Case study · feature documentary
One project documented to full depth: from testimony to workflow to an AI-assisted review pass to the grade, with the ethical decisions on the record. The film is in post-production; this page grows as it completes.
The film
40 Years of Silence is a feature documentary built around testimony: memory, and the structures that shape what gets said and what remains silent across generations. I am co-producer, and the film is currently in post-production.
A film like this lives or dies on trust. The people who speak in it are giving something that cannot be retaken, and every stage of the workflow — from the first recorded conversation to the final grade — has to protect that. That is what this page documents.
Why one project in depth
Project cards tell you what was made. They do not tell you how decisions were reached, what was ruled out, or where the ethical lines were drawn. This case study is portfolio, research evidence, and book material at once: the workflow feeds the EdD's practice-as-research methodology, and the AI-assisted review pass tests claims made in Filmmaking in the Age of AI: A Practical Guide against a real production.
Where a stage cannot yet be shown in full — the film is unreleased and clearances are still in progress — the decision framework is documented instead, and the material will be added as it clears.
The workflow
Each stage names what was done, what the tools were allowed to touch, and what they were not.
Long-form interviews conducted over extended access, with consent treated as a process rather than a signature: contributors know how their testimony will be used, and that understanding is revisited as the film takes shape. No AI touches this stage. The record itself — what was said, how it was said — is the film's foundation and is preserved exactly as captured.
Machine transcription is used to make hours of testimony searchable — a mechanical task where AI serves the work without shaping it. Every transcript is verified against the recording before it informs a structural decision. The paper edit, where the film's architecture is actually decided, is authored by people: the machine finds the moment; it does not decide what the moment means.
The edit works from the paper structure, testing it against what the material actually does on screen. This is where the film's central tension — what gets said and what remains silent — is built rhythm by rhythm. The editorial judgments here are the authorship of the film, and they stay entirely human.
This production doubles as the test bed for the Lab's editorial review experiment: can AI tools usefully flag structural problems, pacing issues, and narrative inconsistencies in a rough cut before fresh human eyes come in? The tool reads the cut's transcript and structure — never the testimony's meaning, never a contributor's face. Its notes are treated as prompts for the editorial team to argue with, not directions to follow. The 4Ds from the calibration instrument — Delegate, Describe, Discern, Diligence — govern what the tool is asked to do at this stage.
Finishing is under way. The grade serves the testimony: archive and present-day material are held in distinct, honest registers rather than smoothed into one look, so the audience always knows when they are. Grade notes and before/after material will be published here once the film clears for release.
Ethics checkpoints
The same Ethics Checkpoint discipline the book applies chapter by chapter, applied to this film.
Nothing a contributor says, and no image of a contributor, is generated, extended, or synthetically altered. In a film about what four decades of silence did to a family, a synthetic voice would not be a production shortcut; it would be a betrayal of the subject itself.
AI tools read the record; they do not add to it. Transcription and structural review are analysis of material that already exists. The moment a tool would put something new in front of the audience, it is out of scope for this film.
Contributors' consent is revisited as the film evolves, because agreeing to an interview is not the same as agreeing to the film that interview ends up inside. This costs time in post. It is not optional.
Where AI tools were used, that use is documented — on this page and in the film's production records — at the level of specificity a festival, a broadcaster, or an ethics committee would need. Nonfiction earns trust by showing its working.
Still to come
As post-production completes and release clearances allow, this case study will add: the full workflow diagram, timeline extracts from the review pass with the tool's notes and the editorial team's responses, and grade notes with before/after frames. The point of publishing it now, unfinished, is the same as the Lab's: documentation you can watch accumulate cannot be faked retroactively.
I am glad to talk through any stage of this workflow in more depth — for festivals, ethics review, research collaboration, or the book.