Dominance Filtering¶
Comparison method — instantiates Bounded Search Pruning
Removes a candidate only when another candidate is at least as good on every criterion and strictly better on one — a purely relative proof needing no bound or threshold.
Dominance Filtering discards an option only when the case against it is relational and airtight: some other option in the set is at least as good on every criterion that matters and strictly better on at least one. When that holds, the dominated option can never be the right choice under any reasonable weighting of the criteria — anything it could offer, its dominator already matches or beats — so it is removed. The one idea that distinguishes this mechanism is that it references no absolute yardstick at all: no optimistic bound, no incumbent to beat, no qualifying threshold. It compares candidates only to each other, and the proof of exclusion is the dominance relation itself. What survives is the set of options that are not dominated by anything — the trade-off frontier — leaving the genuine choices intact and only the strictly-worse options gone.
Example¶
A buyer is choosing a laptop from a shortlist of a dozen, caring about three things: price (lower better), weight (lower better), and battery life (higher better). Rather than scoring each with weights nobody agrees on, they filter by dominance. Model A costs more than Model B, weighs more, and has shorter battery — B beats or ties A on all three, so A is dominated and struck out; no weighting of the three criteria could ever prefer A over B. The same test removes three other models that some rival matches-or-beats across the board. But the light-but-pricey ultrabook and the cheap-but-heavy workhorse both survive: neither dominates the other, because each wins on a criterion the other loses. The shortlist collapses from twelve to four real trade-offs — and the buyer can be certain nothing worth considering was thrown away, because every discarded model had a strictly better twin still on the list.
How it works¶
- Name the criteria and their directions. Fix the dimensions that matter and which way is better on each; dominance is defined only relative to this set.
- Compare candidates pairwise. Candidate X dominates Y when X is no worse than Y on every criterion and strictly better on at least one.
- Remove the dominated. Any candidate with a dominator is struck; the exclusion's justification is the dominating candidate, recorded as the certificate.
- Keep the frontier. What remains is the non-dominated set — the mutually incomparable options that represent genuine trade-offs, none provably worse than another.
Tuning parameters¶
- Criteria set — which dimensions count as "matters." Adding a criterion makes dominance harder to establish (more ways to escape being dominated), so fewer candidates are filtered; dropping one filters more aggressively but may discard on an incomplete picture.
- Strictness of "no worse" — whether near-ties count as equal. Treating tiny differences as ties enables more filtering but can strike a candidate that was genuinely, if barely, better on something.
- Weak vs strict dominance — filtering on weak dominance (ties allowed everywhere, strict nowhere) removes duplicates too but risks dropping an equivalent alternative worth keeping for other reasons.
- Tolerance bands — allowing a small margin before a difference "counts" prunes more but blurs the frontier's edge.
When it helps, and when it misleads¶
Its strength is a guarantee earned without any absolute measurement: it removes only options that are provably worse than a specific alternative, so the surviving frontier is exactly the set of defensible choices.[n1] It needs no scoring model, no threshold, no incumbent — just the candidates and their attributes — which makes it robust precisely where people cannot agree on weights.
It misleads when the criteria set is incomplete. Dominance is only valid over the dimensions written down; if a candidate is dominated on price, weight, and battery but wins on an unlisted attribute the buyer secretly cares about — repairability, brand, a specific port — filtering it out is a real loss dressed as a proof. The classic misuse is running the filter on a convenient subset of criteria and treating the survivors as the whole truth. The guarding discipline is to make the criteria set genuinely complete before trusting any dominance, and to hold a candidate whenever a criterion that might matter has been left out of the comparison.
How it implements the components¶
dominance_or_feasibility_certificate— the dominance relation is the certificate: the exclusion of Y is justified by naming the specific candidate X that beats-or-ties it everywhere and wins somewhere.objective_and_constraint_reference— the criteria and their directions define what "at least as good on every dimension" means; dominance is only meaningful relative to this reference.
It holds no incumbent_solution and screens against no moving front-runner — that is Bound-Based Candidate Screening — and it computes no optimistic upper_lower_bound on a branch's reachable value, which is Admissible Heuristic Search. Dominance compares candidates only to one another, never to an absolute limit.
Related¶
- Instantiates: Bounded Search Pruning — Dominance Filtering is the purely relative form of safe exclusion, pruning only what another candidate provably beats.
- Sibling mechanisms: Admissible Heuristic Search · Bound-Based Candidate Screening · Constraint Propagation · Diagnostic Tree Pruning · Legal Issue Pruning Matrix · Pruning Audit Log · Branch and Bound · Feasibility Certificate Check
Editorial Notes¶
Form Classification¶
Form family: Decision, Gate & Allocation
Rationale: Dominance Filtering operates as a case-specific gate, selection, routing, prioritization, or resource disposition because it removes a candidate only when another candidate is at least as good on every criterion and strictly better on one — a purely relative proof needing no bound or threshold.
Independent corroboration: The frozen evidence defines Dominance Filtering as 'Removes a candidate only when another candidate is at least as good on every criterion and strictly better on one — a purely relative proof needing no bound or threshold', so its operative form is Decision, Gate & Allocation.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Operations Research
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Multi-objective optimization cohered Pareto-dominance filtering as removing a candidate only when another is no worse on every criterion and better on at least one.
Related originating lineages:
- Economics & Finance — Pareto's welfare analysis supplied the underlying dominance concept.
- Mathematics — Order theory formalized partial orders and maximal non-dominated sets.
Review resolution: Multi-objective optimization cohered the filtering procedure within operations research, drawing materially on Pareto economics and order theory; those are formative alternates within a single methodological lineage rather than independent convergences.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Pareto dominance — one option Pareto-dominates another when it is no worse on every objective and better on at least one. The set of options that nothing dominates is the Pareto frontier, the trade-off surface among which no choice is objectively wrong. The concept traces to the economist Vilfredo Pareto. ↩