Work Neuraxic
Neuraxic · Writing fictionI wanted a story's prose, people, places and rules to develop together, and a way to try another version without losing what I'd already decided. Neuraxic is the writing app I'm shaping for that. A new detail stays a proposal until the author accepts it, and a what-if version keeps the point it started from.
Interface study · the app's own screens with a made-up story world · five tests passed on September 19, 2026
Neuraxic keeps a story's facts as claims, single statements such as "the lantern falls dark as the water recedes." Each claim is either a proposal or canon, the version of the story the author has settled on.
The app finds rules, names and quantities in the text and offers each one in the margin. Its own code does this the same way every time, with no AI model.
Sealing a proposal makes it canon. On screen, an open seal marks a proposal and a gold seal marks canon.
The same subject opens as a draft, a list of claims or a map of its relationships.
An alternate version starts from a fixed point in the main story, so later changes there don't leak into it.
A detail appearing on screen doesn't make that decision for the author. The Claims screen states the rule in the app's own words: "A proposal becomes canon only when you seal it. Every seal writes a receipt." The Seal button runs the same step the tests further down call promotion.
These are four of the app's real screens, filled with a made-up setting called The Tideglass Reach. Only the data is invented.
The Library shows each story world as a book on a shelf. The covers are placeholders the app draws with code; no image model made them.
A tome groups one story world, and a loose leaf is an idea not yet placed in one. The Library holds the tomes, and the Catalog also lists the loose leaves. Open any element and it becomes the focus, with its draft, claims and relationship map as views of the same subject.
Dark green surfaces, parchment-toned text and warm gold make the app feel like a place to read and think. The marks carry meaning: open and gold seals separate proposals from canon, and ember-colored findings draw attention to concerns.
The same made-up world, from the shelf down to a single subject and back. The recording is silent; its captions and the transcript describe each step.
Neuraxic calls each version of a story's facts a continuity. The five tests answer two questions: what can a version of the story see, and what has the author accepted? The diagrams explain the tests with a second made-up setting, Bellweather Reach. They aren't screens from the app.
Step 1 of 5
The main story holds one claim: the beacon is lit. Ask what this version knows and that claim comes back, even though it is still a proposal.
Coming back in the answer is what "visible" means. Visible and canon are separate questions.
The beacon is lit.
Proposal Recorded before the split
Ask what had been recorded as of an earlier moment, and anything recorded later stays out.
The author starts an alternate version. Neuraxic records the point in the main story where it splits off.
The what-if can see the main story's claims up to that point. Nothing is copied or accepted along the way.
The beacon is lit.
↓ The what-if sees the main story up to the split
The beacon is lit.
Visible from the main story
Later, the main story adds a claim: the causeway is closed. Ask both versions what they know at that later moment.
The main story includes the new claim. The what-if still sees the beacon, but not the causeway claim added after the split. The main story's edit is neither rejected nor erased.
The beacon is lit.
The causeway is closed.
New main-story claim
The beacon is lit.
The causeway claim came after the split.
Another test starts with a canon claim, then records a claim that gives a different value for the same detail of the same thing in the same version.
Here, "the beacon is unlit" conflicts with the canon claim that it is lit. Neuraxic reports the conflict and keeps the new claim as a proposal. The idea stays available, and canon stays unchanged until the author decides.
Canon
The beacon is lit.
Recorded first
Proposal
The beacon is unlit.
Conflict reported · stays a proposal
When the author promotes a proposal to canon, Neuraxic writes a receipt that records who did it. Send the same promotion again with the same identifying key and it returns the original receipt.
One test saves the first promotion, then opens a fresh session before retrying. The receipt still matches.
A claim the author decides to accept
↓ The author promotes it
The wider design separates what's true in the story world, what the story shows, and what characters or readers believe. A character can be wrong while the author knows the answer.
I decide what each project is for and check what comes back. Here, I wanted a story's prose, people, places and rules to develop together without losing track of what I'd already decided. After the first prototype, I asked for the story to be shown as a map of connected parts while the app stays a good place to write.
AI tools write the code. That includes the five tests on this page.
| Test | What it showed |
|---|---|
| Asking at an earlier point | A question about an earlier point in the record leaves a later claim out of the answer. |
| Starting point | A what-if version keeps the main story's earlier claim and doesn't see one added after the split, while the main story does. |
| Conflicting proposal | A different value for the same detail of the same thing in the same version is reported against the existing canon claim, and the new claim stays a proposal. |
| Promotion in one session | An explicit promotion records who did it and makes the claim canon. Retrying with the same key returns the same receipt. |
| Promotion across sessions | After the first promotion is saved, the same retry in a new session returns the original receipt. |
The screens and the 34.48-second recording were prepared on September 19, 2026 from the app's code at version f29176bf, unchanged, in its flat, low-motion layout and with the fallback fonts its own code names. The made-up data was fed to the screens as sample answers instead of coming from the app's backend. The recording moves through seven views. The screens made twelve requests, all of them reads, and no errors appeared. The app's built-in suggestion finder and map layout ran as they normally do.
The Tideglass Reach and Bellweather Reach were written for this page, and no private story material appears on it.