Copy that passes every check and still reads cold
Clarity, cliches, spam words, CTA strength, compliance. Six gates can all report PASS on a piece nobody wants to finish, because none of them measure whether it sounds like a person talking. v0.6.0 adds the register where the writer is the hero, the method behind it, and four numbers that say whether a draft got there.
For founders writing their own origin story, for anyone whose launch post reads like a memo, and for teams whose copy review is one person saying it feels off and nobody able to say why.
opvs-skills install @di-atomic/copy-engineThen say: “tell my founder story”, “write the launch article”, or “does this sound like AI wrote it”.
Same request, two very different answers
returns a competent essay. I always loved building things so I started a company it was hard now we help customers reads fine. moves nobody. you cannot say why.
It is grammatical, on topic, and forgettable. There is no way to review it except by feel, and feel is exactly what fails you on your own writing. You ship it, it underperforms, and the next draft is a coin flip.
asks you fourteen questions.
the struggle you would
rather not admit
the false belief fighting you
the moment it broke
who you became
then counts four numbers
before it calls the draft done.
The questions come from a named practitioner script with page numbers attached. The numbers come from a corpus, not a preference. Anything you did not answer comes back listed as a gap, never quietly filled in with something plausible.
The bet is simple: the parts of writing everyone calls taste are partly countable, and the countable part is where an agent stops guessing.
The three decisions doing most of the work
Story and explainer are different registers, measured differently
Two modes, not one bar. An explainer is allowed long sentences and heavy second person. A story is not. Running one standard across both flattens whichever one you cared about, which is why the gate asks which register you are in before it judges anything.
The method is cited, not paraphrased
The Epiphany Bridge ships with its five phases, its fourteen questions, and pp.213-222 attached. The Hero’s Two Journeys ships with pp.197-211. Evidence is labelled pattern-level, because a framework being widely taught is not the same as a measured conversion lift, and pretending otherwise is how copy skills lose trust.
It will not invent your results
The two questions that carry performance claims are the two the skill refuses to answer for you. No supplied number means the section is written without it and flagged as missing. A reader who checks one figure and finds it hollow will not check a second.
The spine, and the ending most launch posts never reach
What the register gate actually returns
> verify-narrative-voice --mode story draft.md words=2435 sentences=182 FAIL em-dashes 17 (want 0) FAIL 1st person /1k 16.8 (want 30-50) FAIL 2nd person /1k 11.1 (want 15-30) FAIL avg sentence 13.4 (want 8-11w) RESULT: FAIL (4 bars out of range) → same material, rewritten as a story PASS em-dashes 0 PASS 1st person /1k 39.8 PASS 2nd person /1k 26.8 PASS avg sentence 9.9 RESULT: PASS
That first draft is real, and it had already passed every other gate in the skill. None of those four numbers were visible by reading it. The bars come from a corpus of 83 high-performing posts combined with the measured targets in the brand voice file, and they are register targets rather than a promise about conversion.
It writes and it measures. It cannot answer the fourteen questions for you.
copy-engine is one skill in the system behind Di-Atomic, the marketing and compliance agency that runs cognitoAI, SpiderIQ and OPVS. The material has to be yours, because the whole value of the register is that it is real. If you would rather have the questions asked out loud and the piece written with you, in any of our seven languages, that is the day job.
Book a 30-min call with Di-AtomicJust want the skill? Get @di-atomic/copy-engine free.