LaurieWired's Tailslayer revealed something uncomfortable: DRAM channel placement is software-predictable, which means
ASLR, session keys, and LLM token sampling are all weaker than assumed. If the physical randomness layer is soft, the
internet-connected stack can't be a root of trust.
So we stopped trusting it.
PHANTOM is what you'd get if a flight data recorder, a notary public, and a LoRa mesh had a baby — and the notary used
physics instead of a stamp. Every AI output, position broadcast, and tactical order gets hash-chained to a quantum beacon
and signed with ML-DSA-65 (NIST FIPS 204). Runs over LoRa SX1262 + ESP-NOW. No cloud. No cell towers. No DNS. ~$650 of
commodity hardware.
Phase 5 forks Tailslayer to drive DRAM XOR offsets from a hardware quantum chip instead of a software beacon — making the
Hub's LLM inference both physically unpredictable and measurably faster. Not built yet. That's the research question.
Palantir at home: a cyberdeck and some ESP32s from AliExpress.
So we stopped trusting it.
PHANTOM is what you'd get if a flight data recorder, a notary public, and a LoRa mesh had a baby — and the notary used physics instead of a stamp. Every AI output, position broadcast, and tactical order gets hash-chained to a quantum beacon and signed with ML-DSA-65 (NIST FIPS 204). Runs over LoRa SX1262 + ESP-NOW. No cloud. No cell towers. No DNS. ~$650 of commodity hardware.
Phase 5 forks Tailslayer to drive DRAM XOR offsets from a hardware quantum chip instead of a software beacon — making the Hub's LLM inference both physically unpredictable and measurably faster. Not built yet. That's the research question.
Palantir at home: a cyberdeck and some ESP32s from AliExpress.
https://github.com/seppulcro/phantom (https://github.com/seppulcro/phantom)