T15 Orchestrator - measured compression

Byte-exact compression,
measured by file class.

The T15 Orchestrator is the lossless compression product. It uses SRD math partly together with standard codecs (xz, zstd, brotli, gzip, 7-Zip) and keeps the smallest verified output. Every result restores the original byte-for-byte and is checked with a SHA-256 roundtrip, with a +1 byte never-worse floor on its own output. Compressible data shrinks; already-compressed and random data simply hit the floor. No fake ratios, and nothing here claims to pass an entropy bound.

Byte-exact SHA-256 roundtrip +1 byte never-worse floor single host - reproducible honest ratios by file class

single host (Hetzner CX22) - 2026-05-25 - every input and output hashed and roundtrip-verified to the byte

<- Back to the Orchestrator hub Full benchmark suite

How to read these results

The Orchestrator runs SRD math alongside the standard codecs and keeps whichever byte-exact output is smallest. Because the standard codecs are part of the field, the Orchestrator's output is never larger than the best of them (worst case +1 byte). The outcome depends entirely on the file class.

Compressible data

Shrinks

Structured JSON, logs, periodic and sparse binary. Where exploitable structure exists, the smallest verified output is well below the input size and often below the best standard codec.

Already-compressed data

Ties / floor

MP4, ZIP, JPEG, audio. Little structure is left, so the Orchestrator selects the best codec and matches its size, or returns proof-only at the +1 byte floor. No invented savings.

Random / encrypted data

Hits the floor

High-entropy bytes cannot be compressed by anyone. The Orchestrator preserves exact restore and stops at the +1 byte never-worse floor on its own output - never a fake ratio.

Results by file class

Output sizes in bytes, smaller is better. The Orchestrator row is the smallest byte-exact output it produced. The "best baseline" is whichever standard codec did best on that input. Outright wins, ties (Orchestrator selects the best codec at its exact size) and floor results are all shown - nothing is rounded up to look better than it measured.

A - JSON, 31 KB (structured text - compressible)

Structure-rich: the Orchestrator shrinks well below the input and below the best standard codec.

CodecOutput bytesRatio
gzip-94,8206.6x
7-Zip ultra4,2017.5x
xz-9 -e4,3607.3x
zstd-192,80111.3x
brotli-11 (best baseline)2,51012.6x
T15 Orchestrator1,26524.9x - smaller than brotli here

B - 1 MiB periodic binary (structured - compressible)

Periodic structure is exactly what SRD math exploits; the verified output is a fraction of the best codec.

CodecOutput bytesRatio
gzip-93,140334x
brotli-111,948538x
zstd-191,872560x
7-Zip ultra1,242844x
xz-9 (best baseline)1,108946x
T15 Orchestrator4362,404x - smaller than xz-9 here

C - Apache logs, 3.5 MB (English-like text - ties best codec)

High-entropy text where xz-9 is the strongest standard codec. The Orchestrator selects xz's output and ties it at the exact size - never larger.

CodecOutput bytesRatio
gzip-9325,632 (318 KB)11.3x
zstd-19245,760 (240 KB)14.9x
brotli-11236,544 (231 KB)15.5x
xz-9 -e (best baseline)216,064 (211 KB)17.0x
T15 Orchestrator216,064 (211 KB)17.0x - ties xz, selects the best codec

D - MP4 master, 10 MB (already-compressed container - narrow gain)

A dense container with little structure left. A modest byte-exact gain over the best codec - real, but small.

CodecOutput bytesvs best baseline
brotli-119,830,512+362,132
gzip-99,742,892+274,512
zstd-19 --long9,481,772+13,392
7-Zip ultra9,468,640+260
xz-9 -e (best baseline)9,468,380-
T15 Orchestrator9,430,108-38,272 B (0.40% smaller)

E - Random bytes, 1 MB (incompressible - the never-worse floor)

No codec can compress random data. This checks the floor: the Orchestrator's +1 byte guarantee is a promise about its own output, not a claim of beating every codec on every input.

CodecOutput bytesOverhead
gzip-91,048,768+192 B
7-Zip ultra1,049,234+658 B
xz-9 -e1,049,000+424 B
zstd-191,048,589+13 B
brotli-111,048,581+5 B
T15 Orchestrator1,048,577+1 B - never-worse floor
Reading these numbers honestly. The T15 Orchestrator is entropy-aware. It shrinks where data has exploitable structure (periodic binary, structured JSON), ties the best standard codec on high-entropy text by selecting that codec at its exact size, achieves only a narrow gain on already-dense containers, and stops at the +1 byte floor on random or encrypted input. These are measured results on a single reproducible host - not "records nobody has matched", not "superior to all on every input", and not a claim of going past any entropy bound.
Never-worse floor

+1 byte, guaranteed on the Orchestrator's output

Whatever you feed in, the Orchestrator's container is never more than one byte larger than the input. If no honest reduction exists, it returns a proof-only result at the floor instead of inventing savings. The floor is a guarantee about the Orchestrator's own output - it is not a claim that the Orchestrator beats every standard codec on every input.

Random, encrypted and already-optimal files land here by design. That is the honest answer for data with no exploitable structure.

Byte-exact roundtrip

Restore must match the original SHA-256

Every row above was verified by decompressing the output back to the original and confirming the bytes are identical. There is no lossy mode, no approximation, no hidden mutation - if the restored SHA-256 does not match the input, the result does not ship as a win.

in.bin --> T15 Orchestrator --> out.pdli
out.pdli --> restore --> in.bin'
assert sha256(in.bin) == sha256(in.bin')   # required

Reproducing these benchmarks

Every input and every output above is hashed with SHA-256 and roundtrip-verified (decompress to byte-identical original). The full harness - inputs, scripts, raw result JSON and the exact codec versions used - is available on request so anyone can re-run these on their own host and confirm the byte-exact restore.

Where a timestamp manifest is published, you can verify its anchor with OpenTimestamps:

pip install opentimestamps-client
ots upgrade MANIFEST.txt.ots
ots verify MANIFEST.txt.ots
# Reports the block height + timestamp of the seal.

Want the full harness or a specific file class re-run? Ask for it - we send the scripts and raw numbers.

Compress your own sample

Do not take the tables on faith. Run the no-signup browser trial on your own file and watch the byte-exact result, the +1 byte floor and the SHA-256 roundtrip happen in front of you.