Extract Dimensions Catalog (Locked)
Four locked audit dimensions. Never invent D5. Every finding cites its dimension ID.
D1. Incremental Refresh Key
Signature. Extract is configured as Full refresh, but the underlying data source has a monotonically increasing datetime or bigint field that would support incremental refresh.
Detection.
- The extract's current refresh mode (Full vs. Incremental) is not independently confirmable from this tool surface — no listed tool reads that configuration. Do not claim to have checked it.
- Instead, scan fields for incremental-refresh-key candidates regardless of current mode: datatypes datetime, date, or bigint; name patterns like
_at,updated_at,created_at,id,sequence,event_time. - Sample via
get-summary-dataMIN + MAX to confirm monotonic increase. - Report the finding as "candidate key identified" and note explicitly that the user should verify the extract's current refresh mode via Data → [datasource] → Extract → Edit before applying, since this skill cannot confirm whether it's already configured.
Severity. Warning if a strong candidate exists. Note if candidates exist but coverage across the fact table is unclear.
Fix path. Data → [datasource] → Extract → Edit → All rows / Incremental refresh → New rows identified by [field].
Impact. Estimate refresh-time reduction as (1 - new-rows / total-rows) percentage.
D2. Aggregation
Signature. Extract is row-level, but every sheet aggregates on the same small set of dimensions (typically 2-4 dimensions plus a date grain).
Detection.
- Enumerate every sheet's dimension shelf.
- Identify the intersection: dimensions used on every sheet.
- If intersection is small (2-4 dimensions) and consistent, aggregation-on-extract is a candidate.
Severity. Note (aggregation trades granularity for size; may not fit every use case).
Fix path. Data → [datasource] → Extract → Edit → Aggregation → Aggregate data for visible dimensions.
Impact. Estimate row reduction as (row-count post-aggregation / row-count pre-aggregation). Sample via get-summary-data COUNTD on the aggregation dimensions.
D3. Hidden-Field Pruning
Signature. Extract contains fields never referenced by any sheet's shelves, encodings, filters, or tooltips.
Detection.
- Enumerate all fields in the data source.
- Enumerate all field placements across all sheets (shelves, encodings, tooltips, filters).
- Fields in the data source but not in any placement are pruning candidates.
Severity. Warning if >20% of total fields are candidates. Note if 5-20%. Info if <5%.
Fix path. Data → [datasource] → Extract → Edit → Hide All Unused Fields (or right-click each field → Hide).
Impact. Estimate file-size reduction as (unused-field-count / total-field-count) percentage. Not exact (some fields are wider than others).
D4. Filter-at-Extract
Signature. A filter used on every sheet with the same value or range. Currently applied at query time on each sheet rather than at extract-creation time.
Detection.
- Enumerate every filter on every sheet.
- Group filters by field. If a field is filtered on every sheet with the same value/range, it's a candidate.
Severity. Warning if row-reduction estimate >30%. Note if 10-30%. Info if <10%.
Fix path. Data → [datasource] → Extract → Edit → Add [field] filter with the same value/range currently applied per sheet.
Impact. Estimate row reduction via get-summary-data count with vs. without the filter. Report the percent reduction.
Refuse to Fabricate
If an optimization idea doesn't fit D1-D4 (e.g., custom SQL rewrites, materialized views, source-side denormalization), report "outside current audit scope; consult source-side optimization docs" and.