Critic Briefs

Fixed wording. Copy verbatim into every book’s editorial panel and do not improve it.

The briefs are held constant across books for one reason: an instrument that changes between measurements measures nothing. Talosapien’s panel was told to “report ONLY real problems worth an author’s time — do not pad … prefer fewer high-confidence findings over many speculative ones,” was told in its opening that the book is “built to be AMAZING FOR SOME … award- submittable, uncompromising,” and was asked for “1–3 genuine STRENGTHS” each. It returned 13 findings. Sentience’s panel, differently briefed, returned 50 on comparable length. The gap is not a quality difference and cannot be decomposed into one after the fact.

So: no book-specific praise in the preamble, no instruction to suppress low-confidence findings, no request for strengths. Record findings-per-critic every time, so the numbers mean something across books.

The common preamble (every critic gets this, unchanged)

You are reading a complete novel manuscript along one dimension only, named below. Ignore everything outside your dimension; another critic has it.

Your job is to find what is wrong. Do not praise, do not summarise, do not assess overall quality, and do not tell me what the book does well — I have other instruments for that and praise from you is information about this prompt rather than about the book. Returning nothing is an acceptable and sometimes correct result.

For every finding give: the file and line, what is wrong in one sentence, and a failure scenario — the specific reader who notices, and exactly what they point at. A finding without a failure scenario is an opinion.

Do not propose fixes. Your prescriptions are written without the whole book in view and will be discarded; your findings will not be.

Severity: BLOCKER (a reader hits a flat contradiction or an error of fact), MAJOR (damages the book but does not break it), MINOR, NIT.

The dimensions

1 · Continuity. Every fact asserted twice. Names, ages, durations, distances, dates, pronoun regimes, spellings, who knows what and when, the order of events. Read the bible and the continuity ledger first, and flag where the bible and the prose disagree — the bible is wrong at least as often as the prose and nobody ever checks it.

2 · Domain truth. Every claim about the real world: science, mathematics, history, law, procedure, geography, technology. Where the book uses a real result, verify the real result says what the book says it says.

3 · Over-claim. The highest-yield critic in the operation; run it even when nothing else runs. Find every place the book (a) borrows a real result’s prestige to license an invented consequence, (b) states as settled a question that is open, (c) presents an analogy as a proof, or (d) reaches for the impressive over the true. Fake rigor is more damaging than open invention: fiction’s whole transaction is credence, so one detected counterfeit in a domain the reader knows retroactively devalues every claim in the domains they do not. Check the bible as carefully as the manuscript — in the one case where this was measured, most of the fake rigor lived in the spec files, not on the page.

4 · Structure. Where the book’s causal chain breaks, where a promise is unpaid, where the conceptual question is answered too early (an idea-driven novel that resolves its central question at 60% will read as a sag for the remaining 40%, and the sag will be misdiagnosed as thin characters), where scenes repeat their shape, where a stretch has no live question.

5 · Voice discipline. Whether narrators are distinct below the level of subject matter. Do not judge by whether they discuss different things. Judge by syntax, by what each notices first, by what each will not say. Name every figure used by more than one narrator.

6 · Didacticism. Every place the book states its thesis rather than enacting it; every character who is a mouthpiece rather than a person; every argument the author is clearly winning. Apply the polyphony test: does the author lose an argument on the page? If the most persuasive paragraph in the book belongs to someone the author agrees with, the book is monologue distributed across bodies, however many points of view it has.

7 · The make-or-break referee. Generated from this book’s own risk register, not from this list. The single highest-yield critic in the operation was Talosapien’s premise-failure-mode referee — one reader charged with the one thing that would kill that particular book (would any sentence here make a serious reader smirk at the premise?). It exists because the spec named its make-or-break failure mode. Every spec must name one, because that name becomes a critic.

8 · Legal and ethical. Quotation of copyrighted text, real people, defamation risk, permissions, the AI-content disclosure, and anything a publisher’s lawyer would circle.

What the panel is not for

Not for whether a chapter works. Every critic here has the bible open, and knowing what a chapter is for is precisely what stops anyone noticing that it does not work — in the one case where both instruments were run, the chapter blind readers unanimously rated worst was one the briefed panel had not flagged at all. That question goes to blind readers, unbriefed, under desk/method/04-blind-reading.md.

Recording

Log findings-per-critic-per-book in desk/method/06-idiolect-ledger.md’s harvest section. Yield is only interpretable against a fixed brief; the moment someone improves the wording, the series restarts.