GIST compresses 80 bytes of sensor meaning into 2 bytes — a codebook pointer — using a dendritic tree lookup. The trunk selects a branch by the dominant signal axes; the branch narrows to the exact leaf cluster. One table lookup. O(1) decode on a $0.50 microcontroller with 2 KB of RAM.

Validated across 15 sensor domains: bearing vibration, cardiac rhythms, network traffic, human activity, audio, turbofan degradation, gearbox fault, brain signals, and more. Six domains exceed SF 0.998. All 15 exceed SF 0.90. Fidelity is 0.975 at 1% BER and 0.916 at −10 dB SNR with holographic PHY — no cliff effect, no catastrophic drop.

No neural networks at runtime. No cloud. No fragmentation. Every steady-state frame fits Sigfox's 12-byte uplink. HMAC-SHA256 per frame, key ratchet every 10K messages — 0 forgeries accepted in 100,000 adversarial attempts. Post-quantum codebook signing available via ML-DSA-44 (NIST FIPS 204).

The pipeline adapts to any sensor without rewriting firmware. A universal feature extractor projects raw signals from any domain into the same 20-dimensional space. Data geometry is auto-detected from the eigenspectrum — SOLID for tightly clustered signals like heartbeats and vibration, PLASMA for diffuse multi-sensor arrays — and the codebook topology shifts to match. Codebooks swap live over-the-air via chunked transfer with CRC-32 integrity checks, so a redeployed asset picks up an updated model in the field without reflashing firmware. Decode quality scales with receiver hardware: a $0.50 Cortex-M0 runs a flat table lookup; a Cortex-M4 adds a residual MLP correction for 7–15% better reconstruction at the same 2-byte wire format.