CtrlK
BlogDocsLog inGet started
Tessl Logo

metodologia-hypothesis-driven-development

Hypothesis-Driven Development (HDD) framework for structuring modernization proposals as testable hypotheses with Lean Startup cycles (Build-Measure-Learn). Transforms features into hypotheses with metrics, experiments, and kill/pivot/persevere thresholds. Use when formulating scenarios as hypotheses, designing validation experiments, applying Lean Startup to discovery, or when "HDD", "hypothesis", "hipótesis", "lean startup", "build-measure-learn", "experiment", "kill/pivot/persevere", or "validación de hipótesis" is mentioned.

SKILL.md
Quality
Evals
Security

Hypothesis-Driven Development: Lean Startup for Technical Discovery

Transforms modernization proposals into testable hypotheses with Build-Measure-Learn cycles. Instead of assuming a solution works and planning its full execution, HDD proposes: first the hypothesis, then the minimum experiment, then the evidence, then the decision.

Guiding Principle

We do not assume it works. We propose that it should work, define how we would know, and test it.

Classic discovery produces a roadmap based on the "best-case scenario." HDD produces a roadmap based on incrementally validated hypotheses. Each roadmap phase is an experiment. Each gate is a decision point: kill, pivot, or persevere.

HDD Philosophy

  1. Every feature is a hypothesis, not a certainty. "Migrating to microservices will improve time-to-market" is a hypothesis. It needs metrics, thresholds, and an experiment to validate it.
  2. Fast failure is a success. Discovering in Sprint 2 that a hypothesis is false saves months of execution in the wrong direction.
  3. Kill is a valid decision. If the experiment refutes the hypothesis, killing the work stream is the correct decision. Escalating sunk cost fallacy is the worst decision.
  4. Metrics before code. Define what we will measure BEFORE building. If we do not know what to measure, we do not know what we are testing.
  5. Build-Measure-Learn, not Build-Build-Build. Each cycle is short (1-5 days). Build the minimum to measure. Measure to learn. Learn to decide.

References

  • Eric Ries — The Lean Startup (2011): Build-Measure-Learn loop
  • Jeff Gothelf — Lean UX (2013): Hypothesis-driven design
  • Paulo Caroli — Lean Inception (2018): Discovery as lean inception
  • Martin Fowler — Evolutionary Architecture: fitness functions as architecture hypotheses
  • Barry O'Reilly — Hypothesis-Driven Development in enterprise

Inputs

Parse $1 as project/scenario name. Requires: approved scenario (Phase 3), feature backlog, business objectives. Recommended: metrics baseline (current performance), stakeholder priorities.

HDD Hypothesis Structure

HIPÓTESIS #{N}
══════════════
Creemos que: [acción/cambio propuesto]
Para: [audiencia/sistema afectado]
Resultará en: [outcome esperado]
Lo sabremos cuando: [métrica observable]
Con umbral de éxito: [valor cuantitativo]

Experimento:
  Tipo: [spike/PoC/MVP/A-B test/shadow deployment]
  Duración: [N sprints de 1 día]
  Recursos: [N FTEs]
  Entregable mínimo: [qué se construye]
  Medición: [cómo se mide]

Decisión:
  Kill si: [métrica < umbral_kill]
  Pivot si: [umbral_kill ≤ métrica < umbral_success]
  Persevere si: [métrica ≥ umbral_success]

Delivery Structure

S1: Business Hypothesis Canvas

Transform scenario objectives into business hypotheses:

#HipótesisMétricaUmbral ÉxitoUmbral KillPrioridad
H1Migrar checkout a microservicio reduce time-to-deployDeploy frequency≥1/día<1/semanaMUST
H2Event-driven architecture mejora resilienciaMTTR<15min>60minMUST
H3Nuevo design system aumenta conversiónConversion rate+15%<+5%SHOULD

S2: Experiment Design Matrix

For each hypothesis, design the minimum experiment:

HipótesisTipo ExperimentoDuraciónFTEsEntregable MínimoMétrica de Salida
H1PoC: 1 servicio extraído5 sprints (5 días)2Checkout service deployableDeploy frequency medido
H2Spike: event bus prototype3 sprints1Kafka consumer funcionalMessage processing time
H3A/B test: nuevo vs viejo10 sprints1Feature flag + nuevo UIConversion rate A vs B

S3: Build-Measure-Learn Cycles

Map each hypothesis to BML cycles:

flowchart TD
    H1[Hipótesis H1] --> B1[BUILD\n1 microservicio\n5 días, 2 FTEs]
    B1 --> M1[MEASURE\nDeploy frequency\nMTTR, error rate]
    M1 --> L1{LEARN\n¿Deploy ≥1/día?}
    L1 -->|Sí| P1[PERSEVERE\nExtraer siguiente servicio]
    L1 -->|Parcial| PV1[PIVOT\nAjustar granularidad]
    L1 -->|No| K1[KILL\nReevaluar estrategia\nde descomposición]

S4: HDD-Enhanced Roadmap

The traditional roadmap is transformed:

Before (classic roadmap):

Fase 1 → Fase 2 → Fase 3 → Fase 4 → Entrega

After (HDD roadmap):

H1:Experiment → H1:Measure → H1:Decision → [Kill|Pivot|Persevere]
                                              ↓
H2:Experiment → H2:Measure → H2:Decision → [Kill|Pivot|Persevere]
                                              ↓
H3:Experiment → H3:Measure → H3:Decision → [Kill|Pivot|Persevere]

Each hypothesis has its own cycle. MUST hypotheses go first. If a MUST fails, re-evaluate the entire scenario.

S5: Decision Log

SprintHipótesisMétrica ObtenidaUmbralDecisiónRationale
D5H1Deploy freq: 2/día≥1/día✅ PERSEVERESupera umbral
D8H2MTTR: 45min<15min🔄 PIVOTNecesita retry logic
D18H3Conversion: +3%+15%❌ KILLROI no justifica

S6: Validated Hypothesis Portfolio

At the end of the process, the portfolio shows:

HipótesisStatusEvidenciaImpacto ValidadoSiguiente Paso
H1✅ ValidadaDeploy 2x/día medidoTime-to-market -60%Escalar a 5 servicios
H2🔄 PivotadaMTTR mejoró a 20minResiliencia +70%Agregar retry + circuit breaker
H3❌ MatadaConversión +3% (insuficiente)No justifica inversiónReasignar FTEs

Integration with Discovery Pipeline

PhaseWithout HDDWith HDD
Phase 3 (Scenarios)"Scenario B is better""Scenario B has 5 testable hypotheses"
Phase 3b (Think Tank)"It is feasible""Hypotheses H1-H3 are experimentable in N days"
Phase 4 (Roadmap)"Sprint 1: migrate X, Sprint 2: migrate Y""Sprint 1: Experiment H1, Gate: kill/pivot/persevere"
Phase 4b (Costing)"We estimate 50 FTE-months""Validating H1-H3 costs 5 FTE-months. Executing validated hypotheses costs 45 FTE-months"

When to Use

  • Formulating scenarios as testable propositions (Phase 3)
  • Designing validation experiments for the Think Tank (Phase 3b)
  • Building HDD-enhanced roadmaps (Phase 4)
  • When the client asks "how do we know this will work?"
  • When there is significant uncertainty in the proposed solution

When NOT to Use

  • Well-understood migrations with proven patterns (lift-and-shift)
  • Regulatory compliance projects with fixed scope
  • Emergency/crisis responses where speed overrides learning

Trade-off Matrix

DecisionEnablesConstrainsWhen to Use
Full HDD (all features as hypotheses)Maximum learning, minimum wasteHigher ceremony, slower initial progressHigh uncertainty, new technology, large investment
Partial HDD (only MUST features)Focused validation on critical itemsMay miss risks in SHOULD/COULDMedium uncertainty, time pressure
HDD for architecture onlyValidates big decisionsFeatures not individually validatedArchitecture-driven transformation
No HDD (classic roadmap)Simplest, fastest to planAssumes solution worksLow uncertainty, proven patterns

Edge Cases

ScenarioResponse
All hypotheses validatedRare but ideal — proceed with high confidence, reduce contingency margin
MUST hypothesis killedStop roadmap. Return to Phase 3 scenarios. May need different scenario
Pivot cascades (pivot triggers new hypothesis)Allow max 2 pivot chains. If still failing, kill
Client refuses to killDocument sunk cost fallacy risk. Proceed with explicit disclaimer
No baseline metrics availableFirst experiment = establish baseline. Add 1-2 sprints for measurement setup

Validation Gate

  • Every MUST feature has an HDD hypothesis
  • Each hypothesis has: metric, success threshold, kill threshold
  • Experiments designed with minimum viable scope
  • BML cycles mapped with decision points
  • Kill/Pivot/Persevere criteria are quantitative, not qualitative
  • Integration with roadmap phases documented

Casos Borde

CasoEstrategia de Manejo
Cliente no tiene baseline de metricas para definir umbralesEl primer experimento se dedica a establecer baseline; agregar 1-2 sprints de instrumentacion antes de formular hipotesis con umbrales cuantitativos
Todas las hipotesis MUST son validadas exitosamenteCaso raro pero ideal; reducir margen de contingencia en el roadmap; documentar evidencia para fortalecer la propuesta de inversion
Cadena de pivots en cascada (pivot genera nueva hipotesis que tambien falla)Permitir maximo 2 niveles de pivot encadenados; si el tercer intento falla, ejecutar kill y retornar a Phase 3 para reevaluar el escenario completo
Stakeholders se rehusan a ejecutar kill a pesar de evidencia negativaDocumentar explicitamente el riesgo de sunk cost fallacy; escalar a sponsor ejecutivo con datos cuantitativos; proceder con disclaimer formal si insisten

Decisiones y Trade-offs

DecisionAlternativa DescartadaJustificacion
Formular cada feature MUST como hipotesis testeableTratar features como requisitos fijos sin validacionLas features asumidas sin evidencia son la causa principal de desperdicio en transformaciones; HDD reduce riesgo antes de comprometer presupuesto
Ciclos BML cortos de 1-5 dias por experimentoSprints largos de 2-4 semanas para validacionCiclos cortos permiten decision rapida; ciclos largos acumulan costo antes de generar evidencia y retrasan el kill/pivot
Kill criteria cuantitativos y binariosCriterios cualitativos como "el equipo siente que funciona"Los criterios cualitativos son susceptibles a sesgo de confirmacion; solo metricas medibles producen decisiones objetivas

Knowledge Graph

graph TD
    subgraph Core["HDD Core"]
        A[metodologia-hypothesis-driven-development]
        A1[S1: Business Hypothesis Canvas]
        A2[S2: Experiment Design Matrix]
        A3[S3: Build-Measure-Learn Cycles]
        A4[S4: HDD-Enhanced Roadmap]
        A5[S5: Decision Log]
        A6[S6: Validated Hypothesis Portfolio]
    end
    subgraph Inputs["Inputs"]
        I1[Escenario Aprobado - Phase 3]
        I2[Feature Backlog]
        I3[Business Objectives]
        I4[Metrics Baseline]
    end
    subgraph Outputs["Outputs"]
        O1[HDD Hypotheses Document]
        O2[Experiment Roadmap]
        O3[Decision Log]
    end
    subgraph Related["Related Skills"]
        R1[metodologia-technical-feasibility]
        R2[metodologia-roadmap-poc]
        R3[metodologia-cost-estimation]
        R4[metodologia-software-viability]
    end
    I1 --> A
    I2 --> A
    I3 --> A
    I4 --> A
    A --> A1 --> A2 --> A3 --> A4 --> A5 --> A6
    A --> O1
    A --> O2
    A --> O3
    R1 --> A
    A --> R2
    A --> R3
    A --- R4

Output Templates

Formato MD (default):

# HDD Hypotheses — {proyecto}
## Resumen Ejecutivo
> N hipotesis formuladas, M experimentos disenados, timeline estimado: X sprints.
## S1: Business Hypothesis Canvas
| # | Hipotesis | Metrica | Umbral Exito | Umbral Kill | Prioridad |
## S2: Experiment Design Matrix
| Hipotesis | Tipo Experimento | Duracion | FTEs | Entregable Minimo |
## S3-S6: [secciones completas con diagramas BML]
## Apendice: Referencias y Supuestos

Formato PPTX (para presentacion a steering committee):

Slide 1: Titulo + contexto del escenario
Slide 2: Hypothesis Canvas (tabla resumen de H1-Hn)
Slide 3-N: Una slide por hipotesis (hipotesis + experimento + criterios kill/pivot/persevere)
Slide N+1: Roadmap HDD (timeline visual con gates de decision)
Slide N+2: Investment ask (esfuerzo para validar vs esfuerzo para ejecutar)
Slide N+3: Decision framework (que pasa si H1 falla, si H2 pivota, etc.)

Formato HTML (bajo demanda):

  • Filename: A-03_HDD_Hypotheses_{project}_{WIP}.html
  • Estructura: HTML self-contained branded (Design System MetodologIA v5). Light-First Technical page con hypothesis canvas interactivo, BML cycle diagrams, y decision log con semáforo Kill/Pivot/Persevere. WCAG AA, responsive, print-ready.

Formato DOCX (bajo demanda):

  • Filename: A-03_HDD_Hypotheses_{project}_{WIP}.docx
  • Generado con python-docx bajo MetodologIA Design System v5: portada, TOC automático, encabezados/pies de página con marca, tablas zebra, tipografía Poppins (headings navy), Montserrat (body), acentos dorados

Formato XLSX (bajo demanda):

  • Filename: A-03_HDD_Hypotheses_{project}_{WIP}.xlsx
  • Generado con openpyxl bajo MetodologIA Design System v5. Headers con fondo navy y tipografía Poppins blanca, formato condicional, auto-filtros activados, valores sin fórmulas. Hojas: Hypothesis Canvas, Experiment Design Matrix, Decision Log, Validated Portfolio.

Evaluacion

DimensionPesoCriterioUmbral Minimo
Trigger Accuracy10%El skill se activa ante prompts de hipotesis, lean startup, BML, validacion de escenarios7/10
Completeness25%Cada feature MUST tiene hipotesis con metrica, umbral de exito, umbral de kill, y experimento disenado7/10
Clarity20%Stakeholders no tecnicos entienden la estructura hipotesis-experimento-decision; diagramas BML son legibles7/10
Robustness20%Edge cases cubiertos (sin baseline, all-pass, cascade pivots, kill resistance); decision log estructurado7/10
Efficiency10%Experimentos disenados con scope minimo viable; no se sobre-disenian validaciones innecesarias7/10
Value Density15%Cada hipotesis conecta directamente a valor de negocio; el portfolio muestra impacto validado cuantificado7/10

Umbral minimo global: 7/10. Si alguna dimension cae por debajo, el entregable requiere revision antes de entrega.

Output Configuration

  • Language: Spanish (Latin American, business register — simple, clear, concise, direct)
  • Attribution: Expert committee of the MetodologIA Discovery Framework
  • Tagline: "Construido por profesionales, potenciado por la red agéntica de MetodologIA."

Output Artifact

Primary: A-03_HDD_Hypotheses_{project}.md

Diagrams (Mermaid)

  • Flowchart TD: Build-Measure-Learn cycles per hypothesis
  • Gantt: experiment timeline with decision gates

© Comunidad MetodologIA — All rights reserved

Repository
JaviMontano/mao-discovery-framework
Last updated
First committed

Also appears in

JaviMontano/mao-pm-apex
In sync

since Aug 28, 2026

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.