We build AuthorDeck the way a lab runs an experiment — on
measured results, not marketing. First we need a yardstick, so we take
the best frontier cloud model — today’s
commercial state of the art in AI storytelling — and treat its
craft as the 100 % ceiling. Every draft is then
graded against one fixed rubric: the strength of the opening, the
clarity of the central conflict, the weight of the
ending, tonal and voice control, genre fidelity, continuity. That
rubric is where the percentage comes from — a repeatable craft
score, marked the same way every time, not a gut feeling.
Then we measure honestly. We take the strongest AI model the world can
run locally — entirely on your own computer, with
no cloud — and score it on that identical rubric. Left to
itself, it lands well short of the ceiling: it wanders, it leaks, it
loses the thread halfway through a scene.
So we built an engine and named it the
DeepNarrative Engine. It wraps that local model in
hand-written craft logic — then we run the very same test again.
The finding: the ceiling rises. The same model, on the
same laptop, scores markedly closer to the cloud state of the art
— private, offline, and yours.
The DeepNarrative Engine (DNE) won’t write your
career for you. It is a craft-control layer — built by a
programmer who was also a working novelist, then sharpened by a
hand-picked crew of engineers, where every prompt the engine fires is
shaped by hand, line by line. Left to itself, a local model
wanders: weak openings, a blurred central
conflict, a misplaced ending, a tone that
drifts mid-scene. The DNE closes each gap with hand-written craft
logic, so instead of whatever the model says first, you get a draft
that holds its shape — clean, complete, and
on-genre.
The deepest part of the DNE is not the machinery. It is the
human wanting underneath the machinery: the private
itch of this, not that; start closer to danger; make the
ending cost something; let the romance heal, but not cheaply; let the
mystery turn on guilt, not cleverness. AuthorDeck treats those wants as
the true steering wheel. You decide which way the story leans, what
kind of door the first page opens, and what aftertaste the final page
leaves behind. The DNE’s craft logic then turns that appetite into
pressure the model can obey — not “write me something good,”
but go this way, begin with this hunger, end with this shape.
And it does not write in one breath. Ask a chatbot and you get a
single pass — whatever comes out first. The DNE sculpts
instead: drafting the whole story, re-reading its own draft, rebuilding
the scenes that went slack, and landing the ending across several
deliberate passes, the way a real writer revises — the first
draft is never the page you keep. It gives the model
rails — genre constraints, leak checks,
continuity pressure, repair passes — not a promise to turn any
model into a genre master. The craft ceiling follows your
model; the DNE’s job is to raise the floor —
cleaner, steadier, far harder to derail. No cloud, no subscription,
nothing leaving your laptop; you pay once, about the price of one nice
meal.
Under it runs a working Failure Museum — more
than thirty named failure modes, each discovered, named, and patched by
a hand-picked specialist team. Find a new one and we patch a text file
overnight, not a costly training run — so the engine keeps
improving as the models improve. On any Mac with
16 GB or more it runs its full multi-pass craft
and hands back readable, structurally sound drafts for daily writing;
on 32–64 GB with a dense model it reaches
the deepest genre separation we offer — the widest
range, the richest sensory and tonal control. Either way, what you hold
is a strong draft — what it still needs is your taste,
your edits, your signature. The shortcut was never skipping
the craft; it is reaching it sooner.
In all-subgenre testing the DNE returned 67 of 67 stories
clean, with zero leaks — reliability you can hold, on
your own machine.
Once, the chief of our crew said: “It feels like teaching an
AI to dream like a human — waking the dreamer that had been
asleep inside our twentieth-century machines. We turned the magic
numbers not until the math was right, but until the machine forgot
it was a machine. It feels like teaching someone to dance who swore
they never could!”