Every factual sentence has to resolve to a quote

The incumbent in this category charges three thousand dollars a year and scores niches against a rubric that its own screen labels self-reported. Nothing in it stops the tool producing a confident, uncited claim, which a paying customer then puts their name on and sells to somebody else.

Groundwork takes what a person knows and turns it into a finished digital product across nine stages, from scoring a niche through researching it across more than a hundred and fifty live sources, drafting the manuscript, and rendering it to PDF and EPUB. Its one hard rule is that every factual sentence must resolve to a verbatim quote from a cited source. Chapters carry claim identifiers that tie each factual sentence back to a ledger of source URLs, so any statement in the finished book can be traced to the thing it came from rather than to the model's confidence about it.

Scoring is held to the same standard. Any dimension without real supporting evidence is forced to zero rather than filled in with something plausible, and candidates are ranked against each other so a score carries meaning instead of floating free.

Forcing a zero is the part that makes the rest work, because the natural behaviour of a language model handed an empty cell is to produce a reasonable-looking number. A pipeline that cannot say it does not know will always sound better and be worth less.