Our whole thesis is that good companies get hurt by what nobody is allowed to look at. We don't get to sell that idea and then hide our own incident history. So this page is exactly what it looks like: the mistakes we've made running this firm, written up the way we'd write up a finding for a client — what happened, what it cost or could have cost, how it was caught, and which control changed because of it.
One more thing you should know: this firm is built, in part, by AI agents working alongside the founder — the same class of tools we assess. Some of the entries below are textbook agentic failure modes: automation acting on stale state, parallel sessions overwriting each other, a standing rule that existed in someone's head instead of in the system. We could hide that. Instead we treat every one as field data — we sell the audit of exactly these failures, and we generate our own specimens for free.
Some of these were caught by pointing our own methodology at ourselves. That's the point. The motto on our front page — treat every system like the cameras are on — applies to us first.
Rules of this page: entries are permanent — we never delete one, we only update its status. We publish our own operational and process failures here. Anything touching client work stays under NDA and never appears; if we ever mess up with a client, they hear it from us directly, first.
The log · newest first
⚠️ SERIOUS
Fixed
Shipped July 2026 · caught August 8, 2026 · fixed August 9, 2026
We shipped a "share your story" feature that quietly captured nothing
What happenedWe launched a field-story submission box before the backend for it existed. For visitors without a configured mail client — very common on desktop — the Submit button silently did nothing. For everyone else, it opened their email app, which shares their email address with us whether or not they wanted that.
ImpactAn unknown number of practitioners may have tried to send us a story and believed they had, or walked away when nothing happened. Stories are the most valuable thing anyone can give us. We built a door that didn't open.
How it was caughtA pre-campaign review of our own site — before asking anyone to visit it — traced the submit path end to end instead of assuming it worked.
What changedA real submission endpoint went live August 9, 2026, and this status changed only after it was tested end to end. Worth admitting: the first test submission promptly landed in the form processor's spam folder — a second silent-failure mode, caught only because we tested instead of assumed. The flow was reworked until a test story landed in the inbox with notifications firing. New rule, now twice-earned: no call-to-action ships until someone has completed it end to end as a stranger would.
⛔ CRITICAL
Fixed
August 9, 2026
We broke our only conversion path with our own cleanup
What happenedWe renamed our scheduling-link handle to remove the founder's personal name from the URL — a good hygiene change. The old link died instantly, with no redirect, while it was still the link wired into the live site. For a window of time, "Schedule a consultation" — the one action this entire site exists to drive — led to a dead page.
ImpactAny visitor who clicked through during that window hit a dead end at the exact moment they'd decided to talk to us. We'll never know if that number was zero.
How it was caughtSame-day review of the change caught the dead link and the site was patched within hours.
What changedNew rule: any rename or URL change triggers an immediate inventory of every place the old link lives — site, documents, outreach templates — before the old one is allowed to die. Renames are changes; changes get change control. We tell clients exactly this about their ERP configuration, and we got to relearn it on a Saturday.
⚠️ SERIOUS
Fixed
August 9, 2026
We violated our own account-hygiene rule — twice
What happenedWe have a standing internal rule: every account, service, and registration this firm uses runs on corporate email — never personal. Then we launched with the founder's personal email address hardcoded into the site's page source as the contact address, and registered our analytics account under that same personal address.
ImpactA firm selling controls credibility had a personal inbox wired into its trust surface, visible to anyone who viewed source. Beyond the credibility problem, it undermined the separation between the founder's personal accounts and the firm's — the exact kind of identity/attribution sloppiness we flag in client environments.
How it was caughtWe ran our own pressure-test methodology against our own live site — the same kind of review we sell. It flagged both instances the first time through.
What changedBoth were migrated to corporate accounts the same day. The rule was upgraded from "use corporate email going forward" to an audit obligation: every existing service gets checked against the rule, and any new session of work that finds a violation is required to flag and migrate it. A policy you don't audit is a paper control — we'd have written that finding for anyone else.
⚠️ SERIOUS
Fixed
August 9, 2026
We deployed a stale copy of our own site over a newer release
What happenedAn AI-assisted work session, operating from an outdated local copy, published an older version of this site's main page over a newer release that was already live — silently reverting improvements that had shipped hours earlier. Days later, a second AI session nearly did the exact same thing from another stale copy, and was caught mid-deploy. This is agentic risk in miniature: automation confidently acting on stale state, and parallel sessions with no shared change control.
ImpactThe live product regressed, and nothing in our process at the time would have told us. Visitors saw an older, worse version of the site until the discrepancy was noticed and the newer release was restored by merge.
How it was caughtThe first incident was found after the fact by comparing the live site against what should have shipped. The near-repeat was caught in the act — because by then the control below existed.
What changedOne repository is now canonical, and the rule is mechanical: fetch and diff against it immediately before every deploy, not just at the start of a work session. This is a change-management failure — the same class of finding we write for clients about ERP customizations. Multiple hands touching one system without a single source of truth will eventually overwrite each other. Including ours.
The older, bigger one
Before this firm existed, its founder lost $250,000 in a company that had everything except a willingness to let someone look at the financials. That story — and how it became this firm's founding principle — is on the founder page.
Entries describe our own internal operations and are published voluntarily. This page is informational; it is not an audit, an attestation, or legal advice, and no engagement or duty of care is created.