Nacre Array: A Segmented Dynamic Array with Thermal State Management and Structural Fracture Planes

Dynamic arrays treat every element identically regardless of access pattern, age, or semantic grouping. This paper investigates a segmented alternative in which different regions of the array may be mutated, cooled, compressed, and split differently over time. We present the Nacre Array, a segmented dynamic array with three properties absent from conventional contiguous arrays: (1) per-element metadata headers that make elements self-describing without external bookkeeping, (2) a four-state thermal machine (Hot/Warm/Cold/Compressed) that allows cold segments to compress in place via LZ4 or Zstd while hot segments remain uncompressed and mutable, and (3) fracture planes at segment boundaries that enable structural splits without copying element payloads. Inspired by nacre (mother-of-pearl), where layered aragonite tablets bonded by organic biopolymer achieve roughly 3,000x the fracture resistance of the raw mineral [9, 10], the structure organizes elements into fixed-capacity segments with per-segment summaries and a tick-driven cooldown finite-state machine. The implementation evaluated here performs mutations locally within a segment, updates cumulative suffix metadata incrementally after insert/remove, splits cumulative metadata alongside segments instead of rebuilding both halves from scratch, and exposes an explicit reheat path for compressed segments through owned reads and mutable accessors. Against Rust's `Vec<T>` on the Criterion harness used in this paper, Nacre splits 23.6x faster than `Vec::split_off` at 100K elements and inserts 8.3x faster at the midpoint, while pure random access is 8.1x slower at 100K with the segment cache disabled. A brief 100K push win over default-growth `Vec` persists, but push throughput is not treated here as a structural Nacre advantage because preallocated `Vec` remains faster. Mixed hot/cold memory-reduction claims are deferred until a reproducible benchmark harness for that workload is checked into the repository.