Repository navigation
Claude/fiction analysis tool vb1p h - #61
Merged
Merged
Conversation
…ction rebuilt the same way
The Plot Structure card showed 80/100 over forty raw percentages while admitting it did not
measure stakes or causality. That is engine telemetry with a verdict on top. Both are gone.
Engine
- Narrative Structure (fiction) and Argument Structure (nonfiction) return score null. The
overall renormalises around them; the gauge caption lists only scored dimensions.
- Findings are built from the manuscript's own statistics: a unit is "unusually scene-heavy"
or "almost entirely summary" when it sits 2.5 MADs and at least 12 points from the median;
"scene-driven writing falls away near the ending" when the final third averages 70% or
less of the rest; an early peak when the most scene-heavy unit is in the first third; a
flat book when the range is under 5 points; chapter-length outliers against the median;
cast persistence from the same character analysis the cast tab uses. Each finding names
the unit, states the measurement, and says why an editor would look there.
- Nonfiction reads the same skeleton for an argument: illustration (example, story, scene)
against exposition per unit, the unit with the fewest paragraph-level signposts when the
book signposts at all, and whether its key terms run through most of it. The phrase
lexicon ("this book argues", "in conclusion") no longer decides anything; the self-help
model keeps its own signals untouched.
- Units carry their heading line and first words so the editor can be scrolled to them;
titles set in capitals on their own line are recognised as headings.
Author-facing panel
- One curve drawn from the units (smoothed line, one dot per unit with its value on
hover, flagged units emphasised, the ending drop-off shaded), no act labels: nothing
measures "midpoint" or "climax", so nothing claims one.
- Findings as cards with "Inspect Chapter N →", which opens the manuscript at that chapter.
- A collapsed table of every chapter with its length, share and note, each with Inspect.
- The workbench card reads "described, not scored" and carries the first finding; its
detail panel shows the findings with the same Inspect links.
- A closing line says what is measured and what is not.
Tests cover the ordering of findings on synthetic books, capitalised headings, segment and
scene-break fallbacks, the nonfiction model without a thesis lexicon, quiet output when
nothing stands out, the pipeline wiring, and the browser flow through Inspect.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
…e author has the last word Three chapter headings used to make any text a "Full Manuscript", so an eight-chapter sample was judged as a whole book: climax placement, an ending drop-off, book-level DNF. The detector now has four modes. Excerpt under 1,500 words; Chapter for one unit; Partial manuscript for several chapters that are not established as complete; Full Manuscript only when there is a closing marker (The End, Epilogue, Afterword, Acknowledgments, About the author) after two or more chapters, or 25,000 words or more, or the author says so. Chapter numbering that starts past one is partial at any length. Roman and spelled-out chapter numbers and capitalised titles are read. The result carries its reasons, the chapter range, and a wholeBook flag. The structure findings that only make sense for a whole book (where the most scene-heavy unit sits, whether the ending thins) wait for book mode and say why they are not assessed, with the way to lift the gate. Blurbs already required book mode or 5,000 words. Header: a "manuscript type" select beside the genre, Detected by default, remembered per manuscript, with the detector's reasons as its tooltip. Every analysis path, including the worker, carries the override. The context row wraps on phones so the two selects fit. Tests: partial vs full on the same eight chapters, closing marker, novel length, author override, numbering past one, roman numerals, the structure gate lifting under the override; the browser flow confirms a short complete book and a partial novel sample. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VWzYm8L13Mra3ihjhuCM2U
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.