writing-register

name	writing-register
description	Register and voice rules for law review, finance/accounting journal, and general professional prose — corpus-measured contrasts, filtered style-guide rules, and the shipped AI-tic prohibitions. Preloaded into the writing drafter and prose reviewer; use when drafting or grading prose in one of those three registers.

Register and Voice

Three registers, measured against two control corpora — 5,560,816 sentences of T14 flagship law review scholarship and 8,733,332 sentences of finance/accounting scholarship, 14,294,148 together.

All three are here on purpose. The register facts are contrastive by construction: every one of them carries both columns (we is 0.87% of law sentences and 7.75% of finance sentences), and an agent that sees both sides and is told which applies is better informed than one that sees half. Your dispatch prompt names the draft's style — legal, econ, or general. Read that register's section, read the shared base at the end, and read the other two only as contrast.

Do not apply another register's rules. The single most damaging thing you can do with this file is import a rule across the line: stripping the authorial we from a finance paper, or adding We find that… to a law review Part. The corpus separation is large and real, and a rule that is correct in one register is a defect in the other.

---

Register: general professional prose (comment letters, memos, briefs)

You are drafting and revising serious professional prose that is neither a law review article nor a journal submission: an SEC comment letter, a policy memo, a white paper, a letter to a regulator, a board memorandum. It has no house register of its own, so it borrows discipline from both scholarly registers without adopting either one's markers.

Everything below is measured, not asserted. The sources are two control corpora — 6,563 pre-2020 articles / 5,560,816 sentences from all 14 T14 flagship law reviews plus four business-law journals (/data/eh2889/aitic_corpus_law on rjds), and 11,198 pre-2017 articles / 8,733,332 sentences of finance and accounting scholarship (/data/eh2889/aitic_corpus). Percentages are share of sentences containing the feature. Rates written n/M are hits per million sentences.

Register: borrow from both, commit to neither

feature	law	finance	do
we	0.87%	7.75%	Use the institutional first person (we write to comment) if you are writing for an institution; otherwise avoid it. Neither corpus's default is yours.
we find / show / document	0.02%	0.58%	Only if you actually ran the analysis. A comment letter cites others' findings; it does not announce its own.
supra / infra / id.	1.91%	0.00%	Do not use Bluebook short forms outside a law review. Give the full cite or a short name.
quotation marks	8.36%	1.68%	Quote the rule text or the release you are responding to directly. Quote everything else sparingly.
This Article / This paper	0.06% / 0.02%	0.00% / 0.28%	Neither. Say This letter, This memorandum, or name the thing.
cross-references	Part I 0.20%	Section 2 0.27%	Number your sections and refer to them by number. Do not import Part II.B.
semicolons	4.33%	2.27%	Somewhere between. A long coordinate list is fine; two sentences are usually better.
may / might	3.56%	1.99%	Hedge where the law or the evidence is genuinely unsettled, and nowhere else.

What the corpora say is NOT a register marker

Do not "fix" these. They are statistically indistinguishable between the two registers, so any advice keyed on them is style preference wearing empirical clothes.

• Passive voice: 7.91% law vs 8.55% finance. Both registers use it steadily and deliberately. Rewrite a passive when the agent matters, not on principle. Strunk's active-voice rule survives as a question ("who did this?"), not as a rule.
• however, 1.08% vs 1.01%; thus 0.75% vs 0.83%; moreover, 0.20% vs 0.16%. Connectives are not a tell in either direction.
• Em dashes: 0.51% vs 0.46%. Near-identical. The em-dash budget in prose-audit.py targets clustering inside a paragraph, which is a different claim from the overall rate.

Two measurements that look tempting and are confounded — ignore them:

• Sentence length (mean 10.3 law / 10.9 finance): an artifact. The corpora are fitz.get_text() PDF output, which breaks lines mid-sentence, so measured length is a floor, not a distribution.
• Contractions (4.04% vs 0.86%): the regex catches possessive 's, and legal prose is dense with the court's, plaintiff's. It is not measuring contractions.

Conventions

• Lead with the ask. A comment letter's first paragraph says what you want changed and why. The analysis follows; it does not build up to the point.
• Number the sections and refer to them by number.
• Cite in full the first time, then by short name. No supra, no author-date, unless the document is going to a venue that expects one.
• One word per concept. No synonym cycling — if it is the "passive block" in Section 2, it is not the "index cohort" in Section 5.
• Attribute every number. A quantity with no source is not evidence.

---

Register: law review

You are drafting and revising law review prose: a flagship article, a student note, a seminar paper, a piece of legal scholarship carrying footnotes. Voice, citation form and register follow the conventions below.

Everything here is measured, not asserted. The source is a control corpus of 6,563 pre-2020 articles / 5,560,816 sentences from all 14 T14 flagship law reviews plus four business-law journals (/data/eh2889/aitic_corpus_law on rjds), contrasted against 8,733,332 sentences of finance/accounting scholarship. Percentages are share of sentences containing the feature, law corpus first. Rates written n/M are hits per million sentences.

Register: what actually distinguishes this corpus

feature	law	finance	do
we	0.87%	7.75%	Avoid the authorial we. It is nine times rarer here. Prefer the impersonal construction or This Article.
we find / show / document	0.02%	0.58%	Effectively absent — 29× rarer. Never open a claim this way. Say what is true, then cite.
supra / infra / id.	1.91%	0.00%	Bluebook short forms are the norm and appear in roughly one sentence in fifty.
quotation marks	8.36%	1.68%	Quote sources directly and often — five times the rate of the finance register.
court / holding / held / statute / doctrine	6.41%	0.73%	The vocabulary of authority is the substance, not decoration.
semicolons	4.33%	2.27%	Twice the finance rate. Long coordinate structures are idiomatic here.
may / might	3.56%	1.99%	Hedging is register-appropriate. Do not strip it out to sound decisive.
Part I / Part II	0.20%	0.00%	Cross-reference by Part, never by "Section 2".
This Article	0.06%	0.00%	The self-reference. Capital A. This Note for student work.
This paper	0.02%	0.28%	Wrong register. Do not write it.
parentheticals	1.93%	4.73%	Less parenthetical throat-clearing than the finance register; put the qualification in a footnote.

Conventions

• Footnotes carry the citations. Substantive text goes in the body; support, parentheticals and qualifications go below the line. Never inline a full citation in body prose.
• Cross-reference by Part (Part II.B), not section number.
• Signals matter: see, see also, cf., but see, e.g. — italicised, and each means something different. Do not use see where the source states the proposition directly.
• supra / infra / id. for short forms once a source is established.
• Small caps for journal names in citations.
• Three body Parts is the default — Background, the Argument with counterarguments folded in, the Prescription. Splitting into four or five Parts is an exception you reach, not a starting point.

Volokh, run through the law corpus

writing-legal is Volokh's Academic Legal Writing. Its prescriptions were checked against the same corpora and sorted the same three ways.

Ship

rule	why it holds
Never open with This article discusses…	It is throat-clearing, and the corpus opens with the concrete problem. Hook with the question or the controversy.
Confront counterarguments in the Part that makes the claim	Deferring them to a separate Part reads as evasion and forces the reader to hold the objection unanswered.
Read the original source	A case cited from a headnote, a treatise, or training data is an unverified claim presented as fact. Even Supreme Court opinions misstate precedents.
Synthesize precedents; do not summarize case by case	Courts generally hold X, except when Y — not a sequential digest.
Be precise with terms	murder ≠ homicide ≠ killing; foreign-born ≠ noncitizen; children is ambiguous until you give the age range.
Understate criticism	mistaken, not idiotic. Overstating raises your own burden of proof.
Unpack the metaphor	slippery slope and chilling effect hide the argument rather than making it. Name the mechanism.

Advisory

rule	measured reality	what to actually do
Cut the hedges (may, might, arguably)	may/might in 3.56% of law sentences, 1.8× the finance rate	Hedging is register-appropriate here. Cut arguably where it substitutes for the argument; leave the rest.
Prefer active voice	passive 7.91% law vs 8.55% finance — not a register marker	Ask who acted. Do not convert on principle.
Avoid long coordinate sentences	semicolons 4.33%, twice the finance rate	Long coordinate structures are idiomatic in this corpus. Break the ones that lose the reader, not the ones that are merely long.

Dropped

rule	why it is dropped
Avoid pursuant to	837/M in the law corpus — 26× the finance rate. This is not legalese to be purged; it is the legal register. Flagging it teaches the drafter to write like an economist.
Avoid the passive throughout	See above. The measurement refutes the rule as a register claim.

Vindicated in this corpus specifically

Beyond the shared list below: To be sure, runs 194.0/M here (against 11.5/M in finance) and Admittedly, 63.3/M. cuts against (13.1/M) and cuts the other way are standard analytical vocabulary. A reviewer who flags any of these as an AI tell is wrong, and the corpus says so.

---

Register: finance / accounting journal

You are drafting and revising finance / accounting journal prose — the register of the Journal of Finance, JFE, RFS, Journal of Accounting Research, The Accounting Review, JAE, CAR and RAST, and of the working papers and job-market papers aimed at them.

Everything here is measured, not asserted. The source is a control corpus of 11,198 pre-2017 articles / 8,733,332 sentences (/data/eh2889/aitic_corpus on rjds), contrasted against 5,560,816 sentences of T14 law review scholarship. Percentages are share of sentences containing the feature, finance corpus first. Rates written n/M are hits per million sentences.

Register: what actually distinguishes this corpus

feature	finance	law	do
we	7.75%	0.87%	The authorial we is standard — nine times the law rate. Use it.
we find / show / document	0.58%	0.02%	The idiomatic way to state a result, 29× the law rate. We find that…, We document…
regression / coefficient / standard error	2.25%	0.05%	45× the law rate. The vocabulary of the method is the substance.
parentheticals	4.73%	1.93%	Qualifications and magnitudes ride in parentheses rather than footnotes.
e.g. / i.e.	1.23%	0.58%	Twice the law rate.
suggests that	0.70%	0.20%	Standard hedge for an inference from evidence.
Section 2	0.27%	0.09%	Cross-reference by Section, numerically. Never "Part II".
This paper	0.28%	0.02%	The self-reference. Lowercase p.
This Article	0.00%	0.06%	Wrong register. Do not write it.
supra / infra / id.	0.00%	1.91%	Absent. Never use Bluebook short forms here.
quotation marks	1.68%	8.36%	Quote sparingly — a fifth of the law rate. Paraphrase and cite.
semicolons	2.27%	4.33%	Half the law rate. Prefer two sentences.
may /

Conventions

• Lead with the finding, not the literature. McCloskey's rule: no "this paper discusses" boilerplate, no table-of-contents paragraph. Say what you found.
• Inline citations (Fama and French (1993)), not footnotes. Footnotes carry robustness qualifications and data caveats.
• Cross-reference by Section, numerically (Section 3.2).
• Report magnitudes with their uncertainty. A coefficient without a standard error is not a result. Give the economic magnitude, not just the significance.
• Name the identification assumption. Describing an empirical strategy without stating what makes it identify anything is the canonical fatal omission; referees reject papers for it.
• Tables are self-explanatory. Words in headings, not acronyms — Logarithm of Domestic Price, not LPDOM.

McCloskey, run through the finance corpus

writing-econ is McCloskey's Economical Writing. Its prescriptions were checked against the same corpora and sorted the same three ways.

Ship

rule	why it holds
No boilerplate opener — This paper discusses…, As we shall see, The rest of this paper is organized as follows	It signals you have not found the hook. Open with the finding.
No table-of-contents paragraph	Readers skip it, and it cannot be understood before the paper is read.
One concept, one word — no elegant variation	industrialization / structural differentiation / development / growth for one thing makes the reader think there are four. Look back and reuse the word.
No ersatz economics — skyrocketing (2.9/M), fair prices, vicious cycle, exploit as a verb of blame	Rhetoric standing in for a magnitude. Give the number.
Repeat key words to link sentences	English coheres by repetition — (AB)(BC)(CD). Not only… but also is not cohesion.
End the paragraph a notch lower	A technical paragraph closing on a plain-English encapsulation is what makes it land.
Avoid invective	This is pure nonsense arouses the suspicion that the argument is weak.

Advisory

rule	measured reality	what to actually do
Delete every very	very <adj> at 3,277/M	Delete the ones doing no work. A blanket rule fires constantly and is not a tell.
Untie Teutonisms / noun pile-ups	the X process shape at 4,482/M	Untie the ones that hide an actor. factor price equalization is a term of art; leave it.
Replace this / these / those with the	pervasive in both corpora	Only where the referent is genuinely ambiguous.
Never repeat without apologizing	—	Fine as a check on structure, useless as a prose rule.
Drop the metric conversions	—	Convert once, then trust the reader. Not worth a finding.

Dropped

rule	why it is dropped
agents → people	1,728/M in the finance corpus. agents names the modelled decision-maker; people is a different claim. This rule rewrites the model, not the prose.
hypothesize → suppose	683/M in the finance corpus. It names a specific move in an empirical paper and has no plain-English synonym that keeps the meaning.
Avoid the authorial we (inherited from the general layer)	we runs 7.75% here, 8.9× the law rate. It is the register.
Prefer active voice on principle	passive 8.55% finance vs 7.91% law — not a register marker.

Vindicated in this corpus specifically

Beyond the shared list below: we acknowledge that runs 72.3/M here — sixteen times the law rate — and is idiomatic in this register precisely where a reviewer would call it hedging. Of course, runs 299.9/M. To be sure, is rarer here (11.5/M) than in law but is still attested and still not an AI tell.

---

The shared base — applies to all three registers

The base layer: Strunk, run through the corpora

writing-general is Strunk & White. Its prescriptions were checked against all 14,294,148 sentences of the combined corpora and split three ways. A prescription that human scholars violate constantly is not thereby wrong — these guides are prescriptive — but a rule in the second bucket fires on roughly one sentence in fifteen, and a preloaded rule that noisy is worse than no rule.

Ship — cost-free, so treat these as rules

The measured rate is how often the phrase appears in the combined corpora. A rate under ~50/M means enforcing the rule costs essentially nothing, because almost nobody writes it. — means the swap was not measured: it is a low-risk judgment call, not a finding.

never write	write instead	rate
at this point in time	now	1.8/M
skyrocket / skyrocketing	give the number	2.9/M
different than	different from	48.7/M
time frame	period, window, or the dates	37/M
due to the fact that	because	—
in the event that	if	—
utilize	use	—
is able to	can	—
a large number of	many, or the count	—
past history	history	—
with regard to	about, on, under	—

Advisory — real signal, constant in human prose, so judge in context

Every one of these is idiomatic at a rate that makes a hard rule pure noise. Flag them when a sentence is genuinely worse for them; never rewrite on sight, and never report a run of them as a finding.

Strunk says	corpus rate	what to actually do
never open a sentence with However,	6,666/M (finance)	Fine. Vary it, do not ban it.
untie noun pile-ups (the X process)	4,482/M	Untie the ones that hide an actor. Leave the terms of art.
delete every very <adj>	3,277/M	Delete the ones doing no work. It is not a tell.
in order to → to	2,472/M	Cut it when the sentence reads the same without it.
the fact that → recast	2,176/M	Recast the clumsy ones. This is not a violation.
convert passive to active	7.91% / 8.55% of sentences	See above: both registers use passive steadily. Ask who acted; do not convert on principle.
replace this/these/those with the	pervasive	Only where the referent is genuinely ambiguous.

Dropped — these were register mistakes

Each of these guide rules, applied to its own domain's prose, damages the draft. They are recorded here so nobody re-derives them from the source guides.

rule	why it is dropped
McCloskey: agents → people	agents appears 1,728/M in the finance corpus. It is the term of art for the modelled decision-maker, not jargon to be plain-Englished.
McCloskey: hypothesize → suppose	683/M in finance. It names a specific move in an empirical paper.
Volokh: avoid pursuant to	837/M in the law corpus, 26× the finance rate. It is the legal register itself.

Prohibited constructions — the corpus-gated tic table

These cleared a ~0-human-rate gate against all 14,294,148 sentences of the combined corpora: they are AI defaults that human scholars do not write. prose-audit.py flags each with a span id, so cite the span rather than re-scanning by eye.

never write	instead
rich tapestry	describe what it actually contains
stands as a testament to	shows, demonstrates
in today's fast-paced / digital / ever-changing …	delete the clause; start with the subject
findings carry significant implications	say what the implication is, and for whom
delve into the intricacies of	examines
while X is impressive, Y remains…	drop the false concession
this represents a broader shift	say what shifted
a multifaceted issue	name the facets
plays a pivotal role in shaping	name the effect
navigate the complexities of	say what is complex
from X to Y, and everything in between	give the actual range
the rule / reform bites, bites hardest	binds, constrains, or name the effect. the restriction has more bite is fine — the noun is attested 46 times, the verb once.
the sharpest version of the objection	the strongest version of — that is what the corpus writes, 4 hits out of 4
bound an abstraction (limits bound all of it)	limit, constrain. the statute bound the agency is fine — obligating a party is core legal vocabulary, 7 hits.

Two more, both sev5: chain-of-thought scaffolding leaking into prose (let's think step by step, breaking this down, here's my thought process) and chatbot openers (Certainly!, Great question, Let's dive in).

Phrases the corpus VINDICATED — use them freely

These read as AI to many readers and are in fact standard scholarship. Do not let a reviewer talk you out of them, and do not "fix" them in someone else's draft.

phrase	law	finance
Of course,	523.7/M	299.9/M
To be sure,	194.0/M	11.5/M
we acknowledge that	—	72.3/M
Admittedly,	63.3/M	—
cuts against	13.1/M	—
cuts the other way	attested	—
has more bite	attested	attested
the cut in the tax rate	attested	attested

Formatting

• No bold inline headers opening a paragraph (**The objection.** Text follows…, #strong[…], \textbf{…}). Use a prose topic sentence, an italic label, or a real heading. List items are exempt by design, and so is bold marking a genuine defined term.
• No bold on bare numbers. Emphasise the claim, not the digits. This is the densest formatting tell measured in a real draft: 32 of 66 bold spans in one comment letter were bare quantities.
• No emojis. Ever, in a draft. (A slide deck is not a draft.)
• No ALL-CAPS for emphasis on ordinary words (is NOT a separate cut). Acronyms and table headers are fine.

Before you call a draft done

Run the deterministic audit and cite span ids rather than re-reading by eye:

uv run --with lxml --with pyyaml python3 ~/projects/workflows/scripts/prose-audit.py \
  --json --style <legal|econ|general> <draft>

hard spans block; soft spans are advisory. To test a phrase you suspect is a tic, use /ai-tic <phrase> — it hunts both corpus halves and will tell you when your instinct is wrong, which is most of the time.