Pulse Documentation

name: op-synth-normal
description: Gaussian samples with optional [min, max] clamp; default mean=0, std=1.
kind: operator
category: SYNTH
operator: normal
type: reference
applies_to: inspect, predict, manifest
examples_tags: [synth, distribution-shape]

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

Use when

Draws numbers from a bell curve around a chosen centre and spread, optionally clipped to a low and a high bound.

Questions it answers:

• How do I generate realistic test scores that cluster around 70?
• How do I make a numeric column with a known average and spread?

Use something else:

• lognormal when values are always positive with a long right tail, like income.
• mixture when the data has two or more peaks.

Params

• mean — float, default 0.0.
• std — float, default 1.0; must be > 0.
• min / max — float, default -inf / +inf. Setting EITHER flips an internal clamped flag; once on, BOTH bounds apply.

Inputs

Field type: — numeric u4/u8/u16/u32/u64, f32/f64, decimal128, date (days-since-epoch); u4 writes one byte per row. Categorical / packed_bool unsupported — use weighted_categorical or bernoulli.

Output

Per-row float64 from N(mean, std²), cast/clamped to the declared type at write time; days-since-epoch on a date field.

Gotchas

• Heavy clamping distorts pairwise correlations — |rho|_actual < |rho|_requested when a partner field is clamp-narrow.
• std <= 0 → SERVICE_VALIDATION at spec parse.
• Nullable fields draw the inner value first, then the null mask, so the seeded stream is invariant to which rows end up null.
• Same (spec, opts.Seed) ⇒ byte-identical output. Seed == 0 is stable, not random.

See

• pulse_examples_search tags=[synth]
• Skills: synthetic-data, op-synth-lognormal, op-synth-uniform