← Back to the blog
Epistemics · July 14, 2026

The Copenhagen Interpretation of the White Paper

The Copenhagen Interpretation of the White Paper

Has anyone in the room read the two-hundred-page report everyone keeps citing, or has everyone quietly collapsed it into whichever sentence already matched their prior belief?

Before a white paper gets opened, it sits in a comfortable state of suspended meaning. The program officer, the lead researcher, the policymaker, each carries a private draft of what the document is likely to say. That draft comes from priorities, incentives, and old grudges rather than from the two hundred pages themselves. Then somebody opens the file, and the suspended state resolves into a single reading, usually the one already sitting there, the reading the opener came in wanting to confirm. Physicists have a word for this kind of resolution. Borrowing it here is a little cheap, and worth admitting up front, since dressing confirmation bias in the language of quantum mechanics is one of the oldest tricks in the popular-science drawer. The borrowing earns its keep only if the physics underneath describes the same shape of problem, and not only the same mood.

The easy version reaches for the Copenhagen interpretation: a system exists in superposition until measured, and the act of measurement collapses it into one outcome. Applied loosely to a report, it gives a tidy image, meaning suspended until a reader looks. It is also wrong in the one detail on which everything here turns. Plain Copenhagen describes a single collapse, one measurement, one outcome. A report goes through two.

The physicist Eugene Wigner sharpened the problem in 1961, in a scenario that now bears his name: Wigner's friend. A colleague sits sealed inside a laboratory, measures a quantum system, and records a result. Wigner waits outside the door. From the colleague's position inside the room, the measurement is finished and the outcome is settled. From Wigner's position outside, nothing has collapsed yet, since standard quantum theory describes the entire room, colleague included, as still evolving in superposition until Wigner opens the door and looks for himself. Two observers, two honestly held descriptions of the same event, in direct conflict. The scenario sat mostly as a curiosity for decades, until 2019, when a team at Heriot-Watt built a six-photon version of it in the laboratory and confirmed, through a violation of a Bell-type inequality, the conflict is not a trick of notation. Nested observers, under the right conditions, walk away with incompatible facts about the same event, and no further measurement forces them back into agreement.

A report has exactly this shape. Somebody inside the process, an analyst, a research team, measures the world and records a result. Somebody outside the process opens the document later and performs a second, independent act of measurement on what the first one produced. Two observers, two honest collapses, no referee.

By the time a finding reaches the page, in the passive, hedge-softened voice of a results section, it has already survived a long chain of decisions nobody sees: which variables to include, which outliers to exclude, which subgroup to test, which model to run first and which to try only after the first one looked unpromising. "It was found" is a sentence with its agent scrubbed out. Something got found, evidently, by nobody in particular, through no choices in particular.

In 2018, a research team under Raphael Silberzahn and Eric Uhlmann measured how much this hidden agent matters. Twenty-nine independent teams, sixty-one analysts in total, received the identical data set and the identical question: were football referees more likely to issue red cards to dark-skinned players than to light-skinned ones. Left to choose their own defensible statistical approach, the teams produced effect-size estimates from 0.89 to 2.93 in odds-ratio terms, a spread of more than threefold. Twenty teams found a significant effect. Nine did not. The disagreement tracked neither the analysts' prior beliefs about the answer nor their technical seniority, and independent peer reviewers who judged the analyses blind could not tell the stronger work from the weaker by the results alone. Sixty-one honest, competent people, one data set, and underneath it no single fact for anyone to uncover. The first collapse had already happened by the time anyone wrote a sentence about it, and it happened sixty-one different ways.

Suppose, generously, one of those sixty-one analyses becomes the report somebody eventually reads. A second observer waits on the other side of the door, and the second observer is rarely neutral either.

In 1979, Charles Lord, Lee Ross, and Mark Lepper ran what is now one of the most replicated experiments in social psychology. Committed supporters and committed opponents of capital punishment read the same pair of studies: the first seemed to show a deterrent effect, the second seemed to show none. A neutral process would have moved both groups toward the middle, or at minimum left their disagreement unchanged. Instead, both groups moved further apart. Each side rated the study in favor of its prior view as the better designed and the study against it as flawed, and left the experiment more sure of its original position than before it read anything. Identical evidence, in the hands of two honest groups, produced two hardened and opposite convictions. The report did not resolve the disagreement. It became raw material for both sides to build the disagreement higher.

Put the two studies together and the double collapse stops sounding like a physics metaphor and starts looking like a design flaw nobody designed. A report is one particular collapse out of many defensible ones. An analyst produced it under invisible constraints, and a reader will collapse it again under equally invisible ones. Nobody in this chain is lying. Nobody is even being careless, by the standards of their own trade. The distortion needs no villain, which is exactly why it survives peer review, board oversight, and every good-faith safeguard an institution knows how to build.

The fair objection accepts all of this and asks what the alternative looks like. Nobody has the time, or the statistical training, to read sixty-one competing analyses for every dataset behind every report on the desk. Synthesis exists precisely because raw plurality is unusable at any scale worth caring about, and a good executive summary is a mercy, the only usable form at that scale. The objection is fair, and it does not rescue the illusion this piece is chasing.

The collapse itself was never the trouble. The trouble is the pretense that no collapse occurred, the way "it was found" gets presented as though it named a fact from outside all human judgment, when it named one path among sixty-one.

Wigner never settled his own paradox. Depending on which interpretation of quantum theory a physicist favors today, the friend inside the lab and Wigner outside the door are either both right, in some sense neither manages to explain to the other, or one of them is simply mistaken about a reality neither proves from where he is standing. Six decades on, working physicists still argue over which.

If synthesis is inescapable, and it is, what would an institution that takes this double collapse seriously actually do differently? Three practical disciplines suggest themselves, and none of them require reading sixty-one analyses.

One, fund divergence. Instead of commissioning a single monolithic report, commission two independent teams, or at least require a deliberate alternative analytic path, and ask for something unusual: a divergence report that explicitly maps where and why the two readings part company. The value lies in the boundary conditions of the analysis itself, the precise points where the data are too fragile to support a single interpretation, more than in any average of the two conclusions. A foundation that funds divergence buys clarity about where confusion genuinely lives.

Two, mandate a pre-mortem clause. Before the analysis begins, require the lead analyst to file a short statement, something like this: "If this report ends up supporting conclusion X, here are the defensible analytic choices we rejected that would have pointed toward Y." This leaves the first collapse in place, yet it does something almost as useful: it forces the analyst to leave visible traces of the alternatives they walked past, and it turns an invisible judgment call into a testable assumption. The reader outside the door cannot see the road not taken, but the pre-mortem draws a map of it anyway.

Three, restructure the document itself. Replace the unitary summary with three explicit layers. Layer one is the raw data and the code, for technicians who can rerun the analysis and check their own work. Layer two is the methodological choices, the sensitivity checks, and the rejected alternatives, for peer reviewers and fellow researchers who want to judge the quality of the collapse rather than merely its conclusion. Layer three is the narrative synthesis, for decision-makers who need to act. Affix to layer three an epistemic warning in plain language at the top of the page: "This synthesis is one defensible collapse among many. The alternatives are documented in layer two. If you read only this page, what you are trusting is our judgment about which path to collapse into. The data sits one layer down, untouched by you."

These are not complicated disciplines. They cost nothing but a little extra honesty and a little extra formatting. They also cost the comfortable illusion that a single report delivers a single fact. That illusion, as it turns out, was the most expensive part of the whole operation.

Perhaps the honest posture toward a two-hundred-page report sits closer to Wigner's own than to either the analyst's or the reader's. The person inside the sealed room feels sure the measurement is finished. The person outside feels sure his own reading is the one true collapse. The honest posture keeps neither certainty. It stands at the door, hand on the handle, aware that whatever turns up on the other side will look exactly like the certainty you already carried through the door.

Who was the friend in the last report you trusted. And who, quietly, stood outside the door.




REFERENCES

Lord, C. G., Ross, L., and Lepper, M. R. (1979). Biased Assimilation and Attitude Polarization: The Effects of Prior Theories on Subsequently Considered Evidence. Journal of Personality and Social Psychology, 37(11), 2098-2109.

Proietti, M., Pickston, A., Graffitti, F., Barrow, P., Kundys, D., Branciard, C., Ringbauer, M., and Fedrizzi, A. (2019). Experimental Test of Local Observer Independence. Science Advances, 5(9), eaaw9832.

Silberzahn, R., Uhlmann, E. L., Martin, D. P., et al. (2018). Many Analysts, One Data Set: Making Transparent How Variations in Analytic Choices Affect Results. Advances in Methods and Practices in Psychological Science, 1(3), 337-356.

Wigner, E. P. (1961). Remarks on the Mind-Body Question. In I. J. Good (Ed.), The Scientist Speculates: An Anthology of Partly-Baked Ideas. Heinemann.