LENS · CVM-0

The file that is also the dashboard

A .lens file holds a dataset as typed compressed columns, a Merkle root over every section, and an English self-description. Reports run on the compressed columns— a sum touches each delta-chunk once, a category mix each RLE run, a log search each grammar rule — so the work scales with the structure in your data, not its size. Every number is emitted by a program and signed by a certificate.

31.8×
less work: numeric summary on chunk stats vs a row scan
1 file
dataset + query engine + proof system, no database, no server
per-number
every reported value carries a program + output certificate
refuse-on-tamper
one flipped byte and the container won't load

Build it, then report on it

at1 lens build fleet.csv -o fleet.lens
#   20000 rows, 6 cols (ts:delta, service:dict, status:dict,
#   latency:delta, logtext:gram); 178 KB; root=357bae70...
at1 lens report fleet.lens --search retry
#   Summary: latency sum=8.9M avg=447  [cert 08e6.../80ef...]
#     · 628 compressed ops vs ~20000 row ops (31.8x less work)
#   Category mix: status OK=17588 WARN=1810 ERR=602  [cert 57c5...]
#   Log search 'retry': 4080 matches (on the grammar)  [cert ...]

Tier-3 execution

Compute on the rule, not the data: a sum reads chunk totals, a mix reads run lengths, a search walks the grammar. Work scales with compressibility.

Certified numbers

Every value is emitted by a named program and signed (program hash, output hash, container root). Re-run and the certificate is identical.

Honest scope (M1)

This is the CLI reference VM: text reports, one file, local. Browser host, live ingest, and the compact self-carrying interpreter are the next milestones, not claimed here.

Wins scale with structure in your data; speed is stated as ops-ratios (durable), wall-clock as same-substrate. “Secure” means integrity + reproducibility, not confidentiality.