What is your data actually worth? — at1 appraise
“Data is an asset” has no measurement standard. Raw byte-count is the usual proxy, and it lies: it double-counts duplicates and rewards padding. at1 appraise values a dataset by its information content— its compressed size under a pinned reference compressor (lzma preset 6) — giving a number for M&A due diligence, data marketplaces, and insurers underwriting data as an asset.
Use it
# value one dataset by its information content
at1 appraise value dataset.parquet
# raw bytes : 1,142,180
# MDL valuation : 145,216 bytes of information
# information density : 12.7%
# rank a portfolio by information content
at1 appraise rank a.csv b.csv c.csvThe three properties byte-count fails
- Ranks information — zeros < structured < random, by actual information content.
- Duplication-invariant — 10×-duplicated data values about the same as 1×; it’s the same information.
- Padding-invariant — padding adds bytes, not value, and the metric ignores it.
Compressor-approximated complexity is an upperbound on information content, and a standard requires pinning the reference. Information content is not the same as semantic value — random bytes appraise highest and are worth least — so pair this with domain judgment, not instead of it. To see where a file’s compressibility comes from, run the compressibility audit.