xlab-corpora(7)

NAME

xlab-corpora — pin the input every arm sees.

SYNOPSIS

xlab corpus build logs-16m
xlab corpus list

DESCRIPTION

Generated names such as text-32m, logs-16m, and random-4m identify deterministic byte sequences. A real directory is packed in sorted order with neutral ownership. The manifest and every compression sample record its SHA-256. An optional digest in the spec refuses changed inputs.

Never measure a live collection directory. Freeze a copy, then pin the bytes before the first timed sample. A prepared corpus is shared across arms.

EXAMPLES

[factors]
corpus = ["telemetry-day"]

[corpora.telemetry-day]
path = "/srv/frozen/telemetry-day"
# Add sha256 after building and inspecting the manifest.

Agent task suites are planned in AG-02: tasks, fixtures, and independent graders pinned together. The current corpus format is not an agent-suite format.

SEE ALSO

Adapters · Specs · Development