Skip to content

Claude/fiction analysis tool vb1p h - #58

Merged
mekado11 merged 5 commits into
gh-pagesfrom
claude/fiction-analysis-tool-vb1pH
Sep 21, 2026
Merged

mekado11 merged 5 commits into
gh-pagesfrom
claude/fiction-analysis-tool-vb1pH

Conversation

@mekado11

Copy link
Copy Markdown
Owner

No description provided.

…ngine wrote

Every sentence the Blurb Generator showed was real, but the claims made about
them were not. A self-help book was given a protagonist (the literal fallback
string "the protagonist"), a tone ("Dark/Suspense" from more than five dark
words in seventy thousand), a six-question story framework filled by the first
sentence in each slice containing a cue word ("How does it get worse?" was the
first middle sentence containing "but"), and five blurbs stitched together
with fiction filler nowhere in the book ("must face a choice that will change
everything", "one chance to survive").

Nonfiction now gets blurb material, not blurbs: the author's own sentences by
the job a back cover needs them to do (the promise, the problem it names, who
it is for, the author's stake), each with the section it came from, plus the
method and the proof from Document Intelligence's per-chapter extraction. No
prose is written for the author.

Fiction keeps the six questions, and each answer now states the basis it was
chosen on ("first sentence in the middle containing 'but'"; "no sentence in
the first quarter contains a change cue"). The five shapes assemble only the
manuscript's sentences; a beat the book did not supply is shown as a gap to
write, and every shape leaves its closing line to the author. Tone is a rate
per 1,000 words; a leaning needs 1.5/1K, and a handful of dark words in a long
book is not "dark". A protagonist is named only when a recurring name exists.

Also: qualified names now rank by mentions (evidence decides whether a word is
a name; mentions decide its place), so a lead who opens most of her sentences
no longer ranks below a secondary character who sits mid-sentence.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
… on a gap

Hook First and Character Portrait may end on a manuscript sentence when every
beat was found; the previous assertion demanded a gap from all five and was
stricter than the design. Gates the earlier commit, which was pushed with this
one assertion red.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
"blameville · chapter-005 · author_named_concept" showed the author an internal
chapter id and a source key. Rows now read "blameville · Chapter 5: Blame · a
term the author coined": chapter titles come from the parsed manuscript (with
a readable fallback), and each provenance key has a plain-English label.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
The header had no hierarchy: logo, Library, title, save state and four buttons
at one weight on one row, with a second row that said the genre twice. It is
now one row. The manuscript's title leads at 20px Lora with its eyebrow; the
reading context and counts sit under it as a meta line; the mark is the brand
and Library a quiet link beside it; Save is the one filled action, Intelligence
the outlined one, and the two menus are text. The header is a plane over the
page, with one low, wide shadow, rather than a bevelled block. The status pill
no longer repeats a genre the select already shows.

Also fixes a reference to renderAll's local activeGenre from updateScoresOnly.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
…ll's activeGenre

The previous commit's message claimed this fix; the edit had not applied.
activeGenre is a const local to renderAll, so the reanalysis loop's status
update referenced an out-of-scope name. The chosen genre is now read from the
selects in place.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
@vercel

vercel Bot commented Sep 21, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
manuscriptlens Ready Ready Preview Sep 21, 2026 2:40am UTC

@mekado11
mekado11 merged commit 7a99633 into gh-pages Sep 21, 2026
5 checks passed

This branch was successfully deployed

1 active deployment
Preview — 942d801b Deployed Sep 21, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants