Ideas · Builder Notes
Hardening Kognita 0.4 after an external cross-review
What an independent cross-review of Kognita alpha found in identity, evidence integrity, replay and resume, and how each finding was fixed with a test.
Before releasing 0.4 I wanted independent reviewers to attack it. They found real gaps.
Each finding got a fix and a named test. The principal now always comes from the bound client configuration. SQLite triggers refuse updates and deletes on evidence rows, and a forward-only head checkpoint detects tail truncation. An empty use-case register denies. Held calls re-read policies that came into force during the hold, using the holding pack's context.
If the pack context cannot be restored on resume, every policy that came into force during the hold escalates and nothing is released. 0.4 shipped with these findings already closed.
The triggers are SQLite-specific, so other databases rely on verification failing closed.
If you're building one
Ask for an adversarial review before a release, and publish the finding, fix and test side by side.
Sources
Kognita is an open-source alpha. This note comes from Kognita PR #46, Kognita PR #47, Kognita PR #48, Kognita PR #49, Kognita PR #50 in the Kognita repository.