Re-sort validation
Re-sort validation is a check on a proposed typology of users' situations. After looking through survey or interview responses and forming a hunch that they fall into a few types, the analyst goes back and assigns each individual response to its best-fit type. Types that pull no responses, responses that fit no type, and responses that could go in two all signal the typology needs work before it gets committed to.
Definition
A user value research operation that tests proposed clusters by sorting each raw response into its best-fit cluster, treating the absorption pattern as diagnostic of whether the proposed clusters match the data they were inferred from.
From impression to grounded typology
Pattern recognition over qualitative data — interview notes, survey responses, sticky-note outputs — generates impressionistic candidate clusters: "respondents seem to fall into three types." The impression is a starting point, not a result. Re-sort tests whether the proposed clusters actually absorb the responses they were inferred from.
The mechanics are straightforward. Take each raw response. Assign it to the cluster that fits it best. Read what happens.
Clusters that don't accumulate responses were impression rather than pattern. Responses that don't fit any cluster signal a cluster the impression missed. Responses that fit multiple clusters signal boundaries the impression didn't sharpen. The output is a typology grounded in response-distribution rather than analyst confidence.
The four absorption failures
Four ways absorption can break down, each diagnosing a different problem with the proposed clusters:
- Over-absorbing: a cluster's name technically applies to responses that don't share its defining characteristics. The name absorbed them by word-level match while the underlying situation differs. The cluster has likely fragmented into sub-clusters that should be separated.
- Narrow scope: adjacent responses fail to fit. The cluster's reach is narrower than the pattern suggested.
- Thin sustenance: few responses actually belong. The pattern was weaker than the initial impression suggested; the cluster may not be real.
- Missing pattern: responses fall around something unnamed. The proposed cluster is adjacent to but not the actual pattern.
Best-fit sorting
The discipline is one cluster per response, committed to. If a response could plausibly go in two clusters, the practitioner picks which fits better. The alternative — sorting into both — collapses the cluster boundaries together and defeats the validation. Forced discrimination at the moment of sorting is exactly what surfaces over-absorption and missing patterns; loose-fit sorting lets the practitioner keep responses in multiple clusters mentally and never confronts the failure.
Re-sort vs hot-air
Both diagnostics test whether a named thing is sustained by structure or by convention, but they operate at different layers. Hot-air catches names with no specifiable structure underneath — the name fails when asked what it's made of. Re-sort catches names whose specified structure doesn't match the responses they were inferred from — the name has structure, but the structure doesn't fit the data.
A cluster can pass hot-air (its characteristics are specifiable) and fail re-sort (those characteristics don't absorb the responses cleanly). The two diagnostics run at different stages of cluster work and catch different failure modes.
Example
An analyst interviews 50 situation explorers who recently completed first paid conversion on a project-management tool, asking what drove them to pay. Pattern recognition over the transcripts suggests three clusters: "solo founders coordinating contractors," "team leads onboarding new hires," and "agencies tracking client deliverables."
Re-sorting the 50 transcripts produces:
- Solo founders coordinating contractors: 22 responses — clean cluster.
- Team leads onboarding new hires: 4 responses — thin sustenance.
- Agencies tracking client deliverables: 17 responses — clean cluster.
- Don't fit any cluster: 7 responses, all describing internal consultants evaluating tools on behalf of departments that won't end up using them — missing pattern.
The revised cluster set: solo founders coordinating contractors, agencies tracking client deliverables, and a newly-named "internal evaluators." The team leads cluster gets dropped or held for more responses to decide. None of this was visible from the original pattern-recognition impression.