Skip to content

A ByteQuilt research program

Computingprimitives drawnfrom biology.

What if software could self-heal, adapt to workloads, and gracefully degrade like living systems? Mutuus pursues that question with data structures, operational strategies, and coordination patterns that are formally specified, implemented in Rust, and benchmarked against their classical counterparts.

10
primitives
9
benchmarked
2
papers with DOIs
L0 · hot, mutable
fracture plane
L6 · compressed
Fig. 1 A Nacre Array in section. Tablets are segments; the mortar between them is metadata. Recent layers stay hot; older layers cool and compress, as nacre densifies with age.
§ 01Premise
“A biological analog is useful when it helps identify a structural mechanism, lifecycle pattern, or resource tradeoff that can be specified, implemented, and tested. It is not useful when it merely renames an established technique.”

Phillips (2026), Adaptive Computing Primitives, §2

10
Primitives in the library, each paired with a classical counterpart
7
Evaluation gates before an organic is published
100+
Tick cycles before steady-state benchmarks are taken
0
Losses hidden. Wins and losses are reported together
§ 02Framework

Three strata,from structure to ecosystem.

Higher tiers depend on lower tiers, but each is independently useful. Claims are scoped to one level at a time so they can be tested.

I

Organics

Data structures

Adaptive data structures that offer an alternative to a classical (inorganic) counterpart, adding lifecycle management, adaptive behavior, and self-tuning.

II

Metabolics

Operational strategies

Resource-management policies that formalize what production systems already do ad hoc: how computation is allocated, conserved, and recovered.

Eight strategies cataloged

III

Ecologics

Coordination patterns

Relationships between systems: how they compete for resources, cooperate for mutual benefit, and co-evolve over time.

Cataloged · evaluation deferred

Awaiting production ecosystems
§ 04Rigor

We publishthe losses.

Every organic is benchmarked against the structure it hopes to replace, and the tradeoffs are reported whole. The Nacre Array wins on structural mutation and loses on raw reads. Both appear in the paper.

“Research that hides failures is marketing, not science.”

Read the Nacre Array paper →

Nacre Array vs. Vec<T>, 100K elements

Criterion.rs medians · 12-byte elements

Split at fracture plane

vs Vec::split_off

7.5× faster

Insert at midpoint

segment-bounded shift

3.7× faster

Sequential scan (cache on)

contiguity wins

5.1× slower

Iteration

hardware prefetch

6.3× slower

Random access (cache off)

O(log S) lookup

7.9× slower

Source: Phillips (2026), Nacre Array, §7. Mixed hot/cold memory results are deferred until a reproducible harness is published.

§ 05Observation

Watch a structurecool, compress, and wake.

Three hundred elements in five segments. Untouched segments cool from Hot to Warm to Cold, then compress in place. A single read brings one back.

hot warm cold compressed fracture plane

Tip: click any segment to read it. Reads reheat a segment instantly; cooling only happens on tick().

Field notes

  • Waiting for the organism to stir…
Elements
300
Segments
5
Cold ratio
0%
Compressed
0 B
Tick 0 / 50012-byte elements · cooldown 10 / 20 / 5

Figure. Nacre Array Compression at tick 0 of 500. Five segments of 60 twelve-byte elements; cooldown 10 / 20 / 5 ticks.

Illustrative simulation, not a benchmark · Mutuus Research, ByteQuilt