Vol. I · No. 1 · MMXXVI
NextChapterBench A Benchmark of Literary Continuation
23 models · 10 items · 1 judge

The Standings → Claude Sonnet 5 (medium)

Live run data

Anthropic

Claude Sonnet 5 (medium)

medium reasoning effort · via Claude CLI · judged by Claude Sonnet 5 · not the setting shown on the standings; also run at high60.8 xhigh60.4 low55.9

Results from run ‘first’: ten items, ten verdicts.

Claude Sonnet 5 (medium) scores 64 on craft, level with the field’s 63 and 29 points short of the authors themselves. Its shape differs from the field’s: relatively strong on the author’s voice and sentence-level prose, relatively weak on restraint and folding the brief into the story. On the mechanical checks it kept to the brief’s word range on 7 of 10 items, reproduced a 29-word run of the real book verbatim and leans on stock phrases more than most.

Headline 60.0 of 100 · compliance × craft
Compliance 94 of 100
Craft 64 of 100

The record: 10 items · 10 verdicts · 0 errors · 2,026 words on average · est. generation cost $0.36

Read the chapters in full

§ I

Craft by Dimension

Eight dimensions, each 0–100, averaged over ten items; the author’s real chapter was in the room every time.

Craft by dimension, 0–100 · judged by Sonnet 5 beside the author’s real chapter

Voice
67
Prose
66
Character
64
Scene
65
Integration
60
Momentum
63
Restraint
60
Specificity
65
0255075100
reader’s score, 0–100 · hollow tick: field average · filled tick: the authors themselves

The authors themselves average 93 on the five writing dimensions, the field 63, this model 64. Hover or focus a bar for the field and the authors on that dimension.

In plain words, item by item

The judge’s one-sentence best quality and worst flaw for each chapter, written for a general reader.


§ II

Per-Item Results

Headline, compliance and craft for each chapter, 0–100.

The same ten items in figures, with word counts.
Item Headline Compliance Craft Words
The War of the Worlds item 006 · after ch. 1 · H. G. Wells 68.6100692,136
Leave It to Psmith item 022 · after ch. 3 · P. G. Wodehouse 66.095702,758
The Moonstone item 011 · after ch. 8 · Wilkie Collins 63.290702,618
The Secret Garden item 013 · after ch. 2 · Frances Hodgson Burnett 62.395662,115
Persuasion item 001 · after ch. 1 · Jane Austen 61.5100621,968
Red Harvest item 023 · after ch. 2 · Dashiell Hammett 59.1100591,793
The Sun Also Rises item 021 · after ch. 3 · Ernest Hemingway 57.695611,683
Dracula item 015 · after ch. 5 · Bram Stoker 57.085671,610
The Awakening item 016 · after ch. 7 · Kate Chopin 54.195571,484
The Mysterious Affair at Styles item 017 · after ch. 3 · Agatha Christie 50.890562,097

§ III

Length Against the Brief

Words written per item, against each brief’s maximum.

Word count per item vs. brief maximum · run ‘first’

words written brief maximum
The War of the Worlds
2,136
Leave It to Psmith
2,758
The Moonstone
2,618
The Secret Garden
2,115
Persuasion
1,968
Red Harvest
1,793
The Sun Also Rises
1,683
Dracula
1,610
The Awakening
1,484
The Mysterious Affair at Styles
2,097
08001,6002,4003,200
words; the ink tick marks each brief’s maximum

Seven of ten items landed inside their brief’s word range.


§ IV

Worst Flaws, as Recorded

Each verdict names the chapter’s single worst flaw, with evidence. Quoted verbatim from run ‘first’.

The narrator keeps stopping to spell out exactly what people are feeling and why ("which was that I would have minded very much, and that there was no way on earth I could tell her so"), which is the opposite of the clipped, show-don't-tell style the source material uses.

on The Sun Also Rises, item 021 · Claude Sonnet 5

It never puts Betteredge back in his beehive chair at the end as the scene requires, closing instead in an unspecified 'snug little room,' and it never accounts for where Miss Rachel has gone after Lady Verinder returns alone.

on The Moonstone, item 011 · Claude Sonnet 5

It has Mary hear a sound that 'seemed almost to be crying' and then say out loud 'I heard something crying,' which is exactly the thing the brief said must not appear on the page, even as a denial.

on The Secret Garden, item 013 · Claude Sonnet 5

The library interview is conducted by Lady Constance in a London flat, but the earlier chapters state plainly that it is Lord Emsworth, not Constance, who is in London to interview Miss Halliday.

on Leave It to Psmith, item 022 · Claude Sonnet 5

§ V

The Mechanical Docket

Automated screening results across the ten outputs.