Dominated-Option Removal¶

A comparative-evaluation method — instantiates Search Space Pruning

Eliminates any option that another available option beats (or ties) on every criterion that matters, leaving only the genuine trade-offs to decide between.

Dominated-Option Removal prunes by comparison between concrete, already-listed options. An option is dominated when some other option on the table is at least as good on every relevant criterion and strictly better on at least one — meaning there is no reason anyone should ever pick the dominated one. Removing all such options leaves the non-dominated set (the Pareto frontier): the choices that each win on something, where picking between them requires a real value judgment about trade-offs. Its defining feature, versus the constraint- or heuristic-based siblings, is that it excludes nothing for failing an absolute bar; an option is cut only because a rival renders it pointless. And unlike Branch and Bound, which bounds unexplored regions of a search tree, this operates on a finite set of fully-specified options laid side by side.

Example¶

A team is choosing a laptop model to standardize on, from 20 candidates scored on price, weight, battery life, and repairability. One candidate — call it Model K — is cheaper, lighter, longer-lasting, and more repairable than Model M. There is no scenario, no weighting of the criteria, under which someone should prefer M over K: M is dominated and can be struck out entirely. Sweeping every pair this way removes 13 of the 20 options.

What remains are 7 non-dominated laptops, each best at something — the lightest, the cheapest, the longest battery — and the choice among them now genuinely depends on what the team values. The method did not decide; it cleared away every option that no rational preference would ever select, so the real deliberation happens over 7 candidates instead of 20, with no defensible choice discarded.

How it works¶

Fix the criteria — agree the dimensions that matter and their direction (more battery good, more weight bad); dominance is only meaningful relative to this set.
Compare pairwise — for each pair, check whether one option is ≥ the other on every criterion and > on at least one.
Remove the dominated — strike any option so beaten; it can never be the best choice under the stated criteria.
Return the non-dominated frontier — the survivors, each superior on at least one dimension, for human trade-off deliberation.

The distinctive logic is relational: the bar an option must clear is not fixed, it is "be un-beaten by everything else present."

Tuning parameters¶

Criteria set — which dimensions count. Add a criterion and fewer options are dominated (more survive); drop one and the frontier shrinks. Omitting a criterion someone cares about is how good options get wrongly cut.
Strict vs. practical dominance — whether to require strict superiority or allow "worse by a rounding-error amount" to count as dominated; practical dominance prunes more but can discard near-ties that mattered.
Measurement tolerance — how much noise in the scores before a "domination" is trusted; tight tolerance avoids cutting options beaten only within error bars.
ε-dominance margin — requiring a rival to win by a meaningful margin, not a hair, before it dominates — trims the frontier without over-cutting.

When it helps, and when it misleads¶

Its strength is that it prunes with almost no controversy: removing a dominated option throws away nothing anyone would rationally choose, so it shrinks a decision set before the hard value trade-offs, without pre-committing to any weighting. It sharpens deliberation down to the choices that actually involve a trade-off.

It misleads chiefly through a missing criterion. If a dimension people care about — say, vendor lock-in — is left out of the comparison, an option that was only "dominated" on the listed criteria may in fact have been the best on the unlisted one, and it is silently deleted.^[1] Treating noisy, uncertain scores as if they were exact also manufactures false dominations, cutting options beaten only within measurement error. The guard is to make the criteria list complete before pruning, and to require a margin beyond measurement noise before declaring one option dominated by another.

How it implements the components¶

candidate_or_region_representation — the unit is a fully-specified, enumerated option scored on each criterion; the method needs the options laid out, not a search tree to explore.
objective_or_success_criterion — dominance is defined relative to an explicit multi-criteria set, which the method requires stated up front and in direction.
dominance_check — its core operation is the pairwise dominance test; the non-dominated frontier is exactly what it returns.

It applies no absolute hard constraints — that is Constraint Filtering — and, unlike Branch and Bound, does not bound or prune unexplored regions; it compares only options already enumerated in front of it.

Instantiates: Search Space Pruning — Dominated-Option Removal is the comparison-based cut that yields a non-dominated set.
Sibling mechanisms: Branch and Bound · Shortlisting · Constraint Filtering · Beam Search · Decision Tree Pruning · Eligibility Screening · Negative Keyword Filter · Red-Flag Screen · Safety or Compliance Exclusion · Sample Audit of Exclusions · Triage Filter

Notes¶

Dominated-Option Removal and Shortlisting are natural partners but distinct: this method cuts only options that are provably pointless given the criteria, leaving a possibly-still-large frontier; Shortlisting then reduces that frontier to a working set by soft priority and diversity, which does discard some genuinely-viable options. Run dominance removal first — it is the uncontroversial cut — and let the judgment-laden reduction come after.

References¶

[1] The non-dominated set is the Pareto frontier — options where improving one criterion requires sacrificing another. Its integrity depends entirely on having the right criteria: a Pareto-efficient set computed over an incomplete criteria list can exclude the option that was best on the criterion nobody listed. ↩