Errors¶

When a source fails, gleanyard collect reports an error_kind so you can tell a changed page from a network problem without reading messages.

Kind	Meaning	Typical cause
`structure_changed`	The response is not the shape the reader expects.	A feed changed its keys, an endpoint now returns HTML, a calendar is malformed.
`blocked`	The host refused access.	`robots.txt` disallows the URL, or HTTP 401, 403 or 451, or HTTP 429 that persists or asks for a wait over two minutes (a shorter Retry-After is waited out and retried).
`transport`	The request failed.	A network error or another HTTP error status. Server errors (500, 502, 503, 504) are retried first.
`pagination_changed`	Paging no longer behaves as configured, so records may be missing.	The paging parameter is ignored, pages repeat, or the source grew past `pagination.max_pages`.
`filter_ignored`	The source ignored a filter and returned out-of-bounds data.	A `filter` in the source file finds too many records outside the allowed values; the URL filter stopped working.
`bounds_exceeded`	A configured limit was reached.	More requests, bytes or occurrences than the source's limits allow, or a request to a host the source does not list.
`empty`	The run succeeded but produced no records.	A broken mapping or page, unless the source is `expect_empty: seasonal`.
`model`	The language model failed or returned unusable output.	`llm` mode with no provider key, the optional extra not installed, a provider outage, or a reply that is not `{"items": [...]}`.
`rights`	The source's rights do not allow collection.	`review_status: rejected`, the tier `blocked`, `send_to_model` missing for `llm` mode, or (with `require_reviewed_permission: true`) `private_store` or `normalize` missing.
`no_cache`	`--from-cache` found no stored run.	The source has never been collected, or its rights did not allow storage.

Limits never truncate silently. A calendar with more occurrences than max_occurrences fails with bounds_exceeded rather than returning part of the data.