Rank Candidates

---
name: rank-candidates
description: "Rank, classify, screen, or select typed candidates under explicit criteria and supplied hard constraints. Object schema and decision mode are parameters; pairwise and portfolio cases may jump to dedicated tactics."
---

# rank-candidates

## Purpose

Rank, classify, screen, or select typed candidates under explicit criteria and hard constraints.

## Input contract

```yaml
mode_contracts:
  gap-prioritization: &rank_input
    required: [candidates, criteria, decision_rule]
    optional: [weights, evidence, hard_constraints, object_schema]
    constraints: [criterion_direction_and_missing_value_policy_must_be_explicit]
  direction-selection: *rank_input
  mcda-best-choice: *rank_input
  full-ranking: *rank_input
  category-sorting: *rank_input
  non-compensatory-screening: *rank_input
  rapid-triage: *rank_input
  stakeholder-weighted: *rank_input
```

## Execution protocol

Do not perform called SOP operations inline; each loaded SOP owns its contract and thresholds.

1. Normalize the candidate schema and separate hard constraints from preferences. You MUST load skill `define-criteria` to derive explicit criteria and criterion directions from the objective and candidate set.
   If the candidates cannot be made comparable because the research goal still contains unresolved branches, consider `decompose-research-goal` before ranking.
2. Select `gap-prioritization`, `direction-selection`, `mcda-best-choice`, `full-ranking`, `category-sorting`, `non-compensatory-screening`, `rapid-triage`, or `stakeholder-weighted`.
   If candidates are better compared pairwise than scored absolutely, consider `pairwise-ranking`. If the decision requires selecting a jointly feasible subset rather than ordering independent candidates, `portfolio-optimization` may be the better next tactic. If feasibility is the binding uncertainty rather than relative merit, consider `analyze-constraints-readiness`.
3. You MUST load skill `normalize-comparison-scale` to put heterogeneous criterion values on a declared comparable scale. You MUST load skill `score-object` to score every typed candidate against the supplied rubric and evidence. You MUST load skill `aggregate-ranking` to combine the criterion or comparison outputs under the declared rule.
4. You MUST load skill `assess-sensitivity` to perturb the declared weights or decision inputs, report stability, and return ordered or categorized candidates with rationale.
   If a prioritized gap or direction is ready to become a testable proposition, consider `formulate-hypotheses` as the next tactic.

## Mode branches

For `mcda-best-choice`, `full-ranking`, `category-sorting`, or `stakeholder-weighted`, You MUST load skill `elicit-weights` to produce and validate the criterion-weight vector. For `category-sorting` or `non-compensatory-screening`, You MUST load skill `set-threshold` to justify the decision boundaries.

- `gap-prioritization`: rank research gaps by evidence deficit and expected value so scarce investigation effort reaches the most consequential unknowns first. You MUST load skill `normalize-gap` to normalize heterogeneous gap records before ranking.
- `direction-selection`: narrow competing research directions against explicit scope, evidence coverage, and feasibility constraints before committing to one. You MUST load skill `assess-goal-feasibility` to test candidate directions against resources, obstacles, and timeline.
- `mcda-best-choice`: combine normalized multi-criteria scores with declared weights to select the strongest feasible option while preserving criterion-level rationale.
- `full-ranking`: produce a complete ordered list using at least the required comparison methods, exposing incomparable pairs and method disagreement.
- `category-sorting`: assign candidates to threshold-defined classes when ordinal categories are more defensible than fine-grained ranks.
- `non-compensatory-screening`: apply hard thresholds and vetoes so a severe failure on one criterion cannot be hidden by strengths elsewhere. You MUST load skill `apply-veto-filter` to enforce hard failures. You MUST load skill `check-dominance` to expose dominated and non-dominated alternatives.
- `rapid-triage`: use coarse importance and feasibility passes to reduce a large candidate set quickly, retaining elimination reasons for later review.
- `stakeholder-weighted`: aggregate perspective-specific scores with an explicit consensus rule, showing where stakeholder rankings converge or diverge.

## Output contract

```yaml
mode_contracts:
  gap-prioritization:
    produces: [ranking_or_categories, scores, sensitivity_results, recommendation]
    delta_fields: [findings, decisions, uncertainties, recommended_jumps]
  direction-selection: &screening_output
    produces: [ranking_or_categories, scores, eliminated_candidates, recommendation]
    delta_fields: [findings, decisions, uncertainties, recommended_jumps]
  mcda-best-choice:
    produces: [scores, weights, sensitivity_results, recommendation]
    delta_fields: [findings, decisions, uncertainties, recommended_jumps]
  full-ranking: &weighted_ranking_output
    produces: [ranking_or_categories, scores, weights, sensitivity_results, recommendation]
    delta_fields: [findings, decisions, uncertainties, recommended_jumps]
  category-sorting:
    produces: [ranking_or_categories, scores, weights, recommendation]
    delta_fields: [findings, decisions, uncertainties, recommended_jumps]
  non-compensatory-screening: *screening_output
  rapid-triage: *screening_output
  stakeholder-weighted: *weighted_ranking_output
```

## Thresholds and quality gates

- Full ranking: select >=2 ranking methods when the source protocol calls for method comparison.
- Weight elicitation: select >=2 weighting methods where required.
- Priority sensitivity: perturbation scenarios cover the declared weight space at a justified relative floor, each scenario annotated; retain the observed ranking-stability verdict.
- Direction narrowing imports report candidate/evidence-pool coverage, full-text coverage, independent-source ratio, batch increment, stopping reason, and source references rather than fixed paper/page counts.
- For every relative gate, declare the eligible candidate/evidence universe (denominator) and record the covered candidates or evidence items (numerator), the batch increment, the stopping reason, and source references; a gate is not passable when any audit field is missing.
- Stop direction search when marginal information gain falls below the justified floor across batches; report the resulting saturation state alongside coverage ratio and independent-source ratio.
- Non-compensatory and category modes must expose threshold/veto values; never hide them in prose.

## Failure and counterexamples

Reject rankings with undefined criterion direction, unhandled missing values, or hard constraints treated as compensable scores. Do not collapse pairwise active ranking into scalar ranking.

## Provenance map

20 architecture `old` entries; shared criteria->core->aggregate kernel retained. Portfolio and pairwise operations remain separate nodes, not silently absorbed.

## Legacy context checkpoint / Delta notes

Append candidate set hash, criteria/weights, rule, ranking, sensitivity scenarios, exclusions, and unresolved trade-offs.

## Preserved source criteria ledger

| source | source line | kind | source criterion |
|---|---:|---|---|
| multi-criteria-ranking | 49 | numeric | Each dimension is scored independently (1-5) to avoid cross-contamination between dimensions. Weights are set by AHP (Analytic Hierarchy Process) or specified by the user. Final score = $\Sigma$(dimension score * dimension weight). |
| multi-criteria-ranking | 51 | numeric | **Sensitivity check**: perturb weights by +/-20%; if the ranking is unchanged the conclusion is robust; if the ranking flips it must be flagged as "weight-sensitive". |
| multi-criteria-ranking | 53 | textual | ## Budget Gate |
| multi-criteria-ranking | 57 | numeric | \\| S \\| 5-8 \\| >=3 dimensions \\| Optional \\| Ranking table + attack suggestions for top 2 gaps \\| |
| multi-criteria-ranking | 58 | numeric | \\| M \\| 9-15 \\| >=4 dimensions \\| Required \\| Ranking table + attack suggestions for top 3 gaps \\| |
| multi-criteria-ranking | 59 | numeric | \\| L \\| 16-20 \\| >=5 dimensions \\| Required (multi-weight scenarios) \\| Ranking table + attack suggestions for top 5 gaps + weight-sensitivity report \\| |
| evidence-based-prioritization | 51 | textual | ## Budget Gate |
| evidence-based-prioritization | 55 | numeric | \\| S \\| 3-8 \\| all 6 dimensions \\| >=2 supporting references per gap \\| ranking table + evidence-void report \\| |
| evidence-based-prioritization | 56 | numeric | \\| M \\| 9-15 \\| all 6 dimensions \\| >=3 supporting references per gap \\| ranking table + evidence-void report + attack suggestions for top 3 gaps \\| |
| evidence-based-prioritization | 57 | numeric | \\| L \\| 16-20 \\| all 6 dimensions \\| >=5 supporting references per gap \\| ranking table + detailed evidence map + attack suggestions for top 5 gaps \\| |
| stakeholder-weighted-ranking | 35 | textual | - A consensus must be built across parties, or the ranking differences across perspectives must be shown |
| stakeholder-weighted-ranking | 58 | textual | ## Budget Gate |
| stakeholder-weighted-ranking | 62 | numeric | \\| S \\| 5-10 \\| 2-3 classes \\| Simple average \\| Per-perspective rankings + consensus top-3 \\| |
| stakeholder-weighted-ranking | 63 | numeric | \\| M \\| 11-20 \\| 3-5 classes \\| Borda count \\| Per-perspective rankings + consensus top-5 + divergence analysis \\| |
| stakeholder-weighted-ranking | 64 | numeric-table | \\| L \\| 20+ \\| 5+ classes \\| Weighted Borda + sensitivity \\| Full perspective matrix + consensus ranking + divergence heatmap \\| |
| rapid-triage | 49 | numeric | **Round 2: light scoring (1-3 points, two dimensions)** |
| rapid-triage | 51 | numeric | - Importance (1-3): a rough estimate of field impact |
| rapid-triage | 52 | numeric | - Feasibility (1-3): whether progress can be made within 6 months with existing resources |
| rapid-triage | 56 | textual | **Key insight**: the three round-1 questions must be answered quickly (no more than 30 seconds per gap); no deep analysis allowed. Speed is the core value of this strategy. |
| rapid-triage | 58 | textual | ## Budget Gate |
| rapid-triage | 60 | numeric-table | \\| Tier \\| Input gap count \\| Round-1 retention rate \\| Round-2 output \\| Final output \\| |
| rapid-triage | 62 | numeric | \\| S \\| 50-80 \\| <=60% \\| top-15 \\| Candidate set + elimination-rationale summary \\| |
| rapid-triage | 63 | numeric | \\| M \\| 81-150 \\| <=50% \\| top-20 \\| Candidate set + elimination-rationale summary \\| |
| rapid-triage | 64 | numeric | \\| L \\| 150+ \\| <=40% \\| top-30 \\| Candidate set + elimination-rationale summary + category statistics \\| |
| rapid-triage | 71 | numeric | 4. Call the `importance-scoring` SOP on the Keep set (1-3 coarse score) |
| rapid-triage | 72 | numeric | 5. Call the `feasibility-scoring` SOP on the Keep set (1-3 coarse score) |
| priority-sensitivity-testing | 28 | numeric | This tactic first establishes baseline weights (AHP or equal weights), then systematically perturbs the weights (+/-20%), observes the ranking changes, and finally gives a stability verdict. |
| priority-sensitivity-testing | 35 | numeric | \\| weight-perturbation \\| Apply +/-20% perturbations to each dimension weight and recompute the ranking \\| Second step, systematic perturbation \\| |
| priority-sensitivity-testing | 42 | numeric | 2. weight-perturbation: apply +20% and -20% perturbations to each dimension in turn (the remaining dimensions are adjusted proportionally to keep the sum at 1), producing a ranking for each perturbation scenario |
| priority-sensitivity-testing | 47 | numeric | - Perturb only the highest-weight dimension (+/-20%), producing 2 perturbation scenarios |
| priority-sensitivity-testing | 52 | numeric | - weight-perturbation expands the perturbation range to +/-30% and adds extreme scenarios (one dimension's weight set to 0) |
| priority-sensitivity-testing | 55 | textual | ## Minimum Yield |
| priority-sensitivity-testing | 58 | numeric | - Ranking results for at least 3 perturbation scenarios (each scenario annotated with its perturbation content) |
| priority-sensitivity-testing | 60 | numeric | - A final stability verdict: **Stable** (top N unchanged across all scenarios) / **Partially Sensitive** (1-2 changes in the top N) / **Highly Sensitive** (more than 2 changes in the top N) |
| best-option-selection | 17 | numeric | - Moderate number of candidates (3-15) |
| best-option-selection | 21 | numeric | \\| Base SOP \\| Target \\| +/-10% Range \\| |
| best-option-selection | 23 | numeric | \\| criterion-definition \\| 5-8 criteria \\| 4-9 \\| |
| best-option-selection | 24 | numeric-table | \\| weight-elicitation-sop \\| 1 weight vector \\| 1 \\| |
| best-option-selection | 25 | numeric-table | \\| alternative-scoring \\| 1 score matrix \\| 1 \\| |
| best-option-selection | 26 | numeric-table | \\| normalization \\| 1 normalized matrix \\| 1 \\| |
| best-option-selection | 27 | numeric-table | \\| scoring-synthesis \\| 1 recommendation \\| 1 \\| |
| full-ranking | 22 | numeric | \\| Base SOP \\| Target \\| +/-10% Range \\| |
| full-ranking | 24 | numeric | \\| criterion-definition \\| 5-8 criteria \\| 4-9 \\| |
| full-ranking | 25 | numeric-table | \\| weight-elicitation-sop \\| 1 weight vector \\| 1 \\| |
| full-ranking | 26 | numeric-table | \\| alternative-scoring \\| 1 score matrix \\| 1 \\| |
| full-ranking | 27 | numeric-table | \\| normalization \\| 1 normalized matrix \\| 1 \\| |
| full-ranking | 28 | numeric-table | \\| rank-comparison \\| 1 agreement matrix \\| 1 \\| |
| full-ranking | 29 | numeric-table | \\| scoring-synthesis \\| 1 full ranking \\| 1 \\| |
| full-ranking | 64 | numeric | 2. Select >=2 ranking methods (recommended: PROMETHEE II + MAVT) |
| full-ranking | 66 | textual | 4. Pay attention to incomparable pairs in partial orders (specific to ELECTRE III) |
| category-sorting | 22 | numeric | \\| Base SOP \\| Target \\| +/-10% Range \\| |
| category-sorting | 24 | numeric | \\| criterion-definition \\| 5-8 criteria \\| 4-9 \\| |
| category-sorting | 25 | numeric-table | \\| weight-elicitation-sop \\| 1 weight vector \\| 1 \\| |
| category-sorting | 26 | numeric-table | \\| threshold-setting \\| 1 threshold set \\| 1 \\| |
| category-sorting | 27 | numeric-table | \\| alternative-scoring \\| 1 score matrix \\| 1 \\| |
| category-sorting | 28 | numeric-table | \\| scoring-synthesis \\| 1 classification \\| 1 \\| |
| non-compensatory-screening | 27 | numeric | \\| Base SOP \\| Target \\| +/-10% Range \\| |
| non-compensatory-screening | 29 | numeric | \\| criterion-definition \\| 3-5 screening criteria \\| 2-6 \\| |
| non-compensatory-screening | 30 | numeric-table | \\| threshold-setting \\| 1 threshold set \\| 1 \\| |
| non-compensatory-screening | 31 | numeric-table | \\| conjunctive-filter \\| 1 pass/fail list \\| 1 \\| |
| non-compensatory-screening | 32 | numeric-table | \\| dominance-check \\| 1 dominance report \\| 1 \\| |
| non-compensatory-screening | 65 | textual | 2. Invoke threshold-setting to define minimum thresholds for each criterion |
| non-compensatory-screening | 75 | textual | **Screening Rule:** Conjunctive rule (all criteria must be met) |
| weight-elicitation | 23 | numeric | \\| Base SOP \\| Target \\| +/-10% Range \\| |
| weight-elicitation | 25 | numeric | \\| criterion-definition \\| 5-8 criteria \\| 4-9 \\| |
| weight-elicitation | 26 | numeric | \\| weight-elicitation-sop \\| >=2 methods \\| 2-3 \\| |
| weight-elicitation | 27 | numeric-table | \\| rank-comparison \\| 1 comparison \\| 1 \\| |
| weight-elicitation | 60 | numeric | 2. Select >=2 weighting methods (recommended: AHP + BWM or Swing + Simos) |
| weight-elicitation | 76 | numeric | - AHP CR: [value] (< 0.1 pass) |
| direction-narrowing | 36 | numeric | - `broad-paper-search`: at least 80 papers scanned |
| direction-narrowing | 37 | numeric | - `deep-web-search`: at least 30 web pages read in full |
| present-and-ask | 29 | numeric | User's selected 1-2 fields of interest + reasoning. |

## Context checkpoint / Delta notes

Append candidate set hash, criteria/weights, rule, ranking, sensitivity scenarios, exclusions, and unresolved trade-offs.