Trade-offs and opportunity cost

A PM runs a prioritization session. Every candidate gets a score. The top three go on the roadmap. Everyone leaves satisfied.

Eight weeks later, two of the three have slipped, the third shipped thin, and the enterprise account manager is asking why nothing happened. The PM looks back at the session notes and finds scores, ranks, and a roadmap. Nowhere is there a sentence saying what was given up.

Here is the specific move that failed. Asked to choose between fixing large- workspace response times and building enterprise summaries, the PM said: "we'll do performance first, then summaries." That sentence feels like a decision. It is not. It is a sequence, and it hides the actual trade-off, which is that summaries are being delayed by however long performance takes — an amount nobody estimated, against a bar nobody set.

The failure is hard to see because a prioritized list looks exactly like a set of decisions. Ranking is not choosing. A choice exists only when something specific stops being possible, and someone can name what it was. A list where everything survives in some form has decided nothing; it has only re-ordered the disappointment.

Follow the top path and notice that nothing is ever given up — which is exactly why the cost surfaces two months later, in someone else's inbox:

flowchart LR
  L["Scores and ranks"] --> R["Roadmap<br/>everything survives"]
  R --> S["'Performance first,<br/>then summaries'"]
  S --> H["Displacement hidden<br/>inside a sequence"]
  H --> W["Eight weeks on:<br/>two slipped, one thin"]
  S -.->|"what makes<br/>it a choice"| N["Delayed by N weeks<br/>Cost X, borne by Y<br/>Accepted because Z"]

Ask after any prioritization: what is now not happening, who was expecting it, and when will we tell them? If nobody can answer with a name and a date, the session produced a ranking, not a decision.

A real trade-off is visible on five faces. Most teams show one — usually cost — and call it analysis.

Face	The question	What hides when it is missing
Displaced option	What specifically stops happening, and who was expecting it	The chosen option looks free
Resource	Which scarce capacity is consumed, and is it substitutable	"We'll find the time" replaces a real constraint
Risk	What gets worse, or more likely, if this is wrong	Downside lands on a population nobody named
Future flexibility	Which options close, and when they reopen	You discover in month three that you locked something in month one
Reversal cost	What it costs to undo this in four weeks	Reversible and one-way choices get treated identically

Opportunity cost is measured against the best displaced alternative

Not against doing nothing. This is the most common error in product prioritization, and it is invisible from inside.

When a team says "the performance work is worth it because slow workspaces cost us paid seats," they are comparing the work to inaction. The real comparison is against the best thing those same engineers would otherwise have built. If that alternative was worth more, the performance work is a loss even though it is positive on its own.

This is why PJ-02 comes first. Two directions cannot be compared. Two thresholds can. If you cannot state what each option buys, in what amount, for which population, you cannot say which one you are giving up.

Capacity is not fungible

Teams reason as if engineering time is a pool of hours. It is not. The two engineers who can do a dependency migration may be the same two who can do the storage-layer work behind large-workspace performance. Moving "40% of capacity" between two projects can mean moving 100% of the only people who can do either.

Before you write a trade-off, name the specific scarce resource: not hours, but the people, the review bandwidth, the design attention, or the single decision-maker whose time everything queues behind.

Sequencing is a trade-off, not an escape from one

"First A, then B" displaces B by the duration of A. That delay has a cost, and the cost is usually paid by a population that was not in the room. Sequencing is often the right answer. It is never a way to avoid stating what was given up.

Write it as: B is delayed by approximately N weeks. The cost of that delay is X, borne by Y. We accept it because Z.

Here is that sentence applied to the two Noted signals, next to the version most teams actually say:

The difference is that the strong version can be refused. The account manager can read it, recognise the description of their own loss, and argue that the delay is worth more than the P95 recovery. That argument is the decision.

One caution about the strong version: "the storage-layer engineers" is an exercise input. The case does not say who can do the response-time work. When you write yours, name the real people or write <unknown>. Do not borrow a team from an example.

Boundary

Not every choice deserves a trade-off record.

For a cheap, reversible, low-consequence choice, the record costs more than the decision. A team that documents every trade-off will document them badly, and will slow down exactly where speed was free. The dividing line is consequence multiplied by reversal cost — which is the subject of PJ-04.

The second limit is more dangerous. A trade-off record written after the decision is not analysis; it is a defence. Once a choice is made, every face of it can be described in a way that makes the choice look correct. If you find yourself writing a trade-off record to explain a commitment already announced, label it as a rationale, not as a decision record, and do not let it enter the evidence base as if it were one.

Read the Noted case if you have not already.

Take Signal 3: P95 response time rose 34% for large workspaces, and put it directly against Signal 2: three enterprise customers asked for automated meeting summaries.

Take it as the exercise premise that both need the same engineering team this cycle. The case does not say so, but it is the condition under which a trade-off exists at all. Both have advocates. The case gives you:

• Large workspaces are more than 500 documents, about 4% of workspaces, holding a disproportionate share of paid seats. No support tickets mention speed. The signal came from monitoring.
• The three summary requests came through one account manager within one month. Their combined contract value is roughly 20% of current revenue. The underlying claim about lost decisions and owners has never been tested. Meeting type is not captured in event data.

Two axes separate these two options: how well the mechanism is actually observed, and how far the consequence reaches. One reading looks like this:

quadrantChart
  title Two Noted signals, one engineering team
  x-axis Mechanism untested --> Mechanism observed
  y-axis Consequence contained --> Consequence reaches paid revenue
  quadrant-1 Strong case to act now
  quadrant-2 Pressure without evidence
  quadrant-3 Park it
  quadrant-4 Small stakes, cheap certainty
  Large-workspace P95: [0.78, 0.6]
  Enterprise summaries: [0.2, 0.72]

The placement is the argument, not a fact from the case. Summaries sit high only if you treat 20% of revenue as exposed, and nothing in the case says those contracts are at risk — move that point down and the chart reverses. P95 sits right because monitoring observed it, but no support ticket mentions speed, so you could argue it right and low instead. Neither point can be placed at all until both options are thresholds rather than directions.

Your task. Choose one. Then make the choice real.

1. State each option as a threshold, using your PJ-02 work. What does each one buy, in what amount, for which population, by when? If you cannot write both, say which one you cannot write and what that tells you.
2. Name the scarce resource specifically. Not "engineering capacity" — which people, and what else are they the only ones who can do.
3. Choose. One option, stated as a choice.
4. Write the displaced option in the words of the person who loses it. For summaries, that is the account manager. For performance, that is whoever is watching paid-seat retention. Write it so they would agree it is fair.
5. State what the delay costs, with a rough magnitude and who bears it. If your answer is "we do it next cycle," say what next cycle then loses.
6. Name which future options close and when they reopen.
7. State the reversal cost: what it costs to switch back in four weeks.
8. Name the one thing that, if you learned it next week, would flip your choice.

Expect to be pushed on: whether "do both, just staged" is hiding in your plan; whether you weighed the 4% of workspaces by count or by paid seats, and whether you said which; and whether the 20% revenue figure entered as evidence of value or as pressure, given that nothing in the case says those contracts are at risk.

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

• Both options are written as thresholds — or the answer admits that the summaries option cannot be written as one yet, because its underlying claim is untested, and says what that admission does to the comparison. Either is strong. Pretending both are thresholds is not.
• The scarce resource is people or a named bottleneck, not hours. If the case does not say who can do the work, the answer writes <unknown> rather than inventing a team, and says what that gap does to the plan's credibility.
• One option is chosen. The other is displaced with a length of delay, a cost, and a named bearer — written so the account manager, or the person watching paid-seat retention, would accept the description of their own loss.
• Future flexibility and reversal cost are answered as separate questions: which options close, when they reopen, and what switching back in four weeks costs in work already done.
• The flip condition is a single learnable fact with a source — something the support lead's recordings, monitoring, or the account manager could deliver next week — not "if priorities change".
• The most common weak move is "performance first, then summaries." It is weak because it is a sequence, not a choice: nothing is given up on paper, so the cost surfaces two months later, in someone else's inbox, with no record of who accepted it.

Produce artifacts/PJ-03-trade-off-record.md using the template below.

Write it for the person who loses. The displaced option should be described well enough that its advocate would accept the description, even while disagreeing with the conclusion. That is the difference between a record and a defence.

This extends your Product Decision Case. PJ-02 said what good looks like for one thing. This artifact is the first time two definitions of good compete, and one of them loses.

# Trade-off record · <chosen option> over <displaced option>

- **Date:** <YYYY-MM-DD>
- **Product:** <product>
- **Decided before or after commitment:** <before | after — if after, this is a rationale>
- **Owner of the consequence:** <person or role>
- **Specialist authority:** <who holds expertise this choice depends on>

## The options, as thresholds

| Option | What it buys | Amount | Population | By when |
|---|---|---|---|---|
| <option A> | <change> | <threshold> | <behaviour-and-context boundary> | <horizon> |
| <option B> | <change> | <threshold> | <behaviour-and-context boundary> | <horizon> |

<If you could not write one of these rows, say which and why. An option you
cannot state as a threshold is one you cannot honestly compare.>

## The scarce resource

- **What is actually constrained:** <the specific people, review bandwidth, design attention, or decision-maker — not "engineering capacity">
- **Substitutable?** <can anyone else do this work, and at what cost in time or quality>
- **What else this resource is the only source of:** <the hidden queue behind it>

## The choice

<One sentence. The option being taken.>

## Displaced option

**Stated by the person who loses it:**

<Write the displaced option's case in the words of its advocate. If they would
not recognise it as their argument, rewrite it.>

- **Who was expecting it:** <name or role>
- **When they will be told:** <date>
- **Delay, if this is a sequence:** <approximately N weeks>
- **Cost of that delay:** <magnitude> — borne by <who>
- **Why that cost is accepted:** <one or two sentences>

## Risk

| If this choice is wrong | What gets worse | Who bears it | How early it becomes visible |
|---|---|---|---|
| <failure mode> | <consequence> | <population> | <date or event> |

## Future flexibility

| Option that closes | When it closes | When it reopens | What reopening costs |
|---|---|---|---|
| <option> | <date or event> | <date or event, or "not on this horizon"> | <cost> |

## Reversal cost

- **Cost to switch back in four weeks:** <what is lost — work, momentum, trust, or contract position>
- **Reversibility:** <reversible | costly to reverse | one-way>

## What would flip this

<One thing you could learn in the next two weeks that would change the choice.
Name the observation, not the topic. If nothing would flip it, this was not a
trade-off — it was a foregone conclusion, and you should say so.>

## Gaps

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