Predict decisions, not just labels
A model score earns its place by improving a declared decision.
An ETA model improves RMSE while increasing missed-delivery cost.
This chapter moves from write the decision sentence to apply constraints. We will make the idea concrete in a small code example before placing it back inside a production system with a declared latency, cost, and operator-capacity budget.
Choose a policy on validation data, then check queue capacity, protected slices, latency, and a safe no-model fallback.
Verify the behavior illustrated in review_policy.py, then measure decision utility, slice reliability, and system latency.
ROC-AUC alone on 0.1% prevalence hides unusable precision and an impossible review queue. An unsafe release must degrade to a named baseline or the last known-good version.
Build the idea one layer at a time.
A prediction becomes useful only when it changes an action. Start with the decision and work backward to the score the decision needs.
“Predict churn” omits the population, decision moment, forecast horizon, intervention, and value of each outcome. Two teams can model the same label correctly yet build incompatible products because one needs a weekly retention offer and the other needs a real-time support escalation.
A score should cross a threshold only when expected benefit exceeds expected cost, subject to capacity and harm constraints. This connects probability quality to an explicit policy rather than treating 0.5 as a law of nature.
- decision utility
- prediction cutoff
- censoring
- temporal splits
Follow the reasoning before touching the code.
- 01
Write the decision sentence
Name who is scored, at what cutoff, over which horizon, what action follows, and which outcomes carry value or harm.
- 02
Estimate action value
Combine event probability, recoverable loss, intervention effectiveness, and action cost in the same monetary or utility unit.
- 03
Apply constraints
Choose a policy on validation data, then check queue capacity, protected slices, latency, and a safe no-model fallback.
transactions = [
{"id": "a", "risk": 0.30, "loss": 100},
{"id": "b", "risk": 0.08, "loss": 500},
{"id": "c", "risk": 0.70, "loss": 20},
]
review_cost = 12
recovery_rate = 0.60
for tx in transactions:
expected_benefit = tx["risk"] * tx["loss"] * recovery_rate
action = "review" if expected_benefit > review_cost else "pass"
print(tx["id"], action, round(expected_benefit - review_cost, 2))Turn risk scores into review decisions using expected value.
What to notice
- Transaction b is reviewed despite its lower probability because the loss at stake is much larger.
- The policy threshold is not one fixed probability; it changes with recoverable value and review cost.
YOUR TURN Pause and predict before revealing the answer.
If reviewers recover only 20% of loss, which transactions still have positive expected value?
Need a hint?
Recompute risk × loss × 0.20 − 12 for each row.
Check your reasoning
None remain positive. Even transaction b, the closest case, yields 0.08 × 500 × 0.20 − 12 = -4, so all should pass under this policy.
Engineering foundations
Connect the chapter to the course’s quantitative spine.
A model score earns its place by improving a declared decision. The cost-aware threshold selection below is the shared course-level lens that helps quantify the surrounding system. Name each symbol, check its units or shape, and use only values that were available at the production cutoff.
Cost-aware threshold selection
| Symbol | Meaning / shape / unit |
|---|---|
t | decision threshold |
FN, FP | error counts at t |
c_FN, c_FP | business cost per error |
Open derivation and numerical substitution
Start from the production quantity being optimized, substitute the observed values with their declared units, then isolate the model-controlled term. Preserve shape annotations at each step so broadcasting or aggregation cannot silently change the result.
- Write the named inputs: t, FN, FP, c_FN, c_FP.
- Substitute one small, hand-checkable batch before vectorizing.
- Calculate an independent reference value and compare within a declared tolerance.
# equation → code contract
inputs = validate_shapes_and_units(batch)
value = compute_c02(inputs)
assert is_finite(value)Change one assumption and make the tradeoff visible.
Use the worked chapter as your prediction, then change one input in the lab and explain why the output moves. The lab runs predefined TypeScript and never executes learner code, so the focus stays on the concept rather than setup.
Cost-aware threshold
Move the cutoff over a fixed set of scored transactions.
Assumption: Fixed 12-event dataset; false positive costs 2 units and false negative costs 12.
Open nonvisual data table
| Item | Computed state | Interpretation |
|---|---|---|
| 0.92 | fraud | review |
| 0.84 | fraud | review |
| 0.76 | legit | review |
| 0.68 | fraud | review |
| 0.59 | legit | review |
| 0.51 | fraud | review |
| 0.43 | legit | pass |
| 0.37 | legit | pass |
| 0.31 | fraud | pass |
| 0.22 | legit | pass |
| 0.14 | legit | pass |
| 0.08 | legit | pass |
Trace the complete operating path.
Carry the chapter’s central idea into the system: A model score earns its place by improving a declared decision. The path below shows where that reasoning needs an artifact, a test, an owner, or a fallback.
- 01
Turn “Write the decision sentence” into a written invariant and a failing test.
- 02
Promote review_policy.py from illustration to a versioned implementation with representative fixtures.
- 03
Use “Apply constraints” as the release checkpoint: Choose a policy on validation data, then check queue capacity, protected slices, latency, and a safe no-model fallback.
- 04
Serve or materialize behind explicit a declared latency, cost, and operator-capacity budget.
- 05
Join telemetry to mature outcomes and retain a rollback path.
Observability
Join service health, input quality, prediction distributions, slice behavior, and mature outcomes by exact version.
Cost
Measure storage, preprocessing, compute, queueing, and human review under a representative arrival pattern.
Failure modes
ROC-AUC alone on 0.1% prevalence hides unusable precision and an impossible review queue. Add a detector, owner, mitigation, and stop condition for this class of failure.
Alternatives
Compare a rule, a simpler statistical baseline, and a different system boundary before adding model complexity.
Explain the contract, not just the vocabulary.
Late-delivery evaluation contract
Specify a cutoff-safe target, validation design, cost/capacity threshold, calibration report, uncertainty, and controlled experiment.
- Decision and metric contract
- Leakage-safe split tests
- Threshold and calibration analysis
- A/B guardrails and rollback
Read primary material with a purpose.
Return to the opening failure.
Store the decision contract, cutoff, split logic, threshold policy, uncertainty, and experiment design beside the model.
For this lesson, the release evidence is the worked code example, a hand-checked formal result, deterministic simulation output, a ≥80% checkpoint, the production rubric, and a named fallback. The course resolves when the system can produce a metric contract, leakage-safe validation, calibration analysis, and experiment plan.