Pulse Documentation

name: op-synth-regex
description: String samples generated from a Perl/RE2 regex pattern; walks regexp/syntax AST with bounded repetition.
kind: operator
category: SYNTH
operator: regex
type: reference
applies_to: inspect, predict, manifest
examples_tags: [synth, data-quality]

Synth distributions emit per-row values; no Response.Components.

Use when

Generates text matching a pattern, such as order codes like AB-1234, for realistic-looking IDs and labels.

Questions it answers:

• How do I generate postcodes that look real?
• How do I fill a SKU column with codes like SKU-00042?

Use something else:

• weighted_categorical when the values come from a short known list.
• monotonic_from when IDs must be unique.
• constant when every row should carry the same text.

Params

• pattern — string, required, non-empty. Perl/RE2 source via regexp/syntax.Parse.
• max_repeat — int, default 8, floored at 1. Bounds unbounded * / + / {m,}.

Inputs

Field type: — categorical_u8/u16/u32. Distinct generated strings become dictionary entries.

Output

Per-row string from the parsed AST: literals copy, char classes draw uniformly, alternations pick a branch, repeats expand under max_repeat. Stored as a dictionary index. Ops: OpLiteral, OpCharClass, OpAnyChar(NotNL), OpCapture, OpConcat, OpAlternate, OpStar, OpPlus, OpQuest, OpRepeat, anchors (emit nothing).

Gotchas

• No backreferences (\1, (?P=name)); invalid pattern → SERVICE_VALIDATION with the syntax error.
• OpAnyChar emits printable ASCII (33–126), for cross-platform stability.
• Unbounded * / + clamps to max_repeat; bounded {m,n} to min(n, min + max_repeat).
• An empty character class produces no character. Write-time dictionary cardinality must fit the declared categorical width — wide patterns + tall cohorts overflow categorical_u8.

See

• pulse_examples_search tags=[synth]
• Skills: synthetic-data, op-synth-weighted-categorical, op-synth-constant