How to Manage Dreams and Insights on Dreaming
The Combination Test Hypothesis — On the Nightly Sorting Work of the Brain
Prologue
This essay began from a single observation about my own dreams. Some dreams vanish the moment I wake, no matter how hard I try to hold them. Others stay for days without any effort. Some I cannot even remember having had. No pattern emerges between the dreams that stay and the dreams that disappear. It looks random.
This randomness itself was strange. Memories made during waking hours do not evaporate this quickly. I remember what I had for lunch yesterday. I remember a scene from a movie I watched a week ago. But a dream I had at dawn this morning is gone before the afternoon. Same brain. Why the difference. This was the starting point.
This essay addresses two things at once. First, a hypothesis about why this phenomenon occurs. Second, what to do about dreams if the hypothesis holds. Insight and management do not separate here. That is the stance of this work.
Abstract
The hypothesis proposed in this essay is simple. Dreams are a combination test run by the brain each night, and because their results do not need to be preserved, they evaporate. The difference between dreams that remain and dreams that disappear lies not in content but in whether the dreamer consciously captured them after waking. The volatility of dreams is not a flaw but a design. This perspective fills in the gaps left by the major existing hypotheses — Hobson's activation-synthesis, Revonsuo's threat simulation, Zadra's NEXTUP, Hoel's overfitted brain — in a single frame. Yang et al.'s 2024 Science paper on hippocampal sharp-wave ripple tagging directly measured the 'automatic selection by intensity' that forms the core of this hypothesis at the neurophysiological level, and Siclari's 2017 Nature Neuroscience paper on the posterior hot zone narrowed down the neural conditions under which dreams become accessible to consciousness. The four axes of management that follow from the hypothesis are emotion tag management, dream journaling, natural awakening, and pre-sleep emotion management. These are not a prescription that replaces clinical treatment but a way of embedding the principles of dream management into the structure of daily life.
1. The Landscape of Existing Hypotheses
Scientific explanations of dreaming are split into several branches. Each is partial, and none explains the whole.
The activation-synthesis hypothesis, proposed by Hobson and McCarley in 1977, views dreams as the cortex's post hoc interpretation of random brainstem activity. It is strong at explaining the bizarreness of dreams but fails to explain why dream content shows certain biases. Revonsuo's threat simulation theory, published in Behavioral and Brain Sciences in 2000, views dreams as training for threat rehearsal. Its central evidence is the observation that 66.4% of 592 dream reports analyzed by his team contained at least one threatening event. However, this hypothesis fails to account for the vast majority of ordinary dreams that have nothing to do with threat. Zadra and Stickgold's NEXTUP model, introduced in 2021, views dreams as 'network exploration for what comes next'. Elegant, but still at the hypothesis stage. Hoel's overfitted brain hypothesis, published in Patterns in 2021, views dreams as noise that prevents neural overfitting. It has drawn interest from AI researchers, but experimental verification is scarce.
And the most mainstream view, the memory consolidation hypothesis. A 2024 review by Bloxham and Horton in Consciousness and Cognition summarizes where this current stands. Evidence for a link between memory reactivation during sleep and the dream experience has accumulated, but the causal relationship and specific mechanism remain in debate.
One question all these hypotheses fail to answer: why do most dreams disappear. Each speaks to the 'function' of dreams, but no theory explains the 'volatility' of dreams itself. If there is a function, why do the results not remain. And a more practical question — if dreams are not simply lost but are lost by something, can we not influence that something and thereby manage dreams. This is the gap this essay enters.
2. The Material of Dreams
A place one has never been to cannot appear in a dream. This is the first observation. The material of dreams is entirely made of what one has seen, heard, and felt at some point in the past. Fragments stored deep in the brain are pulled out.
Combinations, however, are free. Things that cannot possibly exist together in reality can exist together in dreams. The dead sit at the same table with the living. A room inside a building one has never entered is the room of the house one grew up in. The material is real, the combinations are arbitrary. This is half of why dreams do not make sense.
The other half is narrativization. That is addressed later.
3. The Finiteness of Brain Capacity
The amount of information the brain can process in a day is finite. The long-term storage capacity of the brain is close to unlimited — estimates place it around 2.5 petabytes. But working memory is capped at four to seven chunks at any moment, and roughly 50% of newly encoded information is lost within an hour, 70% within 24 hours. Long-term storage is nearly unlimited, but the gate to it is narrow.
Selection happens at this narrow gate. When a cup is full and new water is poured in, only small amounts leak in by chance. The rest does not fit. So sorting is needed during sleep. The brain decides what to keep and what to discard.
The criterion for the decision is 'intensity'. The same information, when seen, heard, and written down, becomes three times more intense as a memory. Number of channels, repetition, emotion, surprise — these all raise intensity. Intense things stay, weak things are discarded. This selection happens automatically, without conscious involvement.
That this is not mere conjecture has recently been demonstrated. In a 2024 paper published in Science, Yang and colleagues directly observed that sharp-wave ripples (SPW-Rs) occurring in the hippocampus of mice during waking 'tag' specific memories, and that these tagged memories are preferentially replayed during subsequent NREM sleep. The brain actually marks certain memories and strengthens them at night. The intuition of automatic selection by intensity has been measured at the neurophysiological level.
4. Dreams Are a Combination Test
Apart from sorting, the brain also runs a separate operation: shuffling memory fragments and running them through random combinations. Strictly speaking it is not random but 'intensity-weighted random'. The more intense a memory, the higher its probability of being drawn. But which fragments end up combining with which cannot be predicted.
Why such a test is needed cannot be pinned down. It is a domain for experiment to settle. It may be optimization for the brain's efficiency. It may be a check on whether new memories can integrate with existing ones without conflict. It may be something else. What can be said is only that the test runs.
What matters is that the results of this test have no reason to be preserved. It is enough that the sorting is confirmed. If every result of the test were stored, the brain would not last a single day. So the results evaporate. The volatility of dreams is not a flaw but a design. This is the core of the hypothesis.
From this view, the existing hypotheses read differently. Activation-synthesis's 'random brainstem signals' points to the randomness of the combination test. NEXTUP's 'network exploration' is one aspect of the combination test itself. Overfitted brain's 'noise-based generalization' is one possible answer to why the test must be random. Each touched a different part of the same elephant. The combination test hypothesis fastens those parts onto a single body.
5. The Dream Itself and the Memory of the Dream Are Different
A crucial distinction enters here. What we call a 'dream' is, for the most part, actually the 'memory of a dream'. The two are not the same.
Melanie Rosen's 'narrative fabrication thesis', published in Frontiers in Psychology in 2013, formalized this distinction in academic terms. A dream report after waking is not an accurate reconstruction of the dream itself. Strange content is rationalized by the waking brain. Gaps in memory are filled in. Fragments are sewn into narrative structure. An experiment by Stickgold and colleagues in 1994 produced a striking result — judges could not distinguish between real dreams and 'artificial dreams' constructed by splicing together pieces of different people's dream reports. That is how fragmented and discontinuous dreams themselves are. When we feel that 'there was a story in the dream,' it is not because the dream was a story, but because the waking brain automatically made one.
This distinction is decisive for the combination test hypothesis. The dream itself is something that briefly surfaces near consciousness and then disappears. But the moment one wakes up and begins to 'remember' and 'think about' the dream, from that moment it is no longer a dream but a waking thought. That is why it remains in memory. It starts to follow the same rules as ordinary memory — intensity, repetition, conscious processing.
If a useful idea comes from a dream, write it down the moment you wake. The act of writing pulls it out of the dream's fate. It is transferred into waking memory, acquires additional intensity, and becomes a candidate for long-term storage. A 2010 study in Consciousness and Cognition confirmed that people who kept a dream journal consistently for six weeks showed significantly higher dream recall rates than controls. Robert Stickgold of Harvard put it precisely: dream recall improves not because dreams become more vivid, but because the brain learns that this material is worth preserving.
That is, the only way to save a dream is to consciously capture it after waking. A dream not captured follows its original fate and disappears. This fact becomes the core basis for the management proposals that follow.
6. Narrativization Is a Habit of the Brain
The first-person perspective of dreams and the sense that they have some kind of storyline follow from the same principle. During waking hours, the brain constantly frames experience as "I see, I act, something happens." This frame runs automatically. So the same frame is automatically applied to the random fragments that surface near consciousness during sleep.
The first-person perspective is not a separate function of dreaming. It is simply the default format used during waking that stays on during sleep. Narrativization also continues to operate after waking. Fragmented raw material is stitched again into a coherent story. This is what the academic term 'confabulation' refers to. What we perceive as a dream is the product of these two layers of narrativization.
The reason this matters becomes apparent in the following phenomenon. When someone who has experienced trauma reports dreaming the same nightmare every night, that too may be a product of narrativization. A perfectly identical dream is neurophysiologically impossible. Similar emotional tones and themes recur, but the concrete fragments of each dream differ. The brain's habit of narrativization, however, bundles these similar dreams into a single 'same dream' category. So the person experiences them as the repetition of one nightmare.
7. The Mechanism of Nightmares
The same event is not stored passively. Depending on how it is processed, the direction and magnitude of its emotional tag change. An event stored after a negative interpretation piles up in the pool as 'bad material'. An event passed through neutral or positive reinterpretation is stored in that direction, or its intensity is diluted. This active tag manipulation directly affects the combination test that runs every night.
When bad material is scarce in a person's pool, the probability that bad material is drawn in the combination test is low. This is a matter of simple probability. If a lottery shop carries few defective tickets, the probability of drawing a defective ticket at that shop is low.
People who dream nightmares repeatedly are in the opposite condition. When bad memories are abundant, nightmares are drawn more often by probability. And because nightmares are intense, they tend to remain in memory even after waking. A nightmare that has become an ordinary memory re-enters the memory pool the next day. Then it is drawn again in the next night's combination test. A vicious loop forms.
This loop has been empirically measured. A 2018 study by Sikka and colleagues in Scientific Reports showed that the level of peace of mind and anxiety in the waking state directly predicts the affective content of dreams. A 2022 follow-up by Sikka and colleagues in Frontiers in Behavioral Neuroscience, which tracked dreams and next-day affect over five days, confirmed that the more negative a person's dreams, the more negative their affect level the next day. In the same year, Barbeau and colleagues reported that individuals who had recently experienced troubling life events showed higher levels of negative emotion in their dreams than those who had not. The connection between waking emotion, dream emotion, and the next day's emotion has been directly measured. Dreams function as both effect and cause of emotion. The repetition of nightmares is not a pathological mechanism but a statistical inevitability.
Post-traumatic nightmares are explained by the same principle. Traumatic memories are extremely intense from the start. A traumatic dream, once experienced, is also highly intense. Both are preferentially drawn from the material pool. Yesterday's dream becomes today's memory and re-enters tonight's pool. The layered loop deepens. The repeated drawing of trauma-related combinations is not a strange phenomenon but a natural consequence.
8. The Dual Function of REM
Why did all mammals preserve REM sleep across hundreds of millions of years of evolution. The answer proposed here is a combination of two functions.
First, the sentinel function. This is the sentinel hypothesis Frederic Snyder proposed in the American Journal of Psychiatry in 1966. Ancestral environments were harsh. Humans had to sleep in trees, on grounds with predators nearby. If only deep NREM continued, threat detection would be impossible. REM is a period when the brain periodically rises to a state close to arousal. It is the brain's way of securing a 'wake-able state' at regular intervals. The primary reason evolution preserved REM is this survival function.
Second, the buffer function. This is an additional hypothesis that has not, to my knowledge, been directly proposed in the academic literature. Asleep and awake are two very different brain settings. The load in the awake state is much larger. Crossing suddenly strains the system. This is why, when one is jolted up by an alarm, the head aches for the rest of the day. This phenomenon, known as 'sleep inertia', has been studied for over two decades and is still not fully understood. Viewing it as the consequence of a skipped buffer explains it intuitively.
REM periods lengthen as the sleep cycles repeat. The first REM lasts about five minutes. As dawn approaches, REM grows to thirty minutes, forty-five minutes, sometimes an hour. In an eight-hour night there are four to five REM periods, with the last being the longest. This pattern matches 'buffer preparation'. As waking approaches, the brain is gradually brought closer to an awake state.
Dreams, apart from these two REM functions, are the combination test running in the arousal-like environment that REM creates. REM is the function, and the dream is the separate operation that happens in that environment. The two overlap temporally but are functionally independent.
9. Answering the First Question
Why do some dreams remain in memory for days while others disappear the moment one wakes.
The answer divides into three categories.
Dreams with no memory at all. The combination test ran but did not register in consciousness. This is the case for most dreams. It is normal. That dreams are not captured is the default state.
Dreams briefly remembered then lost. They were lightly registered in consciousness but were not consciously captured after waking. They follow the combination test's original fate and disappear. Even during the window of recall, they fade at roughly 50% per hour.
Dreams that remain for days. Either they were strongly registered in consciousness, or after waking they were consciously captured and transferred to ordinary memory. Once they become ordinary memory, they follow the same rules as other memories — intensity, repetition, emotion. If the intensity is high enough, they stay longer.
The reason no pattern seemed to exist is that these three states were mixed together. And since the combination itself is intensity-weighted random, hardly any surface regularity can be detected in the selection of its results either. That it 'looked random' was indeed random.
10. Practice — Embedding Dream Management into the Structure of Life
From this hypothesis follows a set of practical proposals. Not proposals to 'save' dreams, but proposals to manage one's memory and emotional system better through dreams.
The first axis, emotion tag management. The practice of not attaching a strong negative tag to events that arrive during the day. Stress itself cannot be eliminated, but the narrative 'my life is ruined' attached to stress can be avoided. Try to understand the event, try to look at it from another angle, let it go if neither works. This slows the rate at which bad material piles up in the pool. The result is a lower probability of nightmares and a shallower vicious loop. This axis manages the emotional bias of the material pool itself.
The second axis, dream journaling. The habit of writing down even a fragment of a dream the moment one wakes. The moment it is captured, the dream is transferred to ordinary memory. This has particular meaning for those whose work involves creativity or insight. Stickgold's work reports meaningful improvements in recall within six weeks. This axis manages preservation.
The third axis, natural awakening. Where possible, waking without an alarm. Letting REM's buffer function complete its work. Where this is not possible in daily life, using a gradual awakening alarm — rising volume, simulated sunlight — is better than an abrupt sound. This axis manages the 'transition process' of brain states.
The fourth axis, pre-sleep emotion management. Multiple studies have confirmed a continuity in which waking emotional state predicts the affective content of dreams, and dream emotion carries into the next day's emotion. What one sees, hears, and reads immediately before sleep becomes the material for that night's combination test. Consuming bad news, stimulating media, or worries just before sleep is equivalent to injecting that content directly into the material pool. This axis manages the composition of that night's material pool.
When the four axes operate together, the effect is greatest. During the day the first axis manages tags. Before sleep the fourth axis manages input. Natural awakening completes the third axis. The moment of waking, the second axis preserves. Dream management is woven tightly into the structure of the day.
These four are not a treatment. They do not replace standard medical care. For serious sleep disorders or post-traumatic stress, professional help comes first. These principles operate on the layer above that. The work is embedding the principles of dream management into the structure of a life.
11. Closing
Most dreams disappear. That is normal. The volatility of dreams is not a flaw but a design. Because the results of the combination test the brain runs each night do not all need to be preserved, they scatter.
This is the combination test hypothesis. Dreams are the brain's nightly sorting operation, and most of its results disappear. The strangeness is not that they disappear. The strangeness is that any of them remain. A dream that remains is a dream that was consciously captured after waking, or one whose emotion was unusually intense, or one whose narrativization was unusually smooth. In other words, a dream that met the conditions to become an ordinary memory.
The first question was why some dreams disappear and others remain. The answer this essay reaches is simple. Disappearing is the default. Remaining is the exception. What makes the exception is not the dream itself but the 'I' who wakes afterward.
And that 'I' can make the exception deliberately. Manage the material pool. Capture the dream right after waking. Complete the buffer. Select what goes in before sleep. These four are the method of managing dreams. Dreams are not an uncontrollable mystery but an object that can be understood and handled.
For comments or discussion: wonbrand.co.kr
References
Barbeau, K., Lafrenière, A., Ben Massaoud, H., Campbell, E., & De Koninck, J. (2022). Dissociated effects of age and recent troubling experiences on nightmares, threats and negative emotions in dreams. Frontiers in Psychiatry, 13, 770380.
Bloxham, A., & Horton, C. L. (2024). Enhancing and advancing the understanding and study of dreaming and memory consolidation. Consciousness and Cognition, 120, 103693.
Crick, F., & Mitchison, G. (1983). The function of dream sleep. Nature, 304(5922), 111-114.
Hobson, J. A., & McCarley, R. W. (1977). The brain as a dream state generator: an activation-synthesis hypothesis of the dream process. American Journal of Psychiatry, 134(12), 1335-1348.
Hoel, E. (2021). The overfitted brain: Dreams evolved to assist generalization. Patterns, 2(5), 100244.
Revonsuo, A. (2000). The reinterpretation of dreams: An evolutionary hypothesis of the function of dreaming. Behavioral and Brain Sciences, 23(6), 877-901.
Rosen, M. G. (2013). What I make up when I wake up: anti-experience views and narrative fabrication of dreams. Frontiers in Psychology, 4, 514.
Siclari, F., Baird, B., Perogamvros, L., Bernardi, G., LaRocque, J. J., Riedner, B., Boly, M., Postle, B. R., & Tononi, G. (2017). The neural correlates of dreaming. Nature Neuroscience, 20(6), 872-878.
Sikka, P., Pesonen, H., & Revonsuo, A. (2018). Peace of mind and anxiety in the waking state are related to the affective content of dreams. Scientific Reports, 8(1), 12762.
Sikka, P., Engelbrektsson, H., Zhang, J., & Gross, J. J. (2022). Negative dream affect is associated with next-day affect level, but not with affect reactivity or affect regulation. Frontiers in Behavioral Neuroscience, 16, 981289.
Snyder, F. (1966). Toward an evolutionary theory of dreaming. American Journal of Psychiatry, 123(2), 121-136.
Stickgold, R., Rittenhouse, C. D., & Hobson, J. A. (1994). Dream splicing: a new technique for assessing thematic coherence in subjective reports of mental activity. Consciousness and Cognition, 3(1), 114-128.
Tononi, G., & Cirelli, C. (2014). Sleep and the price of plasticity. Neuron, 81(1), 12-34.
Yang, W., Sun, C., Huszár, R., Hainmueller, T., Kiselev, K., & Buzsáki, G. (2024). Selection of experience for memory by hippocampal sharp wave ripples. Science, 383(6690), 1478-1483.
Zadra, A., & Stickgold, R. (2021). When Brains Dream: Exploring the Science and Mystery of Sleep. W. W. Norton & Company.
An Seungwon / Wonbrand / https://wonbrand.co.kr