Trade-off Decision Frameworks

Mature methodologies for structuring trade-off decisions, drawn from architecture, decision science, medicine, intelligence analysis and safety engineering. Use as a menu to test in parallel against real decisions.

---

1. Systems & Software Architecture

Method	Core idea	What to steal
ATAM (SEI, Kazman)	Evaluate architecture against quality-attribute scenarios via a utility tree	Sensitivity point = decision strongly affecting one attribute. Trade-off point = decision affecting two attributes in opposite directions. Naming which one you're at removes most of the fog.
CBAM	ATAM + economics	Attach cost, benefit and uncertainty ranges to each branch of the utility tree
Tyree & Akerman (IEEE Software, 2005)	Rich ADR template	Issue, assumptions, constraints, positions (rejected options with argument), implications, related decisions
Y-statements (Zimmermann)	One-sentence decision record	In the context of X, facing Y, we chose Z, to achieve W, accepting V
Kruchten decision ontology	Decision taxonomy + relations	Existence / property / executive decisions; relations: constrains, enables, subsumes, conflicts
ISO/IEC/IEEE 42010	Architecture description standard	Decisions and rationale are first-class elements, not commentary

---

2. Decision Analysis (the parent discipline)

• Howard/Raiffa decision basis — every decision = what you know + what you can do + what you want. Bad write-ups collapse these three.
• Decision Quality chain (Spetzler) — frame · alternatives · information · values · reasoning · commitment. Quality = weakest link. Use as an audit checklist.
• Value-Focused Thinking (Keeney) — establish values/criteria before generating alternatives. Otherwise you only evaluate what was handed to you.
• Even swaps (Hammond, Keeney, Raiffa) — neutralise attributes one at a time to make options directly comparable. No tooling required.
• Kepner-Tregoe Decision Analysis — musts (binary screening) vs wants (weighted scoring), then an explicit adverse-consequences pass.

---

3. Multi-Criteria Decision Analysis (MCDA)

• AHP (Saaty) — pairwise comparison with a consistency ratio check
• MAUT / SMART — utility functions with swing weights (weights must reference the actual range of each criterion)
• TOPSIS — rank by distance to ideal and anti-ideal solutions
• Outranking school — ELECTRE, PROMETHEE (Bernard Roy)

Roy's framing: MCDA structures a debate, it does not compute a truth. Good default posture for vendor comparisons.

---

4. Design Rationale — closest fit to "help me make this trade-off"

• QOC (Questions, Options, Criteria) — a small formal grammar: options assessed against criteria, criteria stated separately from options. Maps almost directly onto a prompt template.
• IBIS / dialogue mapping (Rittel, Conklin) — issue → positions → arguments pro/con. Built for wicked problems.
• Toulmin argument model — claim · grounds · warrant · backing · qualifier · rebuttal. The missing warrant (why the grounds support the claim) is why most ADRs feel hand-wavy.

---

5. Science & Evidence

• GRADE Evidence-to-Decision (medicine) — arguably the best structured trade-off template in existence. Rates separately: benefits vs harms, certainty of evidence, values, resource use, equity, acceptability, feasibility — before stating a recommendation.
• IPCC calibrated language — separate likelihood from confidence; confidence derived from evidence quality × degree of agreement.
• Multiverse / specification-curve analysis — report the conclusion across all defensible analytic choices instead of one path.
• Strong inference (Platt) — design each step to discriminate between competing hypotheses.
• Structured expert elicitation — Cooke's classical method, SHELF, Delphi.

---

6. Intelligence Analysis

• ACH (Analysis of Competing Hypotheses, Heuer) — score evidence against all hypotheses; seek disconfirmation; rank by least-disconfirmed.
• ICD 203 analytic standards — mandates calibrated probability language, explicit assumptions, and separation of judgment from sourcing.
• Key assumptions check, premortem (Klein), red team / devil's advocacy.

---

7. Safety & Assurance Engineering

• Assurance cases with GSN — goal → strategy → solution, with context and assumptions as explicit graph nodes. Toulmin made diagrammatic; excellent at forcing "what would have to be true".
• ALARP — reduce risk until further reduction is grossly disproportionate to benefit.
• STPA, FMEA — systematic hazard and failure-mode enumeration.

---

8. Business & Economics

• Cost of Delay / CD3 (Reinertsen) — quantify the cost of not deciding
• Real options / deferral value — what is it worth to decide later with more information
• Reversibility — one-way vs two-way doors; match rigour to reversibility
• Wardley mapping — situational awareness before choosing
• RAPID / DACI — decision rights, separate from decision content
• Cynefin — match method to problem type (clear / complicated / complex / chaotic)

---

Composite Scaffold

A synthesis to test as an ADR template or prompt fragment:

1. Frame & decision type — reversible? existence / property / executive? Cynefin domain?
2. Criteria and weights — stated before options (Value-Focused Thinking, swing weights)
3. Options — including the null option and rejected ones with argument (Tyree)
4. Assessment matrix — options × criteria (QOC)
5. Sensitivity vs trade-off points — named explicitly (ATAM)
6. Evidence certainty — per claim, calibrated (GRADE, ICD 203)
7. Key assumptions — and what would falsify each (ACH, premortem)
8. Decision, warrant, accepted downside (Y-statement, Toulmin)
9. Consequences — and follow-on decisions unlocked or constrained (Kruchten)

---

Suggested Parallel Test

Run the same real decision through three at once — they fail in different ways and the disagreement is the signal:

• QOC — fast, low ceremony, good for small decisions
• GRADE EtD — heavy, but exposes weak evidence and unstated values
• ACH — best when you suspect you've already anchored on an answer