Limited information ceiling: finite checks (companion to the note "A limited information ceiling for choosing among fixed candidates", 7 October 2026 Pacific, version 1)

Requirements: Python 3 standard library only. No network, no model calls, no data files.

Fast regression tests (a fraction of a second):
  python3 -m unittest test_note_checks

Complete enumeration (a few minutes; deterministic; prints the lines saved in check_output.txt):
  python3 check_ceiling.py
Small exact examples:
  python3 check_small_examples.py

What the files are:
  check_ceiling.py         exhaustive integer-weight tables with exact arithmetic for the ceiling V*, the coverage C and the garbling and conditional-independence statements; a float check of the information bound with a 1e-12 slack; the ternary counterexample; the exact binary additive-gap search.
  check_small_examples.py  the extra-signal and abstention examples of the note, in exact fractions.
  test_note_checks.py      recomputes the printed counterexample tables and a small exhaustive subset.
  check_output.txt, check_small_examples_output.txt  the author's outputs.

Limits: the models have two candidate labels and small signal alphabets on a grid. The checks are evidence, not proof. The ternary additive search uses an integer score grid and gives a lower bound on the best additive value. The 24-task verifier study is unrun and nothing here is a model result.
