▥ CIWAWEIGHT ASSESSMENT

FROM JUDGMENTS TO WEIGHTS

What matters most?

Rank your criteria. Compare hypothetical alternatives.
Find the weights that fit your judgments.

Coalition-Indexed
Weight Assessment

THE METHOD

Weights, with room for uncertainty.

CIWA presents two hypothetical alternatives with numerical performances on every criterion. You choose the alternative you prefer; the corresponding weight constraint is inferred internally. Criteria are ordered from most to least important, with nonnegative weights summing to 1.

The hypothetical alternatives use the performance information you provide. This version supports 3–30 criteria, with an editable block cutoff from 3 to 8 (suggested: 6). Larger sets use recursive prefix-cut comparisons to find near-balanced blocks, followed by maximin comparisons within each block. Split judgments enter the same global weight model. Skipping a split uses its current cut without adding a constraint. Within blocks, it uses explicit prefix implication propagation for compatible answers. Maximin ties maximize the sum of cone sizes, then use lexicographic vector order. Incompatible answers do not trigger logical pruning. The strict preference margin is selected during setup (default 0.001).

The solver minimizes total judgment slack and calculates each weight’s minimum and maximum on the exact optimal face, subject to numerical solver precision. Consistency feedback distinguishes exact compatibility, approximate compatibility below the strict preference margin, and comparisons requiring review. A secondary diagnostic minimizes the largest individual slack among minimum-total-slack solutions. Flagged comparisons can be kept, revised or withdrawn; no answer is changed automatically. The representative maximizes t subject to membership in the optimal admissible region and wⱼ − wⱼ₊₁ ≥ t for j = 1,…,n, with wₙ₊₁ = 0. This maximizes the minimum adjacent weight gap and the smallest weight; the coalition constraints remain part of the admissible region. It is one admissible choice, not a uniquely identified answer. Duplicate and individually dominated directional observations are screened before synthesis.

“Indifference” records equal support for the two alternatives. Any deviation from that equality in the diagnostic solution is flagged for review. “Skip” adds no constraint. A complete assessment can still have nonzero intervals. Intervals are marginal ranges: their endpoints cannot generally be combined into one valid vector.

Generated vectors are convex combinations of feasible extreme vectors in the current admissible region. They are examples, not uniform samples or a unique solution.

Explore admissible weight vectors

Answer consistency