One read, not a patchwork
Older report generators split a manuscript into small chunks, summarised each chunk on its own, then stitched the summaries together. That works for grammar. It falls apart for anything that spans the book, a character’s arc, a promise made in chapter 2 and paid off in chapter 9, because no chunk contains an arc, and nothing built chunk by chunk can judge one. Quarkle’s report agent reads instead of chunking. It opens your chapter or book in full, in one sitting, before writing a word of the report.Written section by section
Rather than handing back one giant block of findings, the agent writes the report as a series of sections, each one saved the moment it’s written. A long Full Assessment covers a dozen or more sections, and saving as it goes means a hiccup partway through doesn’t cost the sections already delivered. The verdict is written last, after every other section, on purpose. It’s the summary of a finished read, not a first impression written before the agent has actually looked at your structure or your continuity.Severity means something
Findings in The Plan sort into three classes by a fixed rule, not a ranking of whatever came up: critical is reserved for a reader-visible failure the book cannot ship with, never a deliberate stylistic choice the reviewer would have written differently. Most competent manuscripts should land with an empty critical list, and Quarkle treats that as the stronger report, not a weaker one. Each finding also offers at most two alternative fixes, not three, on purpose, a model given a third slot will reliably find something to put there whether the book needs it or not.Continuity is verified, not asserted
The Full Assessment tier checks continuity across your book. A continuity finding only makes it into the report after the agent has pulled both passages it’s citing and confirmed they actually contradict, rather than flagging a contradiction it half-remembers from earlier in the read.Some numbers are counted, not judged
Two things in your report come straight from your manuscript, not the model’s read of it: the how it reads stats (sentence length, dialogue share, grade level) and the Presence heatmap. Both are computed after the report finishes, directly from your text, so they’re exact rather than estimated.A quiet, single-shot run
A report doesn’t chat back and forth while it works, and it doesn’t show a section-by-section progress trail either. It’s one long quiet read, then one finished dossier arrives, or a clear error if the run didn’t make it.The everyday how-to lives in Reports.
Next: Keeping the Atlas current
How the Atlas stays current.