Values, Utilities, and What Matters

By the end of this lesson, you will be able to...

• Separate factual uncertainty from disagreement about values, stakes, and constraints.

• Turn hidden priorities into explicit decision criteria without pretending that one score is universally correct.

• Compare options when different stakeholders carry different costs and benefits.

Idea in one sentence: Facts tell us what may happen; values help us decide which consequences matter most and which costs are not acceptable.

Core Insight

Consider the checkout team after the cache incident. The previous lesson estimated a 20-35% chance of another serious failure for the current design. The team has four options:

• buy more capacity;
• build warm failover;
• protect the payment path with a degraded mode;
• delay the campaign and run more tests.

The team agrees on the probabilities. It still disagrees about the choice.

The SRE lead prefers warm failover because another outage could destroy trust. The finance lead prefers more capacity because the campaign deadline and budget matter. The product lead dislikes degraded mode because customers would lose recommendations. The operations lead prefers delay because the unknowns are still large.

This is not necessarily a disagreement about facts. It is a disagreement about consequences.

The team cannot remove that disagreement by collecting one more dashboard. It must make visible what each option protects, what it sacrifices, and which constraints are hard boundaries.

The Naive Idea: Facts Choose by Themselves

Teams often describe a decision as purely technical:

Pick the option with the lowest failure probability.

That rule sounds neutral. It quietly assigns unlimited value to avoiding failure and zero value to cost, delay, user experience, or learning speed.

Another team may use a different hidden rule:

Pick the cheapest option that can survive the campaign.

This rule treats budget as dominant and may accept a risk that users or operators would reject.

Both rules contain values. The problem is not that values exist. The problem is that nobody can inspect them, challenge them, or see who carries the cost.

The Promise and the Boundaries

Start with the promise from the decision frame:

Checkout must not charge a customer twice or lose a successful payment.

That is a hard constraint. It is not merely one preference among many.

Other dimensions are negotiable:

These dimensions are not all measured in the same unit. That does not make them imaginary. It means the team must explain how it compares them instead of hiding the comparison inside a technical score.

Plain meaning:

A value is a consequence or principle that matters when we decide.

In this scenario:

“Do not lose a successful payment” is a value expressed as a hard safety boundary. “Keep recommendations visible” is a product experience value that may be traded against availability.

Technical name:

Utility is a way to represent how much a particular outcome serves the goals and values of a decision-maker in a specific context.

Utility is not a universal measurement of human worth. It is a decision aid. Its usefulness depends on whose goals are represented, which consequences are omitted, and which constraints cannot be traded.

Values, Constraints, and Preferences

These three categories are easy to mix.

Constraint

A condition the option must satisfy. For this team, payment correctness is a constraint. An option that can duplicate charges is rejected even if it is cheap and fast.

Value dimension

A kind of consequence the team cares about: trust, availability, cost, speed, learning, or user experience.

Preference

A ranking among acceptable possibilities. The product lead may prefer full experience to degraded mode, but that preference can lose to the payment-correctness constraint.

Making the categories explicit prevents a common mistake: treating one person's preference as a universal law, or treating a hard safety boundary as a negotiable number.

A Worked Comparison

The team turns its options into a comparison card. The ratings are qualitative on purpose. They are prompts for conversation, not objective measurements.

The plus signs mean “helps this dimension under the current frame.” They do not yet mean “wins.”

Now the team asks who experiences each result.

This table changes the conversation. “Degraded mode is bad” becomes “which features can we remove while preserving payment correctness?” “Delay is expensive” becomes “what revenue do we lose, and what incident cost do we avoid?”

Utility Makes Conflict Visible

Suppose two decision records use the same probability estimate but different priority allocations.

The SRE lead assigns the most attention to trust and availability:

trust 40 | availability 35 | cost 10 | speed 10 | experience 5

The finance lead assigns the most attention to campaign timing and cost:

speed 35 | cost 30 | availability 20 | trust 10 | experience 5

These numbers do not prove that one person is rational and the other is not. They expose why the recommendations diverge. The next conversation can ask whether the weights are authorized, whether the people affected are represented, and which dimensions are actually non-negotiable.

If the team wants to calculate a rough comparison, it can map each option to a small scale such as -2 to +2 for each dimension and combine that with the agreed priority list. But the result should be treated as an audit trail, not as a machine-generated verdict.

The arithmetic is useful when it reveals a disagreement. It is dangerous when it disguises a political, ethical, or organizational choice as neutral math.

Check: Two teams agree that an option has a 25% chance of a five-minute outage. Team A chooses it; Team B rejects it. Does one team necessarily misunderstand the probability?

Think first, then reveal.

Answer: No. They may assign different importance to the outage, its cost, the deadline, or available safeguards. Probability describes what may happen; values help determine what is acceptable.

Choose, Record, and Name the Sacrifice

The checkout team chooses warm failover plus a limited degraded mode. It records:

1. the hard constraint: payment correctness;
2. the central evidence: residual failure risk is still material;
3. the values: availability and trust outrank complete recommendations during the campaign;
4. the sacrifice: some customers will see a simpler experience;
5. the review trigger: restore features only after the failover replay and canary meet the error and recovery thresholds.

The decision is now inspectable. A later reviewer can disagree with the priority order without pretending the team had no reasoning.

This is also where ownership matters. If customers carry the largest harm, their interests cannot disappear because the meeting contains only engineering and finance. If operators carry overnight recovery work, “operational cost” should not be treated as free labor.

Check: A team says, “We chose the cheapest option, so there was no value judgment.” What is missing?

Think first, then reveal.

Answer: Cost was treated as the dominant value, while availability, trust, user harm, and operational burden were either ranked lower or ignored. Cheapest is a value rule, not the absence of values.

What This Changes

Before this lesson, the team might ask:

Which option has the lowest probability of failure?

After this lesson, it can ask:

Which consequences matter, which are hard constraints, who carries each cost, and which trade-off are we willing to record?

This improves both decision quality and disagreement. A person can challenge the values or the evidence separately. The team can say, “We agree on the risk estimate but disagree about whether delay is acceptable,” instead of arguing about a fake technical answer.

The decision still does not become easy. Making values explicit can make conflict sharper. That is a feature when the hidden conflict would otherwise return during an incident.

Trade-offs and Limits

The central trade-off is between explicit value comparison and friction. Naming values improves accountability and makes hidden costs visible, but it takes time and can expose conflicts that a quick technical vote would hide.

Qualitative scores are often enough for a small decision. Detailed utility models may be useful when stakes are large, but they can create false precision. A score cannot automatically represent dignity, trust, fairness, or a cost carried by people who are absent from the meeting.

Values also change with context. A two-hour delay may be acceptable before a campaign and unacceptable during a safety incident. Record the context and decision owner so the criteria do not become timeless slogans.

This lesson does not teach behavioral economics, policy design, or a complete theory of social welfare. It teaches a practical boundary: separate facts, values, constraints, and ownership before pretending that optimization can choose for everyone.

Common Confusions

Confusion: Values are subjective, so they are not useful

Why it is tempting: people want decisions to be settled by facts alone.

Better model: values are inputs to action. Making them explicit allows challenge, authorization, and accountability.

Confusion: A utility score is the true value of an outcome

Why it is tempting: numbers look comparable across columns.

Better model: a score represents a chosen perspective under stated assumptions. It is a tool for exposing trade-offs, not a universal fact.

Confusion: Every value can be traded

Why it is tempting: optimization language treats all dimensions as flexible.

Better model: some constraints are hard boundaries. Payment correctness may reject an option regardless of its low cost.

Confusion: Stakeholder preference and stakeholder impact are the same

Why it is tempting: the loudest person in the room is easy to measure.

Better model: ask who bears the harm, who receives the benefit, and who has authority to set the criterion.

Practice: Make the Migration Criteria Explicit

Return to the event-store migration from the previous lessons. The cheap provider may reduce cost, but regional recovery is unknown. The team must choose among a full cutover, a parallel run, a delay, or staying with the current provider.

Create a decision-criteria table with:

1. one hard constraint;
2. four value dimensions;
3. the stakeholder or group carrying each important cost;
4. a qualitative comparison of the options;
5. one explicit trade-off the final decision must accept.

Then write two short recommendations: one from the finance perspective and one from the operations perspective. Explain why they differ even if they use the same probability estimate.

A good answer should distinguish facts from values, preferences from constraints, and visible costs from costs shifted to users or operators.

Connection to the Next Lesson

After values and probabilities are explicit, the next question is how much commitment each option requires. Reversibility and Optionality shows why a reversible experiment can be rational even when a permanent commitment has a slightly better expected outcome.

Resources

• [BOOK] Thinking, Fast and Slow — Focus: framing effects and the hidden assumptions behind judgments about gains and losses.
• [BOOK] How to Decide — Focus: separating probabilities, preferences, and decision criteria in practical choices.
• [ARTICLE] Stanford Encyclopedia of Philosophy: Decision Theory — Focus: preferences, outcomes, utility, and the limits of formal decision models.

Key Takeaways

• Facts describe possible outcomes; values determine which consequences matter for the decision.
• Separate hard constraints, value dimensions, preferences, and stakeholder impact before comparing options.
• Utility can make a perspective and its trade-offs visible, but it is not a universal moral score.
• The same probability can support different choices when people carry different costs or have different authorized priorities.
• The central trade-off is between explicit value comparison and friction; hidden values do not disappear when a team refuses to name them.