BMad-Inspired Critical Thinking Framework (BCTF)

Version 3.0 — Systems Intelligence and Purpose-Driven Decision-Making

Version: 3.0
Date: October 22, 2026
Status: Proposed operational standard — empirical validation pending
Classification: Independent BMad-inspired methodology
Supersedes: BCTF V1 and V2 as the primary operating specification
Companion: AI-Driven Integrative Design V2 and its Evidence-Based Case Studies Appendix

1. Executive Summary

The BMad-Inspired Critical Thinking Framework (BCTF) is a human-governed methodology for improving reasoning, research, problem-solving, and decision-making through structured AI collaboration.

Version 3 integrates three complementary disciplines:

  1. BMad-inspired workflow orchestration: Specialized agents, structured workflows, persistent context, and iterative review.
  2. Critical thinking and evidence-based reasoning: Problem clarification, evidence evaluation, logical analysis, uncertainty management, and reflective learning.
  3. Purpose-driven systems intelligence: Systems analysis, problem reframing, intervention worthiness, architectural alternatives, and measured outcomes.

The framework's central proposition is:

AI should help humans determine what is true, what matters, what ought to change, and whether proposed actions are justified.

Version 3 establishes a unified operating model with adaptive workflow depth, nine specialized AI roles, explicit human authority, evidence standards, quality gates, and reusable project artifacts.

The framework is designed for research, strategy, organizational decisions, policy analysis, product development, systems engineering, and other domains requiring disciplined judgment.

Its effectiveness relative to simpler methods remains an empirical research question.

2. Foundational Philosophy

2.1 Purpose Before Solutions

Begin by identifying the outcome that matters.

Do not assume the initially proposed technology, artifact, or intervention is necessary.

2.2 Evidence Before Commitment

Consequential decisions require evidence proportionate to their stakes.

Claims must be traceable, uncertainties explicit, and important limitations documented.

2.3 Systems Before Components

Examine relationships, dependencies, and feedback before optimizing isolated components when system interactions are material.

2.4 Reframing Before Optimization

Question inherited assumptions when doing so could materially improve the decision.

Reframing is an option, not an obligation for every task.

2.5 Subtraction Before Addition

Explore whether the desired outcome can be achieved through simplification, elimination, reuse, or non-intervention.

2.6 Diversity Before Convergence

Consider meaningfully different explanations and options before selecting a preferred conclusion.

2.7 Human Judgment Before Automated Authority

AI assists with analysis. Authorized humans remain responsible for consequential decisions.

2.8 Verification Before Confidence

Persuasive writing, agent consensus, and completed templates do not establish correctness.

2.9 Learning After Action

Compare predicted and observed outcomes and revise assumptions when warranted.

2.10 Proportionate Rigor

The effort devoted to reasoning must reflect complexity, uncertainty, risk, reversibility, and the cost of being wrong.

3. Relationship to the Official BMad Method

BCTF is inspired by Brian "BMad" Madison's open-source BMad Method.

BMad provides a model for:

BCTF extends these principles beyond software engineering.

It does not claim to be an official BMad module or a validated extension.

3.1 Public BMad Resources

Official repository:

https://github.com/bmad-code-org/BMAD-METHOD

Documentation:

https://docs.bmad-method.org/

BCTF implementations may use publicly available BMad skills and workflows, including research, elicitation, multi-agent discussions, review, and project-context management.

Exact commands, agent identifiers, and module availability must be verified against the installed BMad release.

3.2 Integration Boundary

Official BMad components should remain identifiable and independently updateable.

BCTF-specific agents, prompts, workflows, and artifacts should be labeled as custom extensions.

The framework should not require modification of official BMad source files when a supported customization mechanism is available.

4. Unified Operating Architecture

The framework consists of four connected capabilities.

4.1 Reasoning Engine

Clarifies questions, investigates evidence, evaluates arguments, and develops defensible conclusions.

4.2 Systems Intelligence

Examines relationships, causal assumptions, feedback loops, system boundaries, and opportunities for higher-leverage interventions.

4.3 Decision Governance

Evaluates alternatives, uncertainty, consequences, stakeholder impacts, and authorization requirements.

4.4 Learning Flywheel

Captures decisions, predictions, experiments, outcomes, and lessons for future use.

4.5 Overall Process

HUMAN PURPOSE / QUESTION
          |
          v
    WORKFLOW TRIAGE
          |
          v
        FRAME
          |
          v
     INVESTIGATE
          |
          v
       CHALLENGE
          |
          v
      SYNTHESIZE
          |
          v
   DECISION READINESS
          |
          +----------------------+
          |                      |
          v                      v
 GENERAL DECISION       INTERVENTION WORTHINESS
          |                      |
          +-----------+----------+
                      |
                      v
               HUMAN DECISION
                      |
                      v
             ACTION / EXPERIMENT
                      |
                      v
                   REFLECT
                      |
                      v
               LEARNING RECORD
                      |
                      +----> FUTURE INQUIRY

Governance, evidence verification, and stakeholder awareness apply throughout.

5. Adaptive Workflow Selection

Not every task requires the same process.

5.1 Mode A — Quick Inquiry

Use for: Low-risk claims, everyday decisions, preliminary exploration.

Activities:

  1. Clarify the question.
  2. Identify relevant evidence.
  3. Examine a plausible alternative.
  4. State a qualified conclusion.

Minimum output:

quick-assessment.md

5.2 Mode B — Structured Inquiry

Use for: Research, business decisions, policy exploration, and moderately consequential choices.

Activities:

  1. Frame the problem.
  2. Develop an evidence base.
  3. Challenge assumptions.
  4. Compare alternatives.
  5. Record a decision or conclusion.
  6. Identify review conditions.

Required outputs:

5.3 Mode C — Systems and Intervention Analysis

Use for: Complex design, organizational change, infrastructure, strategic investment, and systems engineering.

Activities:

  1. Define the desired outcome.
  2. Map the relevant system.
  3. Identify causes and dependencies.
  4. Reframe the problem.
  5. Explore alternative interventions.
  6. Evaluate intervention worthiness.
  7. Recommend an experiment or action.
  8. Measure outcomes.

Additional outputs:

5.4 Mode D — High-Stakes Review

Use for: Decisions involving significant safety, legal, financial, ethical, or societal consequences.

Requires:

BCTF cannot substitute for professional or regulatory review.

5.5 Workflow Selection Criteria

Select the least burdensome workflow that adequately addresses:

Workflow depth may increase as new risks or uncertainties emerge.

6. Six-Phase Critical Thinking Lifecycle

Phase 1 — FRAME

Objective: Determine the real question, desired outcome, and decision context.

Activities:

Questions:

Output:

problem-brief.md

Gate: Human confirms the question and scope.

Phase 2 — INVESTIGATE

Objective: Establish the strongest available evidence.

Activities:

Outputs:

Gate: Material claims are traceable and evidence gaps visible.

Phase 3 — CHALLENGE

Objective: Test whether the emerging interpretation survives meaningful criticism.

Activities:

Outputs:

Gate: Material objections are resolved, mitigated, or explicitly retained.

Phase 4 — SYNTHESIZE

Objective: Integrate evidence and compare plausible conclusions or actions.

Activities:

Outputs:

Gate: The analysis is sufficiently complete for a decision.

Phase 5 — DECIDE

Objective: Make a transparent, authorized judgment.

Possible outcomes:

For research questions, the outcome may instead be a supported conclusion, an unresolved finding, or a recommendation for further investigation.

Outputs:

Gate: The human decision owner approves the outcome.

Phase 6 — REFLECT

Objective: Evaluate outcomes and improve future reasoning.

Activities:

Outputs:

Gate: Learning is captured and appropriate follow-up actions identified.

7. Specialized Agent Architecture

Version 3 defines nine analytical roles.

These are responsibilities, not necessarily nine separately running models.

A small task may use one AI assistant performing several roles.

For consequential work, independent verification should involve genuinely independent evidence and, where appropriate, qualified human reviewers.

7.1 Socratic Facilitator

Mission: Improve the clarity and quality of the initial question.

Primary responsibilities:

Primary phase: FRAME

Primary artifact:

problem-brief.md

Key question:

"What problem are we actually trying to solve?"

Boundaries:

Must not impose an AI-selected objective or indefinitely prolong questioning.

7.2 Evidence Investigator

Mission: Develop a credible, traceable evidence base.

Responsibilities:

Primary phase: INVESTIGATE

Primary artifacts:

Key question:

"What does the strongest available evidence actually support?"

Boundaries:

Must not fabricate citations or present unverified claims as established facts.

7.3 Skeptical Challenger

Mission: Identify weaknesses in arguments and proposed interventions.

Responsibilities:

Primary phase: CHALLENGE

Primary artifact:

challenge-report.md

Key question:

"What would make this conclusion wrong?"

Boundaries:

Must avoid performative disagreement, straw-manning, and false equivalence.

7.4 Creative Explorer

Mission: Expand the range of plausible interpretations and interventions.

Responsibilities:

Primary phases: FRAME, CHALLENGE, SYNTHESIZE

Primary artifact:

alternatives.md

Key question:

"What possibilities have we overlooked?"

Boundaries:

Must distinguish imaginative possibilities from validated findings.

7.5 Decision Strategist

Mission: Translate analysis into transparent recommendations.

Responsibilities:

Primary phases: SYNTHESIZE, DECIDE

Primary artifact:

decision-record.md

Key question:

"Which option best advances the desired outcome under the relevant constraints?"

Boundaries:

Must not hide value judgments inside arbitrary scores or make unauthorized decisions.

7.6 Reflective Auditor

Mission: Improve reasoning through outcome evaluation.

Responsibilities:

Primary phase: REFLECT

Primary artifact:

retrospective.md

Key question:

"What did the results teach us about our reasoning?"

Boundaries:

Must distinguish poor outcomes from poor decision processes.

7.7 Verification and Quality Auditor

Mission: Check factual and procedural reliability.

Responsibilities:

Applies across all phases.

Primary artifact:

verification-report.md

Key question:

"Can the material findings withstand independent checking?"

Boundaries:

Must not treat AI self-review as sufficient independent verification for high-stakes decisions.

7.8 Ethics and Stakeholder Reviewer

Mission: Evaluate consequences for affected people.

Responsibilities:

Primary phases: FRAME, SYNTHESIZE, DECIDE

Primary artifact:

stakeholder-impact.md

Key question:

"Who benefits, who bears the risks, and whose perspective is missing?"

Boundaries:

Must not claim authority to resolve contested human values unilaterally.

7.9 Systems and Leverage Analyst — New in V3

Mission: Identify underlying system structures and potentially high-value intervention points.

Responsibilities:

Primary phases: FRAME, INVESTIGATE, SYNTHESIZE

Primary artifacts:

Key question:

"Could changing the system eliminate the need for the proposed solution?"

Boundaries:

A system map is a model, not proof of causation.

The analyst must document assumptions, uncertainties, and excluded system elements.

8. Common Agent Operating Contract

Every agent must have:

  1. Defined mission
  2. Scope of authority
  3. Required inputs
  4. Expected outputs
  5. Permitted tools
  6. Verification requirements
  7. Escalation conditions
  8. Handoff requirements

Shared rules:

8.1 Agent Handoff Format

# Agent Handoff

Task ID:
Originating Agent:
Receiving Agent:
Approved Question:
Source Artifacts:
Key Findings:
Evidence References:
Known Assumptions:
Unresolved Objections:
Risk Level:
Required Next Action:
Human Approval Status:

9. Systems Intelligence Method

Systems analysis is activated when the problem involves meaningful interdependencies or potential architectural alternatives.

9.1 Purpose

Identify where changes to system structure, relationships, incentives, or goals may create greater value than isolated optimization.

9.2 Four Levels of Inquiry

Level 1 — Purpose

Why does the system exist?

What outcome is required?

Level 2 — System

What interacting factors create the observed outcome?

Level 3 — Architecture

Could the components, processes, or relationships be arranged differently?

Level 4 — Components

Which necessary elements should be optimized?

9.3 Required Analyses

For complex systems problems:

9.4 Design by Subtraction

At least one alternative should investigate:

This requirement may be waived when demonstrably irrelevant.

9.5 Leverage Assessment

Potential intervention points include:

Deeper interventions are not automatically better.

Their value depends on feasibility, consequences, cost, and evidence.

10. Intervention Worthiness Method

The Intervention Worthiness Method is adapted from the Design Worthiness Method described in AI-Driven Integrative Design V2.

Its purpose is to determine whether a proposed change deserves further investment.

10.1 Six Evaluation Lenses

Purpose

What outcome matters?

Necessity

Is intervention required?

Leverage

Where could change have the greatest justified impact?

Value

Who benefits, who bears costs, and what net value is expected?

Feasibility

Can the intervention be implemented within technical, organizational, financial, and regulatory constraints?

Evidence

How strongly do available findings support the intervention?

10.2 Decision Outcomes

Proceed: Evidence and feasibility justify implementation.

Experiment: Critical uncertainties can be tested through a limited intervention.

Reframe: The problem or proposed solution requires reconsideration.

Defer: Additional evidence, resources, or conditions are necessary.

Reject: The intervention does not justify its costs, risks, or opportunity costs.

10.3 Worthiness Gate Requirements

Before approval, record:

The gate must not create artificial numerical precision.

When weighting criteria, document who selected the weights and why.

11. Research and Evidence Standard

11.1 Research Protocol

Substantial research should define:

11.2 Evidence Traceability

Material claims should follow:

Claim → Evidence → Source → Limitations → Confidence rationale

11.3 Evidence Categories

Evidence may include:

No category automatically establishes correctness.

Relevance and methodological quality matter.

11.4 Causal Claims

Claims that an intervention caused an outcome require appropriate evidence.

Correlation alone is insufficient.

Where possible, consider:

11.5 Confidence

Express confidence in proportion to the evidence.

Numerical probabilities should be used only when a defensible basis exists.

Do not invent precision.

11.6 Research Completion

Research may stop when:

12. Governance and Human Authority

12.1 Human Decision Owner

Every consequential decision must have an identifiable human owner.

The owner approves:

12.2 Risk Levels

Low: Basic evidence review.

Moderate: Documented alternatives and decision rationale.

High: Independent verification, stakeholder assessment, and explicit approval.

Critical: Qualified expert review, formal governance, and applicable regulatory controls.

12.3 Quality Gates

Gate 1: Problem framing approved.

Gate 2: Evidence sufficiently verified.

Gate 3: Material objections examined.

Gate 4: Alternatives evaluated.

Gate 5: Intervention Worthiness completed, when applicable.

Gate 6: Human decision authorized.

Gate 7: Outcomes reviewed.

12.4 Verification Findings

Critical findings block release.

High findings require correction or authorized escalation.

12.5 Responsible AI

Apply NIST AI RMF principles:

Treat governance as a continuous responsibility, not a final checklist.

13. Learning Flywheel

The Learning Flywheel converts decisions and experiments into reusable knowledge.

13.1 Cycle

Observe → Reframe → Experiment → Implement → Measure → Learn → Improve

13.2 Knowledge Records

Capture:

13.3 Core Metrics

Evaluate:

Avoid optimizing for document volume or agent activity.

13.4 Learning Integrity

Do not retroactively redefine success to match observed results.

Distinguish:

Record uncertainty and chance where relevant.

14. Project Repository Standard

Recommended structure:

project/
├── README.md
├── AGENTS.md
├── framework/
│   ├── BCTF_V3.md
│   └── references/
│       ├── BCTF_V1.md
│       ├── BCTF_V2.md
│       ├── AI_Driven_Integrative_Design_V2.md
│       └── Integrative_Design_Case_Studies.md
├── research/
│   ├── research-plan.md
│   ├── evidence-register.md
│   └── findings.md
├── analysis/
│   ├── problem-brief.md
│   ├── assumptions.md
│   ├── system-map.md
│   ├── leverage-analysis.md
│   ├── alternatives.md
│   └── challenge-report.md
├── decisions/
│   ├── intervention-worthiness.md
│   ├── decision-record.md
│   └── dissent-log.md
├── governance/
│   ├── risk-register.md
│   ├── verification-report.md
│   └── stakeholder-impact.md
└── learning/
    ├── outcomes.md
    ├── experiments.md
    └── retrospective.md

Only V3 is authoritative for the BCTF operating process.

Supporting documents provide context, historical development, case studies, and specialized methods.

15. Reusable Artifact Templates

15.1 Problem Brief

# Problem Brief

Question:
Decision Owner:
Workflow Mode:
Risk Level:

## Desired Outcome

## Context

## Stakeholders

## System Boundaries

## Assumptions

## Constraints

## Alternatives

## Success Criteria

## Approval

15.2 Intervention Worthiness Assessment

# Intervention Worthiness

Intervention:
Decision Owner:
Date:

## Purpose

## Necessity

## System Leverage

## Value

## Feasibility

## Evidence

## Non-Intervention Alternative

## Stakeholder Impacts

## Key Uncertainties

## Recommendation
Proceed / Experiment / Reframe / Defer / Reject

## Human Decision

## Review Conditions

15.3 Decision Record

# Decision Record

Decision ID:
Owner:
Date:
Status:

## Question

## Evidence

## Alternatives

## Counterarguments

## Assumptions

## Uncertainty

## Tradeoffs

## Decision

## Rationale

## Predicted Outcomes

## Conditions for Reconsideration

## Review Date

15.4 Learning Record

# Learning Record

Decision ID:
Review Date:

## Original Predictions

## Observed Outcomes

## Evidence Collected

## Unexpected Effects

## Assumptions Confirmed

## Assumptions Rejected

## Decision Quality Assessment

## Lessons

## Recommended Changes

16. BMad Implementation Guidance

16.1 Recommended Approach

Use BMad's publicly available workflow and customization capabilities as the orchestration foundation where appropriate.

Build BCTF-specific agents and skills as separate extensions.

16.2 Proposed Custom Skills

These names are proposed, not official BMad commands.

16.3 Recommended Integration Sequence

  1. Verify installed BMad version.
  2. Add BCTF V3 project instructions.
  3. Implement workflow triage.
  4. Implement evidence standards.
  5. Implement Verification Auditor.
  6. Implement Systems and Leverage Analyst.
  7. Add Intervention Worthiness.
  8. Introduce structured multi-agent deliberation.
  9. Validate against representative tasks.
  10. Expand only after measuring value.

17. Worked Example — Customer Service Transformation

Initial Proposal

"Build an AI customer service chatbot."

FRAME

Desired outcome:

Reduce customer effort and resolution time while maintaining service quality.

INVESTIGATE

Collect:

SYSTEMS ANALYSIS

Investigate whether support demand originates from:

CHALLENGE

Ask:

ALTERNATIVES

A. Build a chatbot.

B. Simplify billing.

C. Improve documentation.

D. Remove unnecessary support-generating processes.

E. Combine targeted simplification with limited AI assistance.

F. Make no change until evidence improves.

WORTHINESS ASSESSMENT

Evaluate each option across:

DECISION

Illustrative outcome:

Experiment with eliminating a high-volume source of avoidable support contacts before committing to a broad chatbot implementation.

This is a hypothetical recommendation. An actual decision requires relevant operational evidence.

REFLECT

Measure:

Update the system model based on results.

18. Evaluation and Validation

18.1 Central Research Question

Does BCTF V3 improve reasoning and intervention decisions compared with ordinary AI assistance and BCTF V2?

18.2 Comparison Groups

A proposed evaluation could compare:

  1. Human-only reasoning
  2. General-purpose AI assistance
  3. BCTF V2
  4. BCTF V3

18.3 Evaluation Dimensions

18.4 Proposed Hypotheses

H1: Systems inquiry improves identification of upstream causes.

H2: Intervention Worthiness reduces unjustified commitments to predefined solutions.

H3: Explicit subtraction alternatives increase discovery of simplification opportunities.

H4: Independent verification reduces consequential factual errors.

H5: Structured learning records improve future prediction and decision quality.

H6: Adaptive workflow selection reduces unnecessary process overhead.

These hypotheses require testing.

18.5 Evaluation Safeguards

19. Implementation Roadmap

Phase 1 — Consolidation

Deliverables:

Phase 2 — Minimum Viable Framework

Implement:

Phase 3 — Systems Intelligence

Implement:

Phase 4 — Learning Infrastructure

Implement:

Phase 5 — Comparative Validation

Compare V3 with simpler workflows.

Revise or remove components that do not demonstrate sufficient value.

20. Document Governance and Version Policy

20.1 Version 1

Status: Historical and educational reference.

Purpose:

Not required for routine V3 operation.

20.2 Version 2

Status: Superseded operational specification.

Purpose:

Not required for routine V3 operation.

20.3 Version 3

Status: Authoritative operating specification.

Purpose:

Use this document as the primary project reference.

20.4 Integrative Design Documents

Status: Specialist supporting references.

Purpose:

Do not treat their proposed AI architecture as independently validated.

20.5 Conflict Resolution

When documents disagree:

  1. Follow applicable law, safety requirements, and authorized human instructions.
  2. Follow the current approved V3 operating specification.
  3. Consult official BMad documentation for BMad-specific technical behavior.
  4. Use specialist supporting documents for domain context.
  5. Treat V1 and V2 as historical references.

Changes to V3 must be reviewed and versioned.

21. Outstanding Risks and Limitations

Methodological

AI-Specific

Organizational

Ethical

These limitations require active management rather than rhetorical acknowledgment.

22. Final Operating Standard

BCTF V3 is intended to improve four kinds of judgment:

  1. Epistemic judgment: What should we believe?
  2. Analytical judgment: Why is the situation occurring?
  3. Practical judgment: What should we do?
  4. Reflective judgment: What should we learn?

The framework should not force every question through all four.

A successful implementation:

Final Principle

The goal is not to produce more analysis, more agents, more designs, or more confident answers.

The goal is to help humans make better judgments about what is true, what matters, what should change, and when action is justified.

23. Sources and References

Official BMad

Attached Project Research

Established Methodologies

Supporting Literature

24. Version History

Version Primary contribution Status
1.0 Six-phase critical thinking concept Reference
2.0 Agent contracts, research standards, verification, governance Superseded
3.0 Systems intelligence, intervention worthiness, adaptive workflows, integrated operating standard Current proposed specification

End of BCTF Version 3.0

Independent framework inspired by publicly documented BMad principles. Not an official BMad publication or empirically validated methodology.