The MAFI Framework

A structured method for assessing the mechanistic robustness of therapeutic development programmes.

What The Framework Is

MAFI (Mechanistic Architecture Fragility Intelligence) is a framework for evaluating the mechanistic dependencies that underpin a therapeutic programme's claimed mechanism of action. It decomposes that mechanism into independently falsifiable assumptions, assesses each assumption across five fragility domains, and integrates the domain scores into a fragility tier. The result is a structured view of where a mechanism is robust, where it is fragile, and where failure is most likely to originate.

The framework is rule-based, not judgement-based. Every classification is derived from explicit criteria, and every assessment is traceable to the evidence that supports it. This makes MAFI reproducible - two trained analysts applying the framework to the same programme should reach the same conclusions.

How It Works

The framework proceeds through six analytical stages.

1. Mechanism Reconstruction
The programme's claimed mechanism of action is reconstructed from publicly available scientific and regulatory evidence. This establishes the therapeutic thesis that the assessment will evaluate.

2. Dependency Decomposition
The mechanism is decomposed into its individual mechanistic dependencies - the independently falsifiable assumptions upon which the therapeutic thesis rests. Each dependency is a distinct condition that must hold for the mechanism to succeed.

3. Dependency Architecture
Each dependency is classified by layer - Biological, Translational, Clinical, or Strategic - and by centrality - Core or Peripheral. This establishes where the dependency operates and how essential it is to the programme's mechanism.

4. Fragility Assessment
Each dependency is assessed across five fragility domains: Evidence Proximity, Context Dependence, Historical Failure Prevalence, Recoverability, and Earliest Detectable Phase. Domain scores are integrated into a fragility tier - Robust, Conditional, or Brittle.

5. Critical Failure Node Identification
Dependencies whose failure would produce the greatest adverse effect on the claimed mechanism are identified by combining centrality and fragility. These are the points at which the mechanistic chain is most vulnerable.

6. Benchmarking and Convergence
The programme's dependency profile is positioned against defined reference populations, with outcome-stratified comparison providing interpretive context. Critical dependency configurations are matched against historical precedent to identify whether similar patterns have occurred before.

The Fragility Domains

Each dependency is assessed across five fragility domains.

Evidence Proximity measures how directly an assumption is supported by relevant evidence. High Evidence Proximity means direct human evidence; Low means support is primarily inferential or preclinical.

Context Dependence measures how broadly an assumption is expected to remain valid across patient populations, disease stages, and treatment settings. High Context Dependence means the assumption may fail when conditions vary.

Historical Failure Prevalence measures how frequently comparable assumption classes have historically broken down across therapeutic development. High means the assumption class has a well-established history of failure.

Recoverability measures whether a programme can realistically recover from assumption failure while preserving its core therapeutic thesis. Low Recoverability means failure invalidates the mechanism.

Earliest Detectable Phase measures the earliest stage at which assumption failure could be detected with sufficient confidence to materially challenge the programme. It ranges from Preclinical to Post-Approval.

The Principles

The framework is governed by four principles.

Prospective, not retrospective.
Assumptions are identified from the programme's original therapeutic rationale, not from later events. A programme's outcome - success or failure - does not influence the classification of its assumptions.

Rule-based, not judgement-based.
Every classification is derived from explicit criteria. Domain scores are integrated into fragility tiers using defined rules, not analyst discretion. This makes the framework reproducible.

Comparative, not predictive.
The framework describes where a programme sits relative to comparable programmes. It does not predict outcome. A programme with a fragile profile may still succeed; a programme with a robust profile may still fail.

Evidence-anchored.
Every classification is traceable to the evidence that supports it. The framework's outputs are auditable, and its assessments can be challenged on evidential grounds.

The framework described above is applied consistently across every MAFI assessment. Its classification rules, integration logic, and evidence standards are documented in full in the MAFI Ontology, which is available on request. The ontology is the reference specification for the framework; the reports are its application to individual programmes.