Medical note: This essay uses Addison’s disease and adrenal crisis as a systems-level analogy for thinking about computer failure. It is intended only as conceptual and educational discussion. It is not medical advice, diagnosis, or a treatment guide. Adrenal crisis is a potentially life-threatening medical emergency; suspected adrenal crisis requires urgent medical assessment and treatment according to established emergency guidance and the individual patient’s clinical plan (National Institute for Health and Care Excellence [NICE], 2024; Bornstein et al., 2016).
A few hours after recovering my VPS, I made a slightly absurd joke: “My little computer that runs the website seems to have had an adrenal crisis, and I just saved it by giving it the sysadmin equivalent of a shot.” Obviously, a server cannot have Addison’s disease, and Scaleway rescue mode is not hydrocortisone. But the joke stayed in my head because the resemblance was deeper than I initially expected.
The technical incident had started with a Cloudflare HTTP 522 and an inaccessible WordPress site. Underneath that visible outage, however, I eventually reconstructed something much more systemic. A DNS-over-WARP path had abruptly failed. DNS requests then multiplied, WARP logged enormous numbers of near-identical warnings, the logs consumed gigabytes of a very small root filesystem, ENOSPC appeared, log rotation itself failed, MariaDB began encountering write problems, resolver state deteriorated, and eventually both the public web service and my SSH management path disappeared.
The server did not simply switch from “healthy” to “dead.” It remained partially functional while its ability to remain functional was progressively being consumed.
That was what made me think of an Addisonian crisis—or, in the terminology I will generally use here, an adrenal crisis.
The comparison obviously cannot be made at the level of literal mechanisms. A Linux filesystem has no endocrine system. Cortisol is not free disk space. WARP is not an adrenal cortex, and restarting a daemon is certainly not hormone replacement. Trying to construct a one-to-one correspondence between biological organs and software components would turn a useful analogy into nonsense.
The interesting comparison begins one level higher:
How does a complex system remain viable while disturbances continuously push it away from its normal operating region? How much disturbance can its regulatory mechanisms absorb? And what happens when its capacity to compensate is no longer sufficient?
Those are naturally physiological questions. They also turned out to describe my VPS remarkably well.
What Addison’s disease changes in the regulatory system
Addison’s disease is primary adrenal insufficiency. In primary adrenal insufficiency, dysfunction or damage of the adrenal glands results in inadequate cortisol production and can also impair aldosterone production. Cortisol contributes to the regulation of blood pressure, blood glucose, metabolism and inflammatory activity, while aldosterone is important for sodium, potassium and water balance and therefore for circulatory stability (National Institute of Diabetes and Digestive and Kidney Diseases [NIDDK], n.d.).
Primary adrenal insufficiency is therefore different from secondary or tertiary adrenal insufficiency. Those conditions arise higher in the hypothalamic-pituitary-adrenal axis. Mineralocorticoid function is generally preserved in central adrenal insufficiency because aldosterone is regulated predominantly through the renin-angiotensin-aldosterone system rather than through ACTH in the same way as cortisol (Bornstein et al., 2016).
This distinction becomes important once we stop thinking of cortisol merely as “the stress hormone.” That description is not wrong, exactly, but it is far too compressed for the systems question I am interested in. Cortisol participates in several interacting forms of physiological regulation, and aldosterone adds another important regulatory pathway in primary adrenal insufficiency (NIDDK, n.d.; Bornstein et al., 2016).
So a systems-level description of Addison’s disease should not be:
one hormone is low
↓
symptoms appear
A more useful abstraction is:
part of a regulatory network loses capacity
|
+--> glucocorticoid regulation impaired
|
+--> mineralocorticoid regulation may also be impaired
|
v
the organism has less endogenous capacity
to respond to changing physiological demands
This does not mean cortisol or aldosterone somehow “control the whole body.” They do not. Blood pressure, circulating volume, vascular responsiveness, electrolytes, glucose, renal function, inflammatory activity and neurological function belong to an enormously interconnected physiological system with many overlapping regulatory mechanisms.
That interconnectedness is precisely the point.
When an important regulatory contribution disappears, the consequences do not necessarily remain confined to the component that originally failed. They propagate through relationships among subsystems.
That was already sounding familiar.
Homeostasis does not mean standing still
The concept that first gave me a proper vocabulary for the comparison was homeostasis.
Homeostasis is easy to imagine as though the body were trying to freeze a set of variables at perfect numerical values:
blood pressure = X glucose = Y temperature = Z sodium = S
That is not a very good picture of living regulation. Billman describes homeostasis as a self-regulating process through which an organism maintains internal stability while adapting to changing external conditions. Crucially for this analogy, he emphasizes that homeostasis is dynamic rather than static and that physiological regulation emerges from multiple interacting feedback systems rather than one simple feedback loop (Billman, 2020).
A living organism is changing continuously. A person stands up, sleeps, wakes, eats, exercises, loses water, encounters heat or cold, develops an infection, recovers from one, and passes through circadian changes in hormone secretion. Heart rate changes. Vascular tone changes. Renal handling of water and electrolytes changes. Metabolism changes.
Stability therefore does not require immobility.
It looks more like this:
disturbance
↓
regulated variables begin to change
↓
sensors and regulatory mechanisms respond
↓
other variables change in compensation
↓
the organism remains within
a physiologically viable range
The apparently paradoxical result is that a living system can remain stable because many of its components are continuously changing.
Billman’s account is particularly useful here because it treats physiological regulation as hierarchical and interconnected. Feedback mechanisms can compensate for disturbances, overlap with one another and provide flexibility when the environment changes (Billman, 2020). Stability is therefore an achievement of an active regulatory organization.
Consider something as ordinary as standing up. Gravity changes the distribution of blood. Cardiovascular regulatory mechanisms respond by altering vascular tone and cardiac behaviour so that adequate blood pressure and perfusion can be maintained. The variables move; the regulatory relationships help preserve viable function.
Or consider infection. Inflammatory signalling changes, temperature may change, cardiovascular and metabolic demands can change, and the hypothalamic-pituitary-adrenal axis participates in the physiological response to stress. Severe illness also changes cortisol availability through mechanisms more complicated than the simple textbook story of “stress causes more ACTH and therefore more cortisol”; altered cortisol metabolism and clearance can contribute as well (Dineen et al., 2019).
The organism does not defend one immutable physiological snapshot. It continually reorganizes itself while preserving conditions compatible with continued life.
This is already remarkably close to how I now think about a healthy computer system.
My VPS never occupies one fixed state either:
network traffic changes DNS demand changes process counts change memory pressure changes filesystem usage changes database workload changes logs grow logs rotate connections appear connections disappear services restart
A healthy server is not one on which those variables stop changing. It is one whose regulatory mechanisms allow those changes to occur without pushing essential functions outside an operationally viable region.
Normally, for example:
logs grow
↓
rotation occurs
↓
old logs are compressed or removed
↓
sufficient filesystem capacity remains
Memory management contains another family of regulatory responses:
processes request memory
↓
available memory decreases
↓
cache reclamation / paging / swap may respond
↓
workload remains supportable
Networking has its own:
temporary request failure
↓
retry
↓
path recovers
↓
operation succeeds
I am not claiming that logrotate is literally a homeostatic organ. That would stretch the biological concept until it stopped meaning anything. The useful claim is narrower: both systems contain feedback and compensatory mechanisms whose function is to keep important variables inside workable ranges despite disturbances.
Once I saw the problem at that level, the VPS failure started looking less like “one program broke” and more like a failure of regulation.
Stress reveals capacities that ordinary operation can hide
Addison’s disease becomes particularly revealing from a systems perspective when physiological demand changes.
Under significant physiological stress—such as serious illness, trauma or surgery—the body’s requirement for glucocorticoid activity changes. A person with primary adrenal insufficiency cannot generate the normal endogenous increase in cortisol production through the adrenal cortex, which is why established management includes increased glucocorticoid dosing during significant physiological stress (Bornstein et al., 2016; NICE, 2024).
The details of the stress response deserve some care. It would be too simple to represent critical illness as a single linear chain in which stress produces ACTH, ACTH produces cortisol, and that explains everything. Dineen and colleagues discuss additional changes in cortisol metabolism and clearance during critical illness (Dineen et al., 2019). For the systems analogy, however, the key practical point remains: changing physiological conditions alter the glucocorticoid requirement, and adrenal insufficiency removes the adrenal gland’s ability to autonomously supply the normal adaptive response.
Conceptually, an intact system can be simplified to:
physiological demand rises
↓
integrated stress-response systems react
↓
cortisol availability adapts
↓
increased demand is accommodated
Primary adrenal insufficiency changes an important part of that loop:
physiological demand rises
↓
damaged adrenal cortex cannot provide
the normal adaptive cortisol response
↓
external glucocorticoid replacement
must provide the missing capacity
This is where the systems architecture becomes fascinating. Treatment effectively moves part of an automatic physiological regulatory function outside the organism.
Routine glucocorticoid replacement supplies hormone that the adrenal glands cannot adequately produce. Mineralocorticoid replacement is also required when mineralocorticoid deficiency is present in primary adrenal insufficiency (Bornstein et al., 2016; NICE, 2024). During significant physiological stress, glucocorticoid requirements may rise, so current NICE guidance includes sick-day dosing and provision of emergency hydrocortisone for people at risk of adrenal crisis (NICE, 2024).
Abstractly, part of the regulatory loop now becomes:
physiological disturbance
↓
recognition of increased demand
↓
external intervention
↓
additional regulatory capacity supplied
↓
system remains viable
That is a profound architectural change.
The regulation has not ceased to exist. Part of it has crossed the boundary of the system being regulated.
And that immediately raises another question: how much disturbance can the system tolerate before that additional capacity becomes necessary?
Reserve is not simply unused capacity
Computer engineers are comfortable talking about spare capacity because it is often numerically visible:
filesystem capacity: 10 GB currently used: 6 GB nominal free space: 4 GB
Physiological reserve is much more complicated. Biology, rather inconsiderately, does not expose a command called physiology --show-reserve.
There is no single meter saying:
COMPENSATORY RESERVE: 37%
What we loosely call physiological reserve emerges from interacting capacities. Cardiovascular responses, renal regulation, endocrine responses, autonomic activity, circulating volume, metabolic substrate availability and many other processes contribute to the ability to tolerate disturbance. Homeostatic stability itself arises from interacting and partially redundant regulatory systems rather than a single controller (Billman, 2020).
Most importantly, reserve is meaningful relative to a demand.
A system can function adequately in its ordinary environment while possessing insufficient capacity for an unusual perturbation.
That is particularly visible in adrenal insufficiency. With appropriate treatment, people with adrenal insufficiency can live normal and active lives (NIDDK, n.d.). Yet infection, trauma, surgery, significant gastrointestinal illness or other physiological stresses can increase glucocorticoid requirements. Vomiting or diarrhoea creates an especially dangerous situation because physiological stress and fluid loss may increase at the same time that oral glucocorticoid medication cannot reliably be retained or absorbed (NICE, 2024).
Dineen et al. describe the pathophysiology of adrenal crisis as incompletely understood, but identify an important underlying problem as a mismatch between the cortisol required during physiological stress and the cortisol available (Dineen et al., 2019).
I find that formulation remarkably useful.
It changes the conceptual statement from:
cortisol is low
into:
required regulatory capacity
>
available regulatory capacity
The second expression is relational.
There may be enough capacity for one state of the world and too little for another.
And that is exactly the kind of relationship I had seen in the VPS.
Adrenal crisis is a systems event
Adrenal crisis is a life-threatening medical emergency. It can involve hypotension, volume depletion, electrolyte abnormalities, weakness, altered mental status and circulatory shock or collapse. Established guidance emphasizes immediate hydrocortisone administration and fluid resuscitation when adrenal crisis is suspected, without waiting for diagnostic confirmation when delay would endanger the patient (NICE, 2024; Bornstein et al., 2016).
The exact pathophysiological sequence is more complicated than any simple diagram. Dineen et al. explicitly note that the pathophysiology of adrenal crisis remains incompletely understood. Their discussion nevertheless identifies several interacting elements, including insufficient cortisol availability relative to physiological demand, impaired vascular responsiveness and volume depletion (Dineen et al., 2019).
What interests me here is not using those features diagnostically. It is understanding what the word crisis means when considered as a systems phenomenon.
Imagine, very abstractly, several physiological variables interacting:
circulating volume
↕
blood pressure
↕
vascular responsiveness
↕
electrolyte balance
↕
glucose availability
↕
renal perfusion
↕
neurological function
No single arrow in this diagram contains “adrenal crisis.”
Likewise, cortisol is not a master variable determining every other state directly. The organism contains many overlapping regulatory mechanisms.
Now introduce a substantial perturbation:
infection
|
+--> inflammatory signalling
|
+--> altered metabolic demand
|
+--> possibly fever
|
+--> possibly reduced intake / fluid loss
|
v
greater demand on regulatory capacity
If vomiting or diarrhoea is also present, another interaction can emerge:
vomiting / diarrhoea
|
+--> fluid loss
|
+--> reduced circulating volume
|
+--> oral medication may not be retained
|
v
less capacity to compensate
while physiological demand is increasing
NICE specifically addresses prolonged vomiting or diarrhoea in adrenal insufficiency because inability to absorb oral glucocorticoids requires escalation to parenteral treatment and emergency medical care (NICE, 2024).
Structurally, this is very different from:
component A breaks
↓
system stops
Several relationships can deteriorate together. A disturbance increases demand. A missing endocrine response reduces available adaptive capacity. Fluid loss can worsen haemodynamic stability. Failure to retain oral medication can further reduce available glucocorticoid replacement. The resulting state can then become progressively harder for the organism to correct (Dineen et al., 2019).
The pathology exists partly in the relationships among these processes.
This is where the word decompensation became extremely useful to me.
Compensation is not the same thing as health
During the VPS incident, the DNS-over-WARP failure began around August 11. The externally obvious website failure did not appear until August 16.
For several days, therefore, the server occupied an intermediate state.
It was functioning.
It was also failing.
DNS success had collapsed to essentially zero. Query volume had risen dramatically. WARP was generating roughly 100,000–150,000 DNS warnings per hour. Logs were consuming the filesystem at an extraordinary rate. By August 14, ENOSPC was already appearing. Later, log rotation itself failed.
And yet:
website still reachable
=
externally visible function preserved
If that were my only measurement, I could easily have called the machine healthy.
Physiology made me realize why that conclusion is inadequate.
A regulated system can preserve important functions while compensatory mechanisms absorb disturbances. The fact that compensation is succeeding at this instant does not tell us how much reserve remains, how expensive that compensation has become, or how close the system is to a boundary beyond which compensation will fail. Homeostasis is an active process produced through interacting regulatory mechanisms, not the simple absence of disturbance (Billman, 2020).
At a very abstract level:
disturbance
↓
regulatory response
↓
compensation
↓
visible function preserved
That may remain sustainable if the disturbance is small enough and sufficient regulatory capacity remains.
Persistent or increasing disturbance can create a different trajectory:
disturbance persists
↓
compensation continues
↓
reserve is progressively consumed
↓
operating margin shrinks
↓
system becomes increasingly sensitive
to additional disturbance
Eventually:
required regulatory response
>
available regulatory capacity
↓
decompensation
I am using decompensation here as a systems concept, not proposing a new clinical definition of adrenal crisis. Real adrenal crises vary in presentation and precipitating factors, and the detailed pathophysiology remains incompletely understood (Dineen et al., 2019).
But the systems idea is powerful:
A compensated system may already contain a serious failure. The failure becomes externally obvious only when the remaining regulatory mechanisms can no longer contain it.
That statement described my server almost perfectly.
The VPS had a kind of reserve too
Once I thought about the incident this way, free disk space changed meaning.
Previously I had treated it as an ordinary capacity measurement:
8.9 GB root filesystem some amount used some amount free
During the failure, however, free storage became part of the machine’s ability to tolerate disturbance.
WARP could generate an enormous warning stream without immediately taking down the system because the filesystem still had somewhere to put those warnings:
DNS failures
↓
WARP warnings
↓
filesystem accepts log writes
↓
services continue operating
↓
website remains reachable
Superficially, the machine was coping.
But look at what “coping” meant:
DNS failures continue
↓
warnings continue
↓
free storage decreases
↓
remaining ability to absorb more warnings decreases
↓
failure margin shrinks
The very resource that allowed the machine to remain outwardly operational was being consumed by the process through which it remained operational.
That is much closer to decompensation than to a binary crash.
Then the filesystem reached ENOSPC.
At that point, the consequences changed qualitatively. Logging infrastructure could no longer behave normally. Log rotation failed. WARP’s own rotation failed. MariaDB encountered write problems. Resolver state later deteriorated. The machine was no longer simply absorbing one broken subsystem; the consequences of the original disturbance were beginning to impair mechanisms elsewhere in the system.
This progression is worth separating:
STAGE 1 disturbance exists but available reserve absorbs its consequences STAGE 2 disturbance persists and reserve is progressively consumed STAGE 3 reserve approaches a critical boundary and regulatory mechanisms begin failing STAGE 4 secondary failures alter the operating conditions of still-functioning subsystems STAGE 5 the system can no longer preserve its externally visible functions
This is considerably more informative than:
disk became full therefore server crashed
That statement is technically true at one level and explanatorily almost useless at another.
When a compensatory response begins amplifying the disturbance
The VPS contained another feature that made the physiological comparison more interesting: some behaviours that were normally useful became harmful under persistent failure.
A DNS retry is usually reasonable.
Logging an unexpected network error is reasonable.
Persisting logs for later diagnosis is reasonable.
Rotating old logs is reasonable.
But once WARP’s DNS path had entered a persistent, near-instantaneous ENETUNREACH state, those normal behaviours became coupled:
DNS request
↓
ENETUNREACH
↓
warning logged
↓
caller retries quickly
↓
another ENETUNREACH
↓
another warning
↓
another retry
↓
...
↓
log volume explodes
The feedback structure had changed.
A mechanism that normally helps a system recover from transient errors—retrying—was now increasing the rate at which the persistent error was exercised.
A mechanism that normally helps an administrator understand failure—logging—was now converting each repeated failure into additional filesystem consumption.
Eventually:
failure ↓ response to failure ↓ more resource consumption ↓ less capacity to tolerate failure ↓ more secondary failure
At that point the system was no longer merely failing to compensate. Some of its ordinary responses to failure had become part of the amplification mechanism.
Physiology contains many regulatory responses, and it would be misleading to map any particular mechanism of adrenal crisis directly onto DNS retries or log rotation. The general systems pattern is more defensible: regulatory responses have limits; multiple feedback loops interact; and when a system moves sufficiently far from its normal operating regime, processes that preserve stability under ordinary conditions may no longer be sufficient to restore it (Billman, 2020).
This is also why the description of a mismatch between cortisol need and cortisol availability is so useful here. It directs attention away from a single defective object and toward the relationship between disturbance and the system’s remaining capacity to respond (Dineen et al., 2019).
A green light is not the same thing as a healthy system
One of the strangest findings in the VPS investigation was this pair of observations:
DNS success: approximately 0% Other Error: approximately 100% DNS proxy health status: Healthy
I still find that almost philosophically perfect.
Some narrow internal definition of proxy health may genuinely have been satisfied. The proxy process might have been running. Its listener might have existed. Its task might have passed whatever liveness criterion the software used.
But the actual function the rest of the system depended on—resolving DNS queries—was essentially absent.
Physiological homeostasis provides a much more serious version of this distinction: regulation is produced by the interaction of multiple control systems, and organismal stability cannot be reduced to the state of one regulatory loop (Billman, 2020).
For the server:
process running
≠
service functioning
service functioning
≠
system healthy
And for a complex regulated system more generally:
one locally normal measurement
≠
global stability
The implication for monitoring is obvious. A check that asks only:
Is the process alive?
may completely miss:
Is the function succeeding? Is the system consuming reserve to preserve that function? Is the compensation sustainable? Is the system moving toward or away from a critical boundary?
The same conceptual problem appeared in my UptimeRobot monitoring. It knew when the website finally became unreachable. It did not know that the server had spent days moving toward that failure.
There is no single variable called health
This is where the comparison begins moving toward a more formal systems argument.
Neither an organism nor a computer contains one physically measurable scalar called:
health = 83.7%
For the VPS, I could approximate its operational state using variables such as:
free disk log growth rate DNS success rate DNS error rate memory availability database writability network reachability SSH reachability service states
None of those variables individually is server health.
The useful state emerges from their relationships.
An equally simplified physiological description might contain:
cortisol availability circulating volume arterial pressure vascular responsiveness electrolyte state glucose temperature renal perfusion neurological state ...
Again, no single entry contains “health.” Physiological stability is instead produced through interacting feedback and regulatory systems (Billman, 2020).
We can express the idea abstractly without pretending that the following is a clinical model.
Suppose the organism at time t has a state:
x(t) = [
hormonal state,
circulating volume,
blood pressure,
electrolytes,
glucose,
temperature,
heart rate,
...
]
Its state changes according to its present condition, external disturbances and the regulatory capacities available to it:
x(t + Δt) = F(x(t), u(t), p)
Here, u(t) might contain changing influences such as illness, trauma, fluid loss, exertion, temperature, medication or nutrition. The parameters represented by p include properties of the system that determine how it can respond.
Now imagine a region V containing states compatible with continued physiological viability:
V = {
x :
the organism can continue maintaining
essential physiological organization
}
Ordinary life does not require:
x(t) = constant
Quite the opposite. The state changes continuously:
sleep ↓ wake ↓ eat ↓ walk ↓ exercise ↓ recover ↓ infection ↓ physiological response ↓ recovery
The remarkable property is that regulation usually keeps the trajectory inside a viable region despite those disturbances. That dynamic character of stability is central to modern discussions of homeostasis (Billman, 2020).
Primary adrenal insufficiency changes that dynamical system because an important endogenous adaptive capacity is constrained. Replacement therapy restores essential glucocorticoid—and, when required, mineralocorticoid—function, but significant physiological stress may require an increase in glucocorticoid replacement because the adrenal glands cannot autonomously generate the normal response (Bornstein et al., 2016; NICE, 2024).
The abstract systems question then becomes:
Given the disturbance acting on the system now, is available regulatory capacity sufficient to keep the trajectory inside V?
That was almost exactly the question my VPS had been answering without my realizing it.
Its trajectory was:
DNS route failure
↓
near-total DNS failure
↓
retry amplification
↓
warning amplification
↓
free-space depletion
↓
ENOSPC
↓
failed log rotation
↓
secondary subsystem failures
↓
resolver/network degradation
↓
HTTP and SSH loss
The website remaining reachable during the first part of that trajectory did not prove the server was healthy.
It proved only that:
x(t) had not yet crossed the boundary at which that particular visible function failed
By the time UptimeRobot finally told me that the site was down, the system had already spent days consuming its ability to remain up.
That is the part of the adrenal-crisis analogy I find most valuable.
A crisis can become visible at the end of a process whose decisive change began much earlier.
The interesting question is therefore no longer simply why one component failed. It is why the whole system could compensate for that failure for a while, why that compensation became progressively less sustainable, and what happened when the remaining regulatory reserve was finally insufficient.
That is where the concept of decompensation becomes much more than a metaphor—and where the comparison between physiology and computing becomes genuinely useful.
Decompensation begins with a mismatch between demand and capacity
The concept I wanted to understand more deeply after Part I was decompensation. In ordinary language it sounds simple: compensation works until it does not. Addison’s disease shows why the actual transition is much more interesting.
Primary adrenal insufficiency changes the organism before any crisis occurs. The adrenal cortex has lost an important part of its capacity to produce cortisol and, in many cases, aldosterone. Replacement therapy can restore the hormonal inputs needed for ordinary life, but it does not turn the damaged adrenal cortex back into an intact, continuously adapting endocrine organ (Bornstein et al., 2016; NIDDK, n.d.).
That difference is easy to miss when conditions are stable.
A person may be adequately replaced, haemodynamically stable, eating normally, absorbing medication normally and experiencing no major physiological stress. Under those conditions, the available regulatory support is sufficient for the demand being placed on the system.
Then the environment changes.
ordinary physiological demand
↓
replacement sufficient
↓
stable state
major illness / trauma / surgery / fluid loss
↓
physiological demand rises
↓
required glucocorticoid support rises
↓
the previous replacement level may no longer be sufficient
Dineen, Thompson and Sherlock describe adrenal crisis in terms that fit this systems perspective particularly well: an important part of the problem is a mismatch between cortisol availability and the amount required during physiological stress (Dineen et al., 2019).
That formulation is more illuminating than simply saying “there is too little cortisol.” Too little relative to what?
The answer is: relative to the current state of the organism and the demands being imposed on it.
So the same amount of available cortisol support may be adequate under one set of conditions and inadequate under another.
available regulatory capacity = C ordinary demand = D1 C > D1 system remains compensated severe physiological stress = D2 D2 > C regulatory mismatch develops
This immediately makes adrenal crisis a dynamical problem rather than a static deficiency.
The healthy adrenal response is continuously adaptive
To appreciate what is lost in Addison’s disease, it helps to look briefly at the regulatory architecture of the hypothalamic-pituitary-adrenal axis.
In simplified form:
hypothalamus
↓ CRH
pituitary
↓ ACTH
adrenal cortex
↓
cortisol
Cortisol then participates in feedback regulation of the hypothalamus and pituitary. This is not a simple on/off circuit. Cortisol secretion has circadian and ultradian dynamics, and the system responds to changing physiological conditions. Illness, trauma and other forms of substantial physiological stress alter the demand placed on this regulatory system (Bornstein et al., 2016; Dineen et al., 2019).
Primary adrenal insufficiency changes the final endocrine output stage:
hypothalamus
↓
pituitary
↓ ACTH
damaged adrenal cortex
X
cannot generate an adequate cortisol response
The regulatory signals can demand more output, but the organ responsible for producing that output has lost the necessary capacity.
That is very different from merely having “a low number.”
It is a limitation in adaptive range.
Under ordinary conditions, glucocorticoid replacement supplies cortisol activity externally. When physiological demand rises substantially, however, the replacement strategy has to change because the adrenal glands themselves cannot automatically increase cortisol production. This is the rationale behind stress or sick-day dosing in established adrenal insufficiency (NICE, 2024; Bornstein et al., 2016).
The intact system is approximately:
disturbance increases
↓
endogenous regulatory system detects/responds
↓
hormonal output changes
↓
adaptive capacity increases
In primary adrenal insufficiency:
disturbance increases
↓
endogenous adrenal output cannot increase adequately
↓
external replacement must be adjusted
↓
adaptive capacity restored from outside
This means that part of a normally endogenous control loop has effectively become externalized.
That is one of the most fascinating features of Addison’s disease when viewed as a systems problem.
Treatment restores a missing input but does not recreate the original controller
Hydrocortisone replacement is extraordinarily effective, but from a control-system perspective there is an important distinction between replacing an output and recreating the original adaptive controller.
The healthy adrenal system continually participates in changing hormone availability according to circadian rhythm, physiological state and stress. Oral replacement necessarily works differently. Medication is taken at discrete times and according to an established regimen. During significant physiological stress, the regimen has to be deliberately modified (Bornstein et al., 2016).
Conceptually:
INTACT REGULATION
physiological state
↓
endocrine feedback
↓
continuous endogenous adjustment
REPLACEMENT-DEPENDENT REGULATION
physiological state
↓
recognition that demand has changed
↓
dose adjustment / emergency plan
↓
external hormone delivery
This does not make replacement somehow inadequate as therapy. It means the architecture of regulation has changed.
The closed biological loop has become partly dependent on an external layer: medication availability, appropriate dose adjustment, absorption, and recognition that the physiological situation has changed.
In computer terminology I would call that a change in the control boundary. But the medical phenomenon is the important thing here: the person can be physiologically stable while one adaptive endocrine function is being supplied through an external therapeutic process.
This is why preparedness is part of treatment. NICE recommends sick-day guidance, emergency hydrocortisone kits and training for people with primary adrenal insufficiency because increased physiological demand can require a rapid change in glucocorticoid replacement (NICE, 2024).
That tells us something fundamental about the disease.
The risk is not simply:
cortisol absent
It is:
the organism cannot autonomously scale cortisol availability to match changing demand
Aldosterone makes primary adrenal insufficiency even more systemic
Addison’s disease is particularly interesting because primary adrenal insufficiency may affect mineralocorticoid function as well as glucocorticoid function.
Aldosterone helps regulate sodium and potassium balance and contributes to maintaining extracellular fluid volume and blood pressure. In primary adrenal insufficiency, inadequate aldosterone can therefore contribute to sodium loss, volume depletion, hypotension and hyperkalaemia (NIDDK, n.d.; Bornstein et al., 2016).
This immediately expands the causal network:
adrenal cortical failure
|
+----------------------+
| |
v v
cortisol deficiency aldosterone deficiency
| |
v v
impaired stress sodium loss /
response potassium retention
|
v
volume depletion
|
v
hypotension
This diagram is intentionally simplified, but it demonstrates why adrenal crisis cannot be understood as an isolated hormone measurement.
The relevant physiological state includes endocrine signalling, vascular responsiveness, circulating volume, renal handling of sodium and water, electrolyte balance, glucose metabolism, inflammatory activity and the demands imposed by whatever illness or stressor triggered the deterioration.
Several of these variables can change simultaneously.
And once they begin interacting, the whole state matters more than any single variable considered alone.
Why blood pressure becomes such an important systems variable
Cortisol has important permissive effects on cardiovascular regulation. Adequate glucocorticoid activity contributes to vascular responsiveness to catecholamines and therefore to the maintenance of vascular tone and blood pressure. This helps explain why severe cortisol deficiency can be associated with hypotension and why adrenal crisis can progress to circulatory shock (Dineen et al., 2019).
In primary adrenal insufficiency, this can occur alongside mineralocorticoid deficiency and reduced circulating volume.
So two different regulatory problems can converge on the same macroscopic variable:
insufficient cortisol
↓
reduced vascular responsiveness
↓
difficulty maintaining vascular tone
insufficient aldosterone
↓
renal sodium loss
↓
water loss / reduced volume
↓
difficulty maintaining circulation
both pathways
↓
hypotension / haemodynamic instability
This is one reason the systems view matters. If I look only at the endpoint—low blood pressure—I lose the regulatory structure that produced it.
Blood pressure itself is an emergent result of several interacting factors:
cardiac output
×
systemic vascular resistance
influenced by:
circulating volume
heart rate
contractility
vascular tone
autonomic activity
hormonal regulation
renal regulation
many other variables
An adrenal crisis can disturb several parts of that network at once.
The organism is therefore not dealing with a single missing signal. It is trying to preserve circulation while some of the mechanisms normally supporting circulation have been weakened and physiological demand may simultaneously be increasing.
Why gastrointestinal illness is a particularly revealing trigger
Vomiting and diarrhoea make the dynamics especially easy to see.
The underlying illness is already a physiological stressor. It may produce inflammation, fever or increased metabolic demand. At the same time, gastrointestinal losses can reduce fluid and electrolyte volume.
Then another problem appears: oral glucocorticoid replacement may not be retained or reliably absorbed.
The resulting structure can look like this:
gastrointestinal illness
|
+--> physiological stress increases
|
+--> cortisol requirement increases
|
+--> vomiting / diarrhoea
|
+--> fluid loss
|
+--> sodium loss
|
+--> reduced circulating volume
|
+--> oral medication may not be absorbed
|
v
required support rises
while deliverable support may fall
That is a particularly dangerous configuration because the two sides of the mismatch move in opposite directions.
physiological demand
↑
available / absorbed replacement
↓
haemodynamic reserve
↓
NICE therefore specifically distinguishes situations in which oral glucocorticoids can still be taken from those in which vomiting or prolonged diarrhoea makes oral absorption unreliable; the latter requires escalation to parenteral hydrocortisone and emergency assessment (NICE, 2024).
I am not reproducing that guidance here as a treatment protocol—the medical note at the beginning of this essay remains important. What interests me conceptually is why the route of delivery suddenly matters.
The crisis is attacking both the regulated organism and part of the ordinary mechanism used to support that organism.
That is a classic systemic vulnerability.
The ordinary recovery pathway can itself become unavailable
Under stable conditions:
replacement needed
↓
oral medication
↓
gastrointestinal absorption
↓
systemic circulation
↓
glucocorticoid effect
But gastrointestinal illness can alter that pathway:
replacement needed MORE urgently
↓
oral medication
↓
vomiting / impaired absorption
X
↓
required regulatory input may not arrive
The therapeutic architecture therefore includes another route. Emergency hydrocortisone can be administered intramuscularly or intravenously, bypassing the gastrointestinal absorption problem. NICE explicitly recommends emergency management kits containing injectable hydrocortisone for people with primary adrenal insufficiency and immediate parenteral hydrocortisone when adrenal crisis is suspected (NICE, 2024).
This is one of the strongest systems lessons in the entire comparison:
A recovery mechanism becomes much more robust when it does not depend on the pathway most likely to become unavailable during the failure it is intended to correct.
That statement is general systems language. In Addison’s disease, the actual medical implementation is concrete: when oral delivery becomes unreliable during crisis, a non-oral route provides a way to restore glucocorticoid availability.
The architecture contains an alternative path because crisis can invalidate assumptions that are perfectly safe during ordinary operation.
An adrenal crisis is a convergence of disturbances
At this point, describing adrenal crisis as “acute cortisol deficiency” feels insufficient to me. It is true, but it hides the interesting part.
A better systems picture is:
precipitating illness / stress
|
v
physiological demand increases
|
v
cortisol requirement increases
|
X
adrenal cortex cannot increase output
|
v
relative cortisol deficiency
|
+-----+-------------------+
| |
v v
vascular responsiveness metabolic /
impaired inflammatory regulation altered
|
v
hypotension
PLUS, in primary adrenal insufficiency:
aldosterone deficiency
|
v
sodium / water loss
|
v
volume depletion
|
v
further hypotension
PLUS, depending on precipitant:
vomiting / diarrhoea
|
+--> additional fluid loss
|
+--> impaired oral drug delivery
COMBINED
|
v
systemic decompensation
No single arrow here is “the crisis.”
The crisis is the state produced by the interaction of these processes.
This is why Dineen et al. emphasize that the complete pathophysiology of adrenal crisis remains incompletely understood even though its clinical urgency and core endocrine defect are well established (Dineen et al., 2019).
There is an epistemic lesson in that statement too.
Medicine can know with high confidence that a state is dangerous, know which treatment is lifesaving, identify important causal mechanisms, and still lack a complete microscopic account of every transition producing the crisis.
That is not a contradiction.
Explanations have levels.
Electrolytes are not side effects of the crisis
The electrolyte disturbances associated with adrenal insufficiency are another example of why relationships matter.
In primary adrenal insufficiency, mineralocorticoid deficiency can reduce renal sodium retention and impair potassium excretion. Hyponatraemia and hyperkalaemia may therefore appear, although actual clinical presentations vary (Bornstein et al., 2016).
Hyponatraemia can also be influenced by glucocorticoid deficiency through changes in water handling, so even here a single laboratory value may reflect more than one regulatory pathway.
Conceptually:
aldosterone deficiency
↓
less renal sodium retention
↓
sodium / volume loss
cortisol deficiency
↓
altered water regulation
↓
additional contribution to hyponatraemia
These biochemical changes then feed back into organismal function. Sodium and water balance influence circulating volume. Potassium influences membrane excitability and cardiac function. Glucose availability can also become problematic, particularly in vulnerable states.
Again, the interesting object is the network.
A laboratory value is a window into that network, not an isolated malfunction floating outside it.
The precipitating illness and the adrenal crisis are not the same event
This distinction is particularly important.
An infection may trigger adrenal crisis.
But the infection is not itself the adrenal crisis.
Likewise, vomiting may participate in precipitating a crisis, but vomiting is not equivalent to adrenal crisis.
The precipitant changes the demands imposed on an organism whose adaptive endocrine capacity is constrained.
precipitating event
↓
changes physiological demand
↓
interacts with adrenal insufficiency
↓
regulatory mismatch
↓
systemic consequences
↓
adrenal crisis
Once crisis develops, the physiological state has acquired additional problems that were not identical to the precipitating event.
For example:
infection ↓ initial disturbance later state: infection + inadequate cortisol availability + hypotension + volume depletion + electrolyte disturbance + impaired oral replacement + other downstream effects
So removing or treating the precipitating cause does not necessarily instantaneously reverse the entire crisis state.
History matters.
The organism has moved from one state into another.
Path dependence makes crisis more than a snapshot
This is a concept I find increasingly useful: path dependence.
Suppose we describe the physiological state at one moment as S0.
An infection appears:
S0 + infection ↓ S1
Now vomiting causes fluid loss:
S1 + fluid loss ↓ S2
Glucocorticoid requirement rises while oral absorption becomes unreliable:
S2 + regulatory mismatch ↓ S3
Hypotension and further systemic consequences appear:
S3 + haemodynamic deterioration ↓ S4
At S4, simply removing the original precipitating factor would not magically recreate S0.
S4 - infection ≠ S0
The intervening history has changed circulating volume, hormone availability, electrolyte state and other physiological variables.
That is why crisis treatment is necessarily concerned with the current state as well as the initiating cause.
NICE reflects this directly: emergency management includes immediate hydrocortisone, intravenous isotonic saline, monitoring of blood pressure, heart rate, electrolytes and glucose, and identification and treatment of the underlying precipitating cause (NICE, 2024).
Those are several different interventions because several different relationships may need restoration.
Emergency treatment is a state-restoration process
Looked at physiologically, emergency treatment of adrenal crisis is remarkably systemic.
It does not consist only of restoring one hormone number.
Hydrocortisone addresses the acute glucocorticoid deficit. Intravenous isotonic saline addresses volume depletion and haemodynamic instability. Relevant physiological variables are monitored as treatment proceeds, and the precipitating illness or other underlying cause must also be treated (NICE, 2024).
Conceptually:
CRISIS STATE
insufficient glucocorticoid activity
+
volume depletion
+
haemodynamic instability
+
possible electrolyte / glucose disturbance
+
ongoing precipitating stress
↓
INTERVENTION
restore glucocorticoid support
+
restore circulating volume
+
monitor system response
+
treat precipitating cause
↓
progressive return toward physiological stability
The interesting word here is progressive.
Hydrocortisone enters the circulation, but the entire organism does not instantaneously snap back into its pre-crisis state. Circulatory volume has to be restored. Vascular responsiveness changes. Electrolytes and glucose may require observation and correction. The precipitating disease continues to have its own dynamics.
Recovery is therefore a trajectory.
A crisis is over when the ordinary maintenance regime becomes viable again
NICE’s emergency recommendations contain a systems idea that I find particularly elegant. Parenteral hydrocortisone and intravenous fluids are continued while haemodynamic instability persists; transition back toward oral replacement occurs once the person is haemodynamically stable and able to take and absorb oral glucocorticoids (NICE, 2024).
That means the endpoint of emergency treatment is not merely:
one symptom disappears
or:
one laboratory value improves
It is closer to:
the organism has returned sufficiently toward a stable operating regime AND the ordinary maintenance pathway has become reliable again
This is a powerful way to define recovery.
Emergency mode exists because ordinary regulation is insufficient.
Recovery occurs when ordinary maintenance can once again sustain the system.
stable maintenance
↓
major disturbance
↓
ordinary regulation insufficient
↓
emergency intervention
↓
physiological state restored
↓
ordinary maintenance becomes sufficient again
That is much richer than a binary:
sick ↓ treated ↓ well
Adrenal crisis shows why reserve is multidimensional
Earlier I described physiological reserve as something that cannot be reduced to one gauge. Adrenal crisis makes the reason clearer.
Imagine an organism with several forms of reserve:
circulatory reserve endocrine adaptive capacity fluid reserve renal regulatory capacity metabolic reserve autonomic responses ability to absorb medication ability to increase external replacement ...
These are not independent batteries.
They interact.
A loss of fluid volume makes cardiovascular regulation more demanding. Fever and infection increase metabolic and endocrine demand. Vomiting may simultaneously reduce volume and disrupt medication delivery. Cortisol deficiency can reduce vascular responsiveness just when maintaining vascular tone becomes more important.
So reserve is better imagined as a property of the whole configuration:
reserve ≠ one spare resource reserve = the system's remaining ability to reorganize itself successfully under additional disturbance
That is why a person may be completely stable in one context and vulnerable in another without any contradiction.
The relevant question is not simply “How much reserve exists?”
It is:
Reserve for which disturbance, at which moment, given which current physiological state?
Compensation can conceal how close the system is to its limits
One of the most important ideas for me is that successful compensation can make a system look healthier than it really is.
If blood pressure remains adequate, oral medication is being absorbed, fluid losses are small and glucocorticoid replacement matches demand, there may be little outward evidence that a crucial endogenous endocrine response is absent.
The regulatory deficit becomes obvious mainly when the system is challenged.
This is a general property of reserve.
low demand
↓
limited capacity still sufficient
↓
normal-looking output
high demand
↓
same limited capacity now insufficient
↓
instability becomes visible
In that sense, stress does not necessarily create the underlying vulnerability.
It reveals it.
This explains why studying only steady-state operation can give an incomplete understanding of a system. Stability under low demand tells us little about the size of the region over which regulation remains successful.
The boundary between compensation and crisis is not necessarily a sharp line
It is tempting to imagine a perfect threshold:
before 14:03:17 compensated after 14:03:17 crisis
Clinical reality is not usually that clean.
Physiological variables change continuously, while the clinical category adrenal crisis identifies a dangerous state requiring urgent treatment. The transition can develop rapidly, but the underlying processes—rising demand, inadequate hormonal response, volume loss, hypotension and other changes—may evolve over time.
Dineen et al. discuss the difficulty of defining adrenal crisis uniformly and note that several definitions have been used, often combining acute deterioration with hypotension, gastrointestinal symptoms, electrolyte abnormalities or the need for parenteral glucocorticoids (Dineen et al., 2019).
That is interesting from a systems perspective because categories such as:
stable compensated decompensating crisis
are macroscopic descriptions applied to a continuously evolving biological state.
The molecules do not receive a notification saying that the system has officially crossed from “compensated” to “crisis.”
The category exists at the level of the organism and its dynamics.
The absence of a single crisis molecule does not make the crisis unreal
This returns to one of the deeper questions that originally attracted me to the comparison.
Where exactly is adrenal crisis?
It is not located inside one cortisol molecule.
It is not inside one sodium ion.
It is not inside one adrenal cortical cell.
It is not contained by a single blood-pressure measurement.
At progressively lower levels we find:
organism
↓
organs
↓
tissues
↓
cells
↓
receptors / membranes
↓
molecules
↓
ions / atoms
↓
physical interactions
At the bottom, the individual physical constituents continue obeying chemistry and physics.
No sodium ion becomes “pathological” in the moral or intentional sense.
No cortisol molecule knows whether the organism is in crisis.
Yet adrenal crisis is completely real.
It exists at the level of the organization and dynamics of the whole system.
This is not mystical emergence. It simply means that different levels of description capture different properties.
A sodium concentration is a collective property of many particles.
Blood pressure is a systemic property of cardiovascular dynamics.
Haemodynamic stability is a relationship among multiple physiological processes.
Adrenal crisis describes a pathological state of the organism produced by the interaction of those processes.
Nothing requires a microscopic object called CRISIS.
The body does not fail because physics stops working
This becomes almost philosophically strange if I keep descending.
During adrenal crisis, molecules continue interacting according to chemical laws.
Membrane channels still obey electrochemical gradients.
Receptors still bind according to their biochemical properties.
Water still moves according to physical forces.
The heart, blood vessels and kidneys are made of matter obeying exactly the same physical laws as before the crisis.
And yet at another level we say:
the organism is failing to maintain homeostasis
There is no contradiction.
The word failure refers to the organization of those lawful physical processes relative to the conditions necessary for continued organismal viability.
That question will become much more important later, when I turn from physiology to state spaces, physical transitions and eventually the philosophical question of where words such as wrong, failure and function come from.
For now, adrenal crisis gives us a concrete biological case in which the distinction is impossible to ignore.
Homeostasis is therefore a property of relationships
The more I thought about Addison’s disease, the less useful it seemed to identify homeostasis with a list of correct laboratory values.
A better picture is a network:
environment
↕
nervous system
↕
endocrine system
↕
cardiovascular system
↕
kidneys
↕
fluid / electrolyte state
↕
metabolism
↕
immune response
↕
behaviour
↕
external therapeutic support
None of these systems maintains life independently.
The regulatory property is distributed across their interaction.
Billman’s description of homeostasis as a dynamic, self-regulating process is useful precisely because it shifts attention from fixed values to coordinated responses (Billman, 2020).
In Addison’s disease, one part of that network has permanently lost capacity.
Replacement therapy reconnects the missing endocrine function through an external route.
Most of the time, that restored architecture works.
A crisis occurs when the combination of physiological demand, available hormonal support and the state of the other regulatory systems becomes incompatible with stable compensation.
The human being becomes part of the control loop
There is another unusual systems feature here that deserves attention.
With an intact adrenal system, a person does not consciously calculate an ACTH concentration because they have a fever.
The regulation is endogenous.
With adrenal insufficiency, part of the adaptive process may involve conscious recognition and action:
illness occurs
↓
person recognizes increased physiological stress
↓
sick-day rule applied
↓
replacement dose adjusted
And if the situation deteriorates further:
oral route becomes unreliable
↓
crisis risk recognized
↓
emergency injection / medical response
↓
parenteral replacement supplied
NICE’s emphasis on education, emergency kits and training reflects this altered regulatory architecture (NICE, 2024).
A normally automatic physiological function has acquired a human decision-making layer.
That creates extraordinary resilience—the missing response can be deliberately supplied—but it also introduces new dependencies:
recognition knowledge medication availability correct action route of administration access to emergency care
From a human-computer perspective, this is fascinating. The boundaries of the regulated system have expanded.
The relevant system is no longer simply:
body
but, for this particular regulatory function:
body + replacement medication + knowledge + emergency preparation + clinical support
This does not make the disease less biological.
It demonstrates that biological stability can depend on an extended network involving deliberate human intervention.
Preparedness reduces vulnerability but cannot abolish uncertainty
Another important lesson is that knowledge and preparation do not turn a complex biological system into a deterministic machine.
A person can understand sick-day rules, have emergency medication available and manage the condition carefully, yet still face unexpected illness or a rapidly evolving physiological disturbance. Adrenal crisis remains a recognized complication of chronic adrenal insufficiency even in people who have been educated about prevention and emergency management (Dineen et al., 2019).
This matters because resilience is sometimes described too optimistically:
identify risk
↓
create protocol
↓
risk solved
Real resilience looks more like:
identify risk
↓
reduce probability
↓
increase reserve
↓
create fallback paths
↓
improve detection
↓
prepare emergency response
↓
accept that residual uncertainty remains
That applies especially well to biological systems, where the disturbances themselves are variable and the state of the organism changes continuously.
The purpose of an emergency plan is to act before complete collapse
There is also an important temporal principle hidden inside adrenal-crisis prevention.
If emergency intervention were useful only after every compensatory mechanism had completely failed, it would often arrive too late.
The purpose of recognizing deterioration is to intervene while the system is moving toward crisis.
Conceptually:
stable ↓ increased physiological stress ↓ compensatory demand rises ↓ warning signs / inability to maintain oral replacement ↓ intervention ↓ trajectory redirected
The desired outcome is to alter the trajectory before:
progressive instability
↓
severe hypotension
↓
shock / collapse
This is why the distinction between a state and a trajectory matters so much.
A single measurement tells us where the system appears to be now.
A sequence of measurements and clinical changes tells us where it is going.
Only now does the VPS comparison become useful again
After spending most of this discussion inside Addison’s disease, I can return to the computer without forcing the analogy.
The similarity is not:
cortisol = disk space aldosterone = DNS hydrocortisone = rescue mode
Those mappings would be arbitrary.
The actual structural similarity is:
| Systemic feature | Addison’s disease / adrenal crisis | VPS failure |
|---|---|---|
| Normal state | Replacement and physiological regulation maintain stable function | Ordinary routing, logging, storage and service management maintain operation |
| Underlying limitation | Adrenal cortex cannot generate normal adaptive hormone output | Finite resources and coupled network/storage dependencies |
| Disturbance | Illness, trauma, fluid loss or other physiological stress | Persistent DNS-over-WARP routing failure |
| Demand-capacity mismatch | Required glucocorticoid support exceeds available support | Error/retry/log workload exceeds sustainable system capacity |
| Reserve consumption | Haemodynamic, fluid and other compensatory capacities are increasingly stressed | Free disk and other operational margins are progressively consumed |
| Secondary effects | Hypotension, volume depletion, electrolyte and metabolic disturbance | ENOSPC, failed rotation, database errors and resolver degradation |
| Decompensation | Ordinary regulation can no longer maintain physiological stability | Ordinary services can no longer preserve system operation |
| External intervention | Emergency glucocorticoid support, fluids and treatment of precipitating cause | Out-of-band rescue environment and repair of persistent state |
| Recovery | Haemodynamic stability returns and ordinary replacement becomes viable again | Normal boot, DNS, SSH, bounded logging and adequate free storage return |
Seen this way, the VPS is no longer the main subject.
It is a small engineered example that helped me recognize a much more general pattern already visible in physiology:
disturbance
↓
compensation
↓
increasing demand on reserve
↓
loss of regulatory margin
↓
secondary interactions
↓
decompensation
↓
external intervention
↓
restoration of a viable regime
Where the analogy must stop
The comparison remains useful only if its limits are explicit.
A human organism is not engineered like a VPS.
Its regulatory systems are products of biological development and evolution. They operate through cells, tissues, neural circuits, endocrine signalling, metabolism, behaviour and interactions with the environment. They possess enormous redundancy and plasticity that have no simple correspondence in a small Linux server.
An adrenal crisis is also a life-threatening medical event. A website outage is an infrastructure problem. Their human significance is incomparable.
Nor should engineering concepts be used to infer clinical decisions. The analogy runs in the other direction: established physiology helps illuminate general ideas about regulation, reserve, feedback and decompensation.
So I would keep the comparison at this level:
same mechanisms? NO same material substrate? NO same clinical meaning? NO similar abstract regulatory structure? YES, in useful ways
That distinction matters because a good analogy preserves differences while exposing structure.
The question now moves from medicine toward dynamics
Addison’s disease has therefore given me a much more precise idea of what I meant when I said that my server “decompensated.”
Decompensation is not simply the moment something stops.
It is a transition in a regulated system:
the system begins inside
a region where regulation is sufficient
↓
disturbance changes the state
↓
regulation initially contains it
↓
available reserve becomes progressively less adequate
↓
secondary interactions alter the system itself
↓
ordinary regulatory mechanisms
can no longer return the trajectory
toward a viable region
That description works remarkably well for adrenal crisis without claiming that every clinical crisis follows one simple mathematical path.
And it creates the next question almost automatically.
If compensation, reserve and decompensation are properties of a system’s changing state, can we describe them more formally?
Can we imagine all possible physiological states as a space, identify a region in which the organism remains viable, and then describe illness or crisis as a trajectory approaching or crossing the boundary of that region?
What would stability mean in that space?
What would a tipping point mean?
How could positive and negative feedback alter the trajectory?
And if we descend from organism to organ, cell, molecule and finally physical state transitions, at what level does the property we call failure actually appear?
Those questions belong to the mathematical and cybernetic part of the discussion. But Addison’s disease has already supplied the essential intuition: health is not immobility, compensation is not unlimited, and crisis is a property of a regulatory system losing the capacity to keep itself within a viable range.
Replacement solves the deficiency without fully recreating physiological time
One of the most interesting things about Addison’s disease is that successful replacement therapy does not simply return the endocrine system to its previous architecture. The missing cortisol can be supplied pharmacologically, and mineralocorticoid activity can also be replaced when required, but an oral treatment schedule does not reproduce every temporal property of an intact adrenal system. In healthy physiology, cortisol secretion follows a pronounced circadian pattern and also changes in response to physiological demand; conventional hydrocortisone replacement instead delivers medication at discrete times, with the largest dose generally taken after waking and later doses distributed through the day because hydrocortisone has a relatively short plasma half-life (Bornstein et al., 2016). This means that treatment can restore the essential hormonal function while the timing, feedback relationships and responsiveness of the original system remain different.
That distinction matters because biological regulation occurs in time. A hormone concentration is never simply a number detached from the hour of the day, the recent history of the organism, the current illness burden or the other regulatory processes occurring around it. Cortisol normally rises toward the beginning of the active period and falls toward the night, while physiological stress can substantially alter the amount of glucocorticoid support required. Standard replacement therefore has to approximate a function that an intact endocrine system performs dynamically. The Endocrine Society recommends divided hydrocortisone dosing partly in an effort to approximate normal circadian physiology, while also acknowledging the practical limitations of available replacement strategies (Bornstein et al., 2016).
This adds another layer to the systems interpretation developed in the first two parts. The problem in Addison’s disease is not merely that the organism has lost a substance that can be poured back in until a tank is full. It has lost part of a responsive temporal control system. Pharmacological replacement restores a crucial output of that system, often very effectively, but the process of deciding how much hormonal support is appropriate during changing physiological conditions partly moves outside the adrenal gland. Normal life can therefore be stable while the architecture supporting that stability is fundamentally different from the architecture of an intact hypothalamic-pituitary-adrenal system.
For systems thinking, this is a remarkably important distinction:
Replacing a missing output is not necessarily identical to reconstructing the controller that originally generated that output.
That does not diminish replacement therapy. Quite the opposite: it shows why replacement is such a powerful intervention. A lost biological output can be supplied from outside sufficiently well for the organism to regain a viable and often highly functional state. But it also explains why stress dosing, emergency planning and alternative routes of administration remain necessary. The replacement system has to accommodate a world in which demand changes, while the damaged adrenal cortex cannot autonomously perform the corresponding adaptation (NICE, 2024; Bornstein et al., 2016).
Addison’s disease therefore changes where regulation happens
Thinking about treatment in this way changed my idea of where the boundary of a physiological regulatory system should be drawn. If I look only at anatomy, the damaged adrenal cortex is inside the body and the hydrocortisone tablet is outside it. But if I ask instead what maintains physiological stability, then medication, recognition of illness, sick-day rules, emergency injection capability and access to medical care become parts of the effective regulatory architecture. NICE reflects this explicitly by treating education, additional glucocorticoid supplies for physiological stress, emergency management kits and training in their use as integral elements of managing adrenal insufficiency (NICE, 2024).
In an intact endocrine system, much of the adjustment occurs without conscious intervention. A person does not need to notice a fever and then issue a command to the pituitary gland. The biological control system responds through interacting neural, endocrine, immune and metabolic pathways. Addison’s disease changes that arrangement because the adrenal output stage cannot respond adequately. Part of the missing adaptation can therefore become a deliberate action performed through knowledge and medication. An apparently internal physiological feedback problem acquires an external human layer.
Very abstractly, the control structure changes from:
physiological disturbance
↓
endogenous sensing and regulation
↓
adaptive adrenal output
↓
changed cortisol availability
to something more like:
physiological disturbance
↓
endogenous regulation
+
recognition of changed circumstances
+
therapeutic knowledge
+
medication availability
+
appropriate administration
↓
changed glucocorticoid availability
The second diagram should not be interpreted as a complete physiological model, because most regulation obviously remains biological and automatic. Its purpose is to expose one important architectural fact: a function that once depended principally on endogenous endocrine adaptation has become partly supported by an extended system involving treatment and human action.
This is also why emergency preparedness cannot reasonably be treated as something separate from the disease itself. An emergency hydrocortisone kit sitting unused for months may seem operationally irrelevant during ordinary life, yet it exists precisely because a severe disturbance can invalidate the assumptions under which routine oral replacement works. NICE recommends emergency kits containing intramuscular hydrocortisone for people with primary adrenal insufficiency, together with instructions and training, because suspected adrenal crisis may require an immediate response before the ordinary clinical pathway can provide treatment (NICE, 2024). A capability can therefore be central to resilience even when it contributes nothing visible during normal operation.
Resilience is not the same as never being disturbed
This brings me to a word that is often used rather loosely in both medicine and engineering: resilience. If resilience simply meant that nothing ever changed, no living organism could qualify. Biological life is continuously exposed to disturbances, some external and some generated internally. Temperature changes, fluid intake varies, infections occur, metabolic demand rises and falls, and the organism continually reorganizes itself around these changes. Homeostasis is therefore compatible with considerable motion; what matters is whether regulatory processes can keep essential variables within ranges compatible with continued function (Billman, 2020).
Primary adrenal insufficiency makes the distinction especially clear because ordinary functioning can be excellent while one important dimension of adaptive capacity is permanently constrained. With appropriate glucocorticoid and, when necessary, mineralocorticoid replacement, people with adrenal insufficiency can maintain ordinary physiological function and live full and active lives (NIDDK, n.d.). The vulnerability becomes especially important when physiological demand changes abruptly, because the endogenous adrenal response that would normally increase cortisol availability cannot be recruited in the usual way.
Resilience in this context therefore cannot mean that the organism possesses an unlimited ability to absorb stress. A more useful definition is the capacity to experience disturbance, mobilize appropriate regulatory responses and remain within—or return toward—a viable physiological regime. Treatment expands that capacity by replacing missing hormonal support, while education, sick-day dosing and emergency preparation expand it further by creating mechanisms for responding when ordinary replacement is no longer adequate (NICE, 2024).
That suggests a distinction between two superficially similar situations:
DISTURBANCE A
physiological demand increases
↓
available regulation adapts
↓
stable function preserved
↓
demand falls
↓
ordinary state restored
DISTURBANCE B
physiological demand increases
↓
available regulation becomes insufficient
↓
haemodynamic / metabolic consequences accumulate
↓
ability to compensate declines further
↓
crisis
Both begin with disturbance. Only the second develops into decompensation. The interesting explanatory variable is therefore not simply the presence of stress, but the changing relationship between stress and adaptive capacity.
The risk of crisis is hidden during successful compensation
This point is easy to underestimate because compensation is phenomenologically deceptive. When regulation succeeds, the evidence of the underlying limitation can disappear from view. If glucocorticoid replacement is adequate, mineralocorticoid replacement is appropriately managed, medication is absorbed normally and physiological demand remains within a familiar range, there may be little outward indication that the adrenal cortex itself could not generate an appropriate response if conditions suddenly changed. The therapeutic system is doing exactly what successful regulation is supposed to do: preventing the underlying defect from becoming the dominant fact of everyday physiology.
Yet successful compensation does not abolish the structural limitation. It makes that limitation manageable. This is why the concept of reserve should not be confused with current symptoms. A person may be asymptomatic while possessing less endogenous capacity to respond to a particular kind of physiological stress than an otherwise comparable person with intact adrenal function. The relevant vulnerability becomes visible only when demand rises sufficiently, replacement cannot be adjusted quickly enough, medication cannot be absorbed, or several disturbances occur together.
Dineen and colleagues emphasize that adrenal crisis usually arises in the context of an imbalance between cortisol requirement and availability, while also noting that the complete pathophysiology remains incompletely understood (Dineen et al., 2019). This is a useful formulation because it avoids treating crisis as though there were one universal concentration below which a switch marked CRISIS suddenly flips. The physiological meaning of available cortisol depends on what the organism is being asked to do at that moment.
In abstract terms:
State 1: low physiological demand + adequate replacement = compensation succeeds State 2: high physiological demand + same replacement = possible mismatch State 3: high physiological demand + fluid loss + impaired oral absorption + inadequate cortisol availability = rapidly increasing risk of decompensation
The vulnerability has not suddenly been created in State 3. The relationship among the variables has changed enough to expose it.
Adrenal crisis is better understood as a trajectory than as a photograph
This is perhaps the most useful conclusion I have reached from thinking about adrenal crisis as a systems phenomenon. A clinical crisis is of course diagnosed and treated in the present, but the physiological state observed in the emergency department has a history. It may have developed through infection, fever, gastrointestinal losses, increasing cortisol requirement, inadequate replacement, impaired absorption, worsening hypotension and other interacting processes. The sequence and speed differ from case to case, and some crises can evolve very rapidly, but the final state is still the result of a trajectory through physiological state space rather than a timeless snapshot.
This is also why monitoring during treatment is necessarily dynamic. NICE recommends frequent assessment of blood pressure, heart rate, electrolytes and glucose during adrenal crisis and continuation of hydrocortisone and intravenous saline until haemodynamic stability has been restored (NICE, 2024). A single blood-pressure value or sodium measurement can provide important information, but recovery is ultimately about whether the organism is moving toward a more stable regime in which ordinary maintenance becomes possible again.
A useful conceptual distinction is therefore:
STATE: Where is the organism now? TRAJECTORY: Where is the organism moving? REGULATORY CAPACITY: Can the available mechanisms redirect that trajectory toward stability?
The third question is what turns a description of physiology into a description of resilience. Two people could theoretically have similar measurements at one moment while differing substantially in what happens next because their circulating volume, precipitating illness, hormonal support, absorption, treatment timing and other conditions differ. A state vector without dynamics tells only part of the story.
Adrenal crisis therefore invites a very different conception of clinical deterioration from a simple list of abnormal values. Hypotension, hyponatraemia, altered mental state or weakness are important clinical observations, but they are manifestations of a system whose regulatory relationships are changing. The crisis exists at the level of that changing organization.
The most dangerous transition may be a loss of recoverability
When I first began using the word decompensation, I thought mainly about the failure to maintain current physiological values. The deeper idea is that decompensation can also involve a loss of the system’s ability to restore itself using its ordinary regulatory mechanisms. This distinction is particularly visible when gastrointestinal illness interferes with oral glucocorticoid replacement. At exactly the moment when physiological demand can be increasing, the ordinary route used to provide the missing hormone may become unreliable. The disturbance is therefore affecting both the regulated organism and part of the mechanism by which the organism is usually supported.
NICE makes the practical significance explicit: when prolonged vomiting or diarrhoea prevents adequate absorption of oral glucocorticoids, parenteral hydrocortisone and hospital treatment are required, while suspected adrenal crisis should be treated immediately with intramuscular or intravenous hydrocortisone (NICE, 2024). From a systems perspective, this is striking because the intervention does more than increase the amount of support. It changes the pathway through which support enters the system.
Before deterioration:
oral replacement
↓
gastrointestinal absorption
↓
circulation
↓
glucocorticoid action
During severe gastrointestinal disturbance:
oral replacement
↓
vomiting / impaired absorption
X
↓
required hormonal support may not arrive
The emergency pathway changes the topology:
intramuscular / intravenous hydrocortisone
↓
systemic delivery
↓
ordinary gastrointestinal route bypassed
Again, I am describing the architecture, not offering a treatment protocol. What interests me is the general principle exposed by established medical practice: when crisis compromises the normal route of compensation, effective rescue may require a route that does not share the same dependency.
Emergency treatment does not simply reverse one biochemical variable
The systemic character of adrenal crisis becomes even clearer when looking at what emergency management actually tries to restore. Immediate hydrocortisone is essential, but glucocorticoid replacement is accompanied by intravenous isotonic saline because volume depletion and haemodynamic instability are also central concerns. Blood pressure, heart rate, electrolytes and glucose are monitored while the precipitating cause is identified and treated (NICE, 2024). The intervention therefore acts on several relationships that have become disturbed during the crisis.
This is conceptually important because it prevents us from imagining a simple reversible chain in which cortisol falls, a crisis appears, cortisol is replaced and the organism instantly returns to its previous state. The missing hormonal support may be the defining endocrine defect, but the crisis state can include accumulated consequences whose restoration has its own dynamics. Circulating volume may need to be replenished, vascular responsiveness may recover over time, electrolyte abnormalities may require observation, and the underlying infection or other precipitating stressor continues to influence the physiological system.
In a simplified systems diagram:
PRECIPITATING STRESS
↓
increased physiological demand
↓
inadequate cortisol availability
↓
haemodynamic / metabolic consequences
↓
additional deterioration
↓
CRISIS STATE
RECOVERY
glucocorticoid support restored
+
circulating volume restored
+
relevant variables monitored
+
precipitating cause treated
↓
trajectory redirected toward stability
The important idea is that crisis acquires its own state. Once the organism has moved sufficiently far from its previous equilibrium, treatment must address the state that now exists, not merely the event that began the movement.
Cause, precipitant and current physiological state must remain separate
This distinction is especially valuable because everyday language tends to compress causality. We might say that an infection “caused” an adrenal crisis, and clinically that may be a useful shorthand, but the statement hides several layers. The infection is a precipitating stressor. Primary adrenal insufficiency supplies the underlying limitation in endocrine adaptive capacity. The mismatch between demand and glucocorticoid availability participates in producing the crisis, while volume depletion, hypotension, impaired absorption and other secondary processes can then contribute to the evolving physiological state. By the time emergency treatment is required, the immediate problem is no longer identical to the original precipitating event.
The causal structure is closer to:
UNDERLYING VULNERABILITY
primary adrenal insufficiency
|
v
PRECIPITATING STRESS
infection / trauma / surgery / other stress
|
v
DEMAND-CAPACITY MISMATCH
required cortisol support exceeds availability
|
v
SYSTEMIC CONSEQUENCES
haemodynamic / metabolic / fluid disturbances
|
v
CRISIS STATE
This explains why management has to do several things simultaneously: restore glucocorticoid support, restore haemodynamic stability where necessary, and identify and treat the underlying precipitating cause (NICE, 2024). The root cause of vulnerability, the trigger of this particular crisis and the physiological state requiring treatment are related, but they are not interchangeable.
This separation also prevents a common conceptual mistake in systems analysis: assuming that removing the initiating disturbance automatically restores the previous state. If an infection triggered a cascade involving vomiting, volume depletion and adrenal crisis, controlling the infection does not instantaneously replace cortisol, restore circulating volume or reverse every downstream consequence. The path travelled through the crisis has changed the state of the organism.
History therefore becomes part of physiology
This is what I meant in Part II by path dependence, and I think it deserves a fuller treatment. A regulated biological system remembers its recent past in the most concrete possible sense: current hormone concentrations, circulating volume, electrolyte distributions, inflammatory state, metabolic reserves and receptor responses all depend partly on what has already happened. The body does not need a symbolic memory of yesterday’s illness for yesterday’s illness to influence today’s physiological state. The current material configuration is the record of that history.
Imagine a simplified sequence in which an infection raises physiological demand, fever increases fluid loss, appetite and intake decrease, vomiting begins, oral replacement becomes unreliable and circulatory volume falls. At each step, the starting point for the next event has changed. The same additional disturbance therefore has a different effect depending on when it occurs. A small amount of further fluid loss near the beginning of the process may be easily compensated; the identical loss after substantial volume depletion and hypotension may have much greater consequences.
We can represent that idea abstractly:
S0 --disturbance A--> S1 S1 --disturbance B--> S2 S2 --disturbance C--> S3 S3 --disturbance D--> crisis The effect of D depends on S3, not merely on the intrinsic size of D.
This is one reason the language of “the trigger” can be misleading if taken too literally. A final event may appear to precipitate sudden collapse while the system’s susceptibility to that event was created by everything that came before it. The crisis is temporal organization, not simply a collection of simultaneous abnormalities.
Successful treatment is also path dependent
The same principle applies in the opposite direction. Recovery does not teleport the organism from crisis back to its pre-crisis state. Emergency treatment changes several variables, and those changes alter the conditions under which subsequent regulation occurs. Hydrocortisone restores glucocorticoid support; intravenous fluid expands circulating volume; improving haemodynamics alters tissue perfusion; treatment of the underlying illness reduces the continuing disturbance. As these changes accumulate, ordinary mechanisms that were previously insufficient may become effective again.
This is why the NICE transition criterion is so interesting from a systems perspective. Parenteral hydrocortisone is continued until the person is haemodynamically stable and can again take and absorb oral glucocorticoids; saline is likewise continued according to haemodynamic and electrolyte status until stability is restored (NICE, 2024). Recovery is therefore defined partly by the restoration of a state in which the ordinary maintenance pathway has become viable again.
The architecture is:
ordinary maintenance
↓
disturbance exceeds adaptive capacity
↓
crisis
↓
emergency regulatory support
↓
state progressively restored
↓
ordinary maintenance becomes sufficient again
I find this much more useful than thinking of emergency treatment as a temporary “boost.” The purpose is to move the organism into a different dynamical regime: one in which routine replacement and ordinary physiological regulation can once again maintain stability.
Preparedness is an attempt to intervene on the trajectory before the endpoint
Adrenal-crisis prevention is also best understood temporally. Sick-day rules, emergency identification, access to injectable hydrocortisone and education are not primarily attempts to make adrenal insufficiency disappear; they exist because the direction of physiological change matters. If increased demand is recognized early enough and additional glucocorticoid support is supplied appropriately, a trajectory that might otherwise move toward crisis can potentially be redirected while the system remains much closer to its normal operating range (NICE, 2024; Dineen et al., 2019).
That suggests a general structure:
stable state
↓
physiological stress appears
↓
demand begins to rise
↓
early adaptation / sick-day response
↓
trajectory redirected
↓
crisis avoided
The difficult part is that the boundary between ordinary illness and dangerous decompensation is not represented by one universal observable threshold. Real illnesses vary, symptoms overlap, and adrenal crisis itself does not have one perfectly uniform definition across every study or clinical context. Dineen et al. discuss this definitional difficulty while emphasizing the importance of prompt recognition and treatment (Dineen et al., 2019). In practice, established clinical plans and guidelines therefore focus on recognizable risk situations and on acting safely when crisis is suspected rather than waiting for theoretical certainty.
This is an important lesson in uncertainty. A safety-critical response does not always require complete causal knowledge before action. The cost of waiting for perfect epistemic certainty may be much greater than the cost of acting on a well-founded suspicion, which is why NICE explicitly recommends immediate hydrocortisone when adrenal crisis is suspected (NICE, 2024).
Preparedness reduces risk without turning physiology into a deterministic machine
There is a temptation in systems engineering to believe that enough monitoring, documentation and fallback mechanisms can eventually eliminate surprise. Addison’s disease is a useful corrective to that confidence. Education and emergency preparation are extremely important, yet the biological environment remains variable and partly unpredictable. Intercurrent illness can evolve unexpectedly, gastrointestinal absorption can change, physiological stress can increase rapidly, and individuals do not all respond identically. The aim of resilience is therefore not to abolish uncertainty but to make dangerous trajectories more detectable and more recoverable.
This is also why Dineen et al. place such emphasis on education and access to parenteral glucocorticoids while simultaneously acknowledging that adrenal crises continue to occur (Dineen et al., 2019). The existence of an emergency plan is evidence of a mature understanding of uncertainty, not evidence that uncertainty has been eliminated. A resilient system expects that normal assumptions can fail and prepares another response for that possibility.
That distinction can be summarized as:
fragile safety model: "If everyone follows the normal procedure, the crisis should not happen." resilient safety model: "Normal management reduces risk, but abnormal states remain possible, so detection and emergency recovery must also be designed."
Medicine has had to learn this because the consequences of pretending otherwise are immediate. Engineering systems often learn the same lesson only after an incident reveals that all of their “backup” paths depended on the same failed assumption.
The body has no single central homeostasis daemon
Another reason Addison’s disease makes such an interesting systems example is that homeostasis is profoundly distributed. There is no organ whose complete job description is “keep the organism healthy.” The hypothalamus, pituitary, adrenal glands, autonomic nervous system, cardiovascular system, kidneys, immune system, liver, pancreas and many other tissues participate in regulation, often through overlapping feedback loops. The stability of the organism emerges from their interaction (Billman, 2020).
Even the adrenal gland is not a master controller. Cortisol affects many tissues, but those tissues have their own dynamics and regulatory relationships. Aldosterone influences renal sodium and potassium handling, yet the kidneys simultaneously respond to haemodynamic signals, other hormones and local mechanisms. Blood pressure emerges from cardiac output, vascular resistance, circulating volume, autonomic activity and many other factors. Glucose concentration emerges from the interaction of intake, hepatic production, tissue utilization, insulin, glucagon, cortisol and additional regulatory processes. No single component contains the global state.
This distributed architecture is one reason systemic pathology can become difficult to understand by inspecting components individually. During adrenal crisis, one could theoretically find many cells behaving exactly as their local biochemical environment dictates while the organism as a whole moves toward haemodynamic collapse. The problem exists in the relationships among components and in the trajectory those relationships generate.
There is an obvious echo of the VPS here, although I want to keep it secondary. During that incident, individual software components also continued doing locally understandable things: callers retried, WARP logged failures, rsyslog preserved messages, the filesystem accepted writes until it could not, and services continued attempting their normal work. The catastrophic property belonged to the combined feedback structure. The biological case is vastly more complex, but it makes the same abstract point much more clearly: global stability is not guaranteed by local correctness.
Homeostasis includes the capacity to change the response itself
There is another subtlety worth emphasizing. A homeostatic system does not merely oppose every change with an equal and opposite reaction. Its response can depend on context, history and scale. The cardiovascular response appropriate for standing up is different from the response required during severe infection. Fluid regulation after drinking a glass of water is different from regulation during prolonged gastrointestinal loss. The endocrine response required during quiet sleep is different from that required during trauma or surgery. Biological stability is therefore achieved through adaptive regulation, not through blindly holding every variable fixed (Billman, 2020).
Addison’s disease selectively constrains one part of that adaptive flexibility. Basal glucocorticoid replacement may be sufficient during ordinary conditions, but physiological stress requires a change in treatment because the adrenal cortex cannot generate the corresponding change itself (NICE, 2024). This is why saying simply that “cortisol is replaced” misses something conceptually important. What is difficult to reproduce is not only the amount of hormone but the ability to adapt hormone availability continuously to context.
The distinction can be expressed as:
STATIC REPLACEMENT QUESTION "Is enough cortisol present?" DYNAMIC REGULATORY QUESTION "Is cortisol availability appropriate for this organism, in this state, under this level of physiological stress, at this moment?"
The second question is much closer to what biological regulation actually has to solve.
There is no single quantity called compensatory reserve
At this stage the word reserve also needs to be handled carefully. It is useful, but there is a danger of imagining a literal hidden tank from which the body withdraws “reserve units” until the tank reaches zero. Physiological reserve is distributed and multidimensional. Cardiovascular responsiveness, intravascular volume, renal function, metabolic substrate availability, endocrine adaptation, autonomic regulation and many other capacities contribute to whether the organism can tolerate a particular disturbance. A person may therefore possess ample capacity in one dimension while being constrained in another.
Addison’s disease is a clear example because the limitation is selective. Many organs may be structurally intact, and most regulatory systems may function normally, while adrenal glucocorticoid output remains unable to adapt endogenously to stress. Replacement compensates for that deficit, but the margin available during an acute illness depends on much more than adrenal function alone. The severity of the illness, ability to maintain fluid intake, gastrointestinal absorption, timing and adequacy of glucocorticoid replacement, mineralocorticoid status and the behaviour of the cardiovascular system all matter.
It therefore makes more sense to define reserve relationally:
reserve = the remaining capacity of the whole system to absorb additional disturbance and still reorganize toward a viable state
Seen this way, reserve cannot be interpreted without specifying the disturbance. “How much physiological reserve remains?” is incomplete unless we also ask: reserve against what kind of stress, over what period, given what starting state?
A crisis is the loss of a viable relationship among variables
This brings us to the idea that will eventually connect the medical essay to the mathematical one. Suppose we imagine the organism not as a list of organs but as a point in an enormous multidimensional state space. One dimension might represent effective circulating volume, another arterial pressure, another glucocorticoid availability, others sodium, potassium, glucose, temperature, inflammatory state, renal perfusion and thousands of additional quantities. At every moment the organism occupies some point in that space, and physiological regulation moves that point continuously as conditions change.
Most ordinary states occupy a broad region compatible with continued life and function. They do not have to be identical. Blood pressure can rise during exercise, heart rate can fall during sleep, cortisol varies through the day, and glucose changes after meals. Viability therefore corresponds to a region of acceptable dynamic states, not to one ideal coordinate. Homeostasis is partly the capacity to remain within that region or return toward it after disturbance.
Addison’s disease changes the dynamics governing movement through this space because one important regulatory pathway has lost endogenous adaptive capacity. During ordinary replacement, the trajectory can still remain comfortably inside the viable region. During sufficiently severe physiological stress, however, the same limitation can become consequential because the forces moving the state have changed. If glucocorticoid support becomes inadequate, volume falls, vascular responsiveness deteriorates and additional disturbances accumulate, the trajectory can move toward a boundary where ordinary regulation is no longer capable of reversing it.
I can write this schematically without claiming that it is a clinical equation:
x(t + Δt) = F(x(t), u(t), r(t))
where:
x(t) = current physiological state
u(t) = disturbances
illness, trauma, fluid loss, etc.
r(t) = available regulatory support
endogenous mechanisms
+ replacement therapy
+ emergency intervention when required
The important question becomes whether the resulting trajectory remains inside a viability region V:
compensated: x(t) ∈ V and regulation can keep it there decompensating: x(t) ∈ V but trajectory is approaching a boundary and regulatory margin is shrinking crisis: ordinary regulatory mechanisms are insufficient to keep or return the trajectory to a viable regime
This is only a conceptual model, and the physiology of adrenal crisis should not be forced into a simplistic mathematical threshold. But the model helps clarify why decompensation is a systems property. No single molecule crosses the boundary. The trajectory of the organized organism does.
At the molecular level there is no object called adrenal crisis
If I keep descending through levels of description, the comparison becomes even more interesting. At the clinical level we can say with complete seriousness that a person is experiencing an adrenal crisis. At the organ level we can discuss adrenal cortical insufficiency. At the cellular level we can discuss steroidogenic cells, receptors, membrane transport and signalling. At the molecular level we find cortisol molecules, sodium ions, potassium ions, proteins, lipids and water. Descending further, we reach atoms, electrons and ultimately physical interactions described by chemistry and physics.
At no point do we encounter a special particle labelled ADRENAL_CRISIS.
A sodium ion behaves according to electrochemical forces whether the organism is healthy or critically ill. A cortisol molecule does not know whether it was produced by an adrenal gland or injected as hydrocortisone. A receptor follows molecular interaction dynamics. Water continues moving according to physical and osmotic forces. The underlying physical processes remain lawful throughout the crisis.
And yet the crisis is completely real.
This is possible because reality can contain properties that become meaningful only at particular organizational scales. Blood pressure is not a property of one blood molecule. Circulating volume is not a property of one water molecule. Hormonal regulation is not a property of one cortisol molecule. Homeostasis is not a property of one cell. Adrenal crisis describes a pathological dynamical state of the organism produced by relationships among all of these lower-level processes.
The absence of a microscopic “crisis object” therefore does not make crisis subjective or imaginary. It means that the relevant property belongs to a higher level of organization.
The physics can all be correct while the organism is physiologically wrong
This is where the medical case opens a philosophical problem that I do not want to solve inside this article but cannot completely ignore. During adrenal crisis, every molecule continues obeying physics. There is no moment when sodium ions violate electrodynamics or receptors stop obeying chemistry. From the standpoint of fundamental physical law, the transition from stable physiology to hypotension and shock is simply another sequence of permitted state transitions.
Yet medicine describes one trajectory as healthy and another as pathological.
That judgment is not arbitrary. The distinction is grounded in the organization required for the organism to maintain itself. A blood pressure incompatible with adequate perfusion is not merely an unusual number; it threatens the continuation of organismal function. Severe disturbance of circulating volume, electrolytes, glucose or glucocorticoid support matters because the organism depends on particular relationships among these variables to remain viable.
So at different descriptive levels we can truthfully say both:
PHYSICAL DESCRIPTION matter is undergoing lawful state transitions PHYSIOLOGICAL DESCRIPTION the organism is failing to maintain a viable internal organization
There is no contradiction between them. The word failure enters when we describe those physical transitions relative to the functional organization of a living system.
Illness therefore teaches us something unusual about function
Addison’s disease makes this especially visible because the concept of function is impossible to avoid. We say that the adrenal cortex should produce cortisol, that cardiovascular regulation should maintain adequate perfusion, and that endocrine responses should adapt to physiological stress. These statements sound normative: they contain words such as should, adequate and normal. Physics by itself contains none of those terms. Physics tells us what state transitions occur under given conditions; it does not label one trajectory successful and another pathological.
Biology, however, contains systems whose organization depends on continued self-maintenance. If blood pressure falls beyond a certain range for long enough, tissue perfusion becomes inadequate. If the organism cannot restore a compatible state, damage accumulates and survival becomes threatened. In that context, calling the state pathological is not simply an observer’s aesthetic preference. It describes a real relationship between the current dynamics and the requirements of the living system’s continued organization.
Addison’s disease therefore leads naturally toward questions about biological function, teleology, normativity and viability. Those are questions for the later philosophical essay. For the present discussion, the important point is simpler: the meaning of failure appears at the level where organization and continued viability become relevant.
The small VPS analogy helps here, but only at the edge
This is one place where I think the original server incident can briefly return without taking over the medical essay. At the transistor level, nothing in my VPS became metaphysically “wrong” when DNS failed or the disk reached ENOSPC. Transistors changed state according to electronic rules; memory cells stored bits; the processor executed instructions; filesystem code attempted writes and returned errors when no space remained. The catastrophe existed at a higher organizational level where I expected the machine to provide DNS, preserve writable storage, serve a website and remain administratively accessible.
The resemblance to adrenal crisis is therefore not between particular components but between levels of explanation:
PHYSICAL LEVEL
lawful state transitions
↓
ORGANIZATIONAL LEVEL
feedback
regulation
resource relationships
↓
SYSTEM LEVEL
stable / unstable
compensated / decompensated
viable / failed
The biological case is much richer because the organism maintains itself through evolved physiological regulation and because the consequences involve health and life rather than service availability. Still, the computer incident gave me a concrete way to notice a structure that physiology expresses much more profoundly.
Where the analogy finally has to stop
At this point it is worth stating the boundary clearly because an analogy becomes less useful when it starts claiming identity. Addison’s disease is not “a bug in the body.” The adrenal cortex is not a failed microservice, cortisol is not a packet, hydrocortisone is not a software patch, and emergency medicine cannot be understood adequately by importing engineering vocabulary. Biological organisms are developmental, evolutionary, adaptive systems whose regulatory complexity vastly exceeds the little VPS that started this reflection.
The human significance is also completely different. An adrenal crisis can threaten a person’s life. A VPS outage threatens a website and perhaps some data or services. The comparison should never flatten that distinction or use technological language to trivialize illness. The medical discussion has value precisely because established knowledge about adrenal insufficiency gives us a concrete example of regulation, reserve and decompensation whose reality cannot be reduced to a metaphor.
What survives after those differences are respected is a more abstract shared structure:
both are complex systems both exist in changing environments both depend on regulatory relationships both can tolerate some disturbances both possess finite adaptive capacity both can preserve outward function while internal margin is shrinking both can enter feedback regimes that ordinary regulation cannot reverse both may require an intervention outside the failing ordinary pathway both recover when a state is restored in which ordinary regulation becomes viable again
That is enough. The analogy does not need to do more.
What Addison’s disease ultimately changed in my understanding of crisis
Before thinking through this comparison, I tended to imagine failure as an event: something works, a fault occurs, and then it does not work. Addison’s disease suggests a more interesting model. A complex system can contain a permanent limitation and nevertheless remain stable for years because other mechanisms, including external treatment, successfully compensate for it. A disturbance can then alter the demand placed on those mechanisms. For some time the system may continue functioning while using progressively more regulatory capacity. If the mismatch becomes too large, secondary physiological consequences can accumulate, the ordinary maintenance pathway can become unreliable, and the system can enter a state that requires a qualitatively different form of intervention.
Seen this way, adrenal crisis is not adequately described by one missing hormone, one abnormal laboratory value or one final moment of collapse. It is a dynamical state in which the relationship among physiological demand, hormonal support, fluid balance, vascular responsiveness, metabolic regulation and other processes has moved beyond what ordinary compensation can safely contain. Emergency treatment works by changing those relationships strongly enough to redirect the trajectory toward a region where normal maintenance becomes possible again (NICE, 2024; Dineen et al., 2019).
That gives me a much richer vocabulary for thinking about illness:
health
is not immobility
homeostasis
is not one fixed set point
replacement
is not identical to restoring
the original controller
compensation
is not proof of unlimited reserve
reserve
is not one hidden quantity
crisis
is not simply the final visible collapse
recovery
is not merely removal of the first trigger
Instead, the story is about relationships changing through time.
From Addison’s disease to a more general theory of decompensation
The medical analogy has therefore taken me as far as I want to go in this essay. It began with a slightly ridiculous joke about a computer having an adrenal crisis, but following the comparison carefully forced me to distinguish mechanisms from structures. At the mechanistic level, the two cases have almost nothing in common. At the systems level, however, Addison’s disease provides a remarkably rich example of a permanent regulatory limitation that can be compensated under ordinary conditions, exposed under increased demand, amplified by interacting disturbances, rescued through an alternative intervention pathway and returned to a stable maintenance regime.
More importantly, the comparison made me realize that decompensation is not simply a medical word for “getting much worse.” It describes a transition in the relationship between a system and its own regulatory capacity. A compensated system can still contain vulnerability; a stable observable output can coexist with declining margin; and the decisive threshold may be reached when the system loses the ability to redirect its own trajectory using the mechanisms normally available to it.
The next step is therefore mathematical rather than medical. If an organism can occupy many possible physiological states, if only some of those states are compatible with continued viability, and if feedback determines how the state moves under disturbance, then concepts such as state space, viability region, stability, attractor, positive feedback, tipping point and control begin to offer a more formal language for what this essay has described qualitatively.
After that comes an even stranger descent: if the organismal crisis can be decomposed into cellular activity, molecular interactions, ion movements and finally lawful physical state transitions, where exactly does the property called failure enter the description? And if physics itself contains only what happens, where do biological concepts such as function, pathology, successful regulation and “what should happen” come from?
Those questions deserve separate essays because answering them properly requires leaving clinical physiology and moving into cybernetics, dynamical systems, philosophy of biology and eventually normativity. For now, the lesson I want to keep from Addison’s disease is much more concrete and, I think, more useful:
A complex system is healthy not because nothing disturbs it, but because it retains enough regulatory capacity to absorb disturbance, preserve viability and return toward a sustainable state. Decompensation begins when that relationship breaks down; crisis emerges when ordinary regulation is no longer enough; and recovery succeeds when a viable regulatory regime becomes possible again.
That is what the VPS unexpectedly made me notice. Addison’s disease is what made me understand it.
References
Billman, G. E. (2020). Homeostasis: The underappreciated and far too often ignored central organizing principle of physiology. Frontiers in Physiology, 11, 200. https://doi.org/10.3389/fphys.2020.00200
Bornstein, S. R., Allolio, B., Arlt, W., Barthel, A., Don-Wauchope, A., Hammer, G. D., Husebye, E. S., Merke, D. P., Murad, M. H., Stratakis, C. A., & Torpy, D. J. (2016). Diagnosis and treatment of primary adrenal insufficiency: An Endocrine Society clinical practice guideline. The Journal of Clinical Endocrinology & Metabolism, 101(2), 364–389. https://doi.org/10.1210/jc.2015-1710
Dineen, R., Thompson, C. J., & Sherlock, M. (2019). Adrenal crisis: Prevention and management in adult patients. Therapeutic Advances in Endocrinology and Metabolism, 10, 2042018819848218. https://doi.org/10.1177/2042018819848218
National Institute for Health and Care Excellence. (2024). Adrenal insufficiency: Identification and management (NICE guideline NG243). https://www.nice.org.uk/guidance/ng243
National Institute of Diabetes and Digestive and Kidney Diseases. (n.d.). Definition & facts of adrenal insufficiency & Addison’s disease. National Institutes of Health. https://www.niddk.nih.gov/health-information/endocrine-diseases/adrenal-insufficiency-addisons-disease/definition-facts