Putting the science back into the play

The first play explained a flaw in one research example while leaving much of the research’s actual news outside the scene. I asked whether it read like science, whether the folder was being used, and whether a score was really all the research had said.

The writing had a substantial preparation behind it. I had requested a durable script library, sequential numbers and craft research. A separate taxonomy organised language-model subjects, including social issues. A recorded random draw selected Computational methods, and the research dossier kept the same script identity as its eventual drafts.

The dossier retained source texts, admission decisions, findings and search records. It distinguished company promotion from independent research under the rule I had requested. Reported results stayed attributed to the researchers; no experiment had been rerun merely because its paper was read.

Choose the finding before the scene

D001, The Clean Copy, used a family recording and a smoothing control that removed a laugh. The family and outcome were invented. The technical inspiration came from an appendix discussion of an unsuitable output receiving a favourable evaluation.

The scene had a photograph, controls, a lost sound and a relationship. It did not explain enough science. A later version added comparisons and notes, then a fuller public demonstration with a cropped picture, counted sounds and a reveal of the whole sheet.

The multimedia became more exact while the scientific emphasis still needed correction, because staging one defect carefully could not make it the principal finding of a dossier that contained eleven admitted studies.

D004, Where Next?, moved to two related research contributions. In the retained AgentDiscover account, an AI coding agent gained more choice over which earlier trials to inspect and what program to try next, while a separate checker tested the result. Another retained study concerned training aimed at finding proofs within a fixed search budget.

The opening dialogue made the question ordinary:

NELL
Texas A&M researchers gave AI another job: choosing what to try next.

LEON
I thought that was the job.

NELL
Often, people fix the search rules beforehand.

Public source excerpt. Repository snapshot, requiring repository access.

This excerpt is from the written D004 script. It establishes what the characters were meant to explain, not proof that the audience understood it or that the play had been performed. The paper’s researchers remain the people who conducted the work.

Make explanation change the action

Nell, a correspondent, and Leon, her editor, had three minutes before recording. He began with an apparently finished result. She opened the view to show trials, a failure, a checker and a saved record. His later gesture changed what their report would include.

That gave the projected images a dramatic job. A crop could conceal the work behind the result. Revealing the whole photograph supplied evidence that changed the editor’s choice. The scientific mechanism and the human encounter could then move together.

The written package retained exact dialogue, photographs, diagrams, camera intentions, puppet phrases and one elapsed multimedia sequence. Earlier drafts remained beside it. The script number identified both the research and the writing, even as titles and dramatic choices changed.

The figures in the script were teaching illustrations or attributed experimental reports, with their conditions attached. They were not a new benchmark run by the factory. This article records that writing decision rather than offering a fresh assessment of the papers’ broader claims.

The correction was about what the play centred. The source finding had to be understandable in ordinary language and consequential in the performed exchange. A valid written package was the next useful object. The crew still had to turn those pages and images into a film.