Every page here is generated from the same structured YAML canon that compiles the film's AI-video prompts. Nothing on the public site is hand-edited into place; the accuracy and representation rules that govern the fiction are run over the finished output, landing page included. This layer shows the seams on purpose.
Research → canonical documents → structured YAML in data/ → compiled Flow prompts, and now → compiled site. The artefact manifest (data/artefacts.yaml) is the machine-readable list of every site-facing piece; scripts/compile_site.py renders it into a fresh output directory each build. Governance record: ADR-030.
Rather than one flattening "register," each artefact carries its world (diegetic or production), its form (testimony, junket, deposition, document, gallery-clip, texture, process), the ontology of who is speaking or depicted (character, data double, synthetic performer, human creator), and its visual system (state, resistance, or the seam between them). The film's two-world grammar is the visual axis only — it is not a claim about who an artefact serves.
The clips in the gallery are pre-lock generation tests, not finished performance. Identity locks — the gate that clears a character's likeness and voice for final generation — are still open. Nothing synthetic here is presented as a finished or authorised performance, and no synthetic voice speaks as an authority on lived experience.
Every technical or social mechanism the story asserts is tagged against real-world plausibility. The claims ledger lists each one as existing, plausible_extrapolation, speculative, or poetic_nonliteral. Counter-surveillance in the story is drama, not an operational bypass manual.