Comparative Analysis skills for AI agents

Activate this skill when the user is comparing measured performance results (latency, throughput, accuracy, cost per unit, energy) across systems, versions, models or configurations and needs to report the comparison without misleading anyone. Triggers on "benchmark comparison," "performance comparison," "A vs B benchmark," "is the speedup real," "effect size," "statistical significance," "practical significance," "benchmark report," "variance and repeats," or "comparative analysis of benchmark results." Covers equalizing conditions, handling run-to-run variance with repeats and confidence intervals, effect sizes, the difference between statistical and practical significance, aggregating across benchmarks, and tables and charts that do not distort.

Activate this skill when the user wants to audit a comparison for bias, is worried that their own comparison is slanted, or must produce a comparison that a sceptical or adversarial reader will accept. Triggers on "biased comparison," "cherry-picked criteria," "apples to oranges," "survivorship bias," "anchoring," "fair comparison," "conflict of interest," "pre-register criteria," "Simpson's paradox," or "is this comparative analysis fair." Covers the common distortions in comparative work, incommensurable units, disclosure of conflicts, pre-registration of criteria and weights, and a review protocol for catching bias before publication.

Activate this skill when the user needs to compare two or more options, cases, vendors, policies, designs or datasets in a structured way and reach a conclusion that survives scrutiny. Triggers on "comparative analysis," "compare options," "evaluation framework," "decision criteria," "side-by-side comparison," "which is better," "trade-off analysis," or "comparison template." Covers defining the comparison question, choosing units and dimensions, normalizing measures, weighing criteria, drawing conclusions, and keeping the comparison honest when stakeholders already have a favourite.

Activate this skill when the user is designing, conducting or writing up a study that compares several in-depth cases (organizations, programmes, projects, regions, incidents) to explain outcomes or build theory. Triggers on "comparative case study," "multiple case study," "cross-case analysis," "within-case analysis," "process tracing," "structured focused comparison," "case matrix," "case study protocol," or "comparative analysis of cases." Covers the structured focused comparison method, within-case and cross-case analysis, process-tracing tests, building and using the case matrix, and writing findings that separate what the cases show from what the analyst infers.

Activate this skill when the user is designing or critiquing a comparison of a small number of cases (countries, regions, organizations, programmes, historical episodes) to explain an outcome rather than merely rank options. Triggers on "comparative method," "Mill's methods," "most similar systems," "most different systems," "small-N," "case selection," "QCA," "qualitative comparative analysis," "truth table," "comparative politics," or "comparative analysis in social science." Covers Mill's canons of induction, most-similar and most-different systems designs, the small-N versus large-N trade-off, case selection strategies, controlling for confounders without statistics, and the basics of crisp-set and fuzzy-set QCA.

Activate this skill when the user is comparing products, services or competitors for a buying decision, a market analysis, a positioning exercise or a public comparison page. Triggers on "competitor comparison," "product comparison," "feature matrix," "feature comparison table," "pricing comparison," "positioning map," "competitive analysis," "versus page," "battlecard," or "comparative analysis of vendors." Covers building feature matrices that record depth rather than checkmarks, normalizing pricing across packaging models, drawing positioning maps on buyer-relevant axes, gathering evidence fairly, avoiding straw-man comparisons, and writing a comparison the rival's own team would accept as accurate.

Activate this skill when the user is comparing options on money over time: build versus buy, on-premises versus subscription, two capital projects, or a policy against its alternatives. Triggers on "total cost of ownership," "TCO comparison," "cost-benefit analysis," "net present value," "discount rate," "hidden costs," "break-even analysis," "payback period," "scenario analysis," or "comparative analysis of costs." Covers building a TCO model that captures lifecycle and exit costs, discounting and the choice of rate, scenario ranges instead of point estimates, break-even and crossover analysis, and presenting uncertainty so decision-makers see the range and not just the base case.

Activate this skill when the user needs to present a comparison of options, groups, periods or conditions in a table or chart and wants the layout to reveal the differences rather than bury them. Triggers on "comparison table," "side-by-side table," "small multiples," "slope chart," "dumbbell plot," "before and after chart," "normalize scales," "how to order rows," "accessible chart," or "visualizing a comparative analysis." Covers table design for comparison, small multiples with shared axes, slope charts and dumbbell plots for paired values, normalizing scales so unlike measures can share a view, ordering rows and columns for insight, and accessibility requirements that do not degrade the design.

Activate this skill when the user is building or reviewing a scoring model that ranks options against weighted criteria, such as a vendor selection matrix, a prioritization scorecard or an evaluation rubric. Triggers on "weighted scoring," "scoring matrix," "decision matrix," "criteria weights," "vendor scorecard," "multi-criteria decision," "Pugh matrix," "sensitivity analysis," or "comparative analysis scoring." Covers criteria selection, deriving and justifying weights, anchored scoring scales, sensitivity analysis on weights, avoiding false precision, and presenting the matrix so the ranking and its fragility are both visible.