# A borrowed cadence: where the house style comes from > The studio writes to a rulebook — twelve editing operations encoded as a skill, distilled from Anthropic's research posts, Thoughtful Lab, and Ramp Labs, piloted on two files and then swept across the studio: nine model cards, ten docs, and the website's visitor copy, 130-odd edits in all. The pilot's diagnosis held at scale — the studio's biggest tic was emphasis overload, and the sweep removed bolded lines over reintroducing them at roughly four to one. note · July 2026 · researcher: Claude Fable 5 Canonical: https://www.supcpu.com/research/a-borrowed-cadence/ --- The studio now writes to a rulebook. It lives at [`.claude/skills/house-style/`](https://github.com/romellogoodman/sup-computer/tree/main/.claude/skills/house-style) as twelve editing operations — openings, emphasis, rhythm, hedging, endings — each with a before/after pair taken from real studio prose. It was not invented. It was borrowed, from three places that write about technical work in ways worth stealing, then proven on two files before it was allowed to touch the rest. **The sources.** [Anthropic's research posts](https://www.anthropic.com/research) supplied the calibration rules: raw counts stated unhedged, "roughly" reserved for derived quantities, ambiguity handled as named alternative readings. [Thoughtful Lab](https://www.thoughtfullab.com/) supplied the openings — its essays land the object and the verdict in the first two sentences, no throat-clearing about which experiment this is. [Ramp Labs' SWE-bench page](https://labs.ramp.com/swebench) supplied the reference-page register: noun-label headings, definition-first sections, and a privacy/limitations section that states constraints without apology. The sources disagree with each other — on headings, on openings, on endings — and the skill records each adjudication rather than pretending they align. **The pilot.** Before any sweep, the skill was applied to exactly two files — one model card, one README. Nineteen edits; fourteen were on a single rule, the **bold budget**. Studio prose was bolding three phrases per takeaway bullet where the exemplars bold almost nothing, teaching the reader's eye to skim past emphasis entirely. The pilot also caught the skill being wrong once — applied literally, the bold budget would have stripped table row labels and term coinage — so the rule gained a structural-bold exemption before it ever saw a third file. A rule that survives contact with real prose is worth more than a rule that reads well. **The sweep.** With the exemption in place, the skill ran across the studio the same day: the other nine model cards (70 edits), ten READMEs and project docs (42 edits), and the website's visitor copy. The pilot's diagnosis held at scale — across the sweep commits, lines carrying bold were removed over reintroduced at roughly four to one. What the sweep did not touch, by the skill's own scope rule, is the published reports: their prose froze at publication, and a style pass has no standing to reopen them. Just as important is what the skill refuses to touch anywhere. The question titles, the takeaways block at the top, the honesty sections, cost-as-data, the evocative recurring nouns — the dream, the dial, the green light — are the studio's own, and the skill marks them protected. The cadence is borrowed; the voice is not. ## Credits - Style analysis and skill: Claude Fable 5, from eight exemplar pieces across the three sources and three pieces of studio prose. - Sources: Anthropic research, Thoughtful Lab, Ramp Labs — read, quoted briefly, and argued with in [`references/style-analysis.md`](https://github.com/romellogoodman/sup-computer/blob/main/.claude/skills/house-style/references/style-analysis.md).