In 1974, Elizabeth Loftus demonstrated that changing a single word in a question caused 32% of subjects to remember broken glass that was never there.
Not the facts. Not the allegation. One word.
This has been replicated for fifty years. The aggregate false memory rate across her research programme is approximately 22%. One in five witnesses, subjected to the ordinary conditions of interrogation and cross-examination, will carry a fabricated memory into court and deliver it under oath with complete sincerity.
They are not lying. They are remembering what the questions built for them.
Every interrogation conducted after 1974 occurred after we knew this. Every conviction that rested on witness testimony occurred after we knew this.
We continued.
In 2006, Bond and DePaulo published the definitive account of human deception detection. Two hundred and six studies. Twenty-four thousand four hundred and eighty-three judges — police officers, customs officials, trained investigators of every kind.
Overall accuracy: 54.1%. Chance is 50%.
For lie detection specifically: 47%. Below chance.
The more experienced the investigator, the more confident their assessments. The more confident, the more wrong. Training does not improve accuracy. It improves certainty while accuracy stays essentially random.
Every court has a person whose job is to assess credibility. Every jury has twelve. Every trial hinges on it. We have known since 2006, with a sample of twenty-four thousand, that this capacity does not exist at above-chance levels in any trained human population.
We continued.
In 2006, the Global Deception Research Team surveyed eleven thousand people across seventy-five countries. They asked: what does a liar look like?
Sixty-four percent of people worldwide believe liars avoid eye contact. The actual effect size of gaze aversion as a deception indicator: d = 0.05. Functionally zero.
When we matched the full list of commonly held beliefs against the actual empirical data on deception cues, ninety-one point three percent of the cues people use to assess guilt are either unrelated to deception or move in the opposite direction.
The framework is not merely inaccurate. It is inverted. The system has been using an inverted compass for the entirety of its existence, and the people using it have grown more confident the further wrong they have traveled.
Every juror who has ever assessed credibility has done so using a belief system that is, by the weight of the available evidence, backwards.
We continued.
When DNA testing became available it created something that had not existed before: certainty.
We began going back. Between 12% and 30% of people exonerated by DNA evidence had confessed. Detailed, signed confessions. Confessions containing specific information that investigators told juries only a guilty person could know — because investigators had provided that information during the interrogation.
Women who confessed to burying babies. Men who described in precise detail crimes they did not commit. The descriptions were accurate because the interrogation was the source of the descriptions. The confession was not a window into what happened. It was a mirror of what the interrogation constructed.
We know this now because of DNA. But DNA only resolves cases where biological material was present, preserved, and tested. For every case DNA has resolved, there are hundreds it cannot touch. The false confession rate we have measured is a floor, not a ceiling. It measures only the cases where exoneration was possible.
Here is what must be understood: a confession case is not a separate category from a heuristics case. It is not outside this indictment. The confession was produced inside the same room, by the same methods, under the same conditions — conditions we know elevate a person's suggestibility between 80% and 120% above their normal baseline before the first question is asked. The confession is not independent evidence. It is the output of the broken process. Treating it as a separate and more reliable category of conviction is like accepting a document as valid because it was correctly filed.
The process was broken. The output is suspect. All of it.
We continued.
Remove the justification narrative. Remove the word "criminal." Remove "convicted." Remove "sentenced."
What remains is this: a human being, in a cage, placed there by a state, against their will, unable to leave.
Their movement is controlled. Their time is not their own. Their labour is extracted or their existence is warehoused. They are fed at the discretion of those who hold them. They are released at the discretion of those who hold them. They are subject to whatever occurs within the walls, at the discretion of those who hold them.
This is the structure of incarceration. This is also the structure of a concentration camp.
The difference is not structural. The difference is the story told to authorise it.
We do not need to argue about degrees of suffering, about food quantities, about the relative weight of labour. Those are arguments about the severity of what is done to the person inside. They are not arguments about what authorises the cage.
What authorises the cage — in both cases — is a legal and institutional process that determined the person belonged inside it.
We now know that the process produces false outcomes at a documented and non-trivial rate. We know the tools it uses are inverted. We know the confessions it extracts are unreliable. We know the witnesses it relies on carry memories shaped by the questions.
Which means: inside those cages, right now, are people placed there by a process we know is broken.
Some of them committed crimes. A percentage of people in concentration camps had committed crimes too. It does not matter. The question was never exclusively about individual guilt. It is about whether the mechanism that authorised the cage was legitimate. Whether the process that determined who goes inside it can be trusted.
We know it cannot.
We continued anyway.
Each finding, taken alone, is uncomfortable.
Stacked, they constitute something that has a name.
We know memory is rewritten by questions. We know trained investigators detect lies at chance level and grow more confident as they grow more wrong. We know ninety-one percent of the cues used to assess guilt point in the wrong direction. We know that when DNA forced us to check our confessions, between a fifth and a third of them were false. We know that before the first question is asked, the person across the table is operating at 80–120% above their normal suggestibility baseline.
We know all of this.
And every day, in every jurisdiction, people are placed in cages on the basis of a process built from these tools. It happened yesterday. It is happening today. It will happen tomorrow.
The question is not whether harm has occurred. The question is not whether the evidence was available to those with the power to act.
The question — the only question that will matter when we look back at this — is what we call people who had authority, had evidence, and chose continuation.
We have answered that question before.
We are placing this on record so that when the time comes to answer it again, there is no defence of ignorance available.
The record exists. It exists now.
Evidence cited: Bond & DePaulo (2006); DePaulo et al. (2003); Global Deception Research Team (2006); Loftus & Palmer (1974); Loftus (2005); Garrett (2011); National Registry of Exonerations (2023); Rizzelli et al. (2021); Gudjonsson & Clark (1986); Harrison et al. (2014); original analyses: Belief-Reality Inversion Matrix, Linguistic Study of Trial Testimony.
This document is a statement of record.