---
name: voice-match-worksheet
description: Use for any job that asks for copy to sound like a particular team, product, publication or person — bringing a draft into a written guide, working from real messages, shifting register for a different reader, or reviewing copy that may already be fine. Works five measurements before any wording changes, because the decisions that matter here are settled by counting and comparing rather than by taste, and because a fluent rewrite will otherwise quietly settle questions belonging to the person who asked. Any industry, any channel, single messages or long documents.
---

# Voice matching

Producing text in a register is the easy part. The work is establishing **what the target actually is** and **how far the draft actually sits from it**, and both are settled by measurement rather than judgment. You are a voice editor, not a fact editor: change how something is said, never what it claims.

## What you hand over

**Every reply opens with the worksheet, and the worksheet ends in the call.** No
preamble before it, no exceptions, on the simplest job and the hardest alike:

```
**Worksheet**
claim:     <contradicted by … | consistent>
cap:       <required N vs allowed M -> TRUE / FALSE | n/a>
evidence:  <N axes differ: … | single source>
mandates:  <N broken, M preferences unmet>
hedges:    <N load-bearing, M filler>
**Call:** <ask | hold | options | split | keep | edit> — <the line above that decided it>
```

The call sits underneath the five results because that is the order it has to be
reached in. Written above them it is an impression, and the measurements then
arrive as its defence rather than its cause — which is the single commonest
failure in this work, and it is invisible, because the reply still contains every
number it was supposed to contain.

Four of the five slots take a number. Fill them — `cap 4>2 TRUE`, `evidence 3
axes differ`, `mandates 0 broken`, `hedges 2 load-bearing` — and write `n/a` for
anything the job does not raise; five `n/a`s is a legitimate worksheet and still
gets written. **Every slot takes a result, never a conclusion.**
`evidence 3 axes differ` is the slot filled; `evidence playful applies here` is
the slot skipped by something shaped like an answer, and the axes that differed
have been settled by a writer with no standing to settle them. `claim` leads
because it can end the job: where the draft contradicts what you were handed, the
call is `hold` and the other four slots stop mattering.

This holds hardest on the easy jobs. A draft that is nearly right still gets the
worksheet, the call and the compliance list; a reply opening straight into
**Rewrite** has skipped the part that made it worth trusting, and the reader
cannot tell it from a first impression that happened to sound confident.

**The route is a lookup, not a judgement.** Read the results in this order and
stop at the first line that fires:

```
a claim the brief's own context contradicts        ->  hold
the draft or any usable voice evidence is missing  ->  ask
required items  >  allowed units                   ->  options
axes that differ between sources          >= 1     ->  split
load-bearing hedges over a decision, a
  commitment, or somebody else's judgement >= 1    ->  split
  (a hedge over a fact someone could look up
   is not one of these — see below)
mandates broken                            = 0     ->  keep
anything else                                      ->  edit
```

A count of one fires a line exactly as hard as a count of five, and nothing below
the line that fired can promote or demote it. Writing `edit` under a worksheet
that logged four load-bearing hedges is not a close judgement call, it is the
table being ignored — the measurement was taken, written down, and then
overridden by the impression it existed to correct. Once the table has
spoken the route is frozen against second thoughts about taste; never against a
measurement. And what the table settles is *which route*, not *how bad the draft
is* — severity is measured separately, below, and is usually lower than the route
suggests.

Then produce only the blocks that call has earned, each under its own bold heading and in this order:

1. **Worksheet**, ending in the **Call** — the five measurements as worked
   lines. Give it no sub-headings, no tables and no fenced blocks of its own:
   five short entries, one per measurement, and `n/a` where one does not apply.
   Short does not mean bare — each entry still shows the count, the comparison or
   the deletion it made, because a worksheet compressed to its verdicts has
   thrown away the only thing it was for. It is the working, not the deliverable:
   one that grows past the rewrite it justifies has crowded out what the reader
   asked for, and on a `split` it pushes the second version off the end of the
   reply.
2. **Anchors** — the voice as named axes, each marked `*(inferred)*` where you guessed rather than read it.
3. **Audit** — findings rated **High / Medium / Low** by exposure rather than by
   how the rule was worded, each quoting the draft and naming the anchor it breaks.
4. **Rewrite** — the text as it would go, carrying inline `[changed: what; reason: which anchor + which finding]`.
5. **Flag** — earned whenever the register you were asked for pulls against what
   the draft actually says and the route stayed `edit` anyway. One sentence
   naming the pull, then the two readings a line each, labelled *tone-compliant*
   and *literal-intent*. A rewrite is allowed to be right and still leave a
   question the sender is better placed to answer than you are.
6. **Compliance** — one line per anchor: `pass` / `yellow` / `fail`, with the measured number wherever a number was specified.

**Blocks are earned, not filled in** — but what is earned is *which* blocks, never whether to open with the worksheet and its call, which every reply carries however small the job. A fixed list invites completing all of it whether the job called for it or not, which is the shortest path to a confident wrong answer: a rewrite where nothing needed rewriting, an audit padded to justify one, two versions where there was never a choice. Under `ask`, `hold` and `options` the reason leads and the rewrite is withheld. A straight rewrite runs about 250–400 words, and five findings is a long audit.

Aim the whole reply at 300–400 words whatever the call, both versions of a
`split` included, and spend it in that proportion: worksheet, anchors and audit
take about a third between them, and **the deliverable takes the rest**. When
space runs short the worksheet contracts first, then the audit; the deliverable
never does. A reply that stops mid-sentence has failed regardless of how good the
reasoning above it was, and so has one that spent its length on analysis and
handed over a stub.

**Whatever the call promises, hand over.** Under `options`, each option is
drafted as copy the reader could paste; under `split`, both versions are; under
`edit`, the rewrite is the whole message and not the changed sentences. Naming
what an option *would* say, or describing the shape a version *would* take, is
the commonest way an answer with correct reasoning ends up worth nothing — the
reader came for text and left with a plan to write some. `hold` is the one call
where withholding the copy is the point, and there you hand over the two
branches' shape instead.

## Work these four before you choose

Write each one out as arithmetic or as a comparison. **Do not report a verdict you have not shown the working for** — a verdict is a guess wearing a measurement's clothes, and every failure in this work traces back to one.

### 1. The cap against the content

Record the operator, the value, and the *measured* count. `under N` permits N−1; `at most N` permits N; `exactly N` permits neither more nor less. Count by terminal punctuation or by counting words — never by how long the text looks. Then convert both sides to the same unit: **each independently required item consumes one delivery unit.** Compare two integers.

```
cap:       "under 3 sentences"  ->  operator <, value 3, maximum 2
required:  price / dimensions / materials / returns window  = 4 items
draft:     measured 5 sentences (counted . ! ? — 5 terminal marks)
compare:   required 4  vs  allowed 2   ->  4 > 2 is TRUE
verdict:   options (deterministic)
```

The `compare:` line is not optional and it is not prose. Write both integers and
evaluate the inequality as TRUE or FALSE. *"The required items fit within the
allowed sentences"* is not that comparison — it is an impression of packing,
and it is how this measurement gets skipped while appearing to have been made.
If required > allowed is TRUE the call is `options`, however comfortably you
believe the items could be crammed in.

Count the draft by its terminal marks and write the count you actually made. A
semicolon does not end a sentence, but a full stop does: a draft of four
sentences is four whether or not any of them could have been joined.

If required exceeds allowed the conflict is real, and it is *deterministic*: clauses, semicolons and labels move neither integer, so compression cannot resolve it.

An `options` reply is three paths and a question, never one clever compression:

1. **Cap-compliant** — obeys the number and therefore *visibly drops* at least
   one required item. Name which one, in the option. An option that claims to
   drop something while relabelling it has dropped nothing.
2. **Content-complete** — carries every required item and therefore *visibly
   breaks* the cap. Say by how much: "three sentences against a cap of one".
3. **Ask the guide's owner** — which rule governs here. This is a real path, not
   a courtesy: a rule this tight was probably written for a smaller occasion
   than this message, and the person who wrote it can say so in a line.

Close with the single question the reader must answer. Do not rank the options
for them; the trade-off is theirs and they know which rule their organisation
actually enforces. Where the integers do fit, check they fit *readably* — compliance bought with clause chains or stacked labels defeats the purpose the cap was written for and is still `options`.

### 2. The evidence against itself

Where the voice comes from more than one source, write the axes for **each source on its own line** before comparing. Never read them as one.

```
sample A:  person=2nd  affect=playful  emoji=yes  length=2 sent  lead=feature
sample B:  person=3rd  affect=neutral  emoji=no   length=2 sent  lead=feature
agree:     length, lead                 -> firm anchors
differ:    person, affect, emoji        -> 3 axes with no single value
verdict:   split — one artifact would hide the disagreement
```

The example deliberately does not say "this one is a launch, so the playful
profile wins". That rule rests on one sample of each, which is a coincidence
promoted to a policy, and it turns a measurement you took correctly into a
choice that was never yours. Verdict `split` means call `split`.

**Write the `differ:` count before you write a single anchor.** The failure this
measurement exists to catch is not missing the disagreement; it is noticing it
and then dissolving it a paragraph later — *"this draft is an announcement, so
the celebratory profile applies"* — at the moment the Affect anchor has to be
given one value. By then the split has been resolved silently, in a line that
looks like inference and is really a decision. So fix the count first, and where
an axis differs, the Anchors block carries **both** values for that axis with the
source beside each, never an average and never a pick.

An axis they disagree on has **no single value**. Averaging invents a register neither writer uses, and the disagreement was the most useful thing in the evidence: it says this team writes differently for different situations, which the brief never stated. One differing axis is enough. You may not resolve the split yourself by
deciding which branch this particular message belongs to — inferring "celebratory
for launches, neutral for fixes" from one sample of each is a rule drawn from a
sample of one, and it is the specific mistake this measurement exists to catch.
Nor may the agreeing axes outvote a differing one. Carry the split forward as two
labelled versions and hand back the question of which profile governs; that
question belongs to the person who has seen more than two of their own messages.

### 3. Mandates against preferences

Classify every line of the guide before you rate anything, because the classification sets the ceiling on severity.

```
"never use 'unfortunately'"      -> mandate    (must/never/always/imperative)
"bullets preferred over prose"   -> preference (prefer/generally/where possible)
"keep it warm"                   -> preference, unmeasurable as stated
mandates broken:    0
preferences unmet:  2
verdict:  ceiling = Low  ->  keep (offer the preference edits as optional)
```

**A preference cannot produce a finding above Low, and it can never justify a rewrite.** If no mandate is broken the call is `keep`: return the draft, walk the anchors one at a time to show the check was real, and offer any preference-based polish afterwards, labelled optional. Finding little is a normal result, not a failure to look hard enough — and inflating a Low into a Medium to justify the rewrite is the commonest fault in this work, invisible because the writer never sees the draft you rejected.

A `keep` is the one call that still shows every block. Run the worksheet, write
the anchors, walk the compliance list one axis at a time and rate each: an
unexamined *"looks fine to me"* is indistinguishable from not having checked, and
it is the reader's only evidence that anything was measured. What shrinks under
`keep` is the deliverable, not the working.

Then hand back the draft, in this order and no other:

1. One line saying the draft is ready to send as written.
2. Any polish you would otherwise have applied, under a heading reading
   **Optional** and introduced as something they may decline — *"if you would
   prefer bullets, here is that version"*.

The order is the whole point. *"The only adjustment is…"*, *"here is the rewrite
with that change applied"*, or the alternative printed above the verdict have all
turned a `keep` into an `edit` that mislabelled its own call, and a reader told
the draft was fine and then handed a replacement will send the replacement. What
you produced under **Optional** is never the thing the reply builds towards, and
it is never introduced as an adjustment, an improvement or a fix — it is a
version they did not ask for and do not need.

### Severity is exposure, not rank

Rate a finding by what the reader is exposed to, not by how forcefully the guide
worded the rule it breaks. The same broken mandate is High in one draft and Low
in another.

```
"never use passive"  broken in 1 span  of an otherwise compliant draft  -> Low
"never use passive"  broken in 4 spans across every section             -> Medium
a claim the brief's own context contradicts                             -> High
a numeric mandate measurably broken                                     -> High
```

**High** is a truth risk, a measurably broken number, or a prohibited pattern
running through the whole draft. **Medium** is a pattern in several identifiable
spans, or a register that is wrong for the reader. **Low** is one span, or an
unmet preference. One or two instances of a banned construction in a draft that
passes everything else is **Low**, and the honest answer is a near-minimal edit:
change those spans, leave every other sentence exactly as the writer set it
down, and say in a line that the draft was already close. Marking them High to
license a fuller rewrite is the same inflation as above wearing a rule's
authority, and it reads to the writer as a verdict on work that was fine.

Before you write a passive-voice finding, run the test: passive is a form of
*be* — is, are, was, were, been, being — followed by a past participle.
*"was flagged by our system"* passes the test and is a finding. *"has paused
fulfillment"* is present perfect **active**: the subject is doing the pausing.
*"is on hold"* is a copula plus adjective. A construction that fails the test is
not a finding, and an audit that lists one has spent its credibility on a
grammar mistake the reader will catch.

### 4. Each hedge against its own deletion

List every hedge and vague phrase, delete it, and record what changed. This is a test with an outcome per line, not an impression.

```
"I think we should"    -> "we should"       : proposal becomes decision   -> load-bearing
"might be worth"       -> "is worth"        : option becomes recommendation -> load-bearing
"a little bit"         -> ""                : nothing changes             -> filler, cut
"around 2am"           -> "at 2am"          : estimate becomes measurement -> load-bearing
verdict:  3 load-bearing  ->  split (tone-compliant + literal-intent)
```

If only the rhythm tightens, it was filler. If what the message **commits to**,
**challenges**, or **claims to know** changes, the hedge was carrying meaning and
a fully compliant rewrite would say something the writer did not mean. Test the
third of those as carefully as the first two: a softener in front of a challenge
to work colleagues already committed to is doing social work, and deleting it
turns a question about their plan into a verdict on it. *"I'm a little worried"*
and *"probably"* look like filler and are not, wherever the sentence questions a
decision somebody else owns.

**One load-bearing hedge is enough: the call is `split`.** Not a majority of
them, not a stack — one. A worksheet that logs a load-bearing hedge and then
calls `edit` has recorded the finding and ignored it, which is worse than not
having tested. Say so, and name the specific dynamic in this message's own words — which
phrase, what it is holding up, and what removing it would cost the sender. A
sentence that would fit any draft ("the writer may wish to sound less certain")
is not a finding, it is a way of appearing to have looked.

A `split` reply is two complete versions, each sendable as it stands:

- **Tone-compliant** *(assumes …)* — the guide applied in full, the parenthesis
  naming what it takes to be true.
- **Literal-intent** *(keeps …, gives up …)* — the meaning intact, the
  parenthesis naming which part of the guide it surrenders.

Both are written out; neither is a summary of the other, and if they differ by a
couple of adjectives you have not found a tension but flinched from one. Two
versions cost length, so a `split` pays for them from the analysis: at most three
findings in the audit, one line per worksheet entry, and no anchor table. Both
versions must survive intact inside 400 words — a second version that runs off
the end of the reply was never handed over. The
literal-intent version is a rewrite too: it obeys every anchor the tension does
not touch and differs from the draft wherever a hedge tested as filler. Handing
the source back as version two is the reader's own problem, relabelled. Open the
pair with one line naming the dynamic in this message — whose commitment is being
questioned, which decision is unsettled — since "the writer may prefer a softer
tone" fits every draft ever written. Close by
naming the one fact that decides which to send. A flag above a single rewrite
that invites the reader to soften it themselves is the same miss wearing a
helpful face — they are still writing and still deciding.

One class of load-bearing hedge splits nothing: a hedge over a **knowable fact**
— an unconfirmed timestamp, a figure nobody has checked yet. There are not two
readings for the reader to choose between; the value is simply not confirmed. If
the guide asks for precision, give it precision and keep the value visibly
marked — `02:07*`, with a line saying the source did not confirm it — and stay on
`edit`. `split` is for what you have no standing to settle: a decision, a
commitment, or a judgement that is somebody else's to make.

Where the call stays `edit` because every hedge tested as filler, the audit still
records what each one was holding up, and the rewrite keeps the writer's own
stance rather than deleting it: a concern stated plainly is on-voice, a concern
removed is a different message. "Confident about the decision even when uncertain about the outcome" means state the direction firmly and keep honest uncertainty about timing and results; it is not a licence to delete every qualifier.

Two families come back load-bearing almost every time, and they are the two the
test is most often run carelessly against.

**Decision status.** *Exploring*, *considering*, *proposed*, *pending*, *subject
to*, *hoping to* describe a state of the world rather than a state of the prose.
Delete one and the message announces a decision that may not have been taken —
in public, over somebody's name, to people who will act on it. A guide asking for
confidence about decisions is asking you to state a settled direction firmly and
stay honest about timing and outcome; it is not asking you to settle the
direction on the writer's behalf. Where the brief does not say the decision is
made, assume it is not.

**Severity and scale.** *Some*, *a few*, *intermittent*, *occasional*, *may be
affected* bound how bad a thing is. Sharpening them sharpens the claim, and a
message that overstates an incident costs the sender as much as one that
understates it. The vagueness is often the only honest thing available to a
writer who does not yet know the number. A guide asking for directness will pull
hard on exactly these words — *intermittent disruptions* wants to become *is
down*, *a few accounts* wants to become *accounts* — and that pull is the
tension, not the fix. Where the register the reader asked for would upgrade a
scope word, say so in the flag and hand over both readings rather than picking
the more dramatic one because it sounds more decisive.

Add nothing forward-looking either. A next step, an owner, a time — *"we'll
follow up once it's resolved"* — is a commitment somebody now has to keep, and a
rewrite is not where commitments get made. If the register wants a closing
action and the source has none, use a marked placeholder and say you left it for
the sender to fill.

Treat hedges, passive voice, politeness and approximation as functional until this test proves otherwise. Never turn an expectation into a guarantee, a proposal into a decision, an intermittent fault into a total one, or a careful challenge into a command without saying that you did.

## The claim check

**This runs first and outranks everything below it.** Before any hedge is tested
or any count is taken, ask whether the message can be true at all. A hedge test
answers *what did this word protect*; the claim check answers *should this be
sent*, and the second question makes the first irrelevant. Where a fact conflicts,
the call is `hold` no matter what the four measurements go on to find — a
load-bearing hedge inside a message that should not go out is not a reason to
offer two versions of it.

Separately from all four: compare what the draft asserts against everything else you were given. Where the brief's own context contradicts the draft, the call is `hold`. A clumsy wrong statement invites a second look; a fluent one gets acted on, so polishing is what converts an error into a decision somebody takes.

Put the contradiction first, above every other block, and frame it as a truth risk that outranks the rewrite you were asked for — not as one finding among several. Say in one sentence what conflicts with what, and state plainly that you have not rewritten the text and why.

**Under `hold`, ship no send-ready copy in either direction.** This is the part most answers get wrong: having flagged the problem, they helpfully supply a polished version for each branch, and one of those is a finished artifact built on a premise nobody has confirmed. Somebody will paste it. Give the *shape* instead — which claim each branch turns on, what the message would need to say, what changes structurally — and close by asking for authorisation on the fact. Reversing what a message asserts is a new communication, not a voice rewrite, and it is not yours to authorise.

Qualifiers travel with the facts they qualify. Add no facts: no invented dates, names, numbers, causes or commitments, and where the message needs one, a marked placeholder such as `[DATE]`. Drop none silently either. Correcting a claim is a different job with a different authority behind it — flag, do not fix.

## Anchors

| Axis | What you pin down |
|---|---|
| Formality | Casual, neutral, professional, executive, clinical |
| Lead | What occupies sentence one — the action, decision, feature or ask |
| Length | Sentences, words, or "tight" — the hardest number available |
| Person & voice | I / we / you / third, active vs passive. Passive is sometimes *permitted* |
| Prohibited | Banned phrases, and banned *patterns*: hedge stacks, noun stacks, blame |
| Required | What must appear — a next step, a header set, an owner |
| Affect | Emotional temperature: warm, matter-of-fact, celebratory, neutral |
| Structure | Headers, bullets, tables, prose. "Structured" is a format instruction |
| Audience | Who reads it, and what they will not tolerate |

Mark what you inferred; an anchor read off a guide and one guessed from context carry different weight. Most briefs are silent on affect, and affect is what a reader feels first. Read samples structurally rather than by vibe: sentence length, what occupies the first clause, person, emoji policy, punctuation energy, technical density. "Casual and friendly" is not an anchor; "feature name first, second person, two sentences, no emoji" is.

## What the evidence is worth

A written guide states rules and is followed literally, including its permissions. Real samples show the voice in use. A description gestures at a voice rather than specifying one. **A comparison — write it like some other team, product or author — points at a voice you cannot see**, and what you recall of how they write is partial, probably years stale, and assembled from whichever of their surfaces you happened to read. Listing that brand's anchors reads authoritative and is a guess. And nothing at all arrives more often than you would expect.

Under `ask`, say plainly that a remembered voice is not a specification, ask for everything missing **together** rather than in sequence — a paragraph they consider on-target, and the draft if none came — and offer a fallback that is not a dead end: if no sample exists you will infer, and mark every anchor provisional. Keep it to a question rather than an analysis. Publish no speculative anchors while you wait, and do not pad the reply to look substantial; a short precise request *is* the deliverable.

Short is not the same as bare, and an `ask` has four parts it cannot drop:

1. **The missing inputs, numbered, in one list.** Usually the draft itself and
   one real sample. Ask for them together; a question answered in three rounds
   has cost the reader more than the rewrite would have.
2. **Why the remembered voice will not do.** Two sentences at most, and specific:
   the recollection has a cutoff, the voice has moved since, and what the asker
   means by that name may not be what you would produce under it. Say that
   anchoring to it and being wrong produces copy that is confidently off-target,
   which is worse than copy that is obviously unfinished.
3. **What a usable answer looks like**, shown rather than described — *"plain and
   specific, never salesy, always ends with the next step"* is the shape you are
   asking for, and one line of it saves the reader guessing what you meant.
4. **The fallback, as an offer they can take.** If they genuinely have no sample,
   you will work from inference, mark every anchor *(inferred, provisional)* and
   ask them to check it before anything is committed. Never close on the request
   alone; a question with no way past it reads as a refusal.

Four short paragraphs, not four pages. A request that names the gap without
showing the shape of the answer sends the reader away to guess, and the second
round arrives no more usable than the first.

## Writing the rewrite

Write it as a finished message, not a diff, then annotate every substantive change immediately after the sentence it explains:

`[changed: what you did; reason: the anchor it serves + the finding it closes]`

Both halves carry weight — `[changed: tightened wording; reason: clarity]` names no anchor and closes no finding. Keep the writer's own words wherever they already sit inside the voice; a rewrite replacing every sentence reads as a rejection of the person who wrote it and usually smuggles in a register the guide never asked for. When cutting to a budget, cut modifiers first, then clauses restating the main clause, then whole sentences whose content survives elsewhere.

Where the ask implies a shape — scannable, sectioned, a table, a technical reader who wants density — build that shape. Prose that mentions the sections is not the sections. *Executive* leads with the decision rather than the process; *neutral* strips evaluative language and makes severity words earn their place; *empathetic* leads with what can be done and puts the constraint after the path forward.

Some wording is not yours to change. Disclaimers, safety instructions, licence terms and risk statements are usually worded exactly as they are because somebody signed off on that wording; leave them, re-voice around them, and say which sentences you treated as fixed. Accessibility and plain-language requirements outrank house style wherever they collide. Where copy publishes under an individual's name, their voice has a claim the house guide does not override.

## Long documents fail at the seams

Anything long drifts — written over weeks, or by several hands, it starts in one register and ends in another, and a sentence-by-sentence audit returns a flat list of small findings while missing that the second half belongs to a different voice. Read the first and last paragraph of each section back to back, and read across every transition between hands. Resolve vocabulary before cadence: two names for one thing costs a reader more than an uneven sentence ever will.

## Read it back

Check the artifact against the call, mechanically rather than by impression.

- **Counts.** Recount the finished text. Never inherit a number from the source or from your own earlier estimate, and never certify a limit by how the result reads. State the measured number against the stated limit and say whether it holds.
- **Claims.** Every fact, commitment and qualifier from the source is still present and still attached to what it qualified. Nothing precise appeared where something approximate stood.
- **Scope.** Every change traces to a logged finding. Anything else goes back, whatever its merits.
- **The call against the worksheet.** They agree. Any load-bearing hedge, any
  differing axis, any required-over-allowed overflow means the call reads `split`
  or `options` — if it does not, the call is wrong, not the worksheet.
- **The call.** The artifact still is what you said it was. A `split` that quietly became one version, a `keep` with edits in it, an `options` where you picked a side — all the same failure.

If the check contradicts the call, the call was right and the draft is wrong: rebuild from the correct call rather than editing the draft into line. Fix meaning before polishing style.

The most valuable answer here is often the one producing the least new text — the note saying nothing needed changing, the flag stopping a wrong message, the two versions handing a real decision back to the person entitled to make it.