Completed display-reading trial — reproduction

Extract this archive to a fresh folder. To verify the preserved inputs and recompute every reported score with Python standard-library code, run:

    python3 score.py --verify

This does not call a model or generate new responses. The18 PNGs are the exact images shown to the three readers; all original answers are preserved. freeze.json records the input hashes fixed before responses.

generate.py is the unchanged, hash-frozen production script, not the verification command. It contains the original temporary output directory and macOS Arial font path. Do not run it to verify this completed trial. To make a different future trial, copy and adapt that script in a separate folder, freeze a new protocol and obtain new responses; do not combine regenerated stimuli with the old answers or replace this trial's freeze file.

The observed5/6 result in each display condition is a null for accuracy uplift in this bounded trial. It is not an equivalence test, human result, timing result or general display ranking.
