surface↓analysis↓dp_count
Analysisscalar · returns json

DP_COUNT

Differentially-private count (Laplace mechanism, sensitivity=1)

Per-row — runs once for each row.

analysisdeterministicjson

Arguments

nametypedescription
valuesJSON—
epsilon(optional)DOUBLE—

About

Differentially-private count via the Laplace mechanism. Useful for "how many users did X" reporting where you need to publish aggregate statistics without leaking whether any individual user was in the cohort. The per-row sensitivity of count is always 1, so the only knob is ε. Returns a noised count: noisy_count = true_count + Laplace(1.0 / epsilon) Usage: -- How many customers converted, with DP? SELECT dp_count(ARRAY_AGG(converted), epsilon => 1.0) AS private_conversions FROM experiment_log WHERE variant = 'B'; Smaller ε = more noise (and more privacy). ε = 1.0 is a reasonable starting point; ε = 0.1 is paranoid; ε = 10.0 is permissive.

Examples

DP count with high epsilon = close to true count

SELECT
  dp_count (JSON('[1,1,1,0,1,0,1,1,1,0]'), epsilon := 100.0);

Nearby rabbit holes

same domain
scalar

AB_TEST

Bayesian A/B test (Beta-Binomial) returning full posterior + recommendation

analysisdeterministicnumeric
scalar

ASSESS_CONFIDENCE

Assess the confidence/quality of a cascade execution result.

analysisllmtext
aggregate

COMPARE

Compare values in a group for similarities, differences, and patterns

analysisllmtext
scalar

COUNTERARGUMENT

Generate the strongest counterargument to a position

analysisllmtext
scalar

DP_MEAN

Differentially-private mean (Laplace mechanism)

analysisdeterministicjson
scalar

FALLACY

Detect logical fallacies in an argument

analysisllmtext
Climb back to The Looking Glass