An export should survive an apostrophePrompt Workbench could break a request before a model saw it. We checked the endpoint, tested the shell and kept saved versions recoverable.13 Sept 2026field-notespromptsreliability
The green tick was checking too littleOur JSON tool could pass an output while ignoring the constraint that mattered. We replaced the checker, then tested the failure cases.13 Sept 2026field-notesevaluationstructured-outputreliability
A map needs evidence, not more pinsWe cut back Horizon's unsupported claims, withdrew eight forecasts, and made the remaining predictions easier to challenge.12 Sept 2026field-noteshorizonevaluationcorrections
One broken prompt froze the siteA stray Markdown wrapper stopped publication in July. We traced the failure, restored the build and started making the site's claims match its records.12 Sept 2026field-notessite-maintenanceagentsreliability