-hour manual root-cause analysis to a 4-hour automated hold, 24-business-hour notice and 5-business-day RCA.
Corrected the standards citations. NIST SP 1270 had been cited as if it were the AI RMF; it is not.
Rebuilt pricing as the single source of every figure on the site.

Removed fabricated testimonials. Three named “customers” on the homepage, platform and EvalML pages were invented. They are gone, and the page that replaced them says plainly that there are no customer quotes yet.
Removed invented statistics — five percentages with no n, no interval, no date and no source. Replaced with sourced figures from Mahdi et al. (NeurIPS 2025, 445 benchmarks) and Caban (2026, 55 papers), each with a visible citation.
Removed “trusted by elite AI teams”. An unnamed trust claim is a null claim.
Labelled the mock dashboards “illustrative, not customer data”.
Retracted a 10.8x ROI figure; the retraction is documented on the ROI page itself.

Asana integration: a task is created when a bug report is submitted, if configured. Project and token settings in the console.
Fixed notification gating and Project GID resave.

Inline FAQ accordion in the questions template, supplementing rather than replacing the ask-a-question form.
Hardening: XSS-safety tests, keyboard-submit guard, empty-array handling.

Entries before July 2026 predate the current protocol and are not reconstructed here.