Loose text uploads — content spotcheck

The lake holds 5.51M loose text uploads (single files, not archives) — separate from the repo count. We classified the whole pool deterministically by filename extension + sniffed MIME, then read 379 randomly-sampled files (stratified across 18 extensions, fetched from vox-lake-raw) to verify each extension actually contains what it claims. Every file below is real; click to read its head and see why it was labeled.
extpool filespool %readis code
.txt2,254,35240.9%200%
.xml1,054,88019.1%200%
.html536,6009.7%1217%
.srt434,9727.9%200%
.json155,6792.8%220%
.lua153,8942.8%2190%
.py153,5492.8%2181%
.conf79,0181.4%210%
.sh49,8380.9%2584%
.php22,2430.4%2152%
.cpp20,7630.4%2492%
.mq418,5730.3%2688%
.js18,2460.3%1788%
.h8,4060.2%2190%
.c6,0260.1%2492%
.java5,5930.1%2696%
.sql4,8510.1%1493%
.cs4,3300.1%2496%
~496k real source files  ·  9% of 5.51M
Extension purity is content-verified: code extensions run 81–96% real code (only .php dips to ~52%, half being markup/data). The mass of the pool — .txt (41%), .xml (19%), .html (10%), subtitles .srt/.ass/.vtt (9%) — verified as prose, credential/proxy dumps, saved pages, and subtitles: ~0% code. The estimate reweights per-extension purity to the true pool distribution, and is conservative (compiled bytecode, ELF-misnamed, and obfuscated scripts are counted as non-source).
real source 232credential / proxy dump 5prose / notes / offset lists 33subtitles 20json / xml / html / csv data 53config 19minified / packed / obfuscated 15logs 2
Select a file