Tree + Taxonomy Diagnostic Loop

Description

A two-phase protocol pairing an unmoderated tree test (where findability fails) with moderated micro-sessions coded as Slip, Lapse or Mistake (why), joined in a joint display and re-tested.

The framework turns 'navigation was confusing' into decision-ready action. Phase A is an unmoderated tree test (N = 80–150) on a text-only IA of 2–3 levels with 8–12 randomised tasks, yielding direct and indirect success, failure, first-click accuracy, extra clicks, time to first click and misroute hotspots. Phase B is 10–15 moderated 30–40 minute think-aloud micro-sessions on the same tree and tasks, with each breakdown coded by a one-page codebook and two coders on 30–40% of sessions (κ ≥ 0.70). The glue analysis joins phases in a joint display mapping quant signal → taxonomy read → prescribed fix → effort and impact. Decision bars are set before changes (direct ≥ 85%, misroute ≤ 10%, extra clicks ≤ 1.0, first-click ≥ 80%) and the tree test is re-run with the same tasks and similar N, plus 5–8 verification think-alouds.

Attributes

Loss type
Community-proposed
Data
Mixed
Effort
Weeks

Steps

  1. Detect: export the IA as plain text (depth ≤ 3), write 8–12 tasks without label leakage, and run an unmoderated tree test with 80–150 participants, logging first click, final node, path, time and backtracks.
  2. Compute direct success, indirect success, failure (1 − direct − indirect), misroute share per node, first-click accuracy, extra clicks and time to first click.
  3. Diagnose: run 10–15 moderated 30–40 minute think-alouds on the same tree, using prompts such as 'What do you expect to find under this label?'
  4. Code each breakdown as Slip, Lapse or Mistake; double-code 30–40% and compute Cohen's κ, refining the codebook until κ ≥ 0.70.
  5. Prescribe: build a joint display mapping quant signal → mechanism → fix with effort and impact ratings, using the insight pattern 'Because [quant evidence] and think-aloud revealed [mechanism], we recommend [fix] with decision bar [thresholds]'.
  6. Decide: confirm bars set upfront (direct ≥ 85%, misroute ≤ 10%, extra clicks ≤ 1.0, first-click ≥ 80%).
  7. Re-test: re-run the tree test with the same tasks and similar N and 5–8 verification think-alouds to confirm the mechanism is resolved.

Sources

  • Vivek Gopinath (2025), Tree Testing and Error Taxonomy: A Complete Chapter for UX Researchers
  • Vivek Gopinath (2025), Forward-Thinking UX Research Playbook: A Mixed-Methods, Decision-Ready Approach from Prediction to Validation

Also known as

  • Tree Testing and Error Taxonomy framework
  • Two-phase tree test and taxonomy design

Categories

  • Community-Proposed Frameworks