agentleFS
peterbamuhigirepeterbamuhigire/business-plan-skills/skills/meta-strategy/meta-statistics/SKILL.md

meta-statistics

Skill4 starsChanged 18 days agoMITView on GitHub

Use when market sizing, survey results, forecasts, or comparative metrics need statistical rigour. Use the relevant plan-section skill for section drafting.

The file

Show its rules and text

We split this skill into the rules it gives an agent

20 rules. About 0 are common, 18 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈18

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this meta-skill to apply statistical discipline to business-plan evidence.

    line 16
  2. Distinctive:

    Use when market sizing, survey results, forecasts, or comparative metrics need statistical rigour.

    line 20
  3. Distinctive:

    Use when a chart, rate, trend, or average could be misleading if calculated badly.

    line 21
  4. Distinctive:

    Use when the plan needs defensible quantitative presentation.

    line 22
  5. Distinctive:

    Do not use to create false precision from weak data.

    line 26
  6. Distinctive:

    Do not apply advanced methods when the data does not support them.

    line 27
  7. Distinctive:

    Do not use statistics to decorate claims that remain strategically weak.

    line 28
  8. Distinctive:

    Check whether the data quality and structure fit the intended method.

    line 48
  9. Distinctive:

    Use standard deviation to describe spread within one variable.

    line 140
  10. Very specific:

    Read references/statistics-for-business-plans.md Chart Selection for the full decision table.

    line 155
  11. Distinctive:

    Never use 3D charts adds visual noise, distorts magnitudes

    line 167
  12. Distinctive:

    Use direct labels (percentages on slices) not a separate legend

    line 171
  13. Distinctive:

    Keep displays simple remove all decoration that carries no information

    line 178
  14. Very specific:

    Use colours that are clearly distinguishable (not red/orange/brown together)

    line 180
  15. Distinctive:

    When the plan includes a primary customer survey or field research, document the sampling method.

    line 189
  16. Distinctive:

    When the plan cites a survey finding ("70% of customers prefer X"), always attach a confidence interval.

    line 204
  17. Distinctive:

    Use simple or multiple linear regression when 3+ years of historical data exist.

    line 217
  18. Distinctive:

    Use adjusted R2 (not R2) to decide whether additional variables improve the model

    line 222
  19. Distinctive:

    Only predict within the observed range of x (no extrapolation beyond data)

    line 223
  20. Distinctive:

    Every statistic in the plan must pass this three-question test:

    line 233

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- dual-compat-start -->

Meta Statistics

Overview

Use this meta-skill to apply statistical discipline to business-plan evidence. It helps choose the right method, summary, chart, or inference so numbers support decision-making rather than decorative credibility.

Use When

  • Use when market sizing, survey results, forecasts, or comparative metrics need statistical rigour.
  • Use when a chart, rate, trend, or average could be misleading if calculated badly.
  • Use when the plan needs defensible quantitative presentation.

Do Not Use When

  • Do not use to create false precision from weak data.
  • Do not apply advanced methods when the data does not support them.
  • Do not use statistics to decorate claims that remain strategically weak.
  • For meta-statistics, route to the relevant plan-section skill instead when the request is section drafting rather than cross-section analysis.

Required Inputs

InputSource / providerRequired?If absent
Statistics brief and decision audienceClient, plan owner, or approved project filesYesStop before making a recommendation; state the missing decision context.
Claims, assumptions, and supporting evidenceSource register, model, research notes, interviews, or operating recordsYesSeparate known facts from assumptions and return a qualified gap list.
Authority and delivery constraintsRequesting owner and repository instructionsYesRemain read-only and produce a draft or review only.
  • The dataset, figures, or claim being analysed
  • The decision or message the numbers need to support
  • Data quality, sample size, and sourcing context
  • Any adjacent sections that rely on the same metrics

Workflow

  1. Identify what decision or claim the numbers must support.
  2. Check whether the data quality and structure fit the intended method.
  3. Apply the simplest sound statistical treatment that answers the question.
  4. Present the result using the right summary or chart.
  5. Reconcile the statistic with the surrounding commercial narrative.
  6. Flag weak data, weak sampling, or over-interpretation risks.

Decision, stop, and recovery controls

  • Decision point: confirm that the requested output is the statistical review record and that the decision concerns which inference, comparison, or chart the evidence supports.
  • Stop condition: halt the affected conclusion if required evidence is missing (dataset, sampling method, metric definitions, and calculation files) or if the work could lead to this identified risk: turning a biased sample or unstable average into a business claim.
  • Recovery: obtain the missing record or reviewer, repeat the affected check, and update the exception record before release.

Quality Bar

  • The chosen method fits the data and question.
  • Results are interpretable by the intended audience.
  • Charts and metrics clarify rather than distort.
  • Any uncertainty or estimate basis is explicit.

Anti-Patterns

  • Using averages where distributions matter more.
  • Plotting charts that make the wrong comparison easy.
  • Presenting regression or inference without adequate data.
  • Treating estimated data as measured fact.
  • Treating a generic statistics template as a conclusion. Correction: tie each choice to the named audience, evidence, and operating constraint.
  • Applying the wrong neighbouring route to meta statistics. Correction: confirm the decision and route to the named neighbour before analysis.
  • Treating an assumption as verified evidence. Correction: label it, cite its source or owner, and assign a verification action.
  • Recommending action without a decision threshold. Correction: state the measurable acceptance condition and review trigger.
  • Recording an unavailable check as passed. Correction: mark it not assessed and state the consequence for the decision.
  • Mutating or publishing during an analysis-only task. Correction: remain read-only until the owner gives explicit authority.

Outputs

ArtefactConsumerObservable acceptance condition
Statistics deliverableNamed decision-maker or plan authorThe recommended choice, assumptions, countercase, and next action are explicit.
Evidence and exception registerReviewer, funder, board, or implementation ownerEvery load-bearing claim is sourced or labelled as an assumption; missing checks are not shown as passes.
  • Statistically sound summaries, methods, or visualisation guidance
  • Corrected metrics or interpretation notes
  • Any data-quality caveats affecting confidence

Overview

This skill transforms raw numbers into credible, decision-quality evidence. Every figure in a business plan must be:

  1. Correctly computed right method for the data type
  2. Properly presented chart/table chosen for the message, not decoration
  3. Verifiably sourced primary source cited, or flagged as estimate with basis

Source books: Anderson et al. Essentials of Statistics for Business and Economics (7th ed., Cengage 2013); Keller Statistics for Management and Economics (9th abbrev. ed., Cengage 2011).


Section-by-Section Application Guide

Plan SectionPrimary StatisticMethodReference
04 Market AnalysisTAM/SAM/SOM sizingFrequency tables + totals from sourced dataSee statistics-for-business-plans.md Market Sizing
04 Market AnalysisMarket growth trendGeometric mean (not arithmetic)Growth Rates
05 Target MarketCustomer segment profilesDescriptive stats + bar/stacked bar chartsDescriptive Stats
06 Competitive AnalysisCompetitor positioningScatter diagram (2 axes: price × quality)Chart Selection
07 Marketing/SalesCustomer survey resultsConfidence intervals + sample designSampling & CIs
10 Financial ProjectionsRevenue forecastingRegression (if 3+ years historical data exist)Regression
10 Financial ProjectionsRisk comparison across marketsCoefficient of variationVariability
10 Financial ProjectionsMulti-year growth rateGeometric mean (CAGR)Growth Rates

1. Growth Rates Always Use Geometric Mean

Rule: Any multi-period growth rate (revenue CAGR, population growth, market expansion) must use the geometric mean never the arithmetic mean.

Why it matters: Arithmetic mean of annual returns overstates true compounded growth.

  • Year 1: +100%, Year 2: −50% Arithmetic mean = +25% (misleading: you broke even)
  • Geometric mean = 0% (correct: $100 $200 $100)

Formula:

CAGR = (End Value / Start Value)^(1/n) − 1
Geometric Mean = n(x1  x2  ...  xn)

When to cite: "Revenue grew at a CAGR of 18.4% over 5 years (geometric mean of annual growth rates, 2019–2024)."


2. Variability and Risk Comparison

Use standard deviation to describe spread within one variable. Use coefficient of variation (CV) to compare variability across variables with different scales or units.

CV formula: CV = (s ÷ x) × 100%

Business plan application:

  • Compare revenue stability of two product lines: Line A (mean UGX 45M/month, CV = 8%) is more predictable than Line B (mean UGX 120M/month, CV = 31%)
  • Compare supply-chain risk across sourcing geographies
  • Compare customer cohort reliability (repeat purchase rate CV)

When to cite: "Supply from Supplier X shows a coefficient of variation of 12% vs. Supplier Y at 34%, indicating greater supply reliability from X (Anderson et al., 2013)."


3. Chart and Table Selection

Read references/statistics-for-business-plans.md Chart Selection for the full decision table. Summary:

Message / Data TypeChart
Market share or fund allocationPie chart (max 5 slices) or bar chart
Revenue trend over timeLine chart
Revenue mix by periodStacked bar chart
Competitor comparison (2 attributes)Scatter diagram
Distribution of customers by revenue sizeHistogram
Multiple KPIs on one screenDashboard (bar + line + KPI card)

Hard rules from Anderson et al. (2013):

  • Never use 3D charts adds visual noise, distorts magnitudes
  • Axis must start at zero when comparing absolute values
  • Label axes with units (UGX millions, % market share)
  • Maximum 5–6 pie slices; use "Other" for remainder
  • Use direct labels (percentages on slices) not a separate legend

4. Data Visualisation Best Practices (Section 2.5, Anderson et al.)

  1. Give the chart a clear, concise title that states the conclusion or key fact
  2. Keep displays simple remove all decoration that carries no information
  3. Label every axis with the variable name and unit of measure
  4. Use colours that are clearly distinguishable (not red/orange/brown together)
  5. Scale axis from zero unless explicitly noting a truncated axis
  6. For dashboards: group related charts, use borders, minimise scrolling
  7. Strategic vs. tactical layers: operational dashboard (granular, real-time); executive dashboard (aggregate KPIs)

5. Sampling Design for Primary Research

When the plan includes a primary customer survey or field research, document the sampling method.

Population TypeMethodMinimum Sample
Homogeneous (single industry/area)Simple random samplingn ≥ 30
Heterogeneous (multiple segments)Stratified random samplingn ≥ 30 per stratum
Geographically dispersed, cost-sensitiveCluster samplingVaries; see Sampling

Sample size for confidence interval: n = (z2 2) / E2 where E = desired margin of error, z = 1.96 for 95% CI


6. Confidence Intervals for Market Research Claims

When the plan cites a survey finding ("70% of customers prefer X"), always attach a confidence interval.

Formula (proportion): p ± 1.96 × √[p(1−p)/n]

Example (n=150, 70% preference): CI = 0.70 ± 1.96 × √(0.70×0.30/150) = 0.70 ± 0.073 = [62.7%, 77.3%]

Write-up: "70% of surveyed customers (n=150) reported preference for X [95% CI: 63–77%] (primary survey, March 2026)."


7. Regression for Forecasting

Use simple or multiple linear regression when 3+ years of historical data exist.

Decision rules:

  • One driver (e.g., population sales): simple regression y = b0 + b1x
  • Multiple drivers (population + income + competition): multiple regression
  • Use adjusted R2 (not R2) to decide whether additional variables improve the model
  • Only predict within the observed range of x (no extrapolation beyond data)

Reporting standard: "A simple linear regression of monthly revenue on customer base (R2 = 0.87, p < 0.001, n=36 months) projects revenue of UGX 82M at a customer base of 2,400."

See references/statistics-for-business-plans.md Regression for worked examples.


8. Verifiability Standards

Every statistic in the plan must pass this three-question test:

QuestionStandard
Source?Named source (UBOS, World Bank, primary survey) with date
Method?State if geometric mean, CV, regression, or raw secondary data
Reproducible?A reader with the same data could arrive at the same figure

Flagging estimates:

  • Sourced fact: "(UBOS, 2024)"
  • Derived calculation: "(calculated from UBOS 2024 data; geometric mean 2019–2024)"
  • Acknowledged estimate: "(author estimate based on industry benchmarks; see assumptions)"

Never acceptable: Unsourced percentages, rounded numbers without basis, growth rates without stating the method.


References

  • references/analytics-ladder-and-kpi-standard.md Analytics ladder, data-quality gate, method selection, market/financial/demand analytics, AI analytics controls, and KPI dashboard standards
  • references/statistics-for-business-plans.md Full decision tables, formulas, worked Uganda/EA examples, chart selection guide, data type definitions
  • Anderson, D.R., Sweeney, D.J., Williams, T.A., Camm, J.D. & Cochran, J.J. (2013). Essentials of Statistics for Business and Economics (7th ed.). Cengage Learning.
  • Keller, G. (2011). Statistics for Management and Economics, Abbreviated (9th ed.). Cengage Learning.

Evidence Produced

EvidenceFormatAcceptance condition
Statistical review record decision traceSources, calculations, assumptions, countercase, and selected actionA reviewer can trace the selected action and rejected alternatives to the cited inputs.
Exception recordFailed and not-assessed checks with owner and due actionThe register exposes every unresolved exception that could lead to turning a biased sample or unstable average into a business claim.

Capability and Permission Boundaries

Default to read-only inspection while producing the statistical review record. Read supplied records and run non-mutating checks; running reproducible analysis without altering source data is permitted only when requested. Do not publish, contact third parties, alter live systems, commit funds, or claim legal, tax, audit, valuation, ESG, or investment assurance without the owner's explicit authorisation and the appropriate reviewer.

Degraded Mode

If dataset, sampling method, metric definitions, and calculation files cannot be obtained, return a qualified statistical review record covering only the checks that remain supportable. Leave this decision unresolved: which inference, comparison, or chart the evidence supports. Record the evidence owner and next check; an inaccessible source, tool, or reviewer is never a pass.

Decision Rules

Decision conditionActionFailure or risk avoided
Evidence is sufficient to decide: which inference, comparison, or chart the evidence supportsRecord the conclusion, source trail, owner, and review trigger in the statistical review record.Risk of turning a biased sample or unstable average into a business claim
Material evidence conflicts or remains uncertainRun the appropriate sensitivity, robustness, or alternative specification and report how the business conclusion changes.Selecting an option without resolving the decision-relevant uncertainty
Required evidence is missing: dataset, sampling method, metric definitions, and calculation filesMark the decision on which inference, comparison, or chart the evidence supports not assessed in the statistical review record, and send it to the analyst and decision owner.Otherwise, the work risks turning a biased sample or unstable average into a business claim

Quality Standards

Accept the statistical review record only when evidence is sufficient for this decision: which inference, comparison, or chart the evidence supports. Assumptions and countercases remain visible, calculations and cross-references reconcile, and the reviewer can see how the recommendation addresses the risk of turning a biased sample or unstable average into a business claim.

Worked Example

A survey average hides opposite results between new and repeat customers. Report the segment estimates and uncertainty, then revise the recommendation instead of relying on the blended mean.

<!-- dual-compat-end -->

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.