GenderMag Walkthrough Process

• GenderMag is a persona-centered walkthrough process that identifies inclusivity bugs by evaluating five empirically validated cognitive facets.
• It employs iterative subgoal evaluation with explicit facet tagging and quantitative metrics to diagnose usability issues across diverse user types.
• Extensions like GenderMag-for-AI and InclusiveMag adapt the methodology to address AI-specific challenges and intersectional inclusivity barriers.

The GenderMag walkthrough process is a systematic, persona-centered inspection methodology designed to uncover “inclusivity bugs”—usability breakdowns that disproportionately impede users whose problem-solving styles statistically cluster by gender. By focusing on five empirically validated cognitive facets, the method operationalizes gender inclusivity as actionable heuristics integrated into cognitive walkthrough analyses. Recent extensions expand its scope to user-facing AI systems (“GenderMag-for-AI”) and other intersectional axes (such as neurodivergence, via InclusiveMag integration), enabling the explicit surfacing and mitigation of both traditional and AI-specific inclusivity failures (Mendez et al., 2019, Anderson et al., 21 Oct 2025, Zaib et al., 5 Dec 2025).

1. Theoretical Foundations and Cognitive Facets

GenderMag is anchored in five key individual-difference facets, each grounded in longitudinal research and shown to correlate with observed gender disparities in software usability:

Facet	Endpoint Values (Abi ↔ Tim)	Description
Motivation	Learning/process-oriented (Abi) ↔ Goal/task-oriented (Tim)	Primary driver for engaging with technology.
Information Processing	Comprehensive/exploratory (Abi) ↔ Schematic/heuristic (Tim)	How users gather and filter information.
Attention Investment	High (Abi) ↔ Low (Tim)	Degree of focus and sustained engagement.
Synthesis (“Self-Efficacy”)	Lower (Abi) ↔ Higher (Tim)	Confidence and perceived capability in unfamiliar computing tasks.
Attitudes Toward Risk	Risk-averse (Abi) ↔ Risk-seeking (Tim)	Tolerance for ambiguous feedback and uncertain outcomes.

Pat, the midpoint persona, anchors intermediary positions. Persona demographic details (age, profession, pronouns, background) can be tailored to context, but facet values remain fixed to retain methodological consistency (Mendez et al., 2019, Anderson et al., 21 Oct 2025).

Empirical rationale: supporting both endpoint personas ensures UI coverage for the full spectrum of problem-solving styles, reflecting observed gender clustering (Anderson et al., 21 Oct 2025, Mendez et al., 2019).

2. Standard GenderMag Walkthrough Workflow

The core GenderMag process follows a staged, reproducible procedure:

1. Preparation:
   • Assemble a walkthrough team (typically 2–4 participants) encompassing facilitator, driver (UI navigator), recorder, and evaluators.
   • Select a persona (Abi, Pat, or Tim) and customize only context-appropriate demographic/background parameters.
2. Scenario and Subgoal Decomposition:
   • Define an overall usage scenario and decompose it into a sequence of subgoals (task steps) for the persona.
3. Iterative Walkthrough Loop (for each subgoal):
   • Subgoal Question: "Will [Persona] form this subgoal as a step toward the overall goal?" (Y/N/Maybe)
   • Pre-Action Question: "Will [Persona] take the intended action at this step? Why?" (Y/N/Maybe)
   • Post-Action Question: "If [Persona] does the right thing, will they know it and recognize progress?" (Y/N/Maybe)
   • For each non-affirmative (No/Maybe) response, flag a potential inclusivity bug; document the cognitive facet(s) responsible.
4. Data Capture and Analysis:
   • Employ printed or electronic forms for each subgoal, capturing answers, rationale, and facet associations.
   • Aggregate bug instances by similarity into bug types; compute content analysis metrics such as inter-rater agreement (e.g., Jaccard index $J(A,B) = |A \cap B| / |A \cup B|$).
5. Issue Tagging and Fix Proposals:
   • Each bug instance is traced to implicated facets, guiding targeted remediation (e.g., preview options for risk aversion, inline help for low self-efficacy).

This process diverges from generic cognitive walkthroughs by enforcing explicit facet citation and a standardized answer palette augmented by "Maybe" to reduce groupthink and capture nuance (Mendez et al., 2019, Anderson et al., 21 Oct 2025).

3. GenderMag-for-AI: Extensions for AI-Powered Interfaces

GenderMag-for-AI introduces systematic modifications to address the unique inclusivity bugs inherent to user-facing AI systems, including breakdowns triggered by user trust (or distrust) in AI outputs (Anderson et al., 21 Oct 2025).

Key adaptations:

• Pre-Action Fork: Each pre-action question is split into “AI-Right” (user trusts previous AI output) and “AI-Wrong” (user suspects a mistake), with parallel analysis of both branches.
• Optional 'Understand?' Interstitial: An intermediate check interrogates whether the persona comprehends the AI output before proceeding.
• Failure Mode Enumeration: For each subgoal, enumerate all salient AI failure modes $F = \{f_1, f_2, ...\}$; for each, analyze changes in persona behavior and interpretation.
• Metrics: Compute per-facet bias detection ratios: $$B_{facet} = \frac{\text{AI-Inclusivity issues per facet}}{\text{total subgoal/task steps}}$$.

This approach exposes over-the-hood AI inclusivity bugs, such as ambiguous input-output mappings, opacity regarding AI reasoning, and interface misalignments under user uncertainty, which remain undetected by standard usability inspection (Anderson et al., 21 Oct 2025).

4. Analyst Roles, Artifacts, and Facilitation Practices

The GenderMag process enforces strict procedural roles and artifact conventions to ensure replicable and unbiased analysis:

• Facilitator: Ensures facet focus, enforces equitable contribution, and maintains session direction.
• Driver: Operates the UI/prototype, allowing all evaluators to concentrate on the analysis.
• Recorder: Documents Y/N/Maybe answers, rationale, facet attributions, and direct participant quotes.
• Evaluators: Provide judgments strictly from the persona’s cognitive profile.

Checklists, color-coded forms, and annotated UI screenshots are employed to maintain clarity and facet salience. Team rotation and first-person reasoning ("Abi will…") are mandated to prevent dominance and erroneous abstraction (Anderson et al., 21 Oct 2025).

Session execution involves structured walkthroughs, iterative bug identification, "find" and "fix" sessions, and empirical validation rounds if resources permit.

5. Integration with InclusiveMag and Intersectional Adaptations

InclusiveMag generalizes the facet-driven walkthrough meta-method to other diversity dimensions. Hybrid methodologies (e.g., for neurodivergent women in SE) combine GenderMag’s facet logic with new or extended dimensions, leading to a formalized scoring process (Zaib et al., 5 Dec 2025, Mendez et al., 2019).

• Facet Extension: Facets are reframed or augmented to accommodate new axes (e.g., Motivation extended by “Desire to Mask”, Information Processing by “Sensory-Load Tolerance”).
• Inclusivity Dimensions: Additional constructs such as Discoverability, Comprehension, Cognitive Load, etc., are layered atop the facet logic.
• Formal Notation:
  • Let $G = \{g_1, ..., g_5\}$ be the set of adapted facets
  • $I = \{i_1, ..., i_k\}$ the set of inclusivity dimensions
  • $T = \{t_1, ..., t_m\}$ task scenarios
  • The score $S : T \times G \times I \rightarrow \mathbb{R}$ with $S(t, g, i) = w_i \cdot v(t, g, i)$ aggregates into persona-level scores $S_P(t) = \sum_{g \in G} \sum_{i \in I} S(t, g, i)$.

Application involves collaborative workshops where practitioners and validators (e.g., ND-women) walk through realistic SE tasks, cluster issues, and synthesize actionable tool, workflow, or policy changes (Zaib et al., 5 Dec 2025).

6. Metrics, Validation, and Empirical Outcomes

Metrics for process efficacy and inclusivity impact include:

• Facet Coverage: The proportion of facet-congruent issues identified before and after intervention, $$B_{facet} = |\text{issues}_{facet}| / |\text{total steps}|$$.
• Empirical Effectiveness: Gender equity improvements in user mental-models (e.g., 45% increase post-intervention in field case studies for AI products) (Anderson et al., 21 Oct 2025).
• Conventional Metrics:
  • True Positive Rate (TPR):

  $$\mathrm{TPR} = \frac{\text{TP}}{\text{TP} + \text{FN}}$$ - False Positive Rate (FPR): $$\mathrm{FPR} = \frac{\text{FP}}{\text{FP} + \text{TN}}$$ with observed TPRs between 75–100% and FPRs under 4% for traditional GenderMag walkthroughs (Mendez et al., 2019).

Best practice entails iterative remediation and retesting, incorporating empirical user validation (e.g., A/B studies), and prioritization by impact on excluded problem-solving styles.

7. Empirical Case Studies and Best Practices

Field evaluations with AI product teams (game, weather, and farm-management domains) validate GenderMag-for-AI’s capacity to uncover and triage unique AI inclusivity bugs. Key examples include diagnosis barriers caused by ambiguous outputs, lack of actionable affordances, and interface passivity in the face of AI uncertainty (Anderson et al., 21 Oct 2025).

Best practices emphasize:

• Rigorous facet education and persona-specific reasoning
• Artifact and UI snapshot annotation focused on atomic subgoals
• Forked data capture for AI-Right/AI-Wrong branches
• Rotated facilitation roles and session structure spanning persona familiarization, bug elicitation/fix prototyping, and iterative retesting

Concrete artifacts include original GenderMag forms, fork templates for AI, and failure mode tables. This procedural infrastructure enables systematic inclusivity audits and continual product improvement cycles for diverse user populations.

---

The GenderMag walkthrough process, and its recent extensions to AI products and intersectional inclusivity (via InclusiveMag), represent a reproducible and empirically grounded approach for surfacing and mitigating inclusivity bugs. Rooted in cognitive science and validated through multi-team studies, it bridges research and practice in the systematic design of equitable and accessible user-facing technologies (Mendez et al., 2019, Anderson et al., 21 Oct 2025, Zaib et al., 5 Dec 2025).