The standard account of grief is backwards. Grief is described as suffering about the past: the loss has happened, and grief is what you feel about it. This framing positions grief as the problem — the thing that makes loss worse, the experience you need to get through before you can return to living. The prescription that follows is: process the grief so you can move on.
The description is not wrong. The direction is wrong. Grief is not about the past. Grief is about the future. It is the process by which the future is freed from the past — the specific mechanism by which a mind calibrated to include a lost person, a lost relationship, a lost version of a life actually updates its model to include the irreversible fact of the loss. Grief is not what makes loss hard. Grief is what makes loss survivable. And the thing that makes loss permanently hard is not grief but its absence.
Simple Picture
A woman’s mother died fourteen months ago. She is functional: back at work, present with her children, apparently adjusted. But something is happening in her that she cannot quite locate. She finds herself dialing her mother’s number before catching herself. She hears a song her mother loved and feels a disorienting blankness. She makes a decision without thinking, realizes she was thinking “I’ll tell Mom about this,” and only then encounters the fact that she cannot. The mother is gone. Her brain is not. Her brain is still running predictions calibrated to a world that includes her mother. Every gap between prediction and reality is a small, private collision.
Fourteen months later, these collisions should be fewer. They are not notably fewer. They are merely suppressed. She has learned to manage her responses before they complete. She has learned to interrupt the reaching before it becomes reaching. She has mistaken this management for adjustment. She has not grieved. She has developed a sophisticated suppression practice. The model is still running, untouched, generating predictions about a world that no longer exists. The future is still encrypted with her mother’s presence. The grief that would have updated the model has not run. The collisions continue, internally, below the level where she registers them as collisions.
The Predictive Processing Mechanism
The brain is a prediction machine. It maintains a hierarchical model of the world and generates top-down expectations about what the sensory stream will deliver. When the prediction matches the input, nothing happens — the prediction was the perception. When there is a mismatch, a prediction error propagates upward, and the model updates.
Loss creates a specific class of prediction errors: every time the model predicts the presence of the lost person and the sensory stream delivers absence, a prediction error fires. Reaching for the phone to call someone who is dead is a prediction error. Seeing something you want to tell someone who is no longer there is a prediction error. Feeling lonely in the way that only their specific presence would resolve is a prediction error. Each error is small. Each asks the model to update.
Grief is the process of accumulating enough of these prediction errors that the model actually updates — revises itself, at the level of its priors, to incorporate the irreversible reality of the loss. What we call the experience of grief — the waves of pain, the crying, the surfacing of memories — is the felt texture of prediction errors propagating through the model and doing the structural work of recalibration. The pain is not incidental to grief. The pain is the update. It is the model changing shape.
Incomplete grief is the suppression of these prediction errors before they can accumulate into a model update. Every time the management reflex interrupts the reaching — every time the composed exterior shuts down the wave before it crests — the prediction error is absorbed without propagating. The model does not update. The priors that include the lost person remain intact. The future continues to be generated by a prediction model that still expects them.
Non-Ergodicity of Loss
Loss is non-ergodic in the deepest sense: it is irreversible. The ensemble average across people who have experienced similar losses produces information about how people function after loss — but it cannot produce information about what this particular loss means in this particular time-series. You are not averaging across possible lives. You are traversing one life, and the one thing the ensemble average cannot tell you is what it costs to carry an unupdated model forward through years.
The cost is specific. The model running predictions that include the lost person cannot correctly price future decisions involving anything the lost person touched. Relationships, vocations, geographic locations, ways of moving through the world — all of them carry predictions still calibrated to a presence that is no longer there. The person who has not grieved their father does not make decisions about authority figures as a free adult. They make decisions as the child still predicting their father’s response, still organized around earning or defying or avoiding the approval that is no longer available to earn, defy, or avoid. The model keeps generating the old predictions. The future keeps being shaped by the old calibrations. The past is not behind them. It is inside their model, generating the present.
Grief is the mechanism by which the individual catches up to the irreversibility of their own time-series. It is how the model is updated to include the fact that the future does not contain what the past did. Without the update, the future is the past in disguise — the same relational patterns, the same avoidances, the same reaching-for-what-is-no-longer-there, now expressed through new people in new situations because the old model does not distinguish.
The Locally-Optimal Suppression
Why does incomplete grief persist? For the same reason all locally optimal strategies persist: avoiding grief is locally optimal. Each suppression event is rewarded. Crying at work has costs. Allowing the wave to crest at dinner has costs. Letting the reaching complete — actually picking up the phone, encountering the absence, sitting with the collision — is more painful in the short term than catching it before it arrives. The suppression is a precision-engineered strategy for avoiding acute pain, and it works exactly as designed in the short term.
The long-term cost is the frozen model. Every successful suppression is a prediction error that did not propagate. The model does not update. The priors remain. The strategies that would have been unnecessary if grief had been allowed to run — the careful emotional management, the organized avoidance of things that would trigger the reaching, the progressive narrowing of what can be felt safely — accumulate as the infrastructure of a life organized around protecting the gap between prediction and reality from becoming visible. Lewis’s casket is the emotional architecture of this infrastructure: the heart locked in for safety, not broken, slowly becoming “unbreakable, impenetrable, irredeemable” — not from too much loss but from the optimization of too-successful defense.
The wound is largely made of this frozen material. The ungrieved losses of early life — the father who was absent in the specific way that mattered, the mother who was present but not really there, the self that was never seen and learned not to need seeing — are not past events. They are current predictions, still generating expectations that the present keeps failing to fulfill, still organizing the emotional field around what cannot be found because it was never there. The wound hurts not because the past was painful but because the model still expects what the past failed to provide. Grief is what would update those expectations. The wound is where grief did not run.
The Annealing Event
Neural annealing is the brain’s mechanism for releasing structural stress: entering high-energy states that allow rigid neural configurations to dissolve and reorganize. Grief is the primary annealing event. The specific quality of the grief state — the vulnerability, the openness, the dissolution of the ordinary management of experience — is the brain entering the high-energy configuration from which reorganization is possible.
This is why grief cannot be scheduled and cannot be compressed. You cannot grieve on the weekend and return to normal operation on Monday because the annealing process is not complete until the model has actually updated — and the model updates on its own schedule, not yours. The corporate return-to-work timeline, the social expectation that grief should wrap up within some culturally specified period, the personal management reflex that interrupts the waves before they complete — all of these treat grief as an experience to be processed and filed rather than a structural event the brain needs to complete. An annealing cycle that is interrupted before it completes is not a partial annealing. It is an interrupted structural event that leaves the configuration partially dissolved, partially reformed, partially still rigid — less coherent than when it started, unable to complete the reorganization that would have produced something stable.
What Peter Levine describes as the “incomplete discharge of the survival response” is this same mechanism in the somatic register. The animal that shakes after a near-death experience is completing the annealing cycle — discharging the activation that the threat generated, allowing the nervous system to return to baseline from a fully mobilized state. The animal that cannot complete the discharge — because the human inhibition against uncontrolled shaking is stronger than the discharge reflex — carries the incomplete mobilization forward as a nervous system that never fully regulated. The grief that cannot complete is the human version of the frozen animal: the activation generated by loss, arrested mid-discharge, carried forward as ambient structural stress.
The Comprehension Trap
There is a specific failure mode that feels like grieving and is not. Comprehension without dissolution applies to grief in its most precise form. The person who has thoroughly understood their grief — who can articulate what was lost, why the loss matters, how it connects to earlier losses, what it says about their attachment patterns — has a sophisticated model of their grief. They have not grieved.
Understanding grief is a meta-level operation: it adds a description of the grief-process to the prediction model without allowing the prediction errors of actual grief to propagate through the model’s priors. The person who understands their grief in three frameworks can narrate it fluently. Their model for “what it is like when the grief is activated” is accurate. The priors that were calibrated to the lost person’s presence are still intact, described by the meta-model and untouched by the prediction errors the description substitutes for. The comprehension is the near enemy of the grief. It provides the feeling of having processed while leaving the processing undone.
The felt sense — Gendlin’s pre-verbal, body-level signal — is the access point to the actual prediction errors rather than the meta-model about them. The body carries the grief the mind has intellectualized. The technique of sitting with what the body knows about the loss, before the mind has organized it into narrative, is the technique of allowing the prediction errors to surface and propagate rather than converting them into models of themselves. The shift that sometimes happens in a focusing session — the physical release, the quality of something that was stuck actually moving — is not new understanding. It is actual annealing: a configuration that was frozen, briefly heated, and reforming into something more true.
Dimwit / Midwit / Better Take
The dimwit take is “push through it — time heals all wounds.”
The midwit take is “grief is a process with stages; go through each stage and you will come out the other side.”
The better take is that grief is not a journey through emotional territory but a structural event in the brain’s prediction model, and its function is not the resolution of feeling but the update of priors. The person who has “been through” grief in the stage-model sense — who has experienced sadness, anger, bargaining, and acceptance — may have completed a phenomenological sequence while leaving the predictive model entirely intact. The model updates not when you have experienced all the feelings but when enough prediction errors have propagated through the priors to actually change them. The measure is not “have I passed through the stages?” but “am I still generating predictions calibrated to a presence that is no longer here?” Grief completes when the future is no longer encrypted with the past — when the model has updated enough that new things can be new rather than carrying the shape of what is gone. Until then, the past is not behind you. It is inside every prediction you make about what the present will contain.
Threads to Pull
Ideas, thinkers, and questions worth pursuing — and why.
- Peter Levine, Waking the Tiger and In an Unspoken Voice — Levine’s somatic model of trauma is the most rigorous account of what this note calls the incomplete discharge. His Somatic Experiencing technique is specifically designed to allow the frozen activation to thaw and complete — to run the discharge that was interrupted. The connection to predictive processing: the body’s incomplete discharge is a prediction error that was never allowed to propagate to the motor level. Completing the discharge is allowing the error to finish propagating. This is the same mechanism, different register.
- Francis Weller, The Wild Edge of Sorrow — Weller argues that grief in Western culture has been almost entirely privatized and pathologized — turned from a communal process that civilization evolved to support into an individual clinical problem to be managed. His five “gates of grief” (including grief for the unlived life, for what culture has failed to give, for collective wounds) expand the notion of what requires grieving beyond personal loss. The gate of the unlived life connects directly to what-optionality-cannot-buy and wound-as-compass: the life not committed to is also a loss requiring grief, and the refusal to grieve it is precisely what keeps the unlived life perpetually present as possibility rather than letting it pass into the past.
- The relationship between grief and annealing intervals — If grief is the primary annealing event, and modern life systematically interrupts grief (return-to-work norms, the management reflex, the colonization of interior space by content), then there should be measurable changes in the average quality of model updates in high-grief-suppression populations. Is there a way to operationalize “prediction model still calibrated to lost person/relationship/life” that could be measured? The clinical literature on “complicated grief” identifies something like this — chronic grief that has not resolved — but frames it as dysfunction rather than incomplete process.
- Why the death of a relationship requires grief even when the relationship was bad — The intuition is that you only grieve what was good. The mechanism says you grieve what your model was calibrated to include, regardless of valence. The person who leaves an abusive relationship is not grieving the relationship’s goodness. They are grieving the loss of the known — the prediction model that included the relationship’s specific texture, even its specific bad texture. The grief is for the calibration, not the quality. This is why leaving a bad relationship sometimes feels worse than it should: the model was calibrated to it, and the calibration must update even when the content being lost was harmful. Failing to recognize this produces the counterintuitive experience of mourning something you hated.
- The interface between grief and mimetic calibration — If the grandmother’s mimetic anchors (formed in 1932) function as temporal diversification against the current gradient, and if those anchors are partly transmitted through the texture of how she has metabolized her losses — what she has grieved, what she carries as frozen prediction, what calibrations she has actually updated — then the quality of grief across generations may be one of the mechanisms by which the family transmits genuine temporal diversity rather than merely surface values. The grandmother who has fully grieved the old world carries a completed model of its passing. The one who has not still carries predictions from 1932 as if they describe the present — which looks like wisdom to her grandchildren but may be frozen prediction rather than genuinely metabolized experience.