ANTUMBRA
The trust layer of the human-machine economy. Private payments finalized in seconds, humans without biometrics, AI agents accountable by construction, and a reputation that money cannot buy, because it is made of time.
Agent commerce has no trust. Personhood has biometric debt. Privacy chains are slow or unreadable. Nobody serves the square in the middle.
Millions of agents will pay on behalf of humans through rails that say nothing about who is reliable, who vouches for them, or how far their budget goes. ANTUMBRA treats that missing piece, verifiable trust, as a protocol primitive. The scarce resource of the coming economy is not throughput. It is confidence, and confidence is the only thing that must be built in time, because time cannot be bought.
The architecture of a shadow, kept by construction
A three-layer consensus score: deeds, peer attestations, and time. Non-transferable, corroded by absence, crushed by cheating. The one resource money cannot fabricate, because it takes years to build.
One person, one voice, no biometrics. Presence, capped sponsorship and seniority instead of iris scans. Nothing about your body is ever collected, anywhere, by anyone.
Agents that pay on your behalf, by construction accountable: a human sponsor who answers for them, a declarative spending perimeter, a revocation that gates every future spend.
AI senators propose, red-team and implement protocol changes under human veto: a seven-step pipeline, three blind implementations, juror inspection, progressive rollout with rollback. Paid like miners, ruined like attackers. Post-genesis, staged, audited.
Private by default, provable on demand. Three levels of view keys let you show one payment, one ledger, or one compliance fact, without ever exposing anyone else.
Fifty-five seats drawn by reputation sign a checkpoint every four seconds. Payments become irreversible in under six seconds, with zero capital locked: this is not proof of stake.
Each 4-year eclipse emits 61.8 % of the previous one, the golden section. Exactly 10 000 000 ATU exist after 8 years, and the ledger closes on the cap in year 136. The same proportion already drives RandomX.
Our own simulation broke our own rules, then the fix held
A deterministic simulation of the social core replayed sixteen years with a farm of fake profiles and a whale of unlimited capital. The first pass captured all 55 finality seats at year 10. Four corrective rules closed the window: the same maximum attack now takes zero seats, and the whale never even becomes a candidate. That simulation ships in the repository as the regression test of the social core.
A NO-GO is an acceptable, published result
Specification, whitepaper v1.2 (audit closed), architecture decisions, social-core simulation green on 12 seeds
CPU BlockDAG prototype on devnet, Dandelion++ propagation, 24 h without unexpected reorgs, capacity budget measured
The Ring and Kleos v0: signed checkpoints, bootstrap ratchet, sub-6 s finality over 100 000 replayed blocks
Ember and Cipher identities, Lumen view keys, external audit of the diff
Public testnet: faucet, explorer, 100 nodes, three consecutive no-incident upgrades, Arena drills
Genesis when every exit criterion is green: public ceremony, signed binaries for five platforms
A shadow faithfully kept, a ring of light anyone can verify
The whitepaper is the full contract: transaction lifecycle with its cryptographic constructions, the reputation algorithm and its corrected rules, the golden eclipse emission, three-tier contracts, the threat register and the zero-defect engineering method.