← Red Eagle project hub

Section 04 · Where we stand · 16 Sep 2026

The whole film is laid out. A third of it can be written.

All 232 scenes exist as structure. Dialogue and staging can only be written where a document supports them, and 143 of those scenes need a Creek source we do not have yet. Getting those sources runs through a consultation that has not happened. So the work moves on the 89 scenes that are clear, and the pictures are being made from those.

Where each lane has got to Measured across the film’s 232 scenes
Scene 1 89 documented Scene 232 Research & consultation Blocked on other people 89 documented 143 need a Creek source we do not have Writing & staging Structure done 89 written 143 laid out, nobody speaking in them Generating image & video Running on the 89 55 made 57 held for consultation 98 not made yet · video not started 22 22 blocked: the software cannot resolve who is in the scene Consultation not held 143 scenes cannot be written, and 57 pictures cannot be made, until it happens Not a calendar. The block is a conversation, not a duration.

Research & consultation

89 documented143 need a Creek source

↓ Consultation not held. 143 scenes cannot be written until it happens.

Writing & staging

89 written143 laid out, nobody speaking

↓ Only the written 89 can be drawn.

Generating image & video

55 made57 held for consultation22 blocked on cast98 not made yet
  1. Lane 01

    Research and consultation

    Blocked on other people

    There is a lot of research. A peer-reviewed paper this film adapts, twelve volumes of Jackson’s correspondence, both treaties in full, nineteenth-century histories and memoirs, modern Creek scholarship, uniform plates, a register of 134 quotations judged real or not across 22 passes.

    It is still not enough, and the reason is specific rather than general. 143 of the 232 scenes need a Creek-side source, and the route to those sources runs through consultation with the Poarch Band of Creek Indians and the Muscogee (Creek) Nation. That has not happened. It is not a task with a duration; it is a conversation on somebody else’s calendar.

    Moving now

    22 acquisitions being chased, six of them questions about the landscape no archive will answer. The list grows when work moves, because a search that finds nothing is a finding.

    Not moving until the consultation

    The Creek-side record for 143 scenes. Nothing downstream of those scenes can be written, staged or drawn.

  2. Lane 02

    Writing and staging

    Structure done, dialogue a fraction

    The screenplay is built as a structure: 29 scenes, 232 separate moments, each with a date, a place, and whose eyes we are watching through. That part is finished.

    The writing inside it is not. Dialogue only gets written where a document carries it, which in practice has meant a letter from Andrew or Rachel Jackson. Where there is no such letter, there is a laid-out scene with nobody speaking in it. Staging directions are in the same position, and so is the work of writing the other forces in the room as characters rather than as absences. That is the fraction: 89 moments can be written now, 143 are waiting on lane 01.

  3. Lane 03

    Generating images and video

    Running on the 89

    Rather than wait, the pictures are being made from the scenes that are already written. 55 exist. Of the rest, 57 are held behind the same consultation as lane 01, 22 are blocked because the software cannot yet resolve who is in the scene, and 98 are simply not made yet. Every one of those is a gate with something behind it. None of them is a decision not to make the picture. No video yet.

    Running it early is what has told us what the film actually needs. The night sky turned out to be computable from an astronomical library rather than described, which also gives a way to line up accounts from different cultures against the same meteor shower. Vegetation and shadow have to be specified per scene or the software invents them. And the real unit of work is not the picture but the reference plate that keeps a place, a season and a face the same between scenes, of which we expect more than four hundred.

    Case studyWhat the pictures cost to get right is its own write-up. Three single-image tests, an honest account of the one that came back backwards, and the eight-second clip they made possible. Two plates, one shot →

Script development

Two ways of measuring the same screenplay, and they do not agree. One of them is the reason this stage exists.

The screenplay is written. Thirty units, 7,290 lines, 570 speaking cues. Across the cast cards, 94 of 102 speaking beats already carry dialogue on the page and eight are blank. There is no empty page to fill here, which means the build is not a drafting engine with a screenplay-sized backlog in front of it.

Measured a second way, the position is different. Of 570 speaking cues, 30 carry a provenance marker and 540 do not. The gate was specified, hardened over twenty recorded rounds and deployed publicly, and it has never been run over the screenplay it was built for. Five hundred and forty lines of dialogue stand on nobody’s authority but the writer’s memory of what the record said.

That is not a defect peculiar to this project. It is the position most organizations are in. Controls get specified, applied to new output, and the body of work that predates them is never measured. Nothing fails, so nothing surfaces.

What the output actually is

Not pages. A screenplay in which every line knows what it is standing on, which is the only claim this project makes. The instrument is a five-rung ladder, so each line declares whether it is documented, aligned, traditional, interpretive or invented, and fifteen deterministic checks that run before a person reads generated text. Three policies sit over one implementation, so the public exhibit, the private drafting tool and the assistant beside it differ as declared data rather than as three copies of the code.

The split, before anything is routed anywhere

216 of the 570 cues, 38%, belong to 22 characters whose depiction is consultation material. They do not go to a model at any rung. They go on a consultation list. The title character is among them and is the largest speaking part in the film, so consultation is not an exception path around the edges of this work. It runs through the middle of it. The remaining 354 cues are open to the audit.

The flag is not an ethnicity field, and the first character it touches proves it. One enslaved character is held not for his own depiction but because he is enslaved in a Muscogee household, and the depiction of Creek slaveholding is itself the consultation question.

The audit runs backwards

Every check takes text and evidence and returns violations. Nothing in them requires the text to have come from a model. A hand-written line and a generated line are the same object to the gate, which is the design rather than an accident of it: the gate polices claims, and a claim does not care who made it.

Each audited cue lands in one of four states. The record supports it as written. It is sound but the rung was never declared. It claims more than the record holds. Or the record says otherwise. Only the last two go to redrafting. The expectation, written down in advance so the first sheet can be scored against it rather than rationalized afterward, is that most of the 540 is undeclared rather than wrong. That distinction is the product.

Where it stands

Staging is built, 235 movements. The date-locked corpus and the gate are working, 15 check categories, all selftests passing. The drafting component is gated with zero beats currently reachable. Casting is the one term of the build that does not exist yet: no character carries the form-of-address field, and the component that needs it refuses rather than defaulting.

Order here is forced by dependency, not by preference. The address field for the near-term characters is one pass and unblocks everything behind it. The backward audit tool is one session and is blocked by that pass. A single pilot unit follows, chosen because it is the one beat the drafting component has already run, so a regression is visible immediately. Reading that pilot sheet and fixing what it exposes is the work that matters. The last time a real sheet was read it found nine misfiring checks, and the time before that a leak in the consultation gate.

The remaining units are mechanical once the pilot is understood, and they are deliberately not scheduled. Rate of closure is the wrong instrument. One sheet that shows how the gate misreads a line of dialogue is worth more than twenty units cleared.

The finished state is a rung and a ledger row behind every cue, or the cue sitting on the consultation list, with the rung distribution across all 570 published. That distribution is the single artifact that would prove the claim this project makes. It does not exist today and cannot be estimated, only measured.

None of this waits on the research being finished, because the research does not finish. It waits on one conversation, and in the meantime the film gets made out of the part that is clear.

Back toRed Eagle project hub Section 02Quality control Section 03Operational design Case studyTwo plates, one shot

Scene counts checked against the storyboard rebuilt 16 September 2026: 232 scenes across 29 units, 143 carrying SRC-CREEK, 55 images made, 57 held under D-08 pending consultation, 22 blocked on cast resolution, 98 not yet made. The plate estimate is a forward expectation, not a count. The image-refinement runs behind lane 03 are written up separately at Two plates, one shot. Source: Aaron Zeanah, “The End of the Alabama Frontier,” International Journal of Interdisciplinary Social Sciences, Vol. 6, No. 4 (2012). The Creek War involved the mass killing of real people whose descendants are living communities today, the Poarch Band of Creek Indians and the Muscogee (Creek) Nation.