Brotli's raw stream leaves integrity checks to the application
Successful Brotli decompression does not prove that a compressed stream is intact. The Google Brotli README warns that changes to some raw ranges can go unnoticed by the decoder. The stream contains neither checksums nor the uncompressed data length.
Google Brotli is a lossless compression project with native build routes, a command-line program and a Python module. Consider it when your integration requires Brotli-format data and can supply the integrity verification and framing it needs. If you expect the raw stream to provide those properties, that expectation needs to change before integration.
What happened when we ran it
The supplied sandbox record covers commit 2e2bcf1 on October 24, 2026. Installation succeeded in 226 seconds, with 36 dependencies recorded and 37 MB on disk. The build succeeded in 2 seconds. The sandbox's pytest run failed with exit 1 after 18 seconds: 299 passed and 22 failed out of 321.
The failure excerpt names CopyStatRegressionTest cases under tests/regression/t01/copystat_regression_test.py. Permission-mode subcases report FileNotFoundError; several other failures name the missing path 'None'. The named cases concern mode propagation, special mode bits, timestamp copying and stdin or stdout handling. The excerpt does not explain why the paths were absent.
The supplied evidence does not establish full repository test coverage across native, CLI and Python paths. It also does not include the failing invocation. These results describe the tests executed by the harness, rather than certifying every supported integration.
The log summary reports 14.68 seconds, while the outer test-step measurement records 18 seconds. Neither duration is a compression benchmark. Pip-audit reported 2 known vulnerabilities, but the supplied summary does not identify the affected package, severities or deployment exposure.
The sandbox had 3 CPUs and 8 GB of RAM. Its image label was ghcr.io/astral-sh/uv:python3.12-bookworm; that label does not verify the interpreter or operating system actually used. The record therefore cannot support a diagnosis that attributes the failures to a particular runtime, platform or codec defect.
Check the Python package and native components you will deploy
The main README documents system packages and native source builds alongside Python installation. The Python README at commit 2e2bcf1 supports Python 3.10 and later.
That Python README also describes building against an installed system Brotli library instead of vendored sources. It requires pkg-config and all three components: brotlicommon, brotlienc and brotlidec. Choosing this route means checking the native components as well as the Python package.
The sandbox build result applies to the measured checkout. It does not establish equivalent behavior for every distribution package or Python wheel. Validate the artifact you intend to deploy. If your CLI workflow depends on preserved permissions or timestamps, include those properties in your checks because the observed failures concern file metadata.
Decide how to verify and frame the data
For a backup or transfer pipeline, define how the receiver will verify the recovered bytes against the intended data. Successful decompression alone is insufficient under the README's stream warning.
If your application needs an expected output size, provide it through the surrounding protocol or storage format. The raw stream described by the README does not contain that metadata. These are integration recommendations derived from the documented format boundary. The supplied lab record contains no corruption, adversarial-input or memory-use experiments.
Choose alternatives by format and workload
Zstandard offers dictionary training for related small records. Its README says training works when there is correlation within a family of small samples. Consider it when both ends can choose the format and your records fit that pattern. The supplied lab record contains no comparative throughput or compression-ratio measurements, so it cannot establish a performance winner.
zlib supports the zlib, DEFLATE and gzip formats. Choose it when an existing integration requires one of those formats.
For a Brotli integration, the next step is to validate a representative round trip with your deployment artifact, check metadata behavior if you use the CLI, and investigate the harness's missing-path errors and audit findings. Those checks remain work to do; this review does not report that they passed.