Cloudflare Workers Observability API Has Four Silent Traps That Skew Log Data

A developer running an independent AI monitoring project on Cloudflare Workers discovered four significant pitfalls in the platform's observability API while debugging 500 errors. The most misleading issue is that a worker's 'outcome' field logs 'ok' even when it returns an HTTP 500, because the field reflects runtime success, not application-level errors. A second critical trap involves querying large time windows, which silently returns only a fraction of actual events — in one test, a 24-hour query returned just 10% of the records that appeared when the same period was queried in smaller 4-hour slices. Two additional traps relate to filter operations: 'exists' matches fields that are present but empty, and 'includes' performs case-insensitive matching, both of which can produce confidently incorrect counts. The developer recommends slicing queries into narrow time windows, checking for results hitting the limit cap, and filtering on response status or error metadata rather than the outcome field.