About

Smart2Raw is written and maintained by Carlos Alberto Terêncio de Bastos, who holds the copyright — which is what makes the dual licence on this site possible. There is no company between you and the project: what you write on the contact page reaches the person who wrote the code.

The idea, in one paragraph

An integer column is eight bytes per element because a type declaration said so, not because the data needed it. Smart2Raw measures the real range and stores the column in the smallest native class that range requires — and then stops. The bytes stay native integers, so every operation runs on them directly, with no decode step to pay. The name inverts the usual promise on purpose: everyone offers to turn raw data into something smart. Here the intelligence goes into the classification, and what comes out is raw.

How the project is run

Three habits, and they are visible in the repository rather than claimed here.

Every number ships with the program that produced it. No benchmark prints a figure it cannot defend; each one asserts what it can before printing, and aborts on a disagreement rather than reporting a pretty result.

Chosen test cases inherit the blind spot of whoever chose them. That is why there is a differential fuzz suite with fixed seeds, and why it exists at all: it found a defect that twenty-five suites of carefully chosen cases had missed.

A defect gets published, not buried. When version 3.4.0 turned out to return truncated values on unsigned columns crossing 2^63 — with no error and a valid CRC — the fix came with the minimal reproducing case, the explanation, and an upgrade notice at the top of the release. It is on the technical scope page today, where anyone evaluating the project will read it.

The trail

Each deposit is a dated, third-party-timestamped record of the method, which is also what gives the project a priority date.

Where to find the project