River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

House Style Guide Template

Style guide templates hand you categories to fill in. This gives each entry a compliance rate from your published copy and a query count.

Free download  ·  No account needed

A house style guide template is a list of categories with blanks in it. Spelling, punctuation, capitalisation, numbers, tone, then a line asking you to record your deviations from whichever base guide you sit on. Filling it in produces a document, and the document is never joined to anything outside itself, so nobody ever learns which entries the publication follows, which it ignores, and which answer a question no writer has ever asked. This pack runs those two joins first and rewrites the guide from what they say.

It ships filled in for The Halstead Ledger, an invented regional daily on the AP Stylebook. Two hundred and fourteen entries, measured against 1,340 published pieces and 1,024,880 words by 34 writers. Overall compliance came out at 86.7 percent of 97,131 rule instances, which reads like a well-run desk. Per entry it is not. Forty-one entries had no occasion to apply them in 18 months and 14 were followed under 60 percent of the time. The worst ran at 37.4 percent across 2,946 instances, with 31 of the 34 bylines breaking it.

Then the second join, which is the one nothing else does. The desk answered 487 style queries over the same period, and collapsed they were 96 distinct questions, the same 96 things answered 487 times. Seven of the 12 most-asked had no entry at all. At the desk's own median of 14 minutes a round trip, the questions with no entry are 66.7 hours. Open the pack in River and the agent runs both joins, or download the blank sheets and work wherever you already do.

214 entries, 1,340 pieces, 487 desk queries, and the 41 entries that describe nothing

Every rate is instances followed over instances found. Every count in the second panel is computed from the two sheets either side of it.

Corpus Compliance

The Halstead Ledger, an invented regional daily on the AP Stylebook. 214 entries measured against 1,340 pieces, 1,024,880 words, 34 bylines. 168 entries are testable by pattern, 46 are judgement calls. Sorted worst first, 12 rows shown.

EntryInstancesFollowedRateBylinesWhose problem
percent spelled out, never the % symbol2,9461,10237.4%31 of 34The paper has voted
Spaced en dash, never an em dash3,4041,38840.8%29 of 34The paper has voted
more than, not over, for quantities81234442.4%27 of 34The paper has voted
Case names in italic964142.7%6 of 34Needs a ruling
theatre, not theater, for the venue27111843.5%28 of 34The paper has voted
toward, not towards48621444.0%26 of 34The paper has voted
council member, not councilman2149644.9%4 of 34A conversation
Vote tallies written out, four to three29113345.7%3 of 34A conversation
Hyphenate compound modifiers before a noun1,9621,10456.3%12 of 34Needs a ruling
Percentages to one decimal place, never two30221470.9%not countedAbove the line
said, not stated or claimed6,2205,84794.0%not countedWorking
email, not e-mail388388100%not countedSettled
Telegram and cable address format00no instancen/aDelete

The byline column is what separates the top three rows from rows seven and eight. All five are followed less than half the time. The first three are broken by most of the desk, which means the entry is wrong. The other two are broken by four writers and three writers, both on one beat, which is a conversation with seven people and not a style question at all.

What the two joins found

Nothing below is available from a filled-in template. Both halves are counts against something outside the guide.

1. The aggregate, and why it is useless

MeasureValue
Rule instances found in the corpus97,131
Instances following the sheet84,258
Overall compliance86.7%

2. The same 168 entries, each on its own rate

BandEntriesWhat it means
No instance in 18 months41Answers nothing, describes nothing
Under 50 percent9The paper does the opposite more often than not
50 to 59 percent5A coin flip with extra steps
60 to 79 percent9Known, inconsistently applied, usually a wording problem
80 to 94 percent27Working, with a few writers to talk to
95 percent or better77Settled. Leave alone
Total testable168Plus 46 judgement calls no pattern can test

3. The 14 unenforced entries, split by byline spread

Bylines breaking itEntriesThe action
More than 25 of 345Change or delete the entry
6 to 254Rule, with reasoning written down
5 or fewer5Talk to those writers. Four of the five are one beat

4. The query log, ranked

The question as writers asked itAskedEntry
Which pronoun when somebody has not stated one?34none
Percent symbol or the word, in a chart or a headline?2937.4% entry
En dash or em dash in a number range?2640.8% entry
Do we name the school when a minor is charged?22none
What do we call the county transit authority?21none
over or more than, for a number?1942.4% entry
Dollar figures in a headline: M or spelled out?17none
Is Ward capitalised when a number follows?16none
Top 12 questions2317 had no entry
All 96 distinct questions48758 had no entry

The five most-asked questions that did have an entry are the five worst-complying entries on the sheet. Writers were not failing to look things up. They looked, saw that the entry and the paper disagreed, and asked.

5. What the round trips cost, at a 14 minute median

QueriesCountHours
All queries, 18 months487113.6
The 12 most-asked questions23153.9
Questions with no entry at all28666.7

6. The rebuild

StepChangeEntries
Starting sheet214
Delete: no instance, no query-41173
Drop as preference: unenforced, not worth enforcing-8165
Add: rulings for questions with no entry+58223
Net+9223

Six of the 14 unenforced entries were kept and enforced rather than dropped, in every case because the misses sat with a handful of writers on one beat. The guide came out nine entries longer, and every entry in it now either answers a question somebody asked or describes what the paper does.

Style Decision Register

The reason column is the point. 18 of the 38 questions that already had an entry were asked again anyway, and 11 of those entries recorded no reasoning. Six of 18 rulings shown.

QuestionAskedRulingBase guideThe reason
Which pronoun when somebody has not stated one?34Use theirs. If unknown and unreachable, the name and no pronounDeviatesGuessing is a factual error about a person and it is the kind readers remember. The name costs one word
Percent symbol or the word?29Symbol in charts, decks, headlines, tables. Spelled out in body copyFollowsThe old entry banned the symbol everywhere, which the graphics desk could not comply with and did not
En dash or em dash in a range?26Spaced em dash everywhere, ranges includedFollowsThe 2019 entry described a typesetting system replaced in 2021. Writers were being asked to fight production
over or more than, for a number?19Either. The entry is deletedFollows27 of 34 bylines broke it and not one ever asked about it. The base guide dropped the restriction; the house sheet did not
Obituary courtesy titles5None, obituaries included, except inside a direct quoteFollowsEvery one of the 314 misses is an obituary. Ruling either way was fine. Leaving it unresolved was not
Vote tallies in figures or words0Figures, unspacedFollowsNobody asked, 31 of 34 bylines already use figures, and the base guide agrees. The entry was the only thing holding words up

The last row is the only ruling on the register with no query behind it, and it exists because the corpus pass surfaced an entry the paper had already abandoned. The obituary row was reopened three weeks later by the obituaries editor with a family-request carve-out, and both decisions sit on the same row, which is what a register is for.

What's in the pack

01

Corpus Compliance sheet

One row per testable entry, with its own compliance rate, what you counted as an instance, and the band it falls in. Sorted worst first, because the bottom of that list is the actual guide and the top is furniture.

02

The byline column

The count of distinct writers who broke each entry. A rule 29 of 34 bylines break and a rule 2 break are the same compliance rate and opposite problems, and no style sheet has a column for telling them apart.

03

Recurring Query Log

Every style question the desk received, collapsed to distinct questions and ranked by how many times each was asked, joined to the entry that answers it or to nothing. The ranking is what decides where the rewrite starts.

04

The round-trip cost

Queries priced at the desk's own median answer time, split by whether an entry existed. In the worked example the unanswered questions alone are 66.7 hours, which is the argument for spending an afternoon writing 58 rulings.

05

Style Decision Register

One row per ruling: the question as writers phrase it, an example done right and one of the thing it rules out, the base guide's position, whether this deviates from it, the reason, and the date.

06

House Style Guide, stating only the departures

The rebuilt guide, carrying only the places you leave the base guide and the cases it does not cover. Every entry in it has either a compliance rate or a query count behind it, and the dead ones are gone.

07

Query Format and Onboarding Note

How to ask, and the one page a new writer reads instead of 223 entries, built from the most-asked questions rather than from what feels important. Both name what is out of scope so questions land in the right place.

08

A space rule every prompt reads first

Measure per entry and never in aggregate, treat zero instances as a finding, spread the misses across bylines before deciding anything, and record a reason on every ruling. Facts belong to the fact-check pass, not here.

How to use it

  1. 1

    Open in River, or download it

    Open the pack and let the agent run both joins, or download the blank sheets and docs and work in your own tools. The template library holds the rest.

  2. 2

    Send the corpus and the query log

    The sheet on its own can only be tidied. Send published copy with bylines attached, and any record of the questions the desk gets asked, which is the more valuable of the two and the one most desks throw away.

  3. 3

    Measure before you edit anything

    Split the sheet into what a pattern can test and what needs a human, then give every testable entry its own rate. Bands and byline spread come next. No recommendations until both are in.

  4. 4

    Then rule from the top of the query log down

    Write a ruling for the most-asked question with no entry, then the next, with a reason on every row. Stop when the tail goes quiet rather than when the categories are full.

Frequently asked questions

Is this template free?

Yes. Download the whole pack as Word documents and CSV sheets, no signup and no credit card. "Edit with AI" is a separate, optional path for anyone who wants the agent to run the two joins and draft the rulings. Nothing happens until you send it a sheet and some copy.

How is this different from a house style guide template?

A template gives you categories and you fill them from memory. This measures the sheet you already have against the copy you already published, entry by entry, and against the questions your desk was actually asked. The GOV.UK style guide conditions some of its own entries on evidence such as search term data. Most house sheets condition nothing.

What if we have no record of the style questions we get asked?

Then you run half the mechanism and the pack says so rather than presenting the corpus pass as the whole answer. Start logging from today. A desk with no query log is deciding what its guide covers by taste, and the log is cheap to keep.

Does a longer guide not defeat the point?

The worked rebuild deleted 41 dead entries, dropped 8 more, and added 58 rulings for questions that had no answer, so the guide came out nine entries longer at 223. Length was never the problem. Entries with no evidence behind them were.

Why does every ruling need a reason written down?

Because a ruling without one is a preference, and a preference gets argued with. Eighteen of the 38 questions that already had an entry were asked again anyway, 11 of them where the entry recorded no reasoning. AP publishes past answers as a searchable archive for the same reason: Ask the Editor holds thousands of them.

Should a house sheet ever deviate from its base guide?

Often, and the useful discipline is stating each departure and dating it. Base guides move: the 58th edition of the AP Stylebook carried more than 200 new and revised entries, with more added online through the year. An undated house sheet drifts against its base silently.

Where does this sit in the rest of the desk's work?

Late. Structure is settled first, by the developmental edit letter, and assertions by the fact-check pass, which grades claims against source material rather than prose against a sheet. Records work sits earlier still, in the FOIA request pack and document dump triage. Deadlines belong to the editorial calendar, where the copyedit books 1.5 hours a story.

Find out which of your rules are real

Download the blank compliance sheet, query log and decision register as Word and CSV files, or open this exact pack in River and let the agent run both joins.

Edit with AI