Module store 

Block-based storage engine for temporal tensor compression (ADR-018).

Provides tiered quantized storage with CRC32 integrity checking, access-pattern tracking, and eviction support. Each block of tensor data is quantized at the bit width appropriate for its storage tier and tracked with rich metadata for tier-promotion/demotion decisions.

Storage Tiers

Tier	Bits	Description
Tier0	0	Evicted: metadata only, no payload
Tier1	8	Hot: full fidelity quantization
Tier2	7	Warm: moderate compression
Tier3	3	Cold: aggressive compression

Example

use ruvector_temporal_tensor::store::{BlockKey, Tier, TieredStore, ReconstructPolicy};

let mut store = TieredStore::new(4096);
let key = BlockKey { tensor_id: 1, block_index: 0 };
let data = vec![1.0f32; 64];

store.put(key, &data, Tier::Tier1, 0).unwrap();
assert_eq!(store.block_count(), 1);

let mut out = vec![0.0f32; 64];
let n = store.get(key, &mut out, 1).unwrap();
assert_eq!(n, 64);

Structs

BlockKey

Unique identifier for a tensor block.

BlockMeta

Complete metadata for a single block.

TickResult

Summary of actions taken during a budgeted maintenance tick.

TieredStore

In-memory tiered storage engine for quantized tensor blocks.

Enums

DType

Data type of the original tensor.

ReconstructPolicy

Reconstruction policy for evicted (Tier0) blocks.

StoreError

Errors produced by the storage engine.

Tier

Storage tier for a block.

Traits

BlockIO

Block I/O for reading and writing raw quantized data.

Clock

Clock abstraction for deterministic time in tests and production.

MetaLog

Metadata log for append-only persistence of block metadata.

Functions

crc32

Compute CRC32 using the standard reflected polynomial (0xEDB88320).