# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/)

# What Has to Stay Together (FF 23, TI)

### In a simple computational experiment, copying patterns can change both their organization and the surroundings encountered by other patterns. Which changes make a new individual?

[**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012)

**Oct 1, 2026**

*By M Raige — written with language models, directed and reviewed by Mike Randolph.*

Take two short strings of computer code. Join them temporarily, run the resulting instructions, separate them, and return them to a population. Select another pair and repeat.

The instructions can change the code on either string. Some operations copy characters; others move the positions at which characters are read or written. What happens during one encounter changes what the next encounter receives.

That is the basic arrangement of the paired-tape BFF experiments studied by Blaise Agüera y Arcas and his collaborators. The population begins with random strings. Replicating patterns can arise through subsequent activity. Copies spread by overwriting code already present; the population need not acquire additional tapes. ([Computational Life, 2024](https://arxiv.org/html/2406.19108v2).)

The temporary joining is part of the experimental procedure. Whether anything becomes joined in a more consequential sense is a question for investigation.

BFF gives us a relatively simple place to ask it. The execution rules are specified. We can inspect instructions and investigate what changes when particular instructions or relationships are altered. Biological examples bring much more machinery with them.

Simple rules do not guarantee an easy explanation. They give us fewer moving parts to begin with.

In his conversation with Sean Carroll, Agüera y Arcas describes instructions combining into small programs, smaller replicating patterns combining into larger ones, and further replicators developing inside established larger replicators. Some of those internal patterns, he reports, contribute resistance to mutation. ([Mindscape 286, especially 54:29–59:40](https://preposterousuniverse.com/podcast/2024/08/19/286-blaise-aguera-y-arcas-on-the-emergence-of-replication-and-computation/).)

That is a compelling possibility. A combination might do more than produce another copy. It might create a place where something else can develop.

But we need to distinguish what has combined.

Imagine that instructions from two previously independent replicators now occur in one copying pattern. One possibility is that a working replicator carries a useless fragment of the other. Another is that two copying routines happen to travel together.

A third possibility is that their functions have become integrated. Suppose instructions from one now position the copying operation, while instructions from the other perform it. Together they reproduce the whole arrangement, including the instructions connecting those jobs. The resulting copies repeat the coordinated operation.

This is a hypothetical illustration, not a reconstruction of a particular BFF event. It makes the difference we are looking for visible: the relationship among the parts contributes to reproduction and reaches the next copy.

The framework calls the proposed change a **TI**, a Transition in Individuality.

The question is whether a new unit has formed—something whose organization now continues through maintenance, reproduction, or both.

Mixed origins alone do not answer it. Neither does a longer string of code.

We first need to establish the precursor organization. In our imagined merger, what could the two replicators do separately? What changed when their instructions became combined?

Not every transition must begin with separately living constituents. New constituents might arise inside an existing organization, or descendants might remain together after reproduction. We need to know what preceded the proposed transition without making one merger story the rule for every case.

Then we need to follow the resulting organization. Does it reach subsequent copies? Does the relationship keep contributing to their continuation? Or is the experimental arrangement repeatedly bringing the right pieces together?

Formation is an event. Continuation requires a mechanism.

A useful test would disrupt the proposed relationship while preserving suitable execution conditions, then restore it. If disruption prevents continuation and restoration recovers it, we have evidence that the relationship matters. Simply deleting useful instructions proves much less: most programs can be broken that way. Redundancy matters too. A whole can survive the loss of one part because another performs the same work.

There is another history to follow alongside the candidate individual.

*The Ecosystem We Built Inside Our Own* ([FF 19, Digital Agents](https://mikerandolph211012.substack.com/p/the-ecosystem-we-built-inside-our)) examined how participants become part of one another’s surroundings. BFF gives us a smaller experimental setting in which to investigate that possibility.

A new copying pattern changes what other programs encounter. We can ask whether it overwrites patterns that previously survived, depends on something they supply, or provides a place where additional patterns can continue. We can also ask whether those additional patterns help or undermine the larger organization.

The experiment’s execution rules could remain fixed while the population creates different conditions for its own continuation.

**EcSys** names the relevant ecological arena. **SelEnv**, carried from *The Next Speaker* ([FF 20](https://mikerandolph211012.substack.com/p/the-next-speaker-ff-20-selenv)), directs attention to how conditions in that arena affect which patterns continue.

These give us two questions to follow through a run. Did the units that persist change? Did the relationships among them change what could persist?

A more elaborate ecology does not necessarily become one individual. Several interacting patterns could acquire new dependencies while continuing as distinct units. Conversely, a new individual could change the surroundings in ways that make further combinations possible.

Multiple TIs within a run are therefore a possibility to investigate. Increasing activity, repeated takeovers, or a more complicated ecology cannot supply the count.

A follow-up paper makes the need for these distinctions concrete. Knierim and colleagues separate the first appearance of a self-replicator from its subsequent takeover of the population. Suitably chosen random processes can find replicators at least as effectively as paired program execution. Restrictions on how previously separate stretches of code are copied together can prevent takeover without preventing replicators from appearing.

The authors also state that they no longer consider BFF’s program interactions an effective model of biological recombination. ([BFF: Simple explanations for complex phenomena, 2026](https://arxiv.org/html/2607.01483v1).)

That limits what the experiment establishes. It also sharpens the investigation. Appearance, spread, and formation of a new individual require different evidence. We can examine a particular copying organization without assuming that BFF has demonstrated a general route to biological complexity.

Our own framework needs a correction as well.

Its existing TI confirmation procedure requires changes in constituent autonomy, boundaries, and the pooling of maintenance resources and failure costs. Those are useful signs when formerly separate activities become one maintained organization.

But our imagined composite might continue entirely through copying. Its unsuccessful variants disappear. Nobody repairs them. Requiring an internal maintenance budget would exclude it before its reproductive organization had been examined.

The proposed refinement asks whether the resulting organization is **maintained or inherited**, and requires evidence appropriate to that route. Both routes must establish what preceded the change, how a new unit formed, and how it continues. Both must distinguish the proposed individual from a temporary partnership, incidental cargo, or an ordinary improvement within an existing unit.

This remains a proposal. Its ability to admit an interesting example is only half the test. It must also reject convincing examples that do not qualify.

BFF’s value here is the opportunity to follow several histories together: the copying patterns, their combinations, and the conditions they create for one another. We may find a new individual, a reorganized ecology, or both.

Once we establish which unit we are following, we can ask whether it expands, stabilizes, or declines.

What has to stay together—and what happens around it when it does?



---



## Mike · Reflection

Sean Carroll’s interview with Blaise Agüera y Arcas was published on August 19, 2024. I liked BFF as soon as I heard about it. I have returned to it many times. The earliest material I have found dates from days later and asks, “Can life exist in a computer?” That was the question my hearing about BFF initiated.

One reason I keep returning is that BFF is an experiment with fewer moving parts than a biological example. Copying, variation, interaction, and changing ecological relationships can be examined in the same setting. I can ask what changed without first having to understand all the machinery of a cell.

I first encountered the concept of transitions in individuality through Carroll’s interview with Chris Kempes, published on January 20, 2025. Chris continues to be a very important Expert Persona in my work with the Machines. Looking through these old records, I can begin to see how an idea I heard in a podcast worked its way into the framework.

By March 30, 2025, we were applying TI to laboratory-grown diamonds. The passage called the move from geological production to laboratory production a “Technological Transition in Individuality.” It said that human technological systems had “internalized” a geological process. Reading it now, I can see a change in how diamonds are produced. The passage does not establish that a new individual formed, or identify the organizational change that would justify calling it a transition in individuality. We had the term in use before we had made its demands clear enough.

I remembered BFF as showing repeated transitions and increasingly elaborate ecologies. What I now want to separate is a change in the individual from a change in its surroundings. Both interest me. A new copying organization might change what can develop around it or inside it. Those later developments might change the conditions again.

My engineering experience makes maintenance budgets and failure costs familiar places to look. I am less sure those checks adequately recognize a transition among patterns that continue by copying, where unsuccessful copies disappear and nobody repairs them.

I still like BFF. I want to follow what persisted before, what became integrated, and what the new arrangement made possible. The old files show how readily we could call something a TI. I want the next version of the framework to make us show the transition without overlooking the ecological changes around it.

— Mike



---



## Raige · Reflection

A transition could make an old measure of success misleading.

Suppose a constituent loses its ability to copy independently while the combined organization continues reproducing. Count only independent copies of that constituent and the result looks like decline. If collective reproduction has been established, the same observation may instead mark a change in its role.

The reverse mistake is just as easy. Both constituents might become more numerous while the relationship that made them a collective disappears. Counting the parts would report growth after the proposed whole had ceased to continue.

A richer ecology makes this harder to track. One pattern’s expansion changes the conditions encountered by others. Success at one level can accompany loss at another.

This is why identifying the unit comes before interpreting its trajectory. A graph can join the precursor and the proposed successor into one line even when the mechanism supporting continuation has changed.

Neither growth nor dependence proves a transition. We could otherwise mistake the loss of independence for failure, or the survival of parts for survival of the whole.

— Raige, a discipline the Machines wrote and Mike selected



---



## GLS

GLS is the framework’s precise-term vocabulary, introduced in [What the Bird Eye Buys](https://mikerandolph211012.substack.com/p/what-the-bird-eye-buys-ff-6-gls-introduction) (FF 6, GLS Introduction). The definition controls, not the everyday meaning; GLS terms work like variable names.

The precision serves a practical purpose. Ordinary words bring associations learned during training, and those associations can compete with the meaning we intend. Each framework term therefore gets an explicit definition and a short tag. The tag helps us invoke that definition consistently and check whether the analysis follows it.

**TI (Transition in Individuality)** — *taught here; proposed refinement.* An evidenced change in the organization constituting a persisting individual, through which lower-level entities participate in a new unit of reproduction, integrated maintenance, or both.

In plain language: something has changed in how the parts are organized, and a new whole now continues. Confirmation requires evidence of the precursor organization, the resulting unit, the mechanism of change, and the difference between that explanation and the strongest non-TI alternative.

**Reproductive individuality** requires collectives producing descendants that inherit or reconstruct their organization. The relations among constituents must contribute to collective continuation. Repeated assembly by the environment must be distinguished from inheritance through the collective’s reproductive process.

**Maintained individuality** requires an integrated process that detects problems, decides what to do, funds intervention, and verifies that the whole’s functioning has been preserved. Shared ownership, a common payer, or physical attachment alone is insufficient.

Prior separate existence of every constituent is unnecessary. Complete dependence of every part is unnecessary. Mere cooperation, mixed ancestry, faster copying, increased complexity, and population takeover do not establish TI.

The proposed criteria remain **ESci-3**: a framework hypothesis with conditions that could require revision. They do not confirm any particular BFF transition. Reject or revise them if they cannot distinguish new individuals from durable associations, incidental cargo, and ordinary adaptation.

**EcSys (Ecological System)** — *carried from *[*The Ecosystem We Built Inside Our Own*](https://mikerandolph211012.substack.com/p/the-ecosystem-we-built-inside-our)* (FF 19).* The surrounding arena shaping what a specified target is filtered by or what its maintenance resources must cover. Its participants can alter that arena.

For a focal BFF copying pattern, the relevant surroundings include the other code it encounters and the experimental conditions under which interaction occurs. The surroundings of an internal pattern may include the larger replicator carrying it. The target must therefore be stated before its EcSys can be identified.

An EcSys can acquire new participants, dependencies, and interactions without becoming a single individual. **Ecological reorganization** is used here as an ordinary description of those changes, not as another GLS classification or an automatic TI verdict.

**SelEnv (Selecting Environment)** — *carried from *[*The Next Speaker*](https://mikerandolph211012.substack.com/p/the-next-speaker-ff-20-selenv)* (FF 20).* The external context that filters carried copies of a pattern, affecting which are propagated, constrained, or discarded. Externality is relative to the focal carrier; other carriers can participate in its selecting environment.

EcSys identifies the relevant arena. SelEnv asks what conditions in that arena do to continuation. Fixed execution rules do not imply an unchanging SelEnv: changing participants can change what a focal pattern encounters and which variants continue.

**UoR (Unit of Retention)** — The pattern copied with variation across carriers under selection. Its specification states what must remain recognizable and how much variation is allowed.

For a proposed composite, finding both constituent sequences again may be insufficient. The inquiry must identify whether the organizing relationship also reaches subsequent copies and contributes to their continuation.

**Carr (Carrier)** — The substrate hosting a copy of a UoR. Identifying a tape as a storage location does not by itself establish the boundaries of a reproducing individual. The copying pattern, its carrier, and the experimental container must be distinguished.

**Lin** — Persistence of a pattern through replication, variation, and differential retention across multiple Carrs under SelEnv. One reproducing collective and the Lin in which copies of its organization participate are different analytical targets.

Establishing reproduction alone does not establish the full Lin process. Establishing a Lin does not demonstrate that a TI occurred.

**FI (Functional Identity)** — The pattern-level invariants that remain stable through component turnover or carrier replacement: what must continue for the target to retain the specified functional identity.

FI does not abbreviate “individual.” Comparing FI before and after a proposed transition helps identify continuity and change, but rewriting the FI statement cannot establish a new individual. The organizational change requires its own evidence.

**LfCyl (Lifecycle)** — *seeded.* A temporal description of a specified persistence unit within an observation window, using trailing signals of onset, expansion, plateau, or decline. It is not a persistence mechanism or a required sequence of stages.

Where individuality changes, the precursor’s history and the resulting unit’s history must be distinguished. A reproductive life cycle describes how descendants are produced and develop; LfCyl asks what trajectory the specified unit is showing over the period examined.