Mission
- Measurable outcome: By the end, you can frame at least three feasible alternatives, compare them against outcome-linked criteria, test sensitivity, and make a bounded recommendation with residual risk.
- Prerequisites: Chapter 3.1 ConOps and requirements; Chapter 2.3 quality attributes.
- Work product: An evidence-based option record.
- Time: 65–85 minutes.
Before you read: Predict → Commit → Connect
Meridian can migrate by big-bang cutover, region-by-region strangler transition, or long dual run. Predict your choice with only current knowledge. Write the three assumptions most likely to reverse it. Connect each assumption to a requirement or quality attribute.
A recommendation is an argument, not a score
Technical judgment does not mean choosing the architecture. It means ensuring the decision compares real alternatives against the right constraints, evidence, uncertainty, and consequences. An option record should let a skeptical reader reconstruct why a recommendation was reasonable at the time.
Start with the decision question and deadline. Preserve the status quo when it is feasible; it reveals the cost of acting and the cost of delay. Add meaningfully different options, not cosmetic variants. Screen out options that violate non-negotiable constraints, but identify the source authority. Teams often call preferences “constraints.”
Accessible diagram label: A decision moves from a bounded question through criteria, alternatives, evidence, trade-offs, and sensitivity to an accountable decision and revisit triggers.
Criteria derive from the ConOps, requirements, outcome chain, and guardrails: settlement integrity, customer continuity, security, time to evidence, reversibility, operational complexity, cost range, and strategic option value. Weights can clarify priorities, but a weighted matrix is not objective truth. Scores often combine unlike measures and hide uncertainty. Show raw evidence beside scores and test whether small weight or estimate changes reverse the ranking.
Accessible diagram label: Meridian’s migration alternatives produce different speed, concentration, coexistence, cost, and divergence consequences before the next reversible commitment.
Evidence has grades. A verbal opinion is different from a calculation; a calculation from a controlled test; a controlled test from observed production behavior. None is automatically sufficient. Production evidence may reflect a narrow condition, and a prototype may be appropriate before production. Label source, date, environment, sample, owner, and limitation.
Use sensitivity questions:
- Which uncertain input most changes the recommendation?
- What if traffic, cost, defect rate, or supplier lead time is twice the expected value?
- Which option fails a guardrail before it loses on average score?
- Which irreversible commitment occurs first?
- What small experiment buys the most decision-relevant information?
- What future option does each choice preserve or destroy?
At Meridian, region-by-region may appear slower than big bang, but it produces real reconciliation evidence and limits blast radius. It also prolongs coexistence, increasing schema and operational complexity. The recommendation should say “choose staged transition if compatibility and rollback tests pass by date X,” not “staged is best practice.” Context determines the answer.
Separate decision quality from outcome luck. A well-supported decision can encounter an unforeseen event; a weak decision can get lucky. Review whether the team framed alternatives, used available evidence, disclosed uncertainty, and established monitoring, not whether hindsight makes the winner obvious.
Decision rights: Who owns what?
Engineering and Architecture own technical feasibility and detailed consequences. Product owns value and priority. Finance validates financial methods. Domain authorities define policy constraints. The TPM owns the option process, cross-team consequence integration, assumption visibility, decision deadline, record, and propagation. A named decider chooses and accepts residual risk within authority.
Consensus can inform a decision but should not erase the decider. Record dissent when it contains material risk or missing evidence.
I do
For Meridian I frame: “Which transition strategy meets Canadian traffic goals while preserving settlement integrity and recoverability?” I compare big bang, staged regions, extended dual run, and status quo. Instead of assigning unexplained 1–5 scores, I show load-test results, rollback rehearsal time, monthly cost ranges, compatibility gaps, and assumptions. Sensitivity shows that one unknown, the dual-write divergence rate, drives the decision. I recommend a two-week shadow experiment before the irreversible routing commitment.
We do
Northstar can delay globally, launch a subset with the old component, or qualify a new supplier. Together define criteria and one decision-relevant experiment. Identify which stakeholder evidence is authoritative for each criterion.
Show the model answer
Model answer
Criteria include safety and regulatory compliance, field reliability, firmware compatibility, traceability, customer promise, country revenue timing, supplier capacity, cost range, reversibility, and service exposure. Hardware/Quality own qualification evidence; Regulatory owns filing impact; Engineering owns compatibility; Product/Sales own commercial consequences; Finance validates the cost model. A bounded pilot and accelerated qualification may reduce uncertainty, but cannot replace mandatory certification. Recommend conditionally and identify the latest decision date before logistics or public commitments become irreversible.
Scoring rubric (0–4)
- 0: Picks the fastest or cheapest option by opinion.
- 1: Lists options and generic pros/cons.
- 2: Uses outcome-linked criteria and named evidence.
- 3: Adds constraints, uncertainty, sensitivity, authority, and residual risk.
- 4: Also proposes a high-information experiment, tracks reversibility, records dissent, and defines revisit triggers.
You do
Use a live decision. Write at least three options including status quo. Ask owners for raw evidence and limitations. Test one alternative assumption and one weight change. Write a recommendation in this form: “Recommend X because evidence A and B best satisfy criteria C, provided assumptions D hold; otherwise choose Y. Residual risks are Z.”
Pause & Recall
From Chapter 3.1, retrieve the distinction among constraint, assumption, and design. From Chapter 2.3, explain why a trade-off matrix must preserve quality scenarios. From Chapter 1.3, who owns the option process and who owns the decision?
Production lens
Decision records decay unless linked to execution and monitoring. Add revisit triggers: failed test, cost threshold, new regulation, capacity change, or missed assumption. When a trigger occurs, reopen the decision without treating learning as disloyalty to the original choice.
Workplace artifact: option record
Decision question / deadline / decider:
Outcome, requirements, and guardrails:
Constraints and authority source:
Options, including status quo:
Criteria and stakeholder consequence:
Raw evidence, source, date, and limitation:
Cost / schedule ranges and assumptions:
Trade-offs by quality attribute:
Sensitivity and scenario tests:
Irreversible commitments / option value:
Dissent or missing evidence:
Recommendation and conditions:
Residual risk and accepter:
Actions, communication, and revisit triggers:Chapter compression
A defensible recommendation frames the decision, compares feasible alternatives, preserves raw evidence, discloses uncertainty, tests sensitivity, and names residual risk. Matrices aid reasoning but do not manufacture objectivity. The TPM integrates the decision; accountable authorities own domain judgments and the final choice.
Retrieval deck
- Q: Why include status quo? A: It exposes the costs and risks of both acting and delaying.
- Q: What is wrong with unexplained weighted scores? A: They can hide assumptions, incomparable measures, uncertainty, and veto constraints.
- Q: What does sensitivity analysis ask? A: Whether plausible changes in inputs or priorities reverse the recommendation.
- Q: What is option value? A: The value of preserving a future choice while uncertainty is resolved.
- Q: Is a bad outcome proof of a bad decision? A: No; evaluate the process and available evidence, while learning from outcomes.
Spaced review
- Now: State the decision question and three Meridian alternatives.
- 1 day: Reconstruct the option flow without notes.
- 3 days: Change one key assumption and test whether the recommendation reverses.
- 7 days: Add one high-information experiment and one revisit trigger.
- 14 days: Compare predicted evidence with observed results and update the record.