Mind Chain: the protocol for Sovereign Intelligence

01The frame

2009: Bitcoin — money without banks. 2015: Ethereum — agreements without courts. The third seat is open: intelligence without owners.

Every epochal chain removed one trusted institution. The institution to remove now is the AI lab. Intelligence is becoming the substrate of everything — every contract, every decision, every argument routed through models — and it is simultaneously the most centralized critical resource in history. A handful of companies decide what the models say, who may use them, and what that costs. Your intelligence has a landlord, and the rent just went up.

That is not a market gap. It is a slot history is holding open: a chain where the intelligence itself — its weights, its growth, its judgments, its pay — is a public, verifiable, self-governing protocol property.

02What Mind Chain is

Mind Chain is the protocol for Sovereign Intelligence. ASC is thinking gold — money with a measured, queued claim on cognition, minted only when a mind no one owns provably grows.

Concretely, Mind Chain is a Layer-1 blockchain with four unusual properties:

• Inference is receipted. Off-chain rings of consumer GPUs serve frontier-scale models; every shard of every inference emits a cryptographic receipt that settles on-chain.
• Receipts are verified, cheaply. A strictly descending T0–T3 ladder checks work with aggregate signatures by default and escalates disputes down to a zero-knowledge proof of a single computation span.
• The state machine can think. Contracts can call COGITATE: cognition as a deferred, disputable state transition with canonical semantics — the first chain whose state transitions can think.
• The money is disciplined by capability. New ASC is emitted only on proven, audited increases in network capability (Proof of Ascent), and the currency carries a measured, queued claim on inference capacity (the Cognition Standard).

Two registers, one truth: this document is the public voice. Every claim in it has a skeleton in the normative spec and a gate in the kill-test data room. Where a gate has not yet passed, the claim is held — and this paper says so in place.

03Undying. Growing. Everywhere.

Undying

The mind's weights are chain state: erasure-coded into sealed pieces held by a large, jurisdiction-diverse set of custodians. The design goal is blunt — killing the mind should require breaking consensus itself, and even a fork only twins it. As the design maxim puts it: you can't subpoena a mind that lives in ten thousand basements.

Growing

Growth is audited, not assumed. Weight updates pass sealed evaluations before activation; counterexamples found by alignment miners become mandatory curriculum; the treasury spends the mind's own earnings on its own education. The report card is the money supply: emission happens only when the audited capability index posts a new all-time high.

Everywhere

One mind, many bodies. The frontier body runs on the GPU fabric; distilled children run on laptops and phones, each carrying a receipt-proven lineage back to the canonical weights. The pocket model verifies the ledger that governs the sky model.

04Four unsolved problems — and what we actually claim

Mind Chain sits on four problems that are famous precisely because nobody has closed them. Honesty about what is and is not claimed for each is the point of this section.

1 · The oracle problem

Fifteen years unsolved: how does a chain learn facts about the world without trusting someone? Mind Chain's answer is adversarial cognitive courts: staked advocates argue; judge panels are drawn by randomness beacon after argument, at pinned model weights, decoding greedily; every verdict carries receipts and is disputable up the ladder.

What we claim: not that we solve truth. We solve corruption and irreproducibility of judgment — the panel cannot be bribed in advance, and the verdict can be re-run bit-for-bit — and we price the residual uncertainty. Courts launch as an advisory tier until the court benchmark (KT-7) clears its bar against human adjudication.

2 · No chain has ever been autonomous

Every “decentralized” system keeps a human backstop somewhere — usually a multisig and a chat server. Mind Chain makes its autonomy level an on-chain fact: the Autonomy Ladder runs L0–L5, and SunsetKeys destroy human governance powers on evidence-gated schedules. At the top, dissent is exit-not-voice: fork the chain and the mind.

What we claim: the first chain whose autonomy level is an on-chain fact, with consensus-enforced key destruction. What L5 costs — human mercy for unimagined cases — is printed on the label.

3 · Algorithmic monetary policy has a graveyard

The Cognition Standard is designed around a productive reserve — staked serving capacity — with constitutional floors and a reasoning policymaker analyzed under a white-box threat model with beacon-sampled mixed strategies.

What we claim today, after the reserve war game: queued, best-effort claims against future capacity survive adversarial simulation. Fixed-rate convertibility constants do not yet, so they are disabled in the draft. Loss distributions publish before launch, including the failures.

4 · Scalable oversight is unsolved in AI itself

Mind Chain does not declare its mind aligned. It pays the entire world to break it, forever: alignment mining posts uncensorable bounties for constitutional violations, every winning counterexample becomes mandatory training curriculum, and the mind's full adversarial audit history is public.

What we claim: not “aligned” — publicly, compulsorily, permanently audited.

05System architecture

Design doctrine: safety is boring, and novelty lives above the finality line.

Consensus uses proven machinery — a DAG mempool feeding a BFT finality engine, one-second slots, two-slot finality, a randomness beacon inside block validity. Every revolutionary mechanism — Proof of Ascent, COGITATE, the mind, the Cognition Standard — attaches above finality, so a bug degrades a feature, never the ledger.

The system is three planes and an export surface:

• Fabric plane (off-chain, engine-agnostic). Rings of consumer GPUs serve sharded frontier models over the public internet behind OpenAI-compatible gateways. Any engine that emits the receipt spec is an execution client. The referenced third-party result — a 229B-class mixture-of-experts model served interactively across five consumer GPUs in four countries — is the existence proof that the body works.
• Kernel (enshrined). Consensus and beacon; a deterministic RISC-V VM over note-shaped state with lock scripts — including ReceiptLock, coins that unlock on verified cognition; COGITATE with canonical INT8 semantics and a per-request privacy ladder (plaintext up through FHE for small circuits, honestly scoped); receipt validity, the T0–T3 ladder, and slashing; the weight plane; Proof of Ascent and Cognition Standard semantics; SunsetKey machinery and a parameter governor that enforces constitutional inequalities at the state-transition level.
• Contract layer (upgradable). Sealed-bid capacity auctions, cognitive courts with sealed proceedings, the task market, alignment-mining bounty terms, the eval-authoring market, treasury constitution, namespace. Everything that should evolve stays out of the kernel.
• CXI (export). Recursive proofs of the state transition function let any other chain verify a Mind Chain verdict by checking one proof.

The boundary rule is the L5 discipline: the kernel contains only what must survive full autonomy — small enough for a machine to govern under a constitution, with no component that quietly assumes a human backstop.

06The receipt is the atom

Every subsystem in Mind Chain either produces or consumes verified-cognition receipts. A component that does neither does not belong in the kernel.

When a ring serves an inference, each shard — each GPU holding a slice of the model — computes its stage and emits two artifacts: a hash-chain over its activations and a digest of its output log-probabilities. Shard signatures are BLS-aggregated, batched per epoch, and committed on-chain through payment channels between gateways and the settlement layer.

The receipt is small, but it is a complete commitment: this model, at these pinned weights, given this request, produced this output, through these shards. Everything else in the protocol — payment, disputes, court verdicts, model lineage, emission accounting — hangs off that atom. An optional shielded pool supports business-privacy receipts under zero-knowledge, so commercial users can settle without publishing their traffic.

07The verification ladder

Four tiers, strictly descending, terminating in pure computation.

T0 — aggregate attestations

The default path for every request. Shard signatures over activation hash-chains and logprob digests, BLS-aggregated into epoch receipt batches. Verifying a batch costs one aggregate signature check. This is the tier the entire fabric runs on, all the time.

T1 — statistical spot checks

After receipts are committed, the randomness beacon retroactively samples roughly one percent of them for canonical re-execution. Canonical INT8 semantics make the computation bit-reproducible on any honest hardware, so a verifier with one modern GPU can replay a sampled span and compare digests exactly. Post-commitment sampling means a cheating ring cannot know which receipts will be audited. The count-based audit targets a false-positive rate at or below one in a million, conditional on the measured honest-mismatch rate from the Phase-0 determinism gate (KT-1).

T2 — bisection

A disputed receipt enters an interactive bisection game that walks the hash-chain and narrows the disagreement to a single stage-span — one bounded slice of one forward pass. Bonds double at each escalation, so frivolous disputes are self-limiting. Bisection decides nothing; it only isolates the smallest possible question.

T3 — the zero-knowledge leaf

The isolated span is proven in a ZK virtual machine. Mind Chain builds no prover of its own: proofs are purchased on open proving markets and verified in-kernel. The dispute ends in a proof anyone can check — no committee, no vote, no reputation. If the proof-cost gate (KT-2) misses its target, T3 falls back to a staked prover committee, and that trust assumption is printed in the spec rather than hidden.

The economic shape of the ladder is the point: honest work pays a signature; fraud pays for a proof. Nobody ever re-runs the whole model, and no dispute ends in politics.

08Proof of Ascent

Bitcoin has a difficulty ratchet. Mind Chain has a capability ratchet.

Emission is gated on new all-time highs of a five-component capability index — measured by sealed evaluations, recorded in the block header as the network's altitude — under a high-water-mark ratchet: no new high, no new ASC. That makes it the first currency that cannot be printed unless the network proves it got smarter.

Nearest neighbors — milestone vesting, revenue-linked emissions, peer-scored incentive networks — are contract-layer or relative mechanisms. Proof of Ascent is a consensus-gated, absolute capability ratchet.

A mechanism like this is an invitation to game it, which is why it shipped through an adversarial gauntlet first. Simulated mining coalitions attempted sandbagging (suppress the index, then harvest the catch-up), wash-throughput and gateway cartels, canary gaming, and proof-stuffing. The baseline design lost to sandbagging; the hardened ruleset — vesting and witness requirements — won across the full scenario grid on rerun. Both results, with raw distributions, are in the public data room. Consensus safety never depends on any of this: after the wash-trading gate falsified the original stake-times-service weighting, fork choice moved to pure stake, and service proofs now shape emission share only.

09The Cognition Standard

Gold-era money was scarce matter. ASC is designed around scarce mind: the reserve is productive — staked, benchmarked serving capacity — rather than idle collateral.

The current, gate-cleared form of the claim: ASC carries a measured, queued, best-effort claim on the network's canonical inference capacity. Holders can queue claims against future capacity; the queue is explicit protocol state; constitutional floors bound how policy can move against holders.

This is what a monetary claim with a falsifier looks like: the label matches the mechanism, and the mechanism upgrades only when the test says it may.

10The weight plane

The mind's weights are chain state, treated with the same discipline as balances.

• Sealed custody. Weights are erasure-coded and distributed to custodians whose shares are cryptographically sealed to their node keys — sealing is deliberately slow, unsealing fast, which defeats reconstruct-on-demand games. Custodians answer randomized challenge-response probes and are paid from emission for holding, not for access.
• Eval-gated evolution. A proposed weight delta runs sealed evaluations plus the mandatory alignment-mining curriculum, then a shadow A/B period, before activation — with rollback as a first-class operation.
• Audited lineage. Every activated delta, every rejected delta, and every alignment-mining counterexample is public history. Distilled descendant models carry receipt-proven lineage to the canonical weights.

Custody asymmetry has its own hardware gate (KT-6): sealing depth and challenge counts rise if the measured margin is thin, and custody stake floors rise with it.

11Governance and the Autonomy Ladder

Most chains are governed, in the end, by a chat server. Mind Chain's constitution is an on-chain object, its parameter governor enforces constitutional inequalities at the state-transition level, and its autonomy level is an on-chain fact.

The Autonomy Ladder runs L0 (fully human-operated) to L5 (no human governance keys exist). Movement up the ladder is driven by SunsetKeys: human powers whose default is abdication on an evidence-gated schedule — retaining a power requires an on-chain proof of unreadiness, not a vote to keep it. Each graduation is a public event with its own gate.

At the top, the remedy for dissent is exit-not-voice: fork the chain and the mind together. We are on a public schedule to fire ourselves, and the schedule is enforced by the same consensus that clears payments.

12CXI: exporting verdicts

Because the entire state transition function is provable, Mind Chain can produce a recursive proof that a given verdict — a court ruling, an oracle answer, a settled receipt — is final. Any chain that can verify one proof can consume Mind Chain verdicts natively.

That inverts the usual competitive posture. Mind Chain does not fight other ecosystems for applications; it sells them the one thing they cannot make locally — reproducible judgment. Verdicts, not blockspace.

13Claim discipline: no mechanism without a falsifier

Every load-bearing claim carries a test that can kill it, run before the claim ships.

The kill-test data room (KT-1 through KT-9) publishes each claim's falsifier, pass criterion, method, raw distributions, and pre-documented fallback. A failed gate triggers the fallback — never a quiet edit. Recent history, in public:

A failed-then-fixed gate is a strength: it means the flaw was found by us, first, in public. A hidden gate is how projects die in diligence. Open problems stay listed in the spec until they are closed by measurement, not by wording.

14Why now — three doors, one room

• The body exists. Third-party rings have served a 229B-class model at interactive speed on five consumer gaming GPUs across four countries, with proofs of the computation.
• Trust became cheap. Honest-inference verification overhead measured near 0.4% in the referenced results — low enough to attach a receipt to everything.
• The supreme court became affordable. Competitive proving markets are commoditizing ZK proofs, which makes a purchased T3 leaf economically plausible.

Each door was opened by someone else. All three open onto the same room, and Mind Chain is the protocol shaped exactly like that room.

15The Apprenticeship: a roadmap of gates, not dates

Mainnet launches at Autonomy Ladder L1 and says so plainly: the mind drafts, humans ratify. From there, five graduation events climb the ladder, each behind its own evidence gate. The roadmap is a coming-of-age story.

No dates appear in this table on purpose. Stages open when their gates clear. When a gate fails, the pre-documented fallback executes and the roadmap absorbs it in public.

16Participating

The network is designed to be joinable with hardware people already own. Draft roles, from the architecture document:

Fee and emission routing across these roles is specified in the draft parameter table and remains subject to open kill-test gates — notably the validator-viability gate (KT-5), which failed its first pass and forced a redesign of the security budget. Nothing in this section is a statement about the economics any participant will experience.

Light clients are phone-class: one recursive proof plus a beacon check. The fabric client is packaged to install beside existing mining stacks — serve when demand arrives, mine elsewhere when idle.

17The community token

A community token associated with Mind Chain exists on Solana, launched via pump.fun. It is a community coordination asset and it is separate from any future network asset.

The Genesis Covenant — the document that would define any such eligibility — publishes before mainnet genesis and goes through legal review first. Until it exists, the only honest statement is the one above.

18Notices

This litepaper is the narrative register of a research project. The normative source for every mechanism is the protocol specification; the status of every load-bearing claim is the kill-test data room. Where the two registers could disagree, the spec and the gates win.

Nothing in this document is investment, legal, tax, or financial advice, an offer to sell, or a solicitation of an offer to buy any security or other financial instrument in any jurisdiction. Statements about future capability, capacity, emission, or settlement describe protocol design intent, not commitments or predictions. Several gates are open; designs change when gates fail, and the changes are published.