> ## Documentation Index > Fetch the complete documentation index at: https://docs.dqlabs.ai/llms.txt > Use this file to discover all available pages before exploring further. # Overview > What Pattern metrics are in Prizm — auto-discover the structural format template of column values to detect format violations and drift. ## What is a Pattern metric? Pattern metrics identify the **structural format template of values in a column** — auto-discovered from actual data. They detect format violations, encoding changes, and structural drift at the value-format level, not just the data-type level. **Type:** Configurable — Pattern metrics are part of the Structural (Advanced) profile. They are enabled per column through the Configure Structural Metrics panel. Pattern metrics answer the question: *Do the values in this column conform to their expected format?* ### Metrics | Metric | Level | Description | | ------------- | --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Short Pattern | Attribute | A compressed representation of the value format where each character class is collapsed to a single token (e.g., `A9A-9999` for a product code like `B2B-1234`). Short patterns capture the essential format shape — ideal for detecting format class changes across a column. | | Long Pattern | Attribute | A character-by-character format representation where each position is mapped to its character class (e.g., `ANNN-NNNN` where `A` = alpha, `N` = numeric). Long patterns provide precise positional matching — useful for fixed-format fields like national IDs, account numbers, and date strings. | ### How pattern monitoring works On the first profiling run, Prizm auto-discovers the dominant short and long patterns in the column and establishes them as the baseline. On subsequent runs, it reports the distribution of patterns observed — including any new patterns not present at baseline. An alert fires when a new pattern appears or when the proportion of values matching the baseline pattern drops below the configured tolerance. Patterns are detected automatically — no user-defined regex is required. For regex-based format validation, see the [Frequency — Regular Expressions](/architecture/metrics/frequency/overview) metric. ### Example use cases **Product SKU format change:** A supplier changes their product code format from `ABC-1234` (Short Pattern: `AAA-9999`) to `AB-12345` (Short Pattern: `AA-99999`). Prizm detects the new pattern immediately and alerts — giving the integration team time to update their SKU parsing logic before the catalog breaks. **Date format inconsistency:** A CSV ingestion pipeline begins mixing ISO date format (`2024-01-15`, Long Pattern: `NNNN-NN-NN`) with US date format (`01/15/2024`, Long Pattern: `NN/NN/NNNN`) in the same `event_date` column. Prizm alerts on the new Long Pattern, flagging the inconsistency before it causes silent type cast failures in downstream models. Pattern metrics require at least one completed profiling run to establish a baseline. New patterns appearing in subsequent runs are flagged immediately — they do not silently replace the baseline. To update the baseline after an intentional format change, acknowledge the alert in Prizm. Enable Pattern metrics per column, set thresholds, and manage baseline updates.