Proofmark · Blog

Receipts, mechanisms, and the case for named answers.

Field notes from Proofmark: how knowledge rots, what makes an answer safe to act on, and why a verified badge should name a person, a version, and a date.

Proofmark field notes · updated 2026-07
Reading order

Your knowledge base is lying to someone right now

Practitioners' verbatim words for wiki rot, and the measured data on what breaks when an assistant answers fluently from a stale page nobody stands behind.

2026-07 · 8 min readRead →

How to guarantee your AI quotes a dead policy

Five repeatable ways to make an assistant cite a document nobody stands behind. Each one struck out by a shipped mechanism.

2026-07 · 5 min readRead →

The auto-verified checkmark, examined

In January 2026 the incumbent let an AI flip the verification checkmark by default; Proofmark keeps the signature human: a named person, a specific version, a date.

2026-07 · 6 min readRead →

Security posture, candidly

Read-only connectors, two-phase authorization, a hash-chained audit log, and crypto-shred erasure: architecture you can inspect, with the gaps stated plainly.

2026-07 · 4 min readRead →