1. Paste a ledger CSV
Compatible with the 29-column AI Answer Evidence Observation Ledger v1.0.
Quoted commas and line breaks are supported. Blank trailing rows are ignored. Validation does not judge the truth or quality of an answer.
Paste an AI-answer evidence ledger CSV to check its structure, controlled values, URLs, citation rules, duplicate IDs, and shared observation context—without uploading the file.
Compatible with the 29-column AI Answer Evidence Observation Ledger v1.0.
Quoted commas and line breaks are supported. Blank trailing rows are ignored. Validation does not judge the truth or quality of an answer.
Errors block structural use; warnings require a human decision.
Validation is deliberately split from measurement. It tests whether the file can support a reproducible review process; it does not turn one answer snapshot into a rank, score source authority, or infer why a system selected a citation.
All 29 headings, required values, booleans, timestamps, URLs, and positive citation positions.
Source roles, publisher relationships, verification states, review states, and taxonomy version.
A cited row needs an observed HTTP(S) URL and position; uncited rows should not carry citation-only values.
Rows sharing one observation ID must agree on prompt, system, surface, timestamp, and market.
Duplicate citation IDs, repeated secondary roles, unresolved review states, and verification gaps.
Mixed synthetic and measured rows are flagged so examples do not silently enter reported totals.
Start with the blank ledger and codebook, annotate source use with the versioned source-role taxonomy, validate the CSV here, then resolve every error and review every warning before aggregation. Preserve the raw observed URL even when a separate normalized key or later verification URL is available.