LTF

Tests whether a claim-based document model — bitemporal facts, exact-text provenance, entity resolution — can hold its own against plain JSON on disk size. First pass (SQLite-backed) lost real-document benchmarks by up to 781%; a rewrite onto a custom binary encoding closed most of the gap. One reported win was later found to be measured against pretty-printed rather than minified JSON — caught and corrected before it went anywhere, then turned into a genuine, re-verified 11.2% reduction by fixing the two causes the correction surfaced. Designed to eventually feed the lab’s own ai_brain memory infrastructure; that integration isn’t built yet.

What it is, plainly

A frontier application built for a Malaysian constraint that existing tools either ignore or addressed last decade. The lab built it because no one else had — and could back it with a build log, not a marketing deck.

Why this exists

LTF isn’t a product for a customer — it’s part of the tooling the lab’s own work depends on. It’s documented here because the lab publishes what it builds, not because it’s for sale.

Where it stands

Status: Prototype.
This is an honest label. No “coming soon” theatre, no inflated readiness claims — if it’s experimental, we say so. If you want the current build log, talk to the lab.