# Evidence methodology

CCFP separates three questions that are often collapsed:

1. Can a product-data master express a fact at the correct grain?
2. Does a particular destination request, accept, or require a representation of that fact?
3. Is the live fact accurate, current, and consistent with the experience a person receives?

Passing one question never proves the others, and none proves ranking, recommendation, distribution, conversion, or growth.

## Evidence workflow

1. Capture the current authoritative platform source and retrieval date.
2. Record the exact platform field, scope, conditionality, and destination context.
3. Map it to a canonical concept without erasing differences in grain or semantics.
4. Assign an evidence status: `DIRECT`, `DERIVED`, `INFERRED`, `EMERGING`, or `UNKNOWN`.
5. Add deterministic tests only where an observable input and expected outcome can be stated.
6. Keep historical results bound to the model, mapping, source, and rule versions used at evaluation time.

## Requirement vocabulary

- `required`: the captured destination source says the field is required in the stated scope;
- `conditional`: required only for stated products, categories, markets, programs, or conditions;
- `recommended`: encouraged but not asserted as a universal eligibility requirement;
- `supported`: represented by the source platform, without destination requirement implications;
- `contextual`: meaningful only with additional market, category, seller, or program context;
- `unknown`: CCFP does not have sufficient current evidence.

## Uncertainty

Unknown is a result, not a defect to hide. CCFP does not convert missing evidence into a confident score. Corrections should update the evidence registry first, then the mapping and rules that depend on it.

## Privacy boundary

The public project contains schemas, field definitions, mappings, rules, methodology, and synthetic examples. Merchant feeds, validation inputs, findings, affiliation, connected systems, and operational states are private by default. Publication requires a separate, explicit choice by the data owner.