Minimax Regret Matrix¶
Decision model — instantiates Regret-Signal Calibration
Lays candidate options against uncertain future states, scores each option's regret in a state as its shortfall from that state's best option, and picks the option whose worst-case regret is smallest.
The Minimax Regret Matrix is the one member of this family that is fully prospective and formal. Rather than examining a single regret that has already landed, it reasons about regret you have not yet incurred, across a grid of choices and futures. For every candidate option and every uncertain state of the world, it computes the regret you would feel in that state — the gap between the option you picked and the best option you could have picked had you known the state — and then it chooses by minimizing the maximum of those regrets. Its defining move is a stance: protect against the worst possible regret rather than chase the highest expected payoff. It is the tool for choosing under genuine uncertainty, when you distrust your probabilities but still have to commit.
Example¶
An operations team must size a new warehouse — small, medium, or large — against uncertain demand that could come in low, medium, or high. They lay out a payoff grid, then convert it to regret: in each demand column they find the best-performing build and subtract every option's payoff from it. A small build has near-zero regret if demand is low but huge regret if demand booms (turned-away orders); a large build is the reverse (idle capacity in a soft market). The medium build is never the best in any single state, but its worst-case regret across the three states is the smallest of the three. Minimax selects it. (Figures here are illustrative.) The decision is defensible precisely because it names the worst regret it is willing to accept and picks the option that keeps that number lowest.
How it works¶
Enumerate the options and the plausible states; fill in a payoff for each cell; within each state (column) find the best payoff and subtract every option's payoff from it to get that cell's regret; take each option's maximum regret across all states; select the option with the smallest of those maxima. The regret tolerance threshold is where the decision rule lives — pure minimax is the maximally regret-averse setting, but the same grid supports relaxations: minimize expected regret if you trust priors, or accept any option whose worst-case regret sits under a tolerated bound.
Tuning parameters¶
- State set and priors — which futures are on the grid, and whether they carry probabilities. Pure minimax uses none; a probability-weighted variant needs them.
- Regret-aversion dial — minimax (worst-case) versus expected-regret versus a fixed tolerance bound. This is the knob that turns a scored grid into a choice.
- Value scale — the single unit all payoffs are expressed in, which determines what the matrix can and cannot see.
- Option/state completeness — how exhaustively the grid is built; a missing state or a spurious option can flip the answer.
When it helps, and when it misleads¶
Its strength is decision-making under deep uncertainty — no trustworthy probabilities, but a real need to bound the worst regret. It turns a fraught choice into an explicit, checkable argument.
Its failure mode is false precision: a tidy matrix launders soft, incommensurable values into hard numbers, and — as the archetype warns — it "should not replace stakeholder judgment when values are hard to quantify." It is also fragile to the option set: because regret is measured relative to the best option present, adding a dominated option can change which choice minimizes worst-case regret, an unsettling violation of the independence of irrelevant alternatives.[1] The classic misuse is treating the matrix output as the decision rather than as one input. The guarding discipline is to keep the value scale honest, stress-test the state set, and let the number inform judgment, not replace it.
How it implements the components¶
regret_gap_measure— every cell holds a regret gap (that state's best payoff minus the chosen option's), computed systematically across the whole grid.regret_tolerance_threshold— the minimax rule is a tolerance criterion on worst-case regret: the dial that converts the scored grid into a committed choice.
It aggregates everything to a single value scale in order to choose, and never decomposes one realized regret across the people who bear it — that stakeholder_regret_map and its multi-dimensional breakdown are its nearest twin Regret Gap Table's job.
Related¶
- Instantiates: Regret-Signal Calibration — the formal prospective chooser for decisions under uncertainty.
- Consumes: Counterfactual Plausibility Screen — to keep the option and state sets credible rather than fanciful.
- Sibling mechanisms: Actionability-Filter After-Action Review · No-Fault Learning Review · Counterfactual Plausibility Screen · Forgone-Alternative Decision Journal · Regret Gap Table · Regret Pre-Mortem · Commitment Reset Memo · Reversal-Window Check · Rumination Timebox
Editorial Notes¶
Form Classification¶
Form family: Decision, Gate & Allocation
Rationale: Minimax Regret Matrix operates as a case-specific gate, selection, routing, prioritization, or resource disposition because it lays candidate options against uncertain future states, scores each option's regret in a state as its shortfall from that state's best option, and picks the option whose worst-case regret is smallest.
Independent corroboration: The frozen evidence defines Minimax Regret Matrix as 'Lays candidate options against uncertain future states, scores each option's regret in a state as its shortfall from that state's best option, and picks the option whose worst-case regret is smallest', so its operative form is Decision, Gate & Allocation.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Operations Research
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Universal
Rationale: Minimax regret is a canonical decision-analysis rule for choice under uncertain states.
Related originating lineages:
- Economics & Finance — Decision theory and opportunity-loss reasoning materially shaped regret matrices.
Review resolution: Both independent reviews place the primary provenance in operations_research. The queued differences (domain_reach_disagreement) concern secondary metadata, not primary lineage. The final retains economics_finance only where a reviewer supplied a formative-lineage rationale; downstream use or broad applicability by itself is not treated as origin. origin_mode=cross_disciplinary_synthesis because the supplied rationales identify formative contributions that are composed in the mechanism's present form. domain_reach=universal records established application breadth separately from provenance. confidence=high preserves the more cautious evidence assessment. encyclopedia_synthesis=false records whether either reviewer identified deliberate corpus-level composition.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Minimax regret criterion — a rule for choice under uncertainty that selects the option minimizing the maximum possible regret, introduced by the statistician Leonard J. Savage in 1951 as an alternative to maximizing expected payoff; because regret is scored relative to the best option available, the ranking can shift when the option set changes. withdrawn registry ↩