From evidence to commitment

A PM finishes a strong evidence cycle. The belief is well argued, the readout is good, and leadership agrees. The roadmap gains a row. Everyone leaves the room feeling that a decision was made.

Three months later, four things are half-built. The feature that was "committed" has a design, no backend, and a slipping date. The migration that nobody formally committed to is being done in evenings. The performance work started and stopped twice. When someone asks what the team decided in that meeting, the honest answer is that the team decided to want something.

Nothing visibly went wrong. The evidence was good. The argument was sound. The failure is that adding a row to a plan is not a commitment. A commitment is a claim about capacity, and capacity is finite, so a real commitment always takes something away from something else. If nothing was removed, nothing was committed — the team simply agreed that the idea was good.

This is hard to see from inside because the emotional signal of commitment arrives on time. Agreement feels like closure. And the cost of a fake commitment is paid later, by the four workstreams that each get a fraction of a team.

Both paths start from the same good evidence and the same agreement. They diverge at one question, which is rarely asked out loud in the room:

flowchart LR
  E["Good evidence<br/>leadership agrees"] --> R["A row is added<br/>to the plan"]
  R --> N{"Was anything<br/>removed?"}
  N -->|"no"| P["Preference<br/>capacity still spoken for"]
  P --> S["Four workstreams<br/>share one team"]
  S --> H["All half-built<br/>dates slip late"]
  N -->|"yes, and it is named"| C["Commitment<br/>the cost is visible now"]

Ask your team: what stopped when we committed to this? If the answer is "nothing, we're fitting it in", you have a preference, not a commitment.

Two separate convictions must both be strong before a commitment is real, and they fail in different ways.

Value conviction is your belief that this change produces the intended outcome for a named population. It is built from evidence about the problem, the population, and the mechanism.

Effort conviction is your belief that you know enough about cost, dependency, and unknown work to hold a scope and a horizon. It is built from technical understanding, not from an estimate someone said out loud.

They combine into four states, and each state has a different correct move.

Value conviction	Effort conviction	Correct move
Strong	Strong	Commit, and name what is displaced
Strong	Weak	Do not commit yet. Spend a bounded slice retiring the effort unknown
Weak	Strong	Do not commit yet. Buy value evidence cheaply; building is the expensive test
Weak	Weak	Do not commit. Choose which conviction is cheaper to raise, and raise that one

The most common failure is the second row treated as the first. The team is confident the thing is worth building, so it commits to a date it has no basis for, and the date does the damage.

A commitment that is real states four things:

1. What it displaces. The specific alternative that now cannot happen, or cannot happen at full size, in this period.
2. What it withdraws. The capacity, attention, or future option that is no longer available — including the option to change your mind cheaply.
3. Who can no longer say yes. The named person or team whose availability is now spent, so that the next request to them is a conflict, not an addition.
4. What ends it. The condition under which you stop, reduce, or re-open the commitment — stated before you start, while you are still honest.

Displacement is the test that separates a commitment from an intention. If you cannot name what is displaced, you have not found the edge of your capacity, and you will find it later at a worse moment.

Here is the same commitment to Noted's enterprise summaries written both ways. Both sentences were said by someone who meant them:

If you cannot say which square you are in before you say what you will do, you are reasoning from the answer backwards:

flowchart TD
  Q["Name what stopped<br/>when you committed"]
  Q -->|"a named alternative,<br/>and its owner was told"| C["A commitment.<br/>Now write the exit condition."]
  Q -->|"nothing —<br/>we are fitting it in"| P["A preference.<br/>Find the edge of capacity first."]
  Q -->|"nothing could stop —<br/>the work is forced"| F["Not a bet.<br/>Use consequence over time,<br/>not value conviction."]

Boundary

This model assumes the alternatives are comparable product bets. It stops being true when the work is not a bet at all.

Some work is forced. A dependency losing support, a legal requirement, a security exposure — for these, "value conviction" is the wrong frame. There is no user outcome to argue about. The real frame is consequence exposure over time: what accumulates if you do not act, and how fast. Pushing forced work through a value-conviction template produces an invented user story to justify work that was never optional, and that invented story then competes for attention as if it were evidence.

The model also has a floor. A small, reversible change that costs two days does not need a displacement analysis. The overhead of formal commitment can exceed the consequence of getting it wrong. Use this when the work is large enough that its displacement is felt by someone who was not in the room.

And note the honest limit of the four-state table: strong effort conviction is not the same as an accurate estimate. It means you know which parts are known and which are not. A team with strong effort conviction can still be wrong about duration. It should not be surprised about kind.

Read the Noted case if you have not already.

Take Signal 2: three enterprise customers asked for automated meeting summaries.

Their combined contract value is roughly 20% of revenue. All three requests came through one account manager within one month. The underlying claim — that enterprise team leads lose decisions, owners, or context after recurring meetings — has never been tested. Meeting type is not captured in event data. The team collaboration environment is in design, not live.

Noted has one team and four signals. It cannot do all four.

Your task. Produce a commitment position for this signal.

1. State the commitment candidate as a change to a named population's job, not as a feature name. "Automated summaries" is a feature. Say what changes for whom.
2. State your value conviction, and state plainly what the evidence cannot establish. Three requests through one channel is not prevalence. Contract value is not evidence about the problem.
3. State your effort conviction. Name at least one dependency you do not control. The team collaboration environment is in design; decide whether your commitment depends on it, and say so.
4. Place yourself in the four-state table, and say which square you are in before you say what you will do.
5. Name what this displaces. Go through signals 1, 3, and 4 explicitly. For each, say whether it is deferred, reduced to a minimum guard, or dropped — and who is told.
6. Name who can no longer say yes, and to what.
7. Write the exit condition: the observation that would make you stop or shrink this commitment.

Step 5 is where most attempts thin out into "we'll fit the rest in". Fill this table before you write any prose. The middle and right columns are yours; the left two are what the case actually gives you.

Signal	What the case states	Deferred, reduced to a guard, or dropped	Who is told, and when
1 · Activation	Fell 11% over six weeks; not segmented by cohort, channel, or platform
3 · Large-workspace P95	Rose 34%; about 4% of workspaces; no support tickets
4 · Dependency end-of-support	10 weeks away; migration estimated at three to five

A blank cell in the right-hand column is a displacement nobody has been told about, which is the same as no displacement at all.

Expect to be pushed on: whether 20% of revenue is evidence about the problem or evidence about account risk; whether "we will still do the migration alongside" is a real displacement or a way of avoiding one; and whether your effort conviction survives the fact that the environment this feature would live in is only in design.

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

• A commitment candidate stated as a change to a named population's job, not as a feature name.
• Value and effort conviction stated separately, each with what its evidence cannot establish. At least one dependency outside your control is named, and the in-design collaboration environment is placed on or off the critical path in so many words.
• A square in the four-state table named before the move is named.
• Every displacement cell for signals 1, 3, and 4 filled with a fate and a person told — nothing "fitted in".
• An exit condition written as an observation, and a named person who can no longer say yes to something specific.
• The most common weak move is carrying the migration "alongside" as if it were free. It is weak because it displaces nothing on paper, so the displacement happens later, unowned, in evenings.

Produce artifacts/DS-01-commitment-brief.md using the template below.

Write it for the people whose work is displaced by this commitment, and for the person who will be asked in six weeks why the other three things did not move. It should carry the choice, the two convictions with their evidence, the uncertainty in each, the displaced alternatives, the owner, and the exit condition.

This entry converts your Product Decision Case from a set of beliefs into a set of obligations. Every later artifact in this phase refers back to it. When DS-05 asks what would make you stop a launch, it is asking you to re-open this file.

# Commitment brief · <short commitment name>

- **Date:** <YYYY-MM-DD>
- **Product:** <product>
- **Commitment owner:** <person who carries the consequence>
- **Specialist authority:** <who holds the technical judgment this depends on>
- **Horizon:** <the period this commitment covers>

## What is being committed to

<A change to a named population's job, stated as an outcome, not a feature name.
"Automated summaries" is a feature. Say what changes, for whom, and in what
situation.>

## Value conviction

- **Position:** <low | moderate | high>
- **Claim:** <the change you believe this produces, and the mechanism>

| Evidence held | Source | Population | Window | What it cannot establish |
|---|---|---|---|---|
| <claim> | <source> | <who> | <when> | <limit> |

**Strongest reason the value claim is wrong:** <the best argument against your
own position, stated as someone who believes it would state it>

## Effort conviction

- **Position:** <low | moderate | high>
- **What is known:** <the parts of the work whose shape you understand>
- **What is not known:** <the parts whose shape you do not understand — kind of
  unknown, not just duration>

| Dependency | Type (sequence / capability / knowledge) | Owned by | Status | Risk if it slips |
|---|---|---|---|---|
| <dependency> | <type> | <team or person> | <live / in design / planned / unknown> | <consequence> |

## Position in the conviction table

<Value strong or weak, effort strong or weak, and therefore which move this is:
commit, retire the effort unknown, buy value evidence, or do not commit. State
the square before you state the action.>

## What this displaces

| Displaced work | Treatment (deferred / reduced to minimum guard / dropped) | Who is told | Consequence of the displacement |
|---|---|---|---|
| <work> | <treatment> | <person or team> | <what accumulates or is lost> |

**Capacity withdrawn:** <who can no longer say yes, and to what>

**Options withdrawn:** <what becomes expensive to change once this starts>

## Forced-work check

<If this work is not a bet — a dependency, a legal requirement, a security
exposure — say so here and replace the value claim with consequence exposure
over time: what accumulates if you do not act, and how fast. Do not invent a
user story for forced work.>

## Exit condition

<The observation and threshold that would make you stop, shrink, or re-open this
commitment. Name the observable and the number. "If it is not working we will
reassess" is not an exit condition.>

- **Who may invoke it:** <person>
- **Earliest date you could learn you were wrong:** <date, and how>

## Not committed

<What this brief deliberately leaves open, so that no one infers more commitment
than you intend — scope beyond the first horizon, populations not covered,
capabilities assumed but not promised.>

## Gaps

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