20: What Happens When the Crystal Can Lose Material?

Until now, the crystal had lived in an impossible world.

Once a cell became occupied, it stayed occupied.

Forever.

The frontier moved outward.

Old material accumulated behind it.

Nothing eroded.

Nothing vanished.

Nothing exposed the interior again.

That assumption had quietly shaped almost every result since Chapter 11.

So Chapter 20 removed it.

We added one rule:

occupied
↓
empty

with some small probability.

Nothing else.

No repair mechanism.

No resource system.

No maintenance controller.

No energy variable.

No metabolism.

No target morphology.

Just loss.

Then we asked what broke.

The Obvious Prediction

The first prediction seemed almost embarrassingly simple.

Suppose the crystal has an effective radius \(r\).

If new construction happens mainly around the outside boundary, then construction opportunity should scale roughly like perimeter:

At small size:

construction > loss

At larger size:

loss > construction

So there ought to be some scale where the two balance.

Perhaps the crystal would stop expanding.

Perhaps it would hover around a finite size.

Perhaps above some loss rate it would collapse.

It was exactly the kind of candidate scaling law we wanted.

And because it sounded so plausible, we wrote the failure condition before running the experiment.

A finite dynamic regime had to satisfy all of these:

late population slope approximately zero

population still substantially above zero

world far from simulation capacity

population meaningfully smaller than decay-off baseline

A plateau caused by extinction would not count.

A plateau caused by reaching the edge of the world would not count.

We wanted a genuine balance between construction and loss.

First Establish the Baseline

With no loss, the crystal still behaved exactly as expected.

The decay-off baseline had a late normalized population slope around:

0.037 per update

and late net growth around:

154 cells per update

The crystal was nowhere near world capacity.

So we had a clean expanding reference.

Then we turned on loss.

The sweep was:

δ = 0.00
δ = 0.02
δ = 0.04
δ = 0.06
δ = 0.08
δ = 0.12
δ = 0.16

The prediction was straightforward:

higher δ
↓
more material removed
↓
growth slows
↓
eventually slope approaches zero

That is not what happened.

Loss Did Not Stop the Expansion

The late normalized population slopes stayed remarkably similar.

Across the entire sweep they remained close to:

0.036 – 0.038

Even at the highest tested loss rate, the crystal was still expanding.

The late mean population did fall.

At δ = 0.16, it was roughly one third smaller than the decay-off baseline.

But the slope did not flatten.

No non-zero loss rate met the predeclared finite-regime condition.

So the primary result was:

FAILED

The obvious perimeter-versus-area argument had not survived contact with the substrate.

That failure was already useful.

But the reason was much more interesting.

Look at the Gross Construction Rate

At first glance, increasing loss should simply subtract material from the crystal.

If the crystal ordinarily adds around 150 cells per update, then adding loss ought to produce something like:

+150 construction
-80 loss
= +70 net

Then perhaps:

+150 construction
-150 loss
= 0 net

But the construction rate itself did not remain fixed.

It exploded.

Late averages looked approximately like this:

At δ = 0.16, the crystal was losing more than five hundred cells per update.

But it was also attaching more than six hundred.

Loss had made the crystal build faster.

That was not in the prediction.

The Missing Variable Was Frontier

The original scaling argument treated the crystal as if it had one meaningful boundary.

Something like:

solid interior
+
outer perimeter

But random loss changes the geometry.

Every time an occupied site disappears, it can create a new empty location surrounded by occupied material.

That location becomes part of the set that ordinary growth can potentially fill.

    flowchart LR
    A[Loss removes<br/>occupied cell] --> B[New empty site<br/>in occupied region]
    B --> C[Site becomes<br/>local boundary]
    C --> D[Ordinary growth sees<br/>attachment opportunity]
    D --> E[Construction rate increases]
  

So:

loss
↓
new empty site
↓
new local boundary
↓
new attachment opportunity

The loss process was manufacturing frontier.

And the boundary measurements showed it dramatically.

Approximate late mean boundary counts were:

So our original statement:

construction opportunity ~ outer perimeter

was simply the wrong geometry.

Once material can disappear, the crystal develops internal construction surfaces.

The active boundary is no longer only outside.

That destroys the naive scaling law.

Where You Lose Material Matters

V1 contained another test.

Suppose the total number of removed cells is held exactly equal.

Does it matter whether they are removed from the surface or the interior?

We synchronized the loss count.

Every update:

surface branch removes K cells
interior branch removes K cells

The total removal budget was identical.

The only difference was placement.

The result was strong.

Late population was about:

11.1%

higher under interior-biased loss than under surface-biased loss.

That cleared the predeclared 10% effect threshold.

And the hole counts were dramatically different:

surface loss      ~2.7 late holes
interior loss    ~29.8 late holes

So equal loss did not have equal consequences.

Where the missing material appeared mattered.

At this point we had a plausible explanation.

Perhaps interior loss creates high-neighbour-count vacancies that ordinary growth rapidly fills.

But that was still only an interpretation.

We needed to measure it directly.

V2: Separate New Construction from Reoccupation

Until loss existed, every attachment meant the same thing.

A cell had never been occupied before.

Then it became occupied.

So:

attachment = new construction

But after material loss, that equivalence breaks.

A cell can now follow this path:

occupied
↓
lost
↓
empty
↓
occupied again

So V2 introduced an observer-only occupancy ledger.

It did not change the crystal.

It merely remembered whether each lattice position had ever been occupied before.

Now every attachment could be classified as one of two things:

FIRST OCCUPATION
a site is occupied for the first time ever

REOCCUPATION
a previously occupied site was lost
and later becomes occupied again

    flowchart TD
    A[Attachment event] --> B{Was site ever occupied before?}
    B -- No --> C[First occupation]
    B -- Yes --> D[Reoccupation]
    D --> E[Site was previously lost]
  

This distinction immediately changes how we should read the V1 numbers.

632 attachments per update does not mean:

632 new places added

It means:

first occupations
+
reoccupations

The crystal may be doing enormous internal turnover while net population grows much more slowly.

Verify the Null First

With no material loss, reoccupation must be impossible.

Nothing ever becomes empty again.

So V2 first ran a structural null.

Across 96 no-loss runs:

reoccupation count = 0

exactly.

All attachments were first occupations.

Good.

Now we could trust the ledger.

The Exact-Count Loss Experiment

We returned to the matched surface-versus-interior design.

Again:

surface loss count = interior loss count

on every update.

The mean cumulative loss count in each branch was about:

890 cells

Then we asked:

What fraction of those losses later became reoccupations?

The answer was much larger than expected.

Almost Everything Came Back

For surface-biased loss:

reoccupation per loss ≈ 0.936

For interior-biased loss:

reoccupation per loss ≈ 0.956

Another way to measure the same phenomenon is to ask what fraction of unique lost sites were ever reoccupied.

Again:

surface     ≈ 93.6%
interior    ≈ 95.7%

That is the central mechanistic result of Chapter 20.

Under these conditions, most removed sites did not remain empty.

They came back.

And usually quickly.

Mean reoccupation delay:

surface     ≈ 1.56 updates
interior    ≈ 1.09 updates

So a typical lost site was often empty for only one or two steps.

But the Primary V2 Hypothesis Still Failed

V2 had not merely predicted that reoccupation would occur.

It made a stronger claim.

We predicted that interior loss would produce a much larger reoccupation rate than surface loss.

The predeclared minimum difference was:

0.15 additional reoccupations per loss

The observed difference was only:

0.0198

It was statistically clear.

The interval was narrow.

The directional test was extremely small.

But it was nowhere near the effect size we had said would matter.

So the primary V2 result was:

FAILED

Again.

And again the failure corrected the interpretation.

Interior loss was not special in the way we expected.

The much larger phenomenon was shared by both conditions.

Almost Every Loss Creates Frontier

V2 also measured how much new frontier was created per removed cell.

The numbers were:

surface     ~0.995 new frontier sites per loss
interior    ~1.000 new frontier sites per loss

That is almost one-for-one.

So the more general mechanism is:

LOCAL MATERIAL LOSS
↓
NEW EMPTY LOCATION
↓
NEW ATTACHMENT OPPORTUNITY

not:

INTERIOR LOSS ONLY
↓
SPECIAL ATTACHMENT OPPORTUNITY

Interior sites were somewhat more rapidly and completely reoccupied.

But that was a second-order difference.

The major effect was universal.

Removing a cell almost always creates a site ordinary growth can see again.

Reoccupation Is Not Repair

This is where the language matters.

It is tempting to say:

The crystal repairs itself.

We have not earned that.

There is no:

damage detector

No:

repair objective

No:

desired target morphology

No:

special reconstruction pathway

The exact same local growth rule runs whether the empty site is:

new territory

or:

a place that used to exist

The crystal does not know the difference.

The observer does.

So the bounded claim is:

Material removal creates attachment opportunities that the ordinary Digital Crystal growth rule rapidly reuses.

That is enough.

And it is more interesting than calling it repair too early.

Population Was Hiding Turnover

The biggest conceptual change in this chapter is not about loss.

It is about measurement.

A population graph shows:

occupied cells at time t

But that is only a net stock.

Once material can disappear and reappear, the same population can conceal very different underlying traffic.

Imagine two systems.

System A:

+100 new cells
0 lost

System B:

+600 attachments
-500 losses

Both end the update at:

+100 net

A population curve treats them as equivalent.

They are not.

System B is undergoing enormous internal turnover.

    flowchart TD
    subgraph A[System A]
    A1[+100 new cells] --> A2[Net +100]
    end
    subgraph B[System B]
    B1[+600 attachments] --> B2[-500 losses]
    B2 --> B3[Net +100]
    end
    A2 -.->|same net population| B3
    B1 -.->|huge gross turnover| B2
  

So another distinction joins our evidence ladder:

NET POPULATION CHANGE
≠
GROSS CONSTRUCTION ACTIVITY

and:

ATTACHMENT
≠
FIRST OCCUPATION

Those distinctions did not exist before material could disappear.

Now they are essential.

The Hole Paradox

There was one result that initially looked contradictory.

Interior-biased loss produced many more visible holes:

surface     ~3.2
interior    ~37.6

Yet interior lost sites were reoccupied slightly more often and much faster.

How can a process with faster refill have more holes?

Because a snapshot counts how many holes exist now.

It does not tell us how long each hole survives.

Imagine:

lose A
lose B
refill A
lose C
refill B
lose D
refill C
...

A snapshot can show many holes even if individual holes are short-lived.

The process may be generating vacancies faster than any one vacancy persists.

Again:

State is not dynamics.

A single picture of the crystal cannot reveal how much material is cycling underneath it.

We keep rediscovering this lesson.

What Happened to the Scaling Law?

We began Chapter 20 with a candidate law:

growth ~ boundary
loss ~ occupied area

therefore:

loss eventually wins

The experiment rejected that reasoning under this substrate.

Why?

Because loss is not an external subtraction from a fixed geometry.

Loss changes the geometry that determines construction opportunity.

The real relationship is closer to:

loss
↓
changes active boundary
↓
changes future construction rate

So loss and construction are dynamically coupled.

That means the original scaling argument treated:

construction rate

as independent of:

loss rate

when it is not.

That was the mistake.

And finding that mistake is more valuable than confirming the original prediction would have been.

A Different Kind of Compensation

There is another tempting word here:

maintenance

We should not use it yet.

But something important has appeared.

A perturbation that removes material automatically creates conditions under which the ordinary growth process tends to replace much of that material.

Conceptually:

loss
↓
creates opportunity for the mechanism
that counteracts loss

That is a form of structural compensation.

Not because the crystal senses damage.

Not because it allocates resources.

Not because it prefers preservation.

But because the geometry of removal feeds directly into the geometry of construction.

    flowchart LR
    A[Loss] --> B[Empty site]
    B --> C[New frontier]
    C --> D[Ordinary attachment]
    D --> E[Reoccupation]
    E --> F[State changes again]
    F --> A
  

That is a substrate-level feedback.

The word feedback is justified here only in the mechanistic sense:

loss changes state
state changes future construction opportunities
construction changes state again

We do not need to make it biological.

What Survived the Hypothesis?

Both primary hypotheses in this chapter failed.

The first predicted a finite near-stationary regime in which loss would eventually balance growth.

That did not happen.

The second predicted that interior-biased loss would produce a scientifically large reoccupation advantage over surface-biased loss.

That also did not happen.

Those failures remain failures.

But underneath them, the experiment exposed a much broader mechanism.

Loss changed the geometry of opportunity

The original scaling argument treated growth and loss almost as independent processes:

growth
~ outer boundary

loss
~ occupied area

That picture failed because material loss changes the very geometry that determines where growth can occur.

A removed cell becomes empty, and if occupied neighbours remain around it, that location becomes an ordinary attachment opportunity.

    flowchart TD
    A[LOSS] --> B[NEW EMPTY SITE]
    B --> C[NEW LOCAL FRONTIER]
    C --> D[NEW ATTACHMENT OPPORTUNITY]
  

The experiment measured this almost one-for-one:

surface loss
~0.995 new frontier sites per loss

interior loss
~1.000 new frontier sites per loss

The crystal did not merely suffer loss.

Loss manufactured frontier.

Phenomenon record

Phenomenon: Loss-generated construction opportunity

Status: SUPPORTED

Current bounded description:

Under the tested Digital Crystal loss conditions, removing occupied material almost always creates a new local frontier opportunity that the ordinary frozen growth rule can potentially reuse.

This extends the Interface Principle from Chapter 18.

Chapter 18 showed that historical state matters when it remains coupled to the active construction interface.

Chapter 20 shows that the interface itself can be created by loss.

So the active interface is not just something the crystal carries outward.

It can also be regenerated internally whenever material disappears.

Reoccupation is the dominant shared phenomenon

Under exact-count matched loss:

surface unique lost sites reoccupied
≈ 93.6%

interior unique lost sites reoccupied
≈ 95.7%

and mean reoccupation delay was approximately:

surface
1.56 updates

interior
1.09 updates

The predeclared interior-minus-surface difference failed because the shared effect was already close to saturation.

The stronger surviving observation is therefore:

Sparse material loss is usually converted into a new attachment opportunity that ordinary growth rapidly reuses.

This is not repair.

No damage detector exists.

No target state is reconstructed.

No special pathway distinguishes:

new territory

from:

territory that used to be occupied

The observer knows a site is being reoccupied.

The crystal does not.

So:

REOCCUPATION
≠
REPAIR

Phenomenon record

Phenomenon: Rapid reuse of lost material

Status: SUPPORTED

Current bounded description:

Under the tested exact-count conditions, more than 93% of lost sites were subsequently reoccupied, typically within one or two updates, through the same local growth rule used for ordinary expansion.

This is a substrate-level consequence of geometry, not evidence of biological maintenance.

The Flux Principle

Chapter 20 also exposes a second major phenomenon.

Before loss existed:

attachment
=
first occupation

After loss exists:

attachment
=
first occupation
+
reoccupation

That means population becomes an incomplete description of the process.

A crystal can have modest net growth while undergoing enormous gross material traffic.

For example:

+632 attachments
-531 losses
----------------
+101 net

A population curve sees:

+101

but the process actually executed:

1163 gross material events

So the chapter establishes:

NET POPULATION CHANGE
≠
GROSS MATERIAL TURNOVER

This gives us the Flux Principle:

Static population or morphology can conceal large ongoing construction, loss and reoccupation flows.

The active process may be much more dynamic than its net state suggests.

State is not dynamics

The hole paradox reinforces the same point.

Interior-biased loss can show more holes in a snapshot while individual lost sites are also being reoccupied faster.

That is possible because:

number of holes now

is not the same thing as:

lifetime of each hole

A system can create and close vacancies rapidly while maintaining many vacancies at any one instant.

So again:

SNAPSHOT
≠
PROCESS

This is the Flux Principle in another form.

Structural compensation, not maintenance

The chapter also reveals a bounded feedback loop:

loss
↓
changes local geometry
↓
creates attachment opportunity
↓
ordinary growth reoccupies sites
↓
changes local geometry again

That is legitimate mechanistic feedback.

But it is not evidence for:

damage sensing
goal-directed preservation
repair objective
self-maintenance

The correct term is structural compensation.

The system counteracts some loss because the geometry of loss creates opportunities for the same process that fills empty sites.

Connection to earlier phenomena

The cross-chapter picture is now becoming much tighter.

Chapter 18:

causal relevance
lives at active interface

Chapter 19:

accessible difference
does not guarantee readout

Chapter 20:

loss creates interface

So the interface is no longer merely:

the outer boundary of a growing object

It is better understood as:

the dynamically generated set of locations where the process currently has an opportunity to change state.

That definition will matter later.

What this phenomenon does not establish

The surviving phenomena do not establish:

• repair,
• homeostasis,
• self-maintenance,
• energy use,
• metabolism,
• a finite sustainable size,
• or life.

They establish something narrower and more fundamental:

Material loss dynamically generates new construction opportunity, and ordinary Digital Crystal growth rapidly reuses most lost sites. At the same time, large gross turnover can remain hidden beneath relatively modest net population change.

Those phenomena should now be tracked independently of the failed finite-regime and interior-superiority hypotheses.

What Chapter 20 Established

The primary V1 hypothesis failed.

The primary V2 hypothesis failed.

That sounds disastrous if experiments are supposed to confirm what we hoped.

It is exactly what we wanted from this book.

V1:

Background loss did not produce the predicted finite near-stationary regime across the tested sweep.

V2:

Interior loss did not increase reoccupation over surface loss by the predeclared scientifically meaningful amount.

But together they exposed a stronger mechanism:

Material loss creates new attachment opportunity, and under the tested exact-count conditions more than 93% of lost sites were subsequently reoccupied, typically within one or two updates, by the ordinary frozen growth rule.

That statement is measured.

It does not require the failed hypotheses to become true.

The New Ladder

The crystal now has a new causal chain:

MATERIAL PRESENT
↓
MATERIAL LOST
↓
EMPTY SITE
↓
NEW FRONTIER
↓
ORDINARY ATTACHMENT
↓
REOCCUPATION

And we can distinguish:

LOSS
≠
PERMANENT LOSS

ATTACHMENT
≠
NEW CONSTRUCTION

REOCCUPATION
≠
REPAIR

NET GROWTH
≠
GROSS MATERIAL TURNOVER

Those are the results of Chapter 20.

Why We Stop Here

We could continue.

Measure exact hole lifetime distributions.

Sweep the placement fraction.

Try different loss schedules.

Separate one-neighbour from six-neighbour vacancies.

Increase the number of groups.

But none of those would cross a new conceptual boundary.

We know the essential mechanism now.

Loss makes reusable frontier.

Ordinary growth reuses it.

The crystal can therefore undergo large material turnover without net population collapse.

That is enough.

No V3.

The Question We Have Now Earned

There is still something artificial about this world.

The crystal currently gets unlimited opportunities to act.

Every update, the growth rule can evaluate every eligible frontier site.

It does not have to choose between:

expanding outward

and:

reoccupying lost material

It can attempt both everywhere.

So rebuilding what vanished is almost free.

That is why Chapter 20 should not end with:

repair

It should end with a constraint.

Suppose the crystal can perform only:

B

construction updates per step.

Now every action has an opportunity cost.

For the first time:

OUTWARD CONSTRUCTION
        competes with
REOCCUPATION

And then a genuinely new question appears:

When computational action is limited, what gets built and what gets preserved?

That is no longer a question about loss alone.

It is a question about allocation under scarcity.

    flowchart LR
    subgraph Unlimited
    U1[Unlimited updates] --> U2[All frontier sites evaluated]
    U2 --> U3[Rebuilding lost material is almost free]
    end
    subgraph Limited
    L1[Limited construction budget B] --> L2[Must choose where to act]
    L2 --> L3[Outward growth vs reoccupation]
    L3 --> L4[Opportunity cost emerges]
    end
    U3 -.->|remove luxury| L1
  

The crystal did not learn to repair itself.

We merely discovered that, while computation was unlimited, rebuilding what disappeared was almost free.

Next, we remove that luxury.