How this Matrix is calculated

Versioned question set and adapter

The active question set is v0.5.3, with 34 questions and 144 substantive answer options. It is generated from reviewed family records; validation fails if the generated artifact drifts from those records. The scoring model is v0.5.3.

For provenance, the internal IDs are alpha-core-0.5.3 and alpha-adapter-0.5.3. They remain stable across historical snapshots.

Question, content, scoring, presentation, and baseline versions travel with a session and result snapshot. This lets a later review distinguish a current 0.5 result from an earlier 0.4 session.

Construct evidence and item weighting

Questions are tied to specific constructs. Each item contributes at most one observation to its construct. Four-, five-, and six-option presentations therefore have equal item weight. Directional evidence is bounded and combined with other usable observations for that construct.

The system keeps raw responses, response states, option metadata, and derived estimates separately so the result can be audited and recomputed from the recorded version.

Variable options and mechanism-only responses

Options are item-specific. A selected option may have a scalar contribution, or it may be evidence about a distinct mechanism. A mechanism-only option has a null scalar: it is retained in the evidence and audit records without inventing a numerical placement on a construct.

Not sure, Skip, uncertain, and unanswered remain distinct from an answered neutral choice. They affect evidence coverage and uncertainty according to the adapter contract.

Cross-checks, snapshots, and baseline

The approved scoring audit covers all 34 items, every option, and six cross-check chains. Cross-construct signals and caveats stay attached to the option record rather than being hidden inside a single coordinate.

Result snapshots retain selected option IDs, response states, estimates, uncertainty, evidence coverage, question-set version, scoring version, methodology version, and baseline version. The current baseline is Developing v0.1 and is explicitly developing. It is not a representative population benchmark. A later baseline can change display context while historical version information remains intact.

The baseline’s internal provenance ID is baseline-alpha-0.1.

Uncertainty limits

The current uncertainty signal reflects the amount and type of usable evidence available in the scoring model. It is a product-resolution indicator, not a claim of a confidence interval, population margin of error, or validated standard error. More responses do not automatically resolve a construct if the evidence is ambiguous or mechanism-specific.

Validation roadmap

Formal validation is ongoing. The research plan calls for:

• Cognitive interviews and pilot testing to check comprehension, symmetry, burden, and response-state meaning.
• Item analysis and reliability to evaluate information and precision across the intended range, without treating internal consistency as proof of a construct.
• Exploratory factor analysis and confirmatory factor analysis on separate development and holdout data to test plausible structures.
• IRT where appropriate for the response operation, with category thresholds, information, and conditional precision reviewed.
• Cross-loading and sensitivity review for wording, order, scenario, policy domain, mode, and personal-incidence effects.
• Test-retest work to separate durable patterns from context, exposure, practice, and real change.
• Measurement invariance and DIF audits across registered groups, languages, modes, forms, and collection waves before group comparisons.
• Locked holdout validation before a candidate model or threshold is treated as confirmatory.