Chemistry trials · v1

Can Narrative Chemistry predict, generate or distinguish films that the ordinary Cineosis taxonomy cannot? The ontology is frozen as v1. Four predictions were fixed before any number was run. One instrument reads every film, and seven films per poem are synthesised blind to the four WYGWYL cuts. Those cuts were made by cutbastard/plan.py, one CLIP-scored rule per cut, not by a person, so what follows measures agreement with plan.py’s strategies, not with human editing.

Verdicts as registered in chemistry/trials.py

How near each film comes to the plan.py cuts mean over 14 poems × 4 plan.py cuts · lower is nearer

Do the plan.py cuts bond the way v1 says? adjacent pairs in each plan.py cut against 2,000 shuffles of its own shots, z-score

Blue: fewer than chance · red: more than chance · gray: as chance. ▲ or ▼ marks p < 0.05. “Same” is two shots of the same sign in a row; v1 has no bond for it and files it under rhyme.

Spectra every film of a poem, read by the same instrument

Blind screening room eight films, one of them the suite cut from plan.py; their names stay hidden until you have rated all eight

choose a film

Each film plays under her voice for the whole poem. Rate each one from 1 to 5 on three questions, and mark the one you think is the suite cut. Reveal is enabled once all eight are rated.