By Philip Smith
Last year Apricity was commissioned to evaluate five years of WASH programming in Yemen. We worked with thousands of data points across four grant projects, multiple regions and three data collection methods. We were rigorous in our approach, and it paid off: we produced coherent, defensible findings.
I have since come to think those findings were also, to a meaningful degree, a product of what I brought to the data, not just what the data contained. Cognitive science tells us the conscious mind processes roughly 40 to 120 bits of information per second, a range converged on by researchers including physiologist Manfred Zimmermann and popularized by Tor Nørretranders in The User Illusion. Total sensory input runs into the millions. The gap is so vast that the brain cannot function as a camera. It functions as an editor, or more precisely, as a prediction engine. It uses prior experiences, patterns, and habits to fill in what it has not directly observed, producing a seamless, convincing account of reality that is, by structural necessity, partial.
Stephen Covey puts the same insight differently: we see the world not as it is, but as we are. Neuroscience now tells us this is not metaphor. In Being You, Anil Seth describes perception as a “controlled hallucination”: what we experience as reality is something the brain actively constructs from its own predictions, and then updates against sense data, rather than the other way round.
What unsettles me is realizing how specifically this applied to my own evaluation work. In the months before data collection in Yemen began, I attended major development conferences where three experienced Global South leaders made persuasive arguments about financial sustainability and local ownership as the hallmarks of mature humanitarian programming. Those conversations were intellectually honest and grounded. They were also, I can now see, loading my prediction engine.
Illustrative: Every evaluator's prediction engine shapes what they see. No single reading catches the whole picture.
When I sat down with five years of Yemen data, the through-line I produced centered on cost-recovery and sustainability. It was coherent, evidenced, and it was also the through-line my prediction engine was primed to produce. This is just how interpretation works. The same logic applies upstream. Every theory of change, every results framework, arrives with a prediction engine already running. The implication of this goes beyond my individual experience.
Consider, hypothetically, how a Yemeni evaluator might have approached the same data. They might have read the role of tribal structures in legitimizing fee-collection differently, the cultural assumptions around mutual aid and state obligation, the particular dynamics of qat consumption on afternoon committee meetings. These are not details any field survey captures. They are tacit, accrued through proximity. They would have shaped the through-line differently, not a correction to mine, but a different reading of the same data.
The case for putting local voices at the center of evaluation is not simply for courtesy or representation. It is, at its core, an epistemological one. Local voices carry knowledge the data cannot capture and the outside evaluator cannot arrive with.
Themrise Khan, co-editor of White Saviors in the Aid Industry, argues the sector’s knowledge problem isn’t only about who is at the table, it’s about whose frameworks count as valid readings of what’s there. That is an epistemological critique, not a representational one.
The interpretations most likely to be accurate are those that emerge from a genuinely plural process of interpretation. Both are expert readings of the same evidence, each shaped by a different prediction engine. Together, they’d have caught things neither of us would have caught alone. I am not outside this problem. I am the western-trained evaluator with the pre-loaded prediction engine.
We talk a lot in this sector about the quality of evidence. We do not talk enough about the quality of the interpretive lenses we bring to it.
Philip Smith is part of the Apricity team. This piece draws on experience gained through that work but reflects their own views and conclusions and does not necessarily represent the views of Apricity.
