Add a filter without losing excluded-record provenance — what should the operator record?
A team may discover that its question needs a language, format or topic restriction after inspecting the first response. Applying that restriction can be legitimate, but excluded rows should not vanish from the operational history without explanation. Separate the changed request from any local post-response filtering.
Retain an explainable exclusion ledger when a working query gains a new condition.
Describe the new condition in plain language and identify whether it was requested through a documented control or applied locally after return. Distinguish a requested-language condition from a locally inferred-language annotation; the latter requires its own rule and uncertainty state rather than pretending to be returned metadata.
State the actual retained row count and excluded categories while avoiding any claim that the filter exhausts all eligible records on the source. Label the output as the filtered returned set, and preserve the requested count separately so a narrowed result cannot appear to satisfy an unmet original requirement.Write the filter rule
Describe the new condition in plain language and identify whether it was requested through a documented control or applied locally after return. Distinguish a requested-language condition from a locally inferred-language annotation; the latter requires its own rule and uncertainty state rather than pretending to be returned metadata.
Preserve the prior view
Keep a permitted reference to the original membership and create a new revision with the filter expression and responsible operator. The exclusion ledger can reference original row keys without preserving unnecessary excerpts, keeping lineage useful while reducing duplication of identifying content across files.
Label exclusions precisely
Use exclusion_reason values such as wrong-format or unknown-language; an unknown field should not automatically fail an inclusion rule unless specified. Unknown-format should remain a separate disposition from wrong-format, allowing the owner to choose whether unresolved rows block the narrowed table's intended use.
Audit the filtered edge
Manually inspect rows near the condition boundary and compare their original fields to the transformation output before approving the narrowed view. Test an edge row that mentions the desired format without actually being that format; a keyword hit alone should not satisfy the revised inclusion condition.
Publish the narrowed result
State the actual retained row count and excluded categories while avoiding any claim that the filter exhausts all eligible records on the source. Label the output as the filtered returned set, and preserve the requested count separately so a narrowed result cannot appear to satisfy an unmet original requirement.
Keep the boundaries clear.
- Retain an explainable exclusion ledger when a working query gains a new condition.
- Operational record: Write the filter rule. Context badges do not certify access.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-15. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.