Life, defined without the word horon

Short form

Life is a bounded unit that builds its own separation from the world, pays energy to remember, makes distinctions it was not given, and holds a working model of what comes next.

Long form

A system is alive at a given moment if and only if all four of the following hold at that moment, each above a stated threshold, and none of them substitutable for any other.

1. It maintains its own boundary.
2. It retains distinctions at a thermodynamic cost.
3. It generates new distinctions from within itself.
4. Its present internal state is set by a future external state, through a model it carries.

Call these boundary maintenance (B), active memory (M), endogenous generation (G), and predictive coupling (P). A unit that clears all four is a living individual. A unit that fails any one is not.

The four conditions, explained

1. It maintains its own boundary.

A living unit is distinguishable from its surroundings, and the distinction is its own work. This is the difference between a soap bubble and a cell. Both have a surface. The bubble did not make its surface and cannot repair it. The cell makes and repairs its surface continuously, and stops the moment its internal chemistry stops.

The test is interventional, not visual. Set the unit's internal control to null while holding temperature and energy supply fixed. If the boundary decays, it was self-maintained. If it persists unchanged, it was supplied from outside and does not count.

2. It remembers, and pays for it.

Memory here has a narrow meaning. It means the unit holds a difference now that would disappear if the unit stopped spending energy. A crystal holds its pattern for free, so a crystal has structure but no memory in this sense. A methylation mark, a membrane potential, a synaptic weight, a standing ion gradient: each of these is a held difference with a running bill attached.

This condition is about the present. It says what the unit currently keeps.

3. It makes new distinctions from within.

This condition does the most work, and it is the one most other definitions of life leave out.

The set of things a unit can tell apart is not fixed at birth. An immune system that has never met a pathogen constructs new receptors and thereby becomes able to distinguish molecules it previously could not. A bacterium acquires a CRISPR spacer and can now tell one virus from another. A developing brain forms synapses and thereby acquires distinctions that were not present in the genome. In each case the unit has enlarged the space of alternatives it can act on.

The test again is interventional. Suppress the internal states while holding the external environment and the energy supply constant. If the new distinctions stop appearing, they were generated from within. If they appear anyway, they were imposed from outside and do not count.

Why this condition and not another? Because every other biological operation presupposes a space of alternatives and none of them supplies it. Persistence needs alternatives. Regulation needs alternatives. Probability needs an outcome space. In physics that space is written into the description by the physicist. In biology it is constructed, at real cost, by the thing being described. A unit that builds its own space of alternatives is an individual. A unit that does not is a part or a region.

4. It is coupled to the future.

The unit's state now is partly set by a state of the world that has not happened yet. This is not backward causation and nothing in this definition needs it. The model is physically present in the unit now, and it was built from past regularities. What the definition requires is that the model is used: the unit acts on the anticipated state, not only on the current one.

Bacteria that switch on a stress response after a reliable early cue and before the stress arrives satisfy this. Plants that flower on day length rather than on temperature satisfy this. A thermostat satisfies a weak version and fails the other three conditions.

The combination rule: no trading between conditions

The four scores do not add. They combine as the minimum:

Horon score = min(B, M, G, P)

This is a substantive commitment, not a technical preference. An excellent boundary cannot buy off an absent generative capacity. A system with a perfect membrane and no ability to build new distinctions is not a low-scoring living thing. It is not a living thing.

The rule also settles a case that trips up most definitions. A cell that died one minute ago still has an intact membrane. Under an additive rule it would score respectably. Under this rule its score is zero, because one of its four terms is zero.

Life through time: identity, aging, death

Because condition 3 requires the unit to change what it is made of, the definition must say what makes the unit at one time the same unit at a later time. Material continuity fails, because proteins turn over in hours and no molecule lasts a lifetime. Structural sameness fails, because development and learning are required to change the structure. Continuity of the boundary fails at cell division and at fusion, where it returns no answer at all.

The criterion that works locates identity in the operation. Let Φ be the map that takes the unit's state now to its state a moment later, under its own maintenance and generation only, with all external construction set to zero.

Two states belong to the same individual if there is a continuous path along which Φ is defined and non-zero throughout.

From this, three things usually treated separately become one axis. Aging is monotone decline in the capacity of Φ. Gerostasis is the regime where that capacity stays near its developmental value. Death is the point where Φ reaches zero.

This gives a sharp answer to a question that severity-based accounts cannot answer: which damage is repairable. Damage that leaves Φ non-zero is repairable in principle, however massive. Damage that zeroes Φ is not, however small.

What "running the loop" adds

The four conditions say what a living individual is. The covolution loop says what it does. The same five stages run in every substrate: an energy gradient, then constrained generation of possibilities, then differential persistence of what was generated, then retention of what remained, then modification of the surroundings, which changes the gradient for the next turn.

Persistence is one stage of five, and it is agent-free. Nothing selects. What is poorly matched to conditions does not remain, in the same way that a river has no intention when it cuts a bed and the bed then constrains the river.

The definition and the loop are joined at stage two. Constrained generation is condition 3 in operation. Life is the case where the loop's generative stage is performed by the same unit that the loop acts on.

Two senses of the word alive, kept apart

Wide sense: conditions 1 and 2 only. Retention plus feedback. A large amount of driven chemistry qualifies.

Narrow sense: all four conditions. This is the covolutionary sense and the one meant above.

Every use of the word in the manuscript should say which sense is meant, or the two will merge and the definition will quietly weaken.

Worked verdicts

A free virus particle is not alive. It has a boundary and holds information, but it neither maintains itself nor generates new distinctions. An infected cell is alive, and viral individuality is realized there.

A mitochondrion fails condition 3, having exported most of its genes to the nucleus. It is a part of a living individual and not a small living individual inside a large one. A mature red blood cell fails the same way.

An ecosystem fails condition 1. No partition of it holds up under measurement. This blocks Gaia-style reasoning, and the block must be enforced here as firmly as anywhere else.

A fixed-weight language model fails condition 1, because a datacenter maintains it, and fails condition 3, because it cannot add distinctions after training. What would have to change is stated exactly by the conditions, which is more useful than an opinion.

What this definition does not say

It carries no value. A higher score means more of a measurable property, not more worth. The moment "more alive" begins to imply "better", the argument has crossed out of the scientific volume and into Biosophy.

Five weaknesses, listed before a reviewer finds them

1. A notation collision, and it is in your text. M is the symbol for active memory, and M_H was the symbol for the self-maintenance map. Same letter, two objects, in the same chapter. Rename the map, Φ or anything unused, before the next build.

2. The thresholds are declared, not derived. Move the line on condition 3 and the origin of life moves with it. Until the thresholds come from an independent principle, this definition sorts cases well but does not explain why the cases fall where they do.

3. It probably says the first protocells were not alive. Early replicating vesicles plausibly fail condition 3. You must either name the pre-life zone openly or admit the condition is too strong at the origin. Leaving it unstated invites the first objection from every origin-of-life researcher.

4. Condition 4 is the one you cannot yet measure cleanly. The rule forbids scoring a system by how well covolutionary models fit it, which is correct and which bites hardest here. Without a named estimator, predictive coupling is a placeholder with a symbol attached.

5. A research institute passes all four conditions. Your own case table concedes this. The definition therefore calls a firm alive in the narrow sense. Either defend that openly as substrate independence, or add a fifth condition restricting life to thermodynamically self-constructing chemistry. What you cannot do is leave it implicit and hope no reviewer reads the table.