Narracode launched on 9 May 2026 as a single markdown file: a harness that unfolds into a
system of separated passes — Initiator, Reading, Structural, Compositional, Reflexive. The premise was
that literary composition could be handled the way Claude Code handles software. Externalise state to files.
Separate generation from critique. Let the critical pass govern the generative one.
Ninety-two days later there are 25 finished stories and 124,081 words, published as
standalone pages with their full working memory beside them — poetics, drafts, structural state,
critiques, version snapshots, attribution. Every story ships its own provenance.
The harness changed as much as the corpus did. It began with three structural memory files
(graph, time-constants, history) and now carries eight, adding obligations, motifs, scene-ledger, character
interiority, and reader-state. AUTO_MODE arrived on 25 May, allowing full-pipeline runs without interaction.
A registry of AI tells — seventeen classes, each with a detector and a target rate — was added on
30 July, along with a tell-scan pass that runs after drafting. Composition and editing drew on models from
three labs: Opus 4.6 through Opus 5, Sonnet 4.6 and Fable 5 (Anthropic); Gemini 3.1 Pro through 3.6 Flash
(Google); GPT-5.5 and Codex (OpenAI).
Nothing here was written by prompt alone. The Seamless Edit method treats the author's manual
edits as the teaching signal: each snapshot diffs the draft against what the model produced, and recurring
patterns are promoted into the project's poetics. The corpus that resulted is also a dataset — paired
pre-edit and post-edit drafts of the same content, across 25 stories.
By the numbers
25
stories
124,081
words
1.9
stories / week
3.7
days between
4,963
mean length
8h 16m
to read it all
One mark per story across the 92 days; height is word count. Three stories landed on 25 May.
The longest silence was 11 days, between
Adjunct: Our Internal and
In Our Image. Median gap: 3 days.
Lengths run from 986 words
(The Chute) to
18,287 (The Wonderful Adventures of Trygve
Aas); the median is 3,572. Output averages
1,349 words a day. Word counts here are recounted from the published pages rather than the drafts,
so they run slightly above the figures listed on the homepage.
David Jhave Johnston is a digital poet working in emergent domains. Author of ReRites
(Anteism, 2019) and Aesthetic Animism (MIT Press, 2016). He is currently an
AI-narrative researcher at the UiB Centre for Digital
Narrative (2023–27) with the Extending Digital Narrative project.
Funding
This work was partially supported by the Research Council of Norway through its Centres of
Excellence scheme, project number 332643 (Center for Digital Narrative), and its SAMKUL project
scheme, project number 335129 (Extending Digital Narrative).