Research question
The starting question was: how can a sleeping person experience vivid images, places, sounds, and events without currently observing them? The review then asked where dream content comes from, how memory contributes, and whether internally generated experience might relate to prediction and model updating.
Current narrower question: Is there direct experimental evidence that an internally represented prediction is violated during REM sleep and that neural activity encodes the resulting discrepancy? The review has not established that sequence.
How the project is being studied
The working claim tracker records individual claims, source details, evidence and reasoning, competing interpretations, status, confidence, and research notes. The review checks primary studies where possible and keeps reported observations distinct from the authors’ models and this project's own inferences.
Human dream reports, memory experiments, recordings from developing rats, lesions, and awake prediction-error tasks answer different questions. A result in one species, developmental stage, or sleep state cannot be silently transferred to another.
How the inquiry progressed
- Dream experience and imagery. The review began with REM-associated dreaming, visual processing, and the question of how an apparently coherent experience can occur with reduced external input. Activity in visual areas is relevant but does not by itself explain dream content or subjective experience.
- Memory and internally generated content. Studies of dream reports and memory reactivation suggested that waking experiences can contribute to dreams, often as altered fragments rather than exact replay. This does not prove that a specific dream updates a memory by a prediction-error mechanism.
- Prediction and REM sensory processing. The review examined predictive-processing proposals alongside studies of sensory responses during sleep. A theoretical model, a predictable neural sequence, or a response to a stimulus is not automatically an encoded prediction–outcome error.
- Developmental motor circuitry. Infant-rat active-sleep twitches, corollary discharge, and reafferent sensory input provide a more concrete route for studying how sensorimotor circuits develop. The 2024 cerebellar study supports coordinated twitch-related input but proposes, rather than directly demonstrates, the formation of an internal model from those signals.
- A causal developmental test. A 2026 study compared climbing-fiber lesions at P12 and P19 before testing cerebellar-dependent activity at P20. The earlier lesion disrupted the later assay more severely. That strengthens the developmental internal-model account, while the lesion does not isolate only twitch-related discharge or measure a REM prediction-error signal.
What the evidence currently supports
Developing-rat recordings show strong responses to active-sleep twitches in cerebellar circuits. The later lesion comparison supports a causal role for early inferior-olive climbing-fiber input in the normal development of a cerebellar-dependent internal-model assay. These are sensorimotor results in developing rats; they are not measurements of human dream imagery.
The convergence of twitch-related corollary discharge and sensory reafference may help build later predictive circuitry. The precise computation and what an “internal model” represents are not directly decoded from the cited developmental experiments.
The review has not identified an experiment that independently establishes a prediction during REM, deliberately changes the expected sensory outcome, and measures a neural signal for the mismatch. Awake human cerebellar experiments provide a useful comparison because they vary expected and actual sensory events, including unexpected presence and omission, but their awake results cannot establish the REM mechanism.
Where the review resumes
- Audit the awake human cerebellar mismatch experiment's design, physiological controls, and Results as a benchmark for what a direct error test actually measures.
- Search for REM experiments that separate a represented prediction from the sensory outcome and compare matched, unexpected-presence, and unexpected-omission conditions.
- Keep the developmental animal evidence distinct from adult REM and from subjective human dreams; revisit competing explanations for neural responses.
- Return to the broader memory and dream-content question only after the intermediate sensorimotor mechanism is evaluated. A connection to dream experience or memory reconsolidation remains unproven.
Selected sources in the evidence trail
These are representative sources from the working review, not its complete bibliography. A citation documents what was examined; it does not endorse every inference in that paper.
- Nir & Tononi (2010), “Dreaming and the brain: from phenomenology to neurophysiology” — dream experience and neurophysiology review.
- Deperrois et al. (2022), “Learning cortical representations through perturbed and adversarial dreaming” — computational model, not a direct test of dreaming humans.
- Tiriac, Del Rio-Bermudez & Blumberg (2014), “Self-generated movements with ‘unexpected’ sensory consequences” — developing-rat active-sleep sensorimotor study.
- Richardson, Sokoloff & Blumberg (2024), “Developmentally Unique Cerebellar Processing Prioritizes Self- over Other-Generated Movements” — developing cerebellar response architecture.
- Richardson, Sokoloff & Blumberg (2026), “A Transient Feature of the Inferior Olive Supports the Development of Cerebellar Internal Models” — developmental lesion comparison.
- Schlerf, Ivry & Diedrichsen (2012), “Encoding of Sensory Prediction Errors in the Human Cerebellum” — awake human mismatch comparison; the source under closer audit next.