Fisheries and Oceans Canada
Diagnosing and repairing the information architecture of a federal government website
Role
Solo UX researcher (extending a team project)
Timeline
Oct to Nov 2026
Methods
Site inventory, card sort, tree testing
Tools
Miro, ProvenByUsers

01 / At a glance
At a glance
25% → 56%
Overall task success
40% → 69%
Success once inside the correct parent
~70%
Predicted before testing
02 / Background
Background
In an Information Design class, my team of five mapped the information architecture of the Fisheries and Oceans Canada website. We inventoried 115 nodes across the site, ran a card sort to understand how users grouped the content, and ran a tree test to evaluate our proposed restructure.
Our redesign performed worse than the live site. Rather than move on, I returned to the project on my own to diagnose why it failed and test whether the structure could be repaired.
03 / The problem
The problem
The task asked participants to find information about Pacific Aquaculture Environmental Risk assessments. Only 25% of participants completed it successfully on the live site’s structure, and our team’s proposed rename made it worse.
The question was whether the problem was the content’s placement in the tree or how the tree signposted it to users.
04 / Diagnosis
Diagnosis
The target page sat inside the Aquaculture section, which was the correct parent taxonomically. But three of its four sibling pages were about fish disease, and six of sixteen participants never opened Aquaculture at all.
Additionally, users who entered Aquaculture were drawn to a decoy: ‘Aquaculture Science,’ which sounded more relevant than it was. Three of sixteen participants ended on Aquaculture environmental risk assessments inside it, making it the single most common wrong answer.
The problem was not placement. It was recognition. Users could not tell from the labels alone that they were in the right part of the site.

05 / Approach
Approach
I designed a within-subjects study: the same 16 participants completed the same task on the baseline structure, then on the revised structure days later.
Baseline test
Ran the original task on the live site’s tree with 16 participants to establish a clean baseline.
The revision
Split the Aquaculture section into two clearly labelled nodes: Fish Health Management and Environmental Impact and Reporting.
Revised test
Ran the same task again on the revised tree to measure whether the new labels improved recognition.
06 / Baseline findings
Baseline findings
4 of 16 participants found the correct page (25% success). Of the 10 who entered Aquaculture, only 4 found it from there (40% conditional success).
Most failures happened because participants never entered Aquaculture at all, or chose the Aquaculture Science decoy once inside.
07 / The revision
The revision
I split the single Aquaculture node into two children: Fish Health Management (grouping the three disease pages) and Environmental Impact and Reporting (grouping the target page with its closest neighbour).
The goal was to give users a clear signal at the moment of choice. Nothing else in the tree changed.

08 / Results
Results
Task success rose from 25% to 56% (4 to 9 of 16). Among participants who entered Aquaculture, success climbed from 40% to 69%, matching the ~70% I predicted before testing.
The revision fixed the recognition problem. Users could now see from the labels that environmental content lived in Aquaculture. But directness stayed low (1 to 2 direct successes) and average time increased, meaning participants still took indirect paths to get there.

09 / What it didn’t solve
What it didn’t solve
The revision improved recognition but not wayfinding. Participants found the right page more often, but most still took indirect routes. Direct successes rose from 1 to only 2, short of the threshold of 3 I set before testing.
Fixing wayfinding would likely require changes higher in the tree or cross-links from related sections. That was outside the scope of a single-node revision.
10 / Reflection
Reflection
This project taught me to separate diagnosis from solution. The team’s original redesign failed because we renamed a section without understanding why users were failing. When I came back and looked at the actual paths, the problem was signposting, not taxonomy.
Working with 16 participants means the numbers are directional, not definitive. But the prediction held, and the method is what I would scale in a professional setting with a larger sample.
11 / Full report
The full 30-page research report covers methodology, data handling, comparative analysis, limitations, and recommendations.


