Provenance

And what they are not

Not one frame of archive has been licensed. Wherever the cut calls for record it holds a card naming exactly what must be cleared, and nothing has been put in its place.

Every picture is either a generated frame or a licensed elemental loop — dust, smoke, embers, sand, water, stars, a turning fan. Both live in the [X] reconstruction register and both are labelled [X] on screen. The elemental loops carry no location claim, no identifiable person and no insignia: they are weather and light, not evidence. They are credited in full below.

Every voice is synthesised, and labelled on screen while it speaks. Boone Cutler’s lines are his own written words read by a stand-in. The reconstruction brief permits that only after the estate’s digital-likeness consent is executed under section 4 of the term sheet, so these are development cuts and his authentic recorded voice replaces every one of those lines before public release.

The sound in the reconstruction register is synthesised too — room tone, the drone, the crowd, the loudspeaker. None of it is a field recording, and none of it is laid under a record cue.

No testimony has been captured. The two testimony cards in the sizzle carry a written placeholder line as type. No synthetic voice has ever been used to speak testimony, and none ever will be.

What the reading is

Every voice in it is synthetic, without exception. Boone Cutler died on 18 September 2025 and no recording of his voice appears anywhere in it. Every chapter says so out loud before it begins, and the label on the reading page stays on screen for as long as audio is playing, naming whichever voice currently holds the microphone.

It is built against a condition that has not been met. The reconstruction brief §3 and the term sheet §4 permit a Boone-alike only once the estate’s digital-likeness consent is executed. It was asked for and it was built, with the condition made visible rather than left silent. If the estate declines, this comes down.

His words are not ours. Nothing in the text was rewritten, tidied, abridged or expanded for the reading. The build proves it rather than promising it: every chapter is reassembled out of the segments that were actually rendered and diffed against the manuscript, and it fails if a single paragraph differs. That check has already caught one fault — a chunker that silently lost characters from three long paragraphs.

Nothing was dropped in the reading either. Speech synthesis does not raise an error when it skips a clause; it returns a clean, shorter reading. So the audio is transcribed back and diffed against the words that went in. The check is calibrated in both directions on every run: it is shown reads it must pass and planted drops it must catch, and it refuses to certify if it cannot demonstrate both.

A quarter of this book is not the author speaking, and the reading says so. Roughly 24% of it is Baghdad Mo’s own first-person testimony, and a further 7% is documents — reports, orders, letters. Reading all of that in one voice would have flattened the one thing the book does structurally, which is hand the microphone over. The Foreword, written and signed by another veteran, and the closing jacket copy are not read in the author’s voice at all.

The room under the reading is synthesised in your browser out of noise and filters. Not one sample, not one field recording. It never follows the story to a place, because nothing in it was recorded in any place — it follows the voice and the part of the book, both of which are properties of the text. Nothing at all plays under a prayer or under a declaration. It can be switched off.