Changelog: 2026-07-05

Blog and changelog dates now render consistently across time zones, and we published a new deep-dive on red/green-flag LLM evaluation.

Highlights

• Fixed blog and changelog date displays to prevent “previous day” shifts for readers in certain time zones.
• Published a new post on red/green-flag LLM evaluation, comparing multiple models against a deterministic baseline.
• Added automatic detection for “cold open” network states to improve startup handling.
• Introduced a new console approvals flow to support smoother operational workflows.
• Shipped a set of updates to our staging workflow in preparation for upcoming releases.

Details

• Consistent date rendering: Dates shown across the site (including blog listings, post pages, changelog surfaces, and share/preview assets) are now normalized to UTC when interpreting bare YYYY-MM-DD values, eliminating off-by-one-day displays.
• New research content: A new article explores a red/green-flag extraction benchmark over SEC 8‑K filings, with side-by-side comparisons across several LLMs and a simple keyword-based baseline for reference.
• Improved startup resilience: The product can now detect a specific “cold open” network condition and respond more gracefully during initialization.
• Approvals workflow: Added a dedicated console approvals flow/handling to better support review and authorization steps within operational tooling.
• Staging updates: Continued work on staging-related behavior to streamline how changes are validated before reaching production.