Strategic sequencing and options

A team agrees on its bets and then orders them the way most teams do: easiest first, or most requested first, or in the order the work became ready.

The first two months look productive. Then the third bet starts and the team discovers that a data model chosen in month one makes it three times harder. A week later, evidence arrives showing the audience assumption behind bet two was wrong — evidence that could have been collected in week two for almost nothing. Two months of work now has to be reconsidered, and one part of it cannot be undone at a price anyone will pay.

Nothing here was a bad decision in isolation. The ordering was the decision, and nobody experienced it as one. Sequence feels like scheduling, which feels like a delivery concern, so it gets settled by readiness and convenience. But the order determines what you know when you make each commitment, and it determines which commitments are still available later. Those are strategic properties, and they are decided by default when nobody decides them on purpose.

Two separate costs come out of the same choice, and neither is visible while the order is being set:

flowchart LR
  O["Order set by readiness<br/>and convenience"] --> S1["Step one commits<br/>a data model"]
  S1 -->|"forecloses"| S3["Step three is now<br/>three times harder"]
  O --> S2["Step two builds on an<br/>untested audience assumption"]
  S2 -->|"evidence arrives late"| EV["A week-two check<br/>would have shown it"]
  S3 --> U["Two months reconsidered,<br/>part of it one-way"]
  EV --> U

The upper branch is a foreclosed option. The lower branch is information that arrived after the commitment it should have informed. Sequence decides both.

Ask this: what will we know after step one that we do not know now, and would it change step two? If step one produces nothing that could change step two, the ordering was chosen for a reason that has nothing to do with strategy.

Sequence is decided by two criteria in tension.

Criterion one: retire the most decision-relevant uncertainty first.

Uncertainty is decision-relevant when different resolutions lead to different next moves. Large uncertainty is not the same as decision-relevant uncertainty. A team can be very unsure about something that would not change any action, and quite confident about something that changes everything.

Uncertainty	Size	Decision-relevant?
Exactly how much a slowdown annoys users	Large	Only if the answer would change whether you fix it
Whether the audience you committed to has the problem at all	Moderate	Yes — a false answer invalidates the position
Which of two implementations is faster	Moderate	Rarely at strategy level; usually an execution choice

Criterion two: do not foreclose options you may need.

Some moves close doors. A data model, a pricing structure, a public commitment, a platform choice, a hire. The cost is not the work itself; it is what becomes unavailable afterwards. Classify each step:

Move type	Property	Sequencing implication
Reversible	Can be undone at low cost	Can go early; low information cost to being wrong
Costly to reverse	Undoing is possible and expensive	Prefer to place after the uncertainty it depends on is retired
One-way	Practically cannot be undone	Place last, or place after the discriminating evidence exists

Options are not free

"Preserve optionality" is the most respectable way to avoid choosing. Keeping an option open has a carrying cost: extra abstraction, delayed learning, a team that cannot specialize, and a product that reads as unfinished to every audience. State the carrying cost of each option you keep. If you cannot name a condition under which you would actually exercise the option, it is not an option — it is an unresolved decision with a better name.

Both of these keep the same door open on Noted. Only one of them is an option:

The difference is not caution. The weak version defers a decision and calls the deferral a strategy.

Dated exposures compress everything

A hard external date does not change the sequence logic, but it removes slack. Work backwards: latest start = date minus the pessimistic estimate minus the time you would need to react if the estimate is wrong. Compare that with your plan. When the latest start is already behind you, the sequencing question is over and the only remaining question is what gets cut.

Boundary

Sequencing produces value only when you can observe a result before the next commitment. If step one's outcome will not be visible until after step two has started, ordering them gives you no information benefit — you have split work, not sequenced it. Check the observation lag before claiming a learning sequence.

Option preservation also stops being useful in two cases. First, when there is no realistic path to exercising the option: an option nobody has the capacity or appetite to take is a permanent cost. Second, when delay is itself the dominant risk. Under a dated exposure or a fast-moving competitive situation, a slower sequence that keeps three options open can be worse than a faster one that keeps one — because the thing you were protecting expires while you protect it.

Read the Noted case if you have not already.

Take Signal 4: a core dependency loses support in 10 weeks, and sequence it against the portfolio you allocated in ST-04.

What the case gives you: engineering estimates the migration at three to five weeks, with wide uncertainty. Two engineers want to migrate now. Two want to spread the work across the quarter. After end-of-support, security patches stop. Nothing breaks immediately.

Laid against the clock, those facts are all you have:

timeline
  title The dated exposure, as the case states it
  Week 0 : Two engineers want to migrate now
         : Two want to spread the work across the quarter
  Weeks 3 to 5 : The migration estimate, with wide uncertainty
  Week 10 : Support ends
          : Security patches stop, and nothing breaks immediately

The timeline deliberately carries no latest-start marker. Deriving one — and deciding how much reaction time you need if the three-to-five week estimate is wrong — is step 4 of your task, and the arithmetic is the argument.

Your task. Produce a sequence for this cycle.

1. List every uncertainty currently sitting under your portfolio. For each, say what you would do differently under two different resolutions. Cross out the ones where the answer is "nothing."
2. Rank the survivors by decision relevance, not by size.
3. Classify each bet in your portfolio as reversible, costly to reverse, or one-way. Name at least one step that forecloses something.
4. Work backwards from the ten-week date using the pessimistic estimate. State the latest start date for the migration and show the arithmetic, including reaction time if the estimate is wrong.
5. Decide the migration question: complete it now, or spread it across the quarter. Defend the choice with the option and exposure logic, not with team preference.
6. Write the sequence as ordered steps. For each step, state what you will know at the end of it, and which later step that knowledge could change.
7. Check the observation lag on every claimed learning step. If the result is not visible before the next commitment, say so and either move it or drop the learning claim.
8. Name each option you are deliberately keeping open, its carrying cost, and the condition under which you would exercise it.

Expect to be pushed on: whether "spread the work" is a sequencing argument or a way to avoid saying what gets cut; whether your first step produces observable information in time to matter; and whether any option you kept open has a real exercise path or is an unresolved decision in disguise.

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

• Lists the uncertainties, crosses out every one where both resolutions lead to the same action, and ranks the survivors by decision relevance rather than by size.
• Classifies each bet as reversible, costly to reverse, or one-way, and names at least one step that forecloses something specific.
• Shows the latest-start arithmetic from the ten-week date, the five-week pessimistic estimate, and a stated reaction time — and lets that arithmetic, not the engineers' split preference, decide now-versus-spread.
• States for each step what will be known at its end and which later step that knowledge could change, and checks the result is visible before the next commitment.
• Gives every option kept open a carrying cost, an exercise condition, and an expiry.
• The most common weak move is "spread the work across the quarter" with nothing cut. It is not a sequence; it is a refusal to say which bet loses the weeks the migration needs.

Produce artifacts/ST-05-strategic-sequence.md using the template below.

Write it for a delivery lead who has to hold the order when something urgent arrives. They should be able to see which steps can move without damage and which ones cannot, and why.

In your Product Decision Case, this artifact adds time to the portfolio. It is also the record that tells a later reviewer whether a bad outcome came from a bad bet or from a bad order.

# Strategic sequence · <cycle or period>

- **Date:** <YYYY-MM-DD>
- **Portfolio this sequences:** <reference to the ST-04 artifact>
- **Owner:** <who can change the order>
- **Delivery lead applying it:** <who holds the order day to day>

## Decision-relevant uncertainties

| Uncertainty | If it resolves one way | If it resolves the other way | Decision-relevant? |
|---|---|---|---|
| <uncertainty> | <action> | <different action> | <yes / no — no means both actions are the same> |

<Keep only the ones marked yes. Rank them by how much of the strategy depends on
the answer, not by how unsure you feel.>

## Reversibility of each step

| Step | Reversible / costly to reverse / one-way | What it forecloses |
|---|---|---|
| <step> | <class> | <what becomes unavailable afterwards> |

## Dated exposures

| Exposure | Hard date | Pessimistic estimate | Reaction time | Latest start | Status |
|---|---|---|---|---|---|
| <exposure> | <date> | <duration> | <duration> | <date> | <ahead / behind> |

<If any latest start is already behind you, say so here and state what gets cut.
Do not carry a plan that the arithmetic has already ruled out.>

## The sequence

| # | Step | Why here | What you will know at the end | Which later step that could change | Observation lag |
|---|---|---|---|---|---|
| 1 | <step> | <retires uncertainty X / dated exposure / dependency> | <what becomes known> | <step number> | <when the result is visible> |
| 2 | <step> | <reason> | <knowledge> | <step> | <lag> |

<If a step's observation lag lands after the next commitment starts, either move
it or remove the learning claim from it.>

## Options held open

| Option | Carrying cost per week | Condition that would make you exercise it | Expiry |
|---|---|---|---|
| <option> | <cost — abstraction, delay, split focus> | <specific condition> | <when the option stops existing> |

## What can move and what cannot

- **Can move without damage:** <steps>
- **Cannot move:** <steps, and the reason — dependency, dated exposure, or
  foreclosure risk>

## Uncertainty

- **Weakest part of this sequence:** <where you are least confident>
- **What a bad outcome would tell you:** <whether it would indicate a bad bet or
  a bad order>

## Revision trigger

<The observation and threshold that would force a re-sequence mid-cycle, and who
would see it first.>

## Gaps

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