NIORet Storage Research | NIORet Systems

What was built and measured

NIORet is developing a native storage architecture designed to preserve and reuse information across related objects, partial overlaps, and evolving datasets. Instead of treating every incoming file as an entirely independent unit, the standalone M7 research engine identifies reusable regions within previously stored objects, records compact reconstruction relationships, and verifies that the original bytes can be recovered exactly.

Measured scaling evidence

Increasing storage advantage across generated workloads

Across three actual executed M3 storage benchmarks, the additional reduction measured against the tested conventional baseline increased from 42.67% at 10,000 generated pages to approximately 59.09% at one million pages.

42.67%10,000 pages

56.12%100,000 pages

≈59.09%1,000,000 pages

Measured percentage advantage

Additional reduction across generated workloads

Actual executed storage benchmarks using generated webpage data. The line joins only the three measured M3 workload sizes; it does not extrapolate beyond them.

Real captured browser data — M7

This standalone storage-subsystem result uses actual captured browser files. It is not a generated dataset. The 14.0% comparison is against the best tested conventional stored size, not against the original uncompressed logical data, and it is not a full-stack performance benchmark for NRL, NSE, or NIORET OS.

Real Forensic Browser capture benchmark — M7
Metric	Measured result
Dataset	4,540 actual Forensic Browser capture files
Logical data	108,549,354 bytes (108.55 MB)
Best tested conventional stored size	27,639,742 bytes (27.64 MB)
NIORet M7 stored size	23,770,997 bytes (23.77 MB)
Conventional logical-to-stored ratio	3.93×
NIORet logical-to-stored ratio	4.57×
Advantage	14.0% less physical stored data than the best tested conventional baseline
Exact reconstruction	PASS — SHA-256 verified

Generated-workload benchmark details — M3

Separate standalone M3 scale experiments used constructed, highly repetitive webpage inputs. “Generated” describes the input data: the storage algorithms actually executed and the stored sizes below were measured. These tests are separate from the M7 real-browser corpus and do not represent the complete integrated NRL/NSE/NIORET OS architecture.

Stored bytes compared

Conventional and NIORet representations

10,000 pages

Conventional

126,279 B

NIORet

72,401 B

53,878 fewer stored bytes

100,000 pages

Conventional

1,262,797 B

NIORet

554,067 B

708,730 fewer stored bytes

1,000,000 pages

Conventional

≈13.25 MB

NIORet

≈5.42 MB

Approximately 7.83 MB fewer stored bytes

Each pair is normalized to its conventional baseline; the printed values preserve the measured absolute stored sizes and widening absolute gap.

Generated-webpage scale tests — M3
Workload	Conventional baseline	NIORet representation	Additional reduction	Reconstruction
10,000 pages	126,279 bytes	72,401 bytes	42.67%	Reported PASS
100,000 pages	1,262,797 bytes	554,067 bytes	56.12%	Reported PASS
1,000,000 pages	Approximately 13.25 MB	Approximately 5.42 MB	Approximately 59.09%	Reported PASS

The million-page values are rounded. NIORet’s architecture is designed to discover and reuse shared information as collections expand, and the measured M3 trend supports further scaling research. Scaling remains workload-dependent on available overlap, new information, indexes, metadata, and reconstruction references; it does not guarantee that percentage savings will increase indefinitely.

How the research architecture works

The M7/M8 research family combines five public-safe architectural ideas:

Reusable internal regions. Previously retained objects can provide candidates that let later related objects reuse portions of earlier data without requiring identical whole-file content.
Partial-overlap discovery. Content signatures and bounded candidate search identify potential reuse across stored objects.
Cost-aware representation. An incoming object can be represented directly or through a reconstruction recipe based on estimated persisted cost, including metadata.
Exact reconstruction. The representation must reproduce the original bytes; the documented tests use SHA-256 comparison.
Incremental ingestion. Follow-on M8 work added per-object journaling, checkpoints, reopen/restart behavior, and rebuilding of ephemeral candidate indexes.

M8 is distinct from the M7 compression comparison. Its restart-roundtrip evidence supports incremental persistence research, but does not establish full crash atomicity under power loss.

Why native storage matters

Nioret Systems treats information identity, ownership, persistent versions, provenance, and temporal relationships as architectural concerns. The objective goes beyond compressing independent files: related information should be able to share reusable structure across time while retaining exact recovery when required.

This makes the research relevant to Forensic Browser histories, native application state, semantic computation, and long-lived local AI information stores. The measured M7 and M3 achievements come from standalone storage research implementations. They support the broader architectural direction, but do not measure the complete integrated language, semantic-execution, or operating-system stack and do not claim universal superiority over every storage system or input.