Witness reports are not ground truth
A witness report is evidence about an event, not an independently known state of the world. The witness is better understood as a human measurement process with error introduced during perception, interpretation, memory, retrieval, and communication. Most of that error process cannot be reconstructed after the event.
Calling a witness “independent” describes a relationship to other witnesses or parties. It does not establish that the report is accurate. Independence of persons does not imply independence of errors.
A report passes through several lossy stages
Rättssäkerheten i brottmål devotes a chapter to the genesis of a statement from perception through the court’s eventual interpretation. It describes memory as reconstructive rather than reproductive and identification evidence as notoriously unreliable.
Each stage creates a different question:
| Stage | Unobserved failure mode | What a later confident account cannot prove |
|---|---|---|
| Perception | Poor lighting, short exposure, divided attention, stress, intoxication, or expectation | That the relevant detail was encoded accurately |
| Interpretation | Filling fragmentary perception with prior assumptions | That the witness observed rather than inferred the detail |
| Memory | Forgetting, reconstruction, source confusion, and post-event information | That the present memory matches the original experience |
| Retrieval | Leading questions, repeated interviews, and interviewer feedback | That the answer is independent of the elicitation method |
| Communication | Social pressure, desired coherence, and adaptation to the listener | That sincerity entails accuracy |
| Adjudication | Demeanor, narrative fluency, and the court’s own expectations | That the court can identify accuracy by observing confidence |
An honest witness can therefore be confidently wrong. The error need not be fabrication, and cross-examination cannot recreate the original viewing conditions or restore an uncontaminated memory.
Apparent corroboration can share one source
Two accounts add more information than one only to the extent that their relevant errors are independent. They may instead share:
- the same poor vantage point or ambiguous event
- conversation before formal interviews
- media descriptions or rumors
- the same suggestive questions and investigator theory
- exposure to the same suspect or photograph
- feedback that rewards agreement
The Gabbert, Memon, and Allan experiment made this dependency visible. Pairs watched different versions of an event and then discussed it. In that experimental condition, 71 percent later reported at least one detail that they had received from the co-witness rather than observed. That figure is not a field prevalence estimate. It demonstrates that two nominal witnesses can become one contaminated information chain while each report still feels like memory.
Repeated retellings from one witness are even less independent. They are not repeated measurements of the event. They are measurements of a memory already changed by prior retrieval and subsequent information.
The scientific position is conditional rather than categorical
It would overcorrect to call every eyewitness report worthless. The scientific hierarchy used to evaluate general causal claims also does not map perfectly onto evidence about one historical event. A witness can possess direct information unavailable from any surviving trace.
The important contrary finding concerns a narrow procedure. Wixted and Wells found a strong relationship between confidence and identification accuracy when an adult’s first identification is obtained under pristine conditions: a fair lineup, double-blind administration, an instruction that the perpetrator may be absent, no contamination, and an immediate confidence statement. They also found that unfair procedures can compromise even a high-confidence identification.
That result does not validate later courtroom confidence, free-form narrative recall, or an account collected after discussion and repeated interviewing. It shows that reliability belongs to a specified collection procedure, not to the social category “witness.”
What would make corroboration stronger
Corroboration becomes more probative when investigators can establish that:
- each account was recorded promptly and in the witness’s own words
- witnesses had no opportunity to exchange information
- interviewers did not know or signal the preferred answer
- the original account and confidence were preserved before feedback
- the accounts converge on details unlikely to arise from a shared source
- material discrepancies are preserved rather than harmonized
- physical, digital, or contemporaneous documentary traces have separately tested provenance and support the same inference
No item supplies ground truth by itself. The objective is to combine evidence streams whose failure modes are visible and genuinely different.
The 2014 National Academies consensus report similarly recommends blind procedures, standard instructions, an immediate confidence record, and video recording of the identification process. Its official download flow rejected automated preservation in this session, so the report remains a remote source rather than a local artifact.