How we test
an idea.
Our research begins with something we observe. We ask what might explain it, make a prediction, and design an experiment that could prove us wrong.
The method.
Evidence can take us back to an earlier step. A failed prediction is a reason to revisit the explanation that led to it.
Observe a problem
Watch people do the work. Where do they lose time, make mistakes, or need help? Record what happens today so there is a baseline for comparison.
Review the evidence
Trace claims back to the original studies. Compare their methods, look for conflicting results, and identify what remains uncertain in this particular setting.
State a hypothesis
Use prior evidence to propose an explanation. Derive a measurable prediction from it and write down what result would count against it.
Run a controlled test
Compare one change against a baseline under the same conditions. Measure the outcome we chose in advance and record failures alongside improvements.
Repeat the experiment
Ask someone else to repeat the test from the recorded inputs and procedure. Then try fresh examples, including difficult cases, to find where the result holds.
Try it in the field
Test a small prototype with the people who would use it. Compare it with their current workflow and listen for problems our lab test missed.
Build, revise,
or stop.
Build when the benefit repeats and matters in use. Revise when the evidence points to a better explanation. Stop when the benefit disappears or the cost to users outweighs it.
What happens in the field becomes the next observation. That is how the loop begins again.
An example
in digital presence.
Can a viewer tell where a digital face is looking without seeing its eyes change shape?
One experiment could compare two gaze animations of the same portrait. One stretches the eye texture. The other moves the iris beneath the eyelids. The face, lighting, direction, and timing would be identical.
The prediction would be fewer reports of distortion with the moving iris. Viewers would see the clips in random order, without labels, and describe the gaze direction and anything that looked unnatural. A second group would repeat the comparison on different screens.
Explore the portrait