Aura should prefer independently compressed chunks over whole-file compression.
Aura0 profile:
chunked zstd
high compression level
chunk directory required
chunk-local decode state
Aura1 profile:
uncompressed for maximum replay speed
or chunked low-level zstd when disk matters
Whole-file compression can produce slightly smaller files, but it makes parallel
conversion, partial validation, and corruption isolation worse. .aura0 should
preserve enough per-chunk metadata to let workers convert chunks to .aura1
files independently.
A one-day BTCUSDT 1m sample was materialized from the current integer-stamp
harness and compressed with the external OpenZL zli CLI as a whole-file
baseline. Every decompressed output matched the original bytes.
file original openzl serial openzl train-inline
.aura 128494 68329 62769
.aura0 33723 32115 30480
.aura1 83945 68784 59091
The result is directionally useful but not a replacement for chunked Aura
compression. OpenZL helps the raw ingest and hot replay files more than .aura0;
after Aura0 has applied candle/residual transforms and bitpacking, whole-file
OpenZL only trims a small amount. Compressed .aura1 can materialize hot bytes
faster than the current generic .aura0 -> .aura1 path, but it stores much more
data than .aura0. Re-run this comparison after a fused Aura0-to-Aura1
transcoder exists.