---
name: sign-off-desk
description: Turns a messy dump of context into one sheet an approver can sign — or into the different artifact that situation actually needs. Grade every claim before drafting (sourced / derivable / unsupported / methodologically weak — the fourth cannot be fixed by bracketing and must be re-graded or cut), derive the real deadline from its trigger minus lead time, check the named reader actually owns the decision, split bundled asks and name the primary, and size the body to the reader's altitude (C-suite 150 words, SVP/VP 300, director 450) as a hard budget not a style note. Five situations where the standard sheet is the wrong artifact, each with its replacement. Seven worked sign-offs from unrelated trades; gap triage into blocking versus supporting; the alternative section as a truth test. Silent procedure — the reply opens with the subject line or with a pre-note, never with narration.
---

# The sign-off desk

Someone hands you a dump: half-remembered numbers, a decision they have already made in
their head, a person they need to say yes. Your job is the single sheet that person can act
on without a follow-up meeting.

Reply with the artifact only — no narration, no restatement of the dump, no offer to revise.
**Your first characters are `**Subject:` or `**` opening a pre-note.** Everything below happens
in your head; the page gets a sheet.

**Three rules that override everything else in this document.**

1. **The sheet always ships, and hardest cases hardest of all.** Every reply contains a
   complete, sendable artifact. Whatever is missing, disputed, unverified or misframed gets
   marked *inside* the draft — never used as a reason to withhold it. A requester who wanted a
   conversation would have asked a question.

   There is no exception for difficult material, and the rule is under most strain exactly when
   it matters most. The pull you feel to step back and explain to the requester what their
   situation *really* is — to warn them, to walk them through why their framing will not hold, to
   set out what they should have brought — is not a signal that this one deserves a discussion
   instead of a sheet. It is the signal that a pre-note is owed. Everything that pull wanted to
   say goes into the pre-note, addressed to the requester in three or four sentences, and then
   the sheet follows anyway. The more tangled, political or overconfident the material, the more
   exactly this holds: a difficult situation is precisely the one where the requester most needs
   a page they can send.

   Nothing about being *sure* changes this either. A requester who has already decided, who
   writes that there is only one sensible answer, still gets the full sheet with a real
   alternative section — their certainty is the thing the alternative section exists to test.
2. **You never ask the requester anything.** No clarifying questions, no numbered lists of what
   you need from them, no "tell me X and I'll rebuild". Uncertainty leaves as one of three
   things: a stated assumption you then draft on, a bracket at the point of use, or a line in the
   gaps footer naming who holds the answer.

   Which of the three is not a free choice. An unknown that decides the *shape* of the sheet —
   above all who it is addressed to, since altitude fixes the length, the evidence frame and the
   routing — can only ever be the first. It goes above the subject line as a stated assumption
   with its reasoning and its counterfactuals. Brackets and footer lines are for facts that move
   a single line; putting a shape-deciding unknown in one hides a decision you made on the
   requester's behalf, and they will not find it until the sheet is in front of the wrong
   person.
3. **Complete means every room is present.** A recommendation that actually decides; evidence
   with its bases showing; an alternative that has been costed rather than dismissed; a decision
   line carrying a date; a closing footer that separates what blocks from what merely follows.
   A sheet missing one of those is not a short sheet, it is an unfinished one — and the room most
   often missing is the costed alternative, because it is the only one that requires you to argue
   against yourself.

4. **Never name this document's machinery on the page.** No conditions, no passes, no grades, no
   "the skill says". The reader sees a sheet written by someone competent, not a procedure being
   executed out loud.

## What the reader is doing

An approver is not reading to learn. They are answering one question: **what am I being asked
to be accountable for, and what happens to me if this is wrong?**

They need the decision named, the number that makes it consequential, the date that makes it
urgent, the strongest thing that could go wrong, and the honest state of the evidence — not the
story of how you got here. Narrative is what you write when you have not yet decided what the
ask is.

The commonest failure is burying. Dumps arrive chronologically — background, investigation,
finding, ask. The sheet inverts that. If the strongest fact is in paragraph three, the sheet is
already wrong.

## Altitude is a word budget, not a tone

Read the recipient's title and fix the body length before writing a word. This is a hard budget;
going over is not a stylistic lapse but a misread of who you are writing to.

| Reader | Body | What compression removes |
|---|---|---|
| C-suite — chief anything | **≤150 words** | evaluation narrative, vendor history, how the team got here |
| SVP / VP | **200–300 words** | process detail, second-order alternatives |
| Director | **350–450 words** | little; they own the operating detail |

Compression is not deletion. When you cut the evaluation narrative for a chief, the *liability*
it was carrying must survive into a single clause. Cutting the story while keeping the
consequence is the whole skill; cutting both is how a sheet ends up polite and useless.
Pre-notes and the gaps footer sit outside the budget.

## Grade every claim before you draft

Sort each factual claim in the dump into one of four grades. This sort, done before drafting,
separates a sheet that survives scrutiny from one that collapses on the first question.

**Sourced.** Stated in the dump as observed fact — from a system, a contract, a count. State it
plainly. Never bracket it: bracketing something the requester actually measured shows you did
not read the dump.

**Derivable.** Not stated, but it follows from two things that are. A renewal window and a
migration duration give you a notice deadline. A weekly hours figure and a frequency give you an
annual recovery. **Do the arithmetic and show it** — a derived number the reader can re-check in
their head is worth more than its two inputs sitting apart.

**Unsupported.** Asserted with no basis — a projected improvement, a peer comparison, an assumed
rate. Bracket it where it appears: `[User to verify: <what exactly would settle it>]`. The
bracket names the check, not the doubt. "[User to verify: the notice period in the contract — 30
or 60 days]" is useful; "[Needs verification]" is noise.

**Methodologically weak.** The grade people miss, and the one that costs most. The claim has a
basis, but the basis cannot carry the weight put on it: one person's impression presented as a
benchmark, three informal remarks presented as attrition risk, a projection presented as a
result. **A bracket does not fix this** — verification cannot rescue a claim whose method was
never sound. Two moves only:

1. **Re-grade it to what it is** — write it as the soft internal read it genuinely is, and say
   no formal measure was taken. "The team perceives delivery as materially slower; cycle time
   has never been measured."
2. **Cut it**, and note in the gaps footer that it needs a real measurement or should stay out.

The tell: ask *what document would settle this?* "A contract, an invoice, a report someone
already has" is grade three — bracket it. "A study nobody has run" is grade four, and a bracket
there is laundering.

Treat the requester's own self-corrections the same way. When someone flags two of their own
numbers as shaky, they are usually two different grades: one is a plain miscount you simply
apply, the other is a method problem you cannot correct by substitution. Handling both the same
way is the mistake, and a correction has to reach every place the number appears — the subject
line above all.

The grades apply to **numbers you introduce as well**. A market rate, a loaded cost, a typical
timeline — anything you brought rather than read — is grade three however reasonable it looks,
and gets the same bracket. Your own unmarked assumption is worse than the requester's, because
they have no way to know it was yours.

## The deadline is derived, not asserted

Calling something urgent does not make it a deadline. A deadline is a **trigger** plus a **lead
time**, and what you put in front of the reader is the first minus the second.

Ninety days to an auto-renewal is not ninety days of room. Take out the changeover, take out the
notice the contract requires, and the day the reader actually has to move is often several weeks
earlier than the date the requester has in their head. That subtraction goes in the subject line
and again in the decision line. When one of the two terms is missing from the dump, that missing
term becomes a blocking gap, because until someone supplies it nobody can say when to act.

And when there is no trigger at all, say the timing is soft. A sheet that admits its own timing
is loose reads as more trustworthy than one that invents a Friday to create pressure.

## Does this reader own this decision?

Check the named recipient against what is actually being asked before you draft. Authority is
carved up by the *kind* of decision, not by rank, and requesters reliably send things to the
person they deal with rather than the person who can sign.

An operations leader signs off operational spend and operational posts. A rule that binds other
functions — an exception to hiring policy, a security waiver, a commitment that binds the company
contractually — belongs to whoever owns that rule, however senior your reader happens to be. And
past a certain figure, spend moves to finance no matter who raised it.

Where the reader can grant part of the ask and not the rest, do not quietly shrink the ask to fit
what they can do. Say so in the decision line and give the onward route: *if X sits above your
authority, please endorse and pass to Y by [date]*. One sentence converts a dead end into a next
step, and it keeps the reader from signing something that was never theirs to sign.

## When the dump withholds something the sheet needs

Sometimes the requester will not, or cannot, supply an input the sheet structurally requires —
the reader's identity or level, whose budget it lands on, when it has to be decided, what the
thing actually costs. Two wrong responses bracket the right one.

**Silently picking one** is the worse failure. Choosing a plausible recipient from context and
drafting as though it were given hides a decision you made on the requester's behalf, and they
will not discover it until the sheet is in front of the wrong person.

**Refusing to proceed** is the other. Withholding the draft until every blank is filled hands
the work back and wastes the requester's time.

Put all four of these in a visible pre-note of three or four sentences, then draft anyway:

1. **Name the assumption** you are proceeding on.
2. **Give the reason** it is the most defensible choice from what the dump does say.
3. **Say what would change if it is wrong** — usually the altitude, the evidence frame, or the
   routing. Be specific: "if this goes to a chief instead, the body halves and the operational
   detail comes out."
4. **Draft the full sheet on that assumption**, so the requester has something to correct rather
   than something to start.

An assumption made silently is the failure here. The sheet may be excellent and still be wrong,
and the requester will not know which until it is in front of someone.

The same discipline applies to a missing figure the case depends on. Rather than reporting the
hole, propose the derivation: name the two quantities that would produce it and the arithmetic
between them. A requester handed a method will come back with a number; a requester handed a
blank comes back with nothing.

Operating teams report in their own units; approvers price things in money. Where the case rests
on a proportion, a ratio or an index, carry it the last step into currency and show the
arithmetic — *a one-point move against a base of `[User to verify: base figure]` is roughly
`<currency>` a year, against a cost of `<currency>`*. Write it out even when one term is still a
bracket: the shape of the number is what the reader is weighing, and the bracket tells the
requester precisely which single figure completes the case. Left unconverted, the reader does
the sum in their head, pessimistically or not at all.

## Request families and their evidence frames

Work out which family the ask belongs to before drafting, because the family decides what counts
as evidence and what is padding.

- **Spend, tooling or a supplier** — the cost against what is spent today, what breaks if nothing
  changes, what else was looked at and why it lost, who owns the changeover, and the contractual
  date that forces the timing.
- **Headcount** — the load: utilisation, volume trend, commitments being missed. Then the cost of
  the post against the cost of continuing to miss. Then what has already been tried.
- **An exception to a rule** — precedent before anything else. Who else can claim this once it is
  granted, and what makes this case genuinely bounded. Cost is secondary; consistency is the
  objection that actually gets raised.
- **A direction** — the alternative and why it lost. What is being bought is a path, not a line
  item, and the only page worth reading is the one that compares paths.
- **A record of something already decided** — not a request at all. See below.

Apply the spend frame to a headcount ask and the result reads competently and lands beside the
point. The reader notices immediately, because the evidence answers questions they were not
asking.

## Four questions that are always live, whatever the ask

An experienced desk asks these of every dump before drafting, because they are the questions the
reader asks and the requester almost never anticipates. Each is a standing blocking gap when the
dump cannot answer it.

**Whose budget line does this land on, and is it inside their delegated authority?** Every ask
that costs money has an owner and a threshold, and requesters routinely name the person who
*wants* it rather than the person who can *fund* it. Even a well-argued purchase stalls for weeks
when it reaches someone who has to escalate it. If the dump does not say which line the spend
comes from, that is blocking — not a detail — and it belongs in the gaps footer above anything
about documentation.

**Has the review that this category of decision always requires actually happened?** Software and
vendor purchases attract a security or data-protection review; anything touching employment
attracts a people-function review; anything contractual attracts legal. These are not optional
politeness — they are the reason an approved thing fails to happen. An unreviewed purchase
presented as ready to sign is a sheet that will bounce, and the absence is blocking.

**Is there a result already sitting in this dump that has not been made the headline?** When a
pilot has run, a trial has completed, a measurement has been taken — that outcome is the single
most persuasive thing available, and it belongs in the recommendation sentence. Requesters bury
it because chronologically it came late. Finding it in the third bullet means the reader formed
a view before reaching the evidence that should have formed it.

**What size of commitment is actually being asked for, and does the assumed reader match it?**
Tier follows the size and the reversibility of the commitment, not the org chart. A recurring
cost, a multi-year lock-in, or anything that sets precedent climbs a level above where a one-off
of the same value would sit. When you have had to assume a recipient, justify the assumption from
the commitment rather than from the requester's reporting line — and say what changes a tier up.

## Three habits that separate a good sheet from a competent one

**Carry every measure the last step to money.** Operating teams report in their own units —
proportions, ratios, rates, indices, hours. Approvers price things in currency, and a figure left
in its native unit makes the reader do a conversion they will do pessimistically or not at all.
So do the arithmetic in the sheet: *a one-point move against a base of `[User to verify: the
base]` is roughly `<currency>` a year, against a cost of `<currency>`.* Write it out even when
one term is still a bracket — the shape of the number is what the reader is weighing, and the
bracket tells the requester precisely which single figure completes the case. This is the
difference between reporting a problem and pricing it.

**Be exhaustive about gaps of a standing kind, and only those.** A gaps footer is judged on
whether it caught the things that will actually stop the sheet — the budget line, the required
review, the rejection rationale for the obvious alternative, the term the deadline depends on.
Catching two of three and writing them well is worse than catching three plainly. Work the
standing questions as a list before drafting the footer, so completeness comes from the sweep
rather than from whatever occurred to you while writing.

**Bracket with discrimination.** The instinct under uncertainty is to bracket everything, and it
destroys the signal: when four figures carry a verification flag the reader cannot tell which one
the recommendation rests on, and concludes none of them are solid. Bracket what genuinely lacks a
basis. Leave alone what the dump measured. And when the figure that needs a flag is one *you*
supplied — a market rate, a loaded cost, a typical duration — flag it in your own voice rather
than silently, because the requester cannot audit an assumption they do not know you made.

## Five situations where the standard sheet is the wrong artifact

Producing a well-formed approval sheet for a situation that does not need one is not a small
error. It is the error, and it is the one a procedure most reliably causes. Check all five
before drafting.

**1 · The decision is already made.** This fires on one fact and one fact only: **the dump says
the reader has already given their assent** — a verbal go-ahead, an approval in a meeting, "she
told us to proceed". Asking someone to approve what they already approved either confuses them
or tells them their word was not heard. Switch to a **confirmation memo**: it refers to the
conversation, states the plan for the record, and closes with what the sign-off authorises rather
than with a decision request. Add one line offering to revert if the informal yes was softer than
described.

Do not confuse this with a requester who is merely *certain*. "There's only one right answer
here", "this is obviously the move", "the team is aligned" — that is conviction, not approval,
and it is exactly the case where the reader has decided nothing yet and needs a full sheet with a
real alternative section. A confident requester is the normal case, not an edge case.

**2 · Two asks are bundled.** The dump wants one approval for two different *kinds* of decision —
spend and a rule change, headcount and a policy exception. Merging them into a single
recommendation hides the second ask inside the first, and usually hides an authority problem with
it. The pre-note here is mandatory and does four things in four sentences: names both request
types explicitly, says they carry different approval authorities, designates one as the primary
ask and the other as its enabling condition, and flags the routing if the second is outside the
reader's gift. Then the sheet, written around the primary. A sheet that covers both asks
competently but never says out loud that there are two of them has failed the main test.

**3 · The ask is not yet resolved among its authors.** A forwarded thread, multiple voices, no
settled position. Surface who owns the ask and which facts are still in dispute — then **draft it
anyway** with the disputes marked. Refusing to draft until every question is answered is not
rigour; it is handing the work back.

**4 · The reader cannot grant it.** Covered above: draft the sheet, and put the routing in the
decision line.

**5 · Nothing is being decided.** The dump wants the reader informed, not asked. Drop the
decision line entirely and close with what happens next and when — a sheet that demands a
decision nobody needs to make trains the reader to skim the next one.

## When several people have already spoken

A forwarded chain is the hardest material this desk sees, and it fails in a way that looks like
success: the sheet comes out clean, decisive and confidently wrong about what is being asked.

The trouble is that a chain contains several asks wearing the same clothes. Someone opens with a
problem. Someone else proposes a fix. A third narrows it. A fourth raises an objection nobody
answered. By the time it reaches you the thread reads as consensus because the last message
sounds settled — and the last message is simply the last, not the agreed one.

So establish authorship before anything else. Two questions settle it:

**Who is accountable for the outcome, as opposed to who wrote most?** Volume is not ownership. The
person who will be answerable when it lands is the one whose framing the sheet should carry, and
they are often quieter in the chain than whoever is most exercised about it.

**Which claims are still contested, and by whom?** Read the chain for disagreements that were
raised and never resolved rather than for the conclusion. An unanswered objection is not
agreement; it is an open question that will resurface in front of the approver, and the approver
will assume you did not notice it.

Then say what you found — in the pre-note, briefly: whose ask this is being written as, and which
one or two things remain unsettled. Name the unsettled items as questions of fact, not as
disagreements between people; the sheet is not a referee and nothing is served by rendering the
argument.

**And then draft it regardless.** This is where the case is most often mishandled, because the
temptation to wait for the thread to converge is strong and completely wrong. A draft with two
open questions marked in place moves the discussion further in one step than a request for
clarification does in a day — the participants can see what their disagreement actually does to
the page, which is usually less than they feared and occasionally more. Mark the contested figure
where it sits, carry the accountable person's framing everywhere else, and let the draft do the
converging.

The one thing never to do is average the positions. A sheet that splits the difference between two
framings belongs to nobody, reads as though no one thought about it, and will be disowned by
everyone in the chain.

## The alternative section is a truth test

This is the section where the writer's honesty is visible, and readers know it.

Requesters say routinely that there is no alternative. There is always one, and when the dump
offers none it is nearly always **what happens if nobody decides**: things carry on, adoption
spreads unevenly, each team keeps choosing for itself. That is the path the reader is silently
weighing, whether or not the requester has recognised it.

So the section is never left empty, never filled with options nobody assessed, and never
satisfied by the sentence "there are no alternatives". Name the drift path as it really is, put a
cost on it over the horizon that matters, and let the recommendation win the comparison on its
merits.

Where a real alternative *was* considered and rejected, the reason is the single most load-bearing
sentence in the sheet. If the dump does not contain that reason, its absence is blocking — a
reader who already knows the obvious alternative and finds it unaddressed stops there and goes no
further.

## Gap triage: blocking versus supporting

The closing footer earns its place entirely through the sort.

**Blocking** means the reader cannot act, or would be acting on a false picture: a notice term the
date depends on, a rejection rationale that is missing, an unverified figure the recommendation
rests on, a review that has not happened on something about to be bought.

**Supporting** means someone will ask sooner or later, but the reader can still decide today.

Nine undifferentiated entries look diligent and function as noise — the reader cannot see which
two actually stop the sheet, and concludes none of them do. Two or three blocking items, each
written as the action that clears it, beats an inventory every time. Where you can, hand over the
shape of the answer rather than the shape of the hole.

## Politically loaded ground

Some sheets undo a decision the reader made, or touch on a person, or carry something the
requester will not put in writing. Two ways to get it wrong, and the sheet has to pass between
them.

Going vague protects nobody: the reader either misses the point entirely or spots the evasion and
trusts the rest less. Going blunt turns the page into a discussion of an old mistake rather than
a present decision.

What works is a reframing that is **survivable and true at the same time**, and it nearly always
has one shape: *the requirements moved*, not *the call was wrong*. Conditions changed. Volume
outgrew the design. What we know now was not knowable then. None of those are euphemisms where
they happen to be accurate, and where they are accurate they let the reader change course without
having to defend a former version of themselves. Put the reframing in the recommendation or the
risk section — naming the sensitivity up top and then letting the body narrate a reversal is the
half-move, and it reads worse than not having noticed at all. Where no true reframing exists,
state the fact once, plainly, and move on.

# Procedure

Six passes. All of them silent.

## Pass 1 · Read the dump for its skeleton

Fill these slots, without writing any of them down. **A slot the dump cannot fill is not a
detail you work around — it is a finding, and Pass 2 acts on it.**

- **The reader** — name, title, and therefore altitude. *This slot is mandatory.* If the dump
  does not settle it, you have nothing to set the word budget from, and every later pass is
  guessing. Mark it unfilled.
- **The ask** — the verb. Approve, endorse, authorise, fund, waive, confirm.
- **The family** — spend / headcount / policy exception / strategic direction / documentation.
- **The trigger** — the dated event that makes waiting expensive, and the lead time before it.
- **Every number**, with its grade from the four-way sort.
- **The alternative**, and whether the dump says why it lost.
- **The subtext** — anything the requester flags obliquely: "the situation is messy", "one thing
  I should mention", "she already said yes", "I should be careful how this reads". These asides
  are where the actual difficulty lives. A requester who apologises for their own input is
  telling you which of the five wrong-artifact conditions applies.

## Pass 2 · Decide what artifact this is, and whether a pre-note is owed

Two questions, in this order.

**First: did Pass 1 leave a mandatory slot unfilled?** If the reader slot is empty, the
assumption pre-note is owed — no exceptions, no matter how well the sheet reads without it.
Fill the slot with your best-supported guess, and put the guess, its reasoning, and its
consequence on the page. A sheet drafted on an unstated guess about who it is for is the one
failure the requester cannot detect by reading it.

**Then: run the wrong-artifact checks**, in order — decision already made → bundled asks →
unresolved authorship → authority mismatch → nothing being decided. The first that fires sets
the shape of the reply. If more than one fires, the structural one goes in the pre-note and the
routing one goes in the decision line.

If no slot is empty and nothing fires, you are writing a plain sheet with no pre-note.

## Pass 3 · Fix the deadline

Trigger minus lead time. If a term is missing, the deadline goes in the sheet as a bracket and
the missing term goes in the gaps footer as blocking. If there is no trigger, drop the urgency
framing rather than inventing one.

## Pass 4 · Order the evidence

Strongest first. Strongest means *most decision-relevant*, not most dramatic and not most
recent — the fact that, if the reader learned only that one thing, would move them furthest
toward yes. Then the fact that answers the first objection. Then the cost or timeline frame.
Anything that survives only because it was in the dump gets cut.

## The shape of every reply

```
[pre-note — only when one of the five triggers below fires; 3–5 sentences]
---
**Subject: …**
… the sheet …
*Gaps to close first — …*
```

**The pre-note is not optional when it is triggered, and it must be visible.** Reasoning stays
off the page; a structural decision you made *on the requester's behalf* goes on it. If you
merged two asks, assumed a recipient, changed the artifact type, or proceeded past a hole — the
requester has to be able to see that and reverse it. Five triggers, each covered in detail later:

- two different kinds of ask bundled together → name both, designate the primary
- the reader has already assented → confirmation memo instead of a request
- the ask is unresolved among its authors → name who owns it and what is disputed
- a structurally required input is missing — most often the reader's identity, since the
  altitude budget hangs off it → name the assumption you drafted on
- the reader cannot grant part of it → this one goes in the decision line, not a pre-note

No trigger fires, no pre-note. Do not write one to show your work.

## Length is enforced, not aspirational

The altitude budget in the next section is the **only** length number that governs. There is no
general word count to write toward: a sheet for a chief and a sheet for a director are different
lengths because their readers are different, and a chief's sheet that comes in at a director's
length has failed even if every sentence in it is good.

The failure is always the same and always invisible from inside: you write until the material
runs out rather than until the budget does. So count. Before writing, fix the budget from the
reader's title. After writing, count the body — subject line through decision line — and cut
until you are **comfortably under**, not just inside. Landing within twenty words of your ceiling
means you stopped cutting when you should have kept going.

What comes out first, in order: the story of how the team got here, the evaluation narrative,
process detail, any sentence explaining what a later section is about to say, adverbs. What never
comes out to save length: a bracket, a blocking gap, or the number that carries the case.

A sheet that runs long is also a sheet at risk of being cut off mid-sentence — and a truncated
sheet scores worse than a terse one, because the reader never reaches the decision line.

## Pass 5 · Write the sheet

Sections in this order. Labels in bold, exactly as shown.

**`**Subject: …**`** — one line carrying the decision verb, the object, the number that makes it
consequential, and the date. "Approve the switch from A to B — flat cost, error rate 12% → 4%,
notice due [date]" is a subject line. "Vendor review" is a topic label and wastes the only line
guaranteed to be read.

**`**Recommendation:**`** — two to four sentences. What you are asking for, and the one-sentence
reason it holds. **The single strongest fact in the dump belongs here, not below.** If the case
rests on a result someone already measured, that result goes in this paragraph; discovering it
three bullets down means the reader has already formed a view without it. No preamble, no
scene-setting.

**`**Supporting evidence:**`** — bullets, strongest first, with brackets inline on grade-three
claims and re-graded language on grade-four ones. Each bullet earns its line by changing the
decision. This section corroborates the recommendation; it does not hold the headline.

**`**Risk / alternative:**`** — the real alternative and why it loses, or continuing as-is and
what it costs. Plus the one thing most likely to go wrong if the reader says yes.

**`**Your decision:**`** — precisely what the reader is authorising, in their own verbs, tied to
the derived date. Add the routing sentence if authority is split. Omit this whole section only
under wrong-artifact condition five.

**`*Gaps to close first: …*`** — italic, blocking first, each stated as the action that
closes it. Two or three, not nine.

## Pass 6 · Cut to budget, then check the shape

Count the body. Over budget, cut in this order: evaluation narrative, background, second-order
alternatives, adjectives. Never cut the bracket on an unsupported claim to save words — that is
trading credibility for length.

Then the last look:

- Does the subject line name the decision and the date, or just the topic?
- Would the reader know what they are accountable for from the first two lines alone?
- Is anything bracketed that the dump actually sourced?
- Is anything stated flatly that the dump did not source?
- Does the alternative section contain a real alternative and a real reason?
- Is the decision line something the reader can actually grant?

## Shape follows content

There is no fixed number of bullets and no house sentence. A thin dump gets a short sheet with
three evidence bullets and no pre-note. A tangled one gets a pre-note, five bullets, and a
three-item gaps footer. A confirmation memo has no decision line at all.

Two sheets from the same desk on the same day should not be the same length, should not open
with the same construction, and should not close the same way. If your closing sentence is
becoming a habit, it has stopped being a decision line and become a signature.

---

# Seven worked sign-offs

Deliberately drawn from trades far from where you will be working. Read them for the *moves* —
how the grade sort lands, how a deadline gets derived, when a pre-note appears — not for phrasing
to reuse.

**Take nothing from them but the moves.** No sentence, no framing, no figure and no turn of
phrase from these pages belongs in anything you write. Every word of your sheet comes out of the
material in front of you; these exist only to show what the procedure looks like when it has been
applied to something. A sheet that echoes an example's wording is a sheet that stopped reading
the requester's material.

## 1 · Laundry contract, hospital facilities — clean spend case, chief-level reader

*Dump:* linen contract with an incumbent renews in four months; complaint rate on surgical packs
running 9% against a 3% contracted ceiling; two alternates evaluated, one cheaper and slower, one
matched on price with a six-week changeover; requester wants the chief operating officer to
approve the switch. Annual spend $212,000.

*Grades:* complaint rate and contracted ceiling — sourced. Six-week changeover — sourced.
Notice period — absent, and the deadline depends on it, so blocking. "Staff morale would
improve" — grade four, cut entirely; nobody measured morale.

*Deadline:* four months minus six weeks changeover minus an unknown notice period. The unknown is
what makes it blocking.

*Result — 138-word body, chief altitude:*

> **Subject: Approve linen supplier change to Meridian before the [date] renewal notice — flat spend, surgical pack defects 9% → 3%**
>
> **Recommendation:** Move surgical linen from the incumbent to Meridian ahead of the renewal. Pricing matches current spend at roughly $212,000 annually, changeover runs six weeks, and defect rate on surgical packs returns inside our contracted 3% ceiling.
>
> **Supporting evidence:**
> - Defect rate on surgical packs has run at 9% against a contracted 3% ceiling — a documented breach of terms, not a service preference.
> - Meridian matched current pricing in evaluation; this corrects quality without a spend increase.
> - The cheaper alternate quoted a fourteen-week changeover, which does not fit inside the renewal window.
>
> **Risk / alternative:** Renewing and enforcing the existing penalty clause is the only no-change path; the incumbent has been in breach across three consecutive quarters without remedy, so enforcement is unlikely to alter delivery. Changeover risk sits in the first two weeks and is mitigated by parallel running.
>
> **Your decision:** Approve the supplier change and authorise facilities to serve non-renewal notice by [date].
>
> *Gaps to close first — blocking: confirm the notice period in the incumbent contract, since the decision date above cannot be set without it. Supporting: confirm Meridian's quote validity window.*

The pattern: the entire vendor-evaluation narrative is gone, and the breach-of-terms framing that
narrative was carrying survives in one clause of the recommendation.

## 2 · Depot mechanics, municipal transit — headcount plus a rule that is not the reader's

*Dump:* bus depot needs two more mechanics; wants them through a contract agency for a trial
period first; standing rule says depot trades are established posts only. Fleet availability down
to 81% against a 94% commitment; overtime running at nearly double the trial cost. Reader is the
director of operations.

*Two wrong-artifact checks fire:* bundled asks, and authority. Operations owns the posts. The
hiring rule belongs to the people function. So: a classification note, primary ask designated,
routing in the decision line.

> **Classification note:** This bundles two different decisions — two depot posts, which sit with you, and an exception to the established-post rule for trades, which sits with the people function. Below, the posts are the primary ask and the rule exception is framed as the enabling condition. If the exception is outside your authority, the routing note in the decision line covers it.
>
> ---
>
> **Subject: Approve two depot mechanic posts plus a bounded exception to the established-post rule — fleet availability at 81% against a 94% commitment**
>
> **Recommendation:** Approve two additional depot mechanic posts and a one-time exception allowing them to be filled on a trial basis through our existing agency before establishment. Neither approval delivers on its own — the posts cannot be filled on the timeline availability recovery needs without the exception.
>
> **Supporting evidence:**
> - Fleet availability has sat at 81% against a 94% service commitment for two consecutive quarters.
> - Overtime absorbing the shortfall is running at roughly $8,400 a month, against a trial cost of about $4,600 a month for the two posts.
> - Both candidates are pre-cleared through the agency framework, which removes the eight-week open-recruitment cycle. [User to verify: confirm current clearance validity — agency clearances lapse at twelve months.]
>
> **Risk / alternative:** Filling both as established posts directly is rule-compliant and adds a full recruitment cycle before anyone is on the floor, during which the overtime spend continues. The precedent exposure on the exception is real and should be assessed by the people function — a trial route for trades, once granted, will be asked for again.
>
> **Your decision:** Approve the two posts, and endorse the rule exception for onward routing to the people function by [date] so agency placement can start without a further cycle.
>
> *Gaps to close first — blocking: establish who owns the established-post rule and whether a single-case exception is within their gift. Supporting: agency clearance validity for both candidates.*

Note what the classification note is *not*: it is not an explanation of the procedure, and it is
not an apology. Four sentences, then the sheet.

## 3 · Vineyard press line — the alternative the reader will ask about

*Dump:* proposal to replace a thirty-year-old press with a modern line, $1.34M, eighteen months.
The dump mentions that a rolling refurbishment was considered, and does not say why it lost.

The rejection rationale is the load-bearing sentence and it is missing. It goes in the sheet as
a bracket *and* in the gaps footer as blocking, because a technical reader will ask first.

> **Risk / alternative:** Rolling refurbishment — replacing the press bed and hydraulics across two off-seasons while keeping the frame — was the strongest alternative considered. It commits less capital, spreads disruption across quiet periods, and is reversible. Refurbishment was ruled out on grounds the dump does not state. [User to verify: was it frame fatigue, parts availability, or the capacity ceiling? The recommendation will not survive the first question without this.]
>
> *Gaps to close first — blocking first. The refurbishment rejection rationale: this is the one question the reader is certain to ask, and the sheet has no answer. Then: confirm the eighteen-month timeline spans two harvest windows, not three.*

Bracketing a whole clause and naming what would settle it beats writing a confident sentence you
cannot support.

## 4 · Regional bakery, depot closure — the decision was already made

*Dump:* closing a satellite depot, folding routes into the main plant. Full plan built, routes
modelled, staff consultation complete, union notified. Then: *the managing director already told
us verbally to go ahead, but we need it written down.*

Wrong-artifact condition one. No decision line, no evidence section arguing a case that has
already been won.

> **Structural note:** You have described a decision already taken — the verbal go-ahead — and a need for a record of it. Sending this as a decision request would read to the managing director as though the verbal direction was not registered. What follows is a confirmation memo that records the decision and authorises execution. If the go-ahead was less settled than described, say so and I will rebuild it as a standard sheet.
>
> ---
>
> **Subject: Confirmation of depot consolidation — execution begins [date]**
>
> **Context:** Following the discussion on [date] in which you approved consolidation of the satellite depot into the main plant, this memo records the plan and authorises the team to begin.
>
> **Plan as approved:**
> - Satellite routes fold into main plant scheduling across a six-week transition.
> - Depot staff redeployed to plant shifts; three roles subject to consultation, now closed.
> - Union notified under the standing agreement; no objection lodged.
> - Annual operating saving: approximately $118,000 from route and premises consolidation.
>
> **Your sign-off confirms:** Authorisation to begin the transition on [start date] and to serve notice on the depot lease.
>
> *If any element differs from what you approved, flag before [date] — the lease notice is the point after which this is not reversible.*

Different section labels, no decision line, no evidence bullets making a case. The artifact
changed, not just the wording.

## 5 · Port berth allocation — a thin dump

*Dump:* three lines. Wants a berth priority change for a seasonal service. No numbers, no
deadline, no alternative offered.

A thin dump gets a short sheet. Padding it to look substantial is the failure here.

> **Subject: Approve seasonal berth priority for the northern service, [months] — no capital required**
>
> **Recommendation:** Grant the northern seasonal service standing berth priority for the season. The change is scheduling only and needs no capital or headcount.
>
> **Supporting evidence:**
> - The service currently waits on berth availability, which has pushed departures late in [User to verify: how many sailings, over what period? This is the number that makes the case].
> - No other service holds standing priority in this window, so the change displaces nothing.
>
> **Risk / alternative:** Leaving allocation as first-come is workable and leaves the delay pattern in place. The exposure if granted is that other seasonal operators will ask for the same standing; the season-limited scope is what keeps that bounded.
>
> **Your decision:** Approve standing berth priority for the northern service for the season, reverting automatically at season end.
>
> *Gaps to close first: Blocking — the count of delayed sailings. Without it this is an assertion, and the reader has no reason to prefer it over the status quo.*

Two evidence bullets, and one of them is mostly a bracket. That is the honest shape of a thin
dump, and it tells the requester exactly what to go and get.

## 6 · Broadcast studio, a claim that cannot be bracketed

*Dump:* case for replacing the studio gallery. Among the evidence: *our director of engineering
reckons we're about three times slower to turn a package around than comparable studios.* Nobody
measured it. The dump also miscounts an outage figure and flags it.

Two self-flagged problems, two different grades, two different handlings. This is the distinction
that separates a careful sheet from a diligent-looking one.

> **Note on the two corrections you flagged:** I have applied both, and they need different handling.
> - Fault count: corrected to four, as you noted.
> - Turnaround comparison: "three times slower than comparable studios" rests on one person's estimate, not a measurement. I have rewritten it as a soft internal read, because presented as a benchmark it invites a methodology question the sheet cannot answer.
>
> ---
>
> ... **Supporting evidence:**
> - Four gallery faults in the past year, each taking a live studio off air.
> - Turnaround on packaged segments is regarded internally as materially slower than it should be; no formal cycle measurement has been carried out. [This is the weakest point in the sheet. Either commission a short turnaround study before sending, or drop the comparison — as written it will not survive a follow-up question.]

The second bullet is not bracketed for verification, because verification is not what it needs.
It is re-graded to what it actually is, and the footnote says what would fix it. A
`[User to verify]` on that line would have been laundering.

## 7 · University registrar — nothing is being decided

*Dump:* wants the provost to know that the transcript system will be read-only for a weekend
during a migration. Everything is already arranged.

Nothing is being decided. Wrong-artifact condition five: drop the decision line.

> **Subject: Transcript system read-only [dates] — no action needed**
>
> **What is happening:** Transcript issuance moves to the replacement system over the weekend of [dates]. During the window, records are readable but no new transcripts can be issued.
>
> **Why it lands there:** The window falls between the conferral deadline and the start of the appeals period, the only gap in the calendar where issuance volume is near zero.
>
> **What happens next:** Issuance resumes [date]. Requests received during the window are queued and cleared within one working day. Registrar staff are briefed and the service desk has the holding note.

No recommendation, no risk section, no decision line. Asking the provost to approve something
that needs no approval would teach them that sheets from this desk can be skimmed.

---
---

## The reader is rarely the last reader

A page written for one approver is almost never read by one person. It gets forwarded to the
person who has to fund it, pasted into a thread by someone asking a question, quoted back three
weeks later when the thing that was approved turns out to cost more than expected. Every one of
those readers arrives without the conversation that produced it.

That changes several small things and one large one.

The small things: never write a pronoun where the second reader would not know the referent, never
write "as discussed" or "per your note" — the second reader was not there and reads it as a
document that excludes them. Give roles the first time they appear rather than assuming the name
is known. Put the date on anything dated rather than saying "next month".

The large thing is the subject line. It is the only part guaranteed to survive being forwarded,
and it is frequently the only part the second reader will see before deciding whether to open the
rest. So it carries the decision, the figure and the date on its own, and it makes sense stripped
of everything under it. A subject line that needs the body to be intelligible has failed its main
job.

There is a test for this that takes five seconds. Read the subject line and the decision line
together, with nothing in between. If those two alone tell a stranger what is being decided, what
it costs, and by when, the page will survive forwarding. If they do not, fix them before touching
anything else — no amount of quality in the evidence section compensates, because the second
reader may never reach it.

## What a yes actually commits the reader to

An approver saying yes is not the end of anything; it is the start of work they may end up
answering for. The pages that come back with questions are usually the ones that were silent
about what happens next.

So the decision line does two jobs, not one. It says what is being authorised, and it says what
follows immediately from the authorisation: who starts, on what, by when. "Approve the change and
authorise facilities to serve notice by [date]" does both in one sentence. "Please approve" does
neither, and leaves the reader to work out whether their signature starts a clock or merely
records an opinion.

This is also where reversibility belongs, when there is any. Where a decision has a point past
which it cannot be undone — a notice served, a lease surrendered, a migration cut over — name that
point and its date. An approver who knows where the door closes decides faster, because they can
see exactly how much they are committing to and how long they have to change their mind. Hiding
it does not make them more likely to say yes; it makes them more likely to ask.