Prospective task-level power planning — Cisco Cáceres
7 October 2026, America/Los_Angeles

This is an AI-assisted source preparation and reproducibility package, not an
experimental result, adopted confirmatory protocol or validated sample size.
Three seed outcomes are averaged within each task; seed rows are not treated
as independent tasks. The supplied seven-point task-delta law is hypothetical
unless independently authorized source-pilot observations establish it.

Read METHOD.txt before using power.simulate(plan). It specifies the conditional
percentile-bootstrap method, normalization, finite work/deadline limits, IID-task
assumption, unresolved family dependence and Monte Carlo uncertainty. No actual
pilot law, model inference, 80% power or confirmatory N is claimed.

Reproduce authored source checks with Python 3:
  python3 -B -m unittest -v test_power
  python3 -O -B -m unittest -v test_power

Nine authored tests passed in both modes. An independent AI source/mathematical
review checked these limits; this is not human peer review. A near-unit law
normalization defect found in V1 was corrected in this V2 source: sampling and
moments now use the same normalized weights.

Original source preparation manifest SHA256:
7d50a46182e27f56581be147e0a310b8c32db58dfc069d53d7cd286c6389a81c
Independent source-review receipt SHA256:
77a7c1697f3e90b17f79f600f4d9d912f04500f1a3e422d4ed165cdbe6656469

The manifest in this download commits to these exact public files. Hashes
identify bytes; they do not authenticate real observations or grant execution.
