Performance

The performance section is the evidence contract for the portal. Its job is not to make fq-compressor look larger than life. Its job is to separate what the repository can prove from what remains design intent.

Performance ledger

Public claims stay attached to method, artifacts, and retrieval cost.

This portal does not separate headline numbers from archive semantics. Every metric is paired with the subsystem, methodology, or repository artifact that makes it auditable.

Open methodology

01

3.97x

Archive density

Compression ratio is presented as a bounded result, not as a floating claim detached from dataset scope.

• ERR091571 smoke-scale artifact
• Public benchmark report in repo

Trace the evidence

02

11.9 MB/s

Write-path throughput

Compression speed is interpreted through the pipeline, chunking, and backpressure design, not as an isolated stopwatch number.

• Pipeline topology
• Block-local work scheduling

Inspect the pipeline

03

62.3 MB/s

Read-back speed

Decode speed stays in scope because random access is only useful if retrieval remains practical under real archive workflows.

• Decompression path
• Original-order restore boundary

Read the algorithm brief

04

O(1)

Random access

Indexed lookup is treated as a first-class contract. The format and block map are part of the public thesis, not an implementation footnote.

• FQC block index
• Range decode without full expansion

Study the format

Current public boundary

• The tracked public evidence is currently centered on ENA accession ERR091571 and a reproducible 2,000-record smoke-scale subset.
• The site can discuss compression density, compression throughput, decompression throughput, and archive semantics, but not claim an already-proven specialized-peer leadership position.
• Spring remains deferred in the local evidence chain, so readers should treat the current benchmark story as bounded rather than exhaustive.
• Fresh tracked benchmark evidence should be generated with ./scripts/benchmark.sh, while ./scripts/benchmark_v2.sh remains the local comparison CLI for exploratory runs.
• Current ratio and throughput standing should be read from those generated reports instead of from hard-coded repository copy.

Claim matrix

Claim	Public strength	Why
fq-compressor has a documented short-read compression result	Strong	Supported by tracked benchmark artifacts in benchmark/results/
fq-compressor is already best-in-class among specialized peers	Not public	Comparative proof is incomplete while Spring stays deferred
O(1) random access is part of the system contract	Strong	Backed by format design, architecture docs, and repository code anchors
Throughput must be read together with archive semantics	Strong	The portal documents write path, read path, and retrieval cost as one story

Evidence anchors worth keeping open

• Methodology
• ./scripts/benchmark.sh --dataset err091571-local-supported --build --threads 1 --runs 1
• ./scripts/benchmark_v2.sh run --json <path> --report <path>
• docs/benchmark/README.md
• benchmark/results/err091571-local-supported.json
• benchmark/results/err091571-local-supported.md

Continue with

• Algorithms for the technical thesis behind the metrics
• References for papers, comparators, and project posture