Direction that enables decisions

A PM writes a direction document. It says the product wins by making documents worth returning to. It names the populations, the two-quarter horizon, and the reason the team believes it. It is reviewed. Everyone agrees. It goes on the wall.

Nine days later the engineering lead asks a question. Three enterprise customers want automated meeting summaries. Large workspaces have slowed down. Both need the same engineers this cycle. Which one? The PM looks at the direction document, finds nothing that settles it, and answers from memory. The next week there is another question, and the PM answers that one too.

Six weeks in, the direction has been read by everyone and used by nobody. Every real trade-off still routes through one person. The PM feels busy and central, which is easy to mistake for being effective.

The failure is hard to see from inside because the artifact looks correct and the team agreed with it. Agreement is the wrong test. A direction that everyone agrees with is usually one that has not yet asked anyone to give something up. Nobody objects to "documents worth returning to." They object to "we will not staff enterprise requests this quarter." The second sentence is direction. The first is a description of a preference.

There is a second failure hiding under the first. Because the PM keeps answering, the team never finds out that the direction is unusable. Each answer patches the gap and hides it. The more responsive the leader, the longer the document stays broken.

Both loops below start from the same trade-off. Only one of them ends anywhere other than your desk:

flowchart LR
  T["Trade-off arrives<br/>summaries vs. workspace speed"] --> Q{"Does the direction<br/>order the two?"}
  Q -->|"no — aspiration only"| L["Leader answers<br/>from memory"]
  L --> H["Gap patched,<br/>and hidden"]
  H -.->|"same shape<br/>returns next week"| T
  Q -->|"yes — ordering<br/>names the loser"| R["Team resolves it<br/>without you"]
  R --> E["Only the novel or<br/>one-way case escalates"]

The dotted edge is what makes this durable. The shape returns, you answer it again, and answering feels like the work.

Ask yourself: in the last month, name one trade-off someone else resolved by pointing at the direction, and name one they resolved by asking me. If the second list is long and the first list is empty, you have published a preference and called it direction.

Direction is not a description of where you are going. It is an instrument other people use to dispose of cases without you.

Four parts. Most direction documents contain the first and stop.

Part	Question it answers	Failure when it is missing
Diagnosis	What situation are we actually in	Direction becomes timeless and cannot expire
Ordering	Which value wins when two collide	Every conflict escalates to the leader
Boundary	What we will not do in this period	Displaced work quietly returns
Revision trigger	What evidence would change this	Direction hardens into a belief nobody may question

Ordering is the part that does the work. It is also the part that is uncomfortable to write, because writing it means naming the thing that loses.

Three levels of direction

Level	Example	What it can resolve
Aspiration	"Documents should be worth returning to"	Nothing. It excludes no option.
Ordering	"Retention of existing paid workspaces beats new-segment expansion this quarter"	A conflict between two funded options
Constraint	"We will not commit engineering capacity to any request from fewer than five customers until the underlying claim has been tested"	A whole class of incoming requests, in advance

Aspiration is not useless. It is the reason the ordering is what it is. But an aspiration alone hands every case back to the leader, because any option can be argued to serve it.

Here is the same direction written both ways, against the live Noted collision between the enterprise summaries request and large-workspace response time:

The difference is not tone or length. The weak version excludes nothing, so it cannot dispose of anything. The strong version gives something up, which is what makes it usable by someone else.

The trade-off test

There is one honest test of a direction, and it takes ten minutes.

Take two live requests that both have real support and cannot both be served this cycle. Hand the direction to someone who was not in the room when it was written. Ask them which request it resolves, and why.

Three possible results:

1. They resolve it, and give your reasoning back to you. The direction works.
2. They resolve it, and give reasoning you disagree with. The direction works, and you have just learned what it actually says. Fix the text, not the person.
3. They say "it depends — what do you think?" The direction is a values statement. Nothing about it will improve until you add ordering.

Result 3 is the common one. It is also the one leaders explain away, usually with "they just need more context." More context delivered verbally is another name for deciding it again yourself.

Direction is written for the person furthest from you

The reader you are writing for is not the person in your weekly meeting. It is the person who gets the question at 4pm on a Thursday, has thirty minutes, and cannot reach you. If your direction only functions when accompanied by your commentary, it is not portable, and it will not survive your absence, your holiday, or your second and third PM.

Boundary

Direction cannot resolve every trade-off, and one that tries becomes a rulebook that is wrong within a month.

Some decisions should still come to you. A genuinely novel case, a one-way consequence, or a conflict between two boundaries the direction itself set — these are not failures of direction. They are the residue that direction is not designed to absorb.

The boundary is this: direction should resolve the recurring class, not the singular case. If a trade-off shape will appear four more times this quarter, it belongs in the direction. If it will appear once, and reversing it would be expensive, deciding it yourself is correct and cheap.

The failure at this boundary runs in both directions. Under-specified direction sends everything to you. Over-specified direction resolves cases that should have been escalated — for example, a rule that says "always prefer paid-seat retention" will happily authorize skipping a security migration, which nobody intended. When you write a constraint, write the class it does not cover in the same breath.

Two properties place a trade-off, and they are not the same property: how often its shape recurs, and what it costs to reverse.

quadrantChart
  title Where a trade-off belongs
  x-axis Appears once --> Recurs this quarter
  y-axis Cheap to reverse --> One-way once started
  quadrant-1 Order it and name the escalation
  quadrant-2 You decide. Cheap and correct.
  quadrant-3 Let whoever holds it decide
  quadrant-4 Direction should dispose of this
  Summaries vs workspace speed: [0.78, 0.32]
  Migration timing: [0.24, 0.82]

The top-left quadrant is the one leaders write rules for and should not. The bottom-right is the one they answer personally and should not.

Read the Noted case if you have not already.

A standing condition for this phase. The case gives you one dedicated PM — you — plus a founder, an engineering lead, and a support lead. Phase 07 adds one condition the case does not contain, and states it openly rather than smuggling it in: Noted has funded two product hires. The first starts in about three weeks. The second starts in about ten weeks. Their names, backgrounds, and strengths are <unknown> and will stay unknown for most of this phase. That is deliberate. Direction and decision rights that only work once you know the person are not direction and decision rights.

Now take the live collision between Signal 2 (three enterprise customers want automated meeting summaries) and Signal 3 (P95 response time rose 34% for large workspaces).

These two compete for the same engineering capacity this cycle. What the case gives you: the three requests arrived through one account manager in one month and represent roughly 20% of revenue; the underlying enterprise claim has never been tested; large workspaces are about 4% of workspaces but hold a disproportionate share of paid seats; no support ticket has mentioned speed.

Your task. Write direction that resolves this collision without you in the room.

1. Write the diagnosis in three sentences. What situation is Noted actually in right now? Not the aspiration — the constraint or dynamic that makes this period different from the last one.
2. Write one ordering sentence: when the interests of existing paid workspaces collide with the interests of a new revenue segment, which wins this quarter, and why. Name the loser explicitly.
3. Write one constraint sentence: a class of incoming request you will not staff this period, stated so that someone can apply it to a request you have never seen.
4. Apply your own direction to the two signals. Write the answer it produces. If the answer surprises you, do not adjust the answer. Adjust the text.
5. Now apply it to Signal 4 (the platform dependency loses support in ten weeks). Your direction was not written for this case. Say whether it resolves it, mis-resolves it, or correctly declines to resolve it. If it mis-resolves it, add the class exclusion.
6. Write the revision trigger: the observation and threshold that would make this ordering wrong.
7. Commit. State the one sentence of your direction that someone on the team will most want to argue with, and say why you are keeping it.

Expect to be pushed on: whether your ordering sentence actually excludes anything or just restates a preference in stronger words; whether the constraint in step 3 can be applied by a person who has never met the enterprise customers; and whether step 5 shows your direction quietly authorizing a delay to a security migration that nobody intended it to authorize.

Committed to a position? Check it against what a strong answer holds

• A diagnosis that describes the constraint or dynamic of this period — what makes it different from the last one — and not an aspiration.
• An ordering sentence that names the loser, and a constraint that a person who has never met the customers could apply to a request you have never seen.
• The direction applied to signals 2 and 3, with the answer it produces written down, and the text changed rather than the answer if the two disagree.
• The direction tested against Signal 4, with either a class exclusion added or a clear account of why it correctly declines to resolve that case.
• A revision trigger with an observation and a threshold. A refusal to write a rule for the singular, one-way case that should come to you.
• The most common weak move is an ordering that restates a preference more forcefully — "we prioritise existing paid workspaces." It is weak because both live signals can still be argued to serve it, so it disposes of nothing.

Produce artifacts/LD-01-decision-enabling-direction.md using the template below.

Write it for the person who joins your team in three weeks and gets a hard question on their fourth day. That person has no history with you, no shared shorthand, and no standing to interrupt you. If the document does not let them resolve a real case alone, it has failed, no matter how accurate it is.

This extends your Product Decision Case upward. Every earlier artifact defended one decision. This one defines the conditions under which a class of future decisions can be made by someone else, which is what makes the rest of this phase possible.

# Direction · <team or product area>

- **Date:** <YYYY-MM-DD>
- **Period this covers:** <start> to <end>
- **Author:** <name or role>
- **Written for:** <the person who will apply this without you in the room>

## Diagnosis

<Three sentences. What situation is this team actually in? Name the constraint
or dynamic that makes this period different from the last one. Not an
aspiration — an assessment.>

## Ordering

<One sentence. When these two things collide, this one wins this period.>

- **What wins:** <the interest that takes precedence>
- **What loses:** <the interest that is displaced — name it, do not soften it>
- **Why, in one line:** <the reason grounded in the diagnosis>

## Constraints for this period

| # | We will not | Applies to a request when | Who can waive it |
|---|---|---|---|
| 1 | <constraint stated so a stranger can apply it> | <observable test, not a judgment call> | <person or role> |
| 2 | <constraint> | <test> | <person or role> |

## What this direction does not resolve

<The trade-off classes this direction deliberately leaves to a person. Include
any case where the consequence is one-way, novel, or where two constraints above
would collide. A direction that claims to resolve everything will authorize
something you did not intend.>

- <class> → decided by <person or role>
- <class> → decided by <person or role>

## Worked cases

<Apply the direction to at least two real, live cases. Write the answer it
produces, including one you did not like.>

| Live case | What the direction resolves | Sentence used |
|---|---|---|
| <case> | <the answer> | <which line above> |
| <case> | <the answer, including "it does not resolve this"> | <which line above> |

## Evidence behind the ordering

| Claim the ordering depends on | Source | Population | Window | What it cannot establish |
|---|---|---|---|---|
| <claim> | <source> | <who> | <when> | <limit> |

## Uncertainty

- **Confidence in the ordering:** <low | moderate | high> — <why, in one line>
- **Strongest argument against it:** <the best case for the displaced option>
- **Who on the team disagrees, and with which sentence:** <name the dissent, or
  `<unknown>` if you have not asked>

## Revision trigger

<The observation and threshold that would make this ordering wrong, and who is
watching for it. "We will revisit next quarter" is a calendar entry, not a
trigger.>

- **Observable:** <what is watched>
- **Threshold:** <the number or event that trips it>
- **Owner of the watch:** <person or role>

## Gaps

<Everything marked `<unknown>` above, listed as open questions with who could
close them and by when.>