Methodology
What stands behind this instrument, and what is still being established.
Every item is written to a named construct, the bank is frozen and versioned, and the limits of a reading are published on every screen that shows one. What is not yet established is the psychometric evidence: reliability, factor structure, invariance and predictive validity. Those need respondents, and the analyses that will test them were written down before the first one existed.
- Items mapped to a construct
- 114 of 114
- Domains, every construct observed
- 8
- Frozen bank version
- 8b9c5245b5a7
Publishing the thresholds before the data is the part worth noticing. An instrument can be marketed on claims alone; this one names in advance the result that would count against it, which is a thing you can hold it to later.
The nine analyses
Each states what is computed, the minimum sample it requires, and what result would count against the instrument. That last column is what makes this a protocol rather than a wish list: an analysis with no disconfirming outcome cannot fail, and an analysis that cannot fail establishes nothing.
Recruiting has not started. Item-level responses come only from paid team engagements whose participants consent separately and after their own result, so the sample builds at the pace of delivered work rather than site traffic. Every row below will move as it does, and the thresholds will not.
A1Item exposure
0 of 30 respondents
How often each item in the adaptive bank was administered, and how far two 40-item forms overlap.
What would count against the instrument: Forms overlapping so little that no common scale links them.
A2Construct coverage in practice
0 of 30 respondents
Whether each construct really receives the three independent observations the design claims.
What would count against the instrument: Constructs routinely scored on fewer observations than the design claims.
A3Item calibration
0 of 400 respondents
Item response theory calibration per construct, on the items with enough exposure.
What would count against the instrument: Items that do not discriminate, or that fit no common dimension within their construct.
A4Dimensionality
0 of 400 respondents
Whether each construct behaves as one thing, by factor analysis on the calibrated subset.
What would count against the instrument: A construct that is not one thing, but two or more factors with no substantive interpretation.
A5Internal consistency
0 of 150 respondents on one form
Omega and alpha on a fixed form, so a reported condition score has a reliability attached.
What would count against the instrument: Omega below 0.70 for a construct reported as a score.
A6Test-retest stability
0 of 20 paired engagements
Whether an organization measured twice, with no intervention between, gets the same reading.
What would count against the instrument: An intraclass correlation below 0.60, which would mean the score moves without the organization moving.
A7Team divergence
0 of 40 engagements
Whether disagreement inside one engagement is larger than the item’s own measurement error.
What would count against the instrument: Divergence indistinguishable from the error of the items it is computed on.
A8Measurement invariance
0 of 400 respondents
Whether the instrument means the same thing across sector and organization size.
What would count against the instrument: Non-invariance, which would mean scores are not comparable across the groups the product compares.
A9Criterion validity
0 of 30 engagements followed to twelve months
Whether the constraint the diagnostic indicated predicts what an organization went on to change, against measures it already kept.
What would count against the instrument: No relationship between the indicated constraint and the measures the organization already tracked, which would mean the reading describes nothing outside itself.
The thresholds, stated in advance
Omega below 0.70 for a construct reported as a score, and an intraclass correlation below 0.60 for a measurement repeated with no intervention between, are the two numbers that would fail. Both are conventional in applied organizational measurement rather than derived, and they are published here so that loosening one to accommodate a result is a visible act rather than a quiet one.
Where the sample comes from
Participants in a team diagnostic may contribute their answers after they have finished and seen their own result. Nothing is contributed without a separate, explicit choice, and the instruments that run entirely in your browser are never asked, because those pages promise your answers are never sent. What a contribution contains, and how to withdraw it.
No instrument has responses yet, so no analysis can be run on any of them.
What a green row would and would not mean
A row marked ready means the sample is large enough for the analysis to be run. It does not mean the analysis has been run, and it certainly does not mean it passed. Results land in the technical note with their sample, their date and the instrument fingerprint they were computed on, including the ones that count against the instrument.
Until then, every score this product reports is a directional reading for a conversation, which is what the technical note has said since version 1.0 and what it will keep saying until there is evidence to say otherwise.