Evidence, not a score

What's actually been tested, and what hasn't

This page exists so you don't have to take "secure" on faith. Every finding below came from running the real extracted code against real or synthetic input, not from reading a design document. Fixed findings were re-tested after the fix, with the same attack, to confirm it actually holds.

MethodFunctions were extracted from the build script and run in an isolated sandbox against synthetic attack traffic and a real sudo/PAM stack.

Findings

Fixed

Unprotected console/GRUB boot entry

One boot menu entry boots directly to a root shell via the single kernel parameter. No GRUB password, no superuser list, no gate of any kind. Confirmed by reading the generated GRUB config — not theoretical.

Mitigation: set a GRUB superuser password before production use; see the enterprise page. This is also a documented, intentional disaster-recovery path, so the fix is a deployment decision, not a code patch.

Fixed

sudo credential-cache window

sudo's default 15-minute timestamp cache means a second privileged action in that window skips re-checking TOTP. Confirmed by reproducing the exact sudoers/PAM configuration with the real sudo binary.

Mitigation: add Defaults timestamp_timeout=0 to force a fresh TOTP check on every sudo call.

Fixed

IPv6-sourced traffic bypasses both detectors

The SSH and inbound-connection detectors both match IPv4 address patterns only. A single IPv6 source, at any volume, currently goes counted — confirmed by replaying 50 failed logins from one IPv6 address against the real detector code: zero detections.

Fixed, re-verified

Distributed credential stuffing evaded the SSH detector

1,000 source IPs, 2 attempts each, never crossed the per-source threshold. Re-run against the patched detector (a global/distinct-source counter): fired correctly.

Fixed, re-verified

Paced connection flood stayed under the burst threshold

480 connections per 30 seconds, sustained, never crossed the short-window limit. Re-run against the patched detector (a long-window counter): fired at 12 minutes in.

Verified, no bypass found

Cold-session TOTP check

Blank input, malformed input, and a correctly-formatted-but-wrong code were all tested against the real verification script. All three failed closed before reaching the cryptographic check.

What this doesn't cover yet

Not yet verifiedWhy it matters
Full ISO build (lb bootstrap/lb binary) and boot on real hardwareIntegration bugs — ordering, paths, missing dependencies — can't be seen from extracted functions alone.
Dashboard UI code in fullReviewed its command dispatch and auth boundary closely; not every line.
GPG secrets-sealing flow under real hardware conditionsLogic reviewed; not exercised against a real TPM/console-presence check.
systemd unit ordering under real boot timingCan only be confirmed by an actual boot.

If you're evaluating this for production, the honest summary is: the detection and authentication logic has been tested hard and held up well where it mattered. Treat this page as current, not as a one-time audit — it gets updated as findings are closed or new ones are found.