xlab-spec(5)

NAME

xlab-spec — declare the comparison before measuring it.

SYNOPSIS

id = "S-example"
question = "Does the candidate reduce time?"
hypothesis = "The candidate is at least 5% faster"
adapter = "simulated"
kind = "screening"
primary = "time_s"
control = "base"
blocks = 12
delta = 0.05

[options]
sd = 0.04

[arms.base]
mean = 1.0

[arms.candidate]
effect = 0.9

DESCRIPTION

An arm is a configuration. Factors cross into cells. A block measures every arm once in a cell, in a seeded order. blocks is the replicate count per cell; the plan interleaves cells. delta is the smallest meaningful ratio-scale effect.

screening produces leads. confirmatory adjusts primary comparisons for multiplicity. characterization reports absolutes and Pareto fronts without requiring a control. pareto names metric pairs to inspect together.

The current analysis requires positive primary values; pass/fail and bounded agent scores need the planned agent statistics. Do not substitute zero for a void.

EXAMPLES

xlab check experiments/S-example.toml
xlab preflight experiments/S-example.toml
xlab plan experiments/S-example.toml --seed 1

See the repository's xlab/examples/ for grids, factors, compression, and manual specs.

SEE ALSO

Adapters · Analysis · Budgets