# AI Answer Evidence Observation Ledger

> A general-purpose, versioned ledger for preserving prompts, answer context, citation occurrences, source roles, verification states, and reviewer decisions before aggregation.

Version: 1.0  
Published: 2026-08-12  
Maintainer: [Corank](https://corank.ai/)  
Downloads: [blank CSV](https://corank-prompt-source-gap-triage.pages.dev/ai-answer-evidence-ledger/template.csv), [synthetic example](https://corank-prompt-source-gap-triage.pages.dev/ai-answer-evidence-ledger/synthetic-example.csv), [JSON Schema](https://corank-prompt-source-gap-triage.pages.dev/ai-answer-evidence-ledger/schema.json)

One ledger row is a citation occurrence within one dated AI-answer observation. It is not a stable ranking. The template does not query an AI system, judge accuracy, infer causation, or claim that a source influenced an answer.

## Workflow

1. Freeze prompt IDs and versions before collection.
2. Record the answer product, surface, timestamp, market, and privacy-safe evidence context.
3. Preserve each citation's exact observed URL, link label, position, and associated claim.
4. Apply the versioned [source-role taxonomy](https://corank-prompt-source-gap-triage.pages.dev/ai-answer-source-role-taxonomy/index.md).
5. Record URL verification in separate fields without overwriting raw evidence.
6. Validate, review, and report denominators before aggregation.

## Core role values

Primary and secondary roles use `entity`, `definition`, `evidence`, `comparison`, `implementation`, `validation`, and `action`. Publisher relationship uses `owned`, `partner`, `independent`, `unknown`, or `not_applicable`. Review status uses `reviewed`, `needs_review`, or `excluded`.

## Quality rules

- A cited row requires an absolute HTTP(S) observed URL and a positive citation position.
- Rows sharing one observation ID must agree on run context.
- Raw observed URLs never receive silent normalization.
- Synthetic fixtures must be labeled and excluded from measured totals.
- Minimize stored answer text and preserve only the evidence needed for review.
- Ambiguous rows remain `needs_review` rather than receiving a forced label.

The ledger cannot prove stable rank, model training inclusion, publisher authority, factual correctness, search demand, commercial impact, or the reason a source was selected. Report prompt-set version, answer systems, surfaces, markets, collection window, reviewers, missing data, and sample size with every summary.