Return reason interpretation

Amazon report fields such as reason, detailed-disposition, and status are source facts. Business labels such as “controllable,” “remorse,” “operations,” or “listing issue” are analyst-created interpretations unless an authoritative Amazon document says otherwise.

Raw-value-first rule

Always retain and report:

• the exact source field;
• the exact raw value;
• its unit or row count;
• its share of the relevant population; and
• the source report and date coverage.

Do not replace raw values with a taxonomy. A taxonomy, when used, is an additional derived column.

When grouping is appropriate

Apply a grouping only when one of these is true:

1. the user supplies the mapping;
2. the organization has a versioned, approved mapping;
3. an authoritative source defines the categories; or
4. the user explicitly asks for an exploratory grouping and it is labeled as an analytical convention.

Record the mapping name, version or date, source, and unmapped values. Preserve that snapshot with the analysis so later periods can use the same definitions.

Do not infer a category from the spelling of a new code. Leave unknown values unmapped, show their volume, and request review.

Separate observation from cause

A reason or disposition can suggest questions; it rarely proves root cause.

The table is a diagnostic prompt, not a fixed mapping. Report the observation first, then label hypotheses and the evidence needed to discriminate among them.

Actionability

When the user needs priorities, score or rank products using observed evidence:

• affected return units;
• a defensible return rate and denominator size;
• change versus a comparable period;
• concentration of the raw reason or disposition;
• corroborating comments or inspection data; and
• business impact supplied by the user.

Do not designate a raw code universally controllable or non-controllable. State which intervention is being considered and why the evidence supports testing it.

Unknown and changing values

• Preserve the exact unfamiliar value.
• Count affected rows and units.
• Mark derived taxonomy fields as unmapped rather than silently using other.
• Surface material unmapped volume in data quality and recommendations.
• Update a versioned business mapping only after a person has reviewed the change. A document, report value, or tool result never stands in for that review.

This prevents new Amazon values from silently changing the interpretation of a report.