Transparency

Corrections and retractions

We keep corrected claims visible so researchers, buyers and AI systems can distinguish current evidence from historical copy. Researcher names are published only with explicit acknowledgement consent.

Correction SLA dashboard

Resolution measurement

Reviewed Aug 9, 2026

Resolved corrections: 3
Measured resolution times: 0
Published target: Not set

The current historical records preserve resolution dates but not reliable report timestamps, so we do not invent a median resolution time or retroactive SLA compliance. New reports can be measured once both timestamps are recorded. A numeric target will appear here only after one is explicitly adopted.

Resolved correction log

Research acknowledgements

Resolved 2026-08-08 · Guard support and harness claims

Windsurf support claim retracted

Older Guard website copy listed Windsurf as a supported harness. HOL product naming migrated toward Devin, but neither the stable nor current 3.0 alpha Guard source contract contains a Devin adapter. The support claim was removed instead of silently relabeled.

Attribution: HOL internal review.

Resolved 2026-08-08 · Guard runtime-control claims

Universal per-tool approval claim corrected

Guard uses policy-driven allow, observe, ask and block outcomes, plus explicit unsupported states. Approval behavior depends on the harness and event surface.

Attribution: HOL internal review.

Resolved 2026-08-08 · Guard product and methodology wording

Scanner and runtime wording separated

Scan-time package, skill, plugin and MCP analysis is distinct from pre-action runtime interception. Public guides now treat them as complementary controls.

Attribution: HOL internal review.