Scoring, interpretation and limits

The instrument is intended for executives and senior leaders describing current organizational conditions from their own perspective. It does not independently observe behavior or verify supporting evidence. Results may be affected by role, visibility, recall, response style, recent events and social desirability.

The production bank contains 88 questions across eight domains. A session selects 35–50 questions; the default is 40. Selection first applies any sector and organization-size eligibility rules, then greedily favors questions that improve coverage of constructs below the target of three observations. Remaining positions favor domain balance. Selected questions are interleaved across domains and their order varies by session.

Different respondents may therefore receive different item sets. Scores can be compared descriptively, but no equivalence study has yet established that every generated form has the same measurement properties.

Every response is an integer from 1 to 5, with 5 always representing the healthiest described condition. Each item is tagged to two or more constructs with a weight from greater than 0 through 1. For construct c, the engine computes:

construct score(c) = Σ(item weight × response value) ÷ Σ(item weight)

The score is a rolling weighted mean on the 1–5 response scale. An item tagged to several constructs is an observation for each tagged construct. The current engine counts observations, but it does not estimate standard error, internal consistency, item discrimination or a confidence interval.

A metric first takes the equal-weight mean of its observed input construct scores. Construct scores with no observations are excluded. Item weights affect the construct scores beneath a metric; the constructs themselves are not differentially weighted when combined into the metric.

This means a partially completed Builder Index is not merely a less complete estimate of a fixed composite. Its composition changes as previously unobserved constructs enter the mean. An early reading and a final reading can therefore differ because the set of included constructs changed, even before considering changes in the answers themselves.

Builder Index and AI Readiness Indicator values linearly rescale the construct mean from 1–5 to 0–100. The 0–100 format is an index without a percentage unit. Execution Friction and Leadership Load invert the healthy-oriented mean so that a lower displayed value is better. All other metrics remain on the 1–5 scale.

Bands and unlock thresholds are design thresholds. They have not been derived from outcome data, normative samples, sensitivity/specificity analysis or externally observed breakpoints.

An unanswered item contributes nothing. Within a metric, an entirely unobserved construct is excluded rather than assigned zero. A metric becomes visible only after both its question-number threshold and minimum observation threshold are met. The engine does not currently offer an “unable to observe” response or impute missing answers.

Evidence sufficiency is answer coverage, not statistical confidence. The labels are: Early for 0–9 answered questions, Developing for 10–20, Established for 21–34, and Complete for 35 or more. “Complete” means the coverage threshold was reached; it does not mean the result is definitive, accurate or validated.

Each metric card also has an evidence-fill display calculated as observations divided by twice the metric's minimum observation requirement, capped at 1. That display is likewise a coverage ratio, not a confidence interval or reliability estimate.

The self-diagnostic does not estimate a causal constraint. It ranks unlocked metrics for which a recommendation exists, excluding the Builder Index. Each metric is normalized to a 0–1 severity value: healthy-oriented metrics use (maximum − score) ÷ range; inverted metrics use (score − minimum) ÷ range. The metric with the highest resulting severity is displayed first and supplies the “Address First” recommendation.

All nine non-Builder-Index metrics currently have recommendation entries, so catalog coverage does not presently narrow the field. Unlock order does. Executive Attention Index can enter the ranking at question 5; Alignment Coherence and Authority Distribution at question 8; Decision Velocity at question 10; Execution Friction and Leadership Load at question 12; AI Readiness Indicator and Risk Judgment at question 20; and Transformation Readiness at question 28. Before a metric unlocks, it cannot be selected regardless of what later answers may show. The displayed result is therefore structurally biased toward earlier-unlocking metrics during partial completion.

Ranking also assumes that a linear normalization makes the different metric scales comparable. That is a design assumption, not an empirically established property. In particular, the operation treats distances between adjacent response-derived values as equal and treats equivalent normalized positions on different composites as commensurable. The current evidence does not establish either claim.

The phrase “primary constraint” should therefore be read as indicated area to investigate first, not as proof that the metric causes other organizational outcomes or that improving it will necessarily produce the greatest effect.

Team reports require at least three respondents. For each unlocked metric, the engine reports the arithmetic mean, minimum, maximum, range and population standard deviation across respondents. Constructs receive the same descriptive summaries.

A divergence zone is flagged when the observed range exceeds 1.5 points on a 1–5 metric, or the equivalent 37.5 points on a 0–100 index. This is a design rule, not an empirically validated threshold. Divergence describes disagreement among respondents; it does not by itself establish that Alignment is the cause of the disagreement.

Version 1.0 has automated tests for scoring behavior, question selection and aggregation. Those tests establish that the software follows its specified rules; they do not establish measurement validity.

No published study has yet established internal consistency, test–retest reliability, inter-form equivalence, factor structure, convergent or discriminant validity, criterion validity, predictive calibration, measurement invariance, normative bands or meaningful-change thresholds for this instrument. Until that work is completed, every output should be treated as developmental and directional.

Future technical-note versions will identify the data set, methods and results behind any validation claim. Changes to items, weights, constructs, thresholds or interpretation rules will be recorded in the version history below.