JR NetworkVision paper

Technical specification

JR Chain

Protocol specification: a Layer 1 that proves and pays for human work, in implementable detail.

Version 0.2 · Draft · November 2026

0Conformance and key words

This document specifies JR Chain and the JR Network protocol that runs on it (Proof of Human Work, JR ID, economics and governance) in enough detail to be implemented. The conceptual account is given in the vision paper.

The specification turns the three pillars of the vision paper into rules: the currency of AI agents (agent accounts and the agent toll, §3.7), the value of the human mind (JR ID and Proof of Human Work, §5–§7) and security beyond the quantum age (hybrid signatures and crypto-agility, §4.2, §4.6).

The main text describes the protocol. Rules are numbered in the form R<section>.<n>. The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “NOT RECOMMENDED”, “MAY” and “OPTIONAL” in this document are to be interpreted as described in BCP 14 [RFC 2119] [RFC 8174] when, and only when, they appear in all capitals, as shown here.

Requirement levelKey words
Absolute requirementMUST, REQUIRED, SHALL
Absolute prohibitionMUST NOT, SHALL NOT
RecommendationSHOULD, RECOMMENDED
DiscouragedSHOULD NOT, NOT RECOMMENDED
Truly optionalMAY, OPTIONAL

Sentences that carry no key word are informative. The Rationale passages in each section and the simulations are non-normative.

1Notation, glossary and clocks

1.1 Notation

  • H(·): a domain-separated cryptographic hash function; Htag(x)=H(JR/‖tag‖x).
  • ‖: concatenation · ⌊x⌋: floor · 𝟙[·]: indicator function.
  • Amounts are integers denominated in the smallest unit, mJR (1JR=109mJR); fractional amounts MUST NOT be used.
  • Epochs are indexed by e, checkpoints by n, accounts by i, items by j and validators by v.

1.2 Glossary

TermDefinition
AccountAn entity on JR Chain identified by an address or a name (name.jr).
JR IDA privacy-preserving proof-of-personhood identity bound to an account (§5).
Agent accountThe class of account used by an AI agent, bound to a JR ID holder (§3.7).
Agent tollThe JR fee that agent accounts pay for every action on the network; part of it is burned and the rest goes to the human-work pool (§3.7).
Assurance tierA tier derived from the JR ID score, from T0 (none) to T3 (highest).
TaskA unit of human work defined by a customer, who locks its fee up front (§6).
ItemThe smallest unit of a task that can be assigned to a single person.
WorkerA user holding a JR ID who answers items.
Human validatorA high-reputation worker who locks stake and resolves disagreements.
JuryRandomly selected human validators who decide the final stage of the dispute game.
Chain validatorA node operator that produces blocks and provides finality.
Gold itemA control item whose answer is known in advance and which cannot be told apart from real items.
Mining sessionA 24-hour mining window started by the user (§9).
Reward rootA Merkle root summarising an epoch's rewards, written to the chain (§9.3).
Sponsor poolA protocol account that covers users' transaction fees.

1.3 Clocks

ClockLengthUsed for
CheckpointSet by consensus (sub-second target)Unit of finality
EpochEPOCH_LEN (default 24 hours)Validator set, reward root, randomness
WeekMonday 00:00 – Sunday 23:59 (UTC+3 on testnet)Team results
Dispute windowDISPUTE_WINDOWItem and reward disputes
Unbonding periodUNBONDINGStake withdrawal

2System architecture

Figure 2.1Layers
ApplicationsJrKripto (JR Mining) · JR Network apps · customer dashboards · AI agents (MCP / API)
JR Network protocolTask market · assignment · consensus · dispute game · reputation · rewards · JR ID (score, tier trees, ZK, agent authority)
JR ChainObject-based parallel execution · DAG BFT consensus · fee market · human lane · hybrid post-quantum signatures
Data availability (DA)Task inputs, answer records, proofs · only commitments (hashes) on-chain

Design principles

  1. Purpose-built, yet programmable. The core operations (tasks, consensus, rewards, JR ID) are protocol modules; Move smart contracts are supported for third-party applications.
  2. Light chain, heavy edges. Task content and answers are not written to the chain; the chain holds only commitments, decisions and payments.
  3. Proof before trust. No participant is trusted on the strength of its reputation; every claim is backed by stake and by a record that can be disputed.
  4. Humans first. Verified humans do not lose fee auctions to bots (§3.5).
  5. No figure without a measurement. Throughput and capacity figures are published only once they have been measured under a disclosed workload and network configuration.

2.1 Node roles

RoleRunsStakeSlashed for
Chain validatorConsensus, execution, threshold randomnessVAL_MIN_STAKEDowntime, double signing
Full nodeState replica, RPC——
DA nodeTask data storage and samplingSet by the DA layerWithholding data
Reward calculatorEpoch reward root (§9.3)Calculator stakeIncorrect root
Identity issuerJR ID scores, tier-tree updatesIssuer stakeIncorrect tree update
Human validatorOn a phone: resolving disagreementsHV_MIN_STAKEGold accuracy falling below the threshold

3Consensus and execution

3.1 Foundations

The execution and consensus layer is derived from Sui's open-source codebase [1][2]. Every piece of state is a typed object, which is either owned by a single account (owned) or shared.

R3.1A transaction MUST declare as inputs every object it will read or write.

R3.2Transactions that touch only owned objects (JR transfers, reward claims) MUST be finalised on the fast path; transactions that touch shared objects (task objects) MUST go through ordered consensus.

R3.3Transactions that touch none of each other's objects MAY be executed in parallel; the result MUST be identical to that of sequential execution.

3.2 Consensus protocol

R3.4Ordering MUST be performed by an uncertified DAG-based BFT protocol (the Mysticeti family [2]).

R3.5The protocol MUST be safe against Byzantine adversaries controlling less than 1/3 of total stake.

R3.6A finalised checkpoint MUST NOT be reverted; reverting one MUST require slashing at least 1/3 of total stake.

f<N3⟹safety and liveness,|Q|≥⌈2N+13⌉
(3.1)

In stake-weighted terms, N is total stake, f is the adversary's stake and Q is the quorum required for a decision. Any two quorums always intersect in at least one honest validator.

Rationale: Sui reported a median consensus latency of about 400 ms on mainnet with Mysticeti (August 2024) [2]. JR Chain's own figure will be measured on testnet under a disclosed workload and then published.

3.3 Chain validators

R3.7The active set MUST consist of at most VAL_MAX validators; the set is updated at the start of every epoch.

R3.8A self-stake of at least VAL_MIN_STAKE is REQUIRED to become a validator.

R3.9JR holders MAY delegate their stake to a validator; delegated stake is added to the validator's weight and bears a proportional share of any slashing.

R3.10Slashing follows the table below.

FaultSlashFurther sanction
Brief downtime (< DOWN_SHORT)σdown1 (small)—
Extended downtimeσdown2Removal from the set
Double signing (equivocation)σeqRemoval from the set
Simultaneous double signing by ≥ 1/3 of total stake100%Permanent ban

3.4 Randomness

R3.11For every epoch, an unpredictable and unbiased randomness value ρe MUST be derived from the validators' threshold signature (a threshold randomness beacon). Task assignment, gold-item mixing and jury selection MUST use ρe as their only source of randomness.

ρe=Hrand(ThresholdSigskgroup(e))
(3.2)

The threshold signature cannot be produced without a quorum of validators taking part, and no single validator can see the result in advance.

3.5 Human lane (priority blockspace)

Adapted from World Chain's idea of “priority blockspace for humans” [4].

R3.12A fraction HUMAN_LANE of every checkpoint's gas capacity MUST be reserved for transactions that carry a valid human-lane ticket. Transactions in this lane are ordered by arrival, not by fee.

R3.13A human-lane ticket is the ZK proof of §5.3 with the statement (tier≥T1,nullifier=Hpbh(sk,month,k)), k<PBH_QUOTA. The same nullifier MUST NOT be used a second time.

R3.14Lane capacity left unused MUST be released to general transactions within the same checkpoint.

Simulation · Human lane: bots cannot outbid humans
0%50%100%0100200300bot demandhuman transactions included
With a human laneFee auction only

Block capacity is 100 units and verified humans demand 25. Without a lane, bots pay higher fees and are included first. The human lane reserves part of the capacity for humans in order of arrival; any unused part opens to everyone in the same block (adapted from World Chain's priority blockspace).

3.6 Fee market

Adapted from EIP-1559 [5].

R3.15The base fee MUST be updated at every checkpoint according to the rule below.

bn+1=max(bmin,bn·(1+gn−g∗g∗·d)),g∗=capacityELASTICITY
(3.3)

d=FEE_D (default 8), ELASTICITY (default 2).

R3.16The base fee MUST be burned; the priority fee, min(prioritymax,feemax−bn), goes to the block producer.

R3.17In sponsored transactions (§4.4) the sponsor pays the fee; the user MUST NOT be required to hold a JR balance.

Simulation · The base fee follows demand
012015304560blockbase fee (relative)
Base feeUsage / target

In every block the base fee rises by at most 12.5% when usage is above target and falls when it is below (d = 8). The base fee is burned, so burn rises with congestion and fees stay predictable.

3.7 Agent accounts and the agent toll

The first pillar of JR Network is that JR is the currency of AI agents: agents pay JR to exist on the network, to talk to one another and to humans, and to get work done, while verified humans act free of charge within a quota. The design adapts Hashcash's principle that “every message should carry a small cost” [22].

R3.18Every account MUST belong to exactly one of two classes: a human account (bound to a valid JR ID commitment) or an agent account. An agent account MUST be bound to exactly one JR ID holder through the authorisation certificate of R5.12.

R3.19Opening an agent account MUST lock a stake of AGENT_BOND JR. The stake is returned once the account has been closed and the UNBONDING period has elapsed; it is slashed if spam or abuse is proven.

R3.20Every action from an agent account (a transaction, a message object, or a Gate, Eye or Reach request) MUST pay an agent toll τ(e) in addition to the base fee.

τ(e)=AGENT_TOLL_BASE·max(1,ρ(e)TOLL_TARGET)2,ρ(e)=agent actions in the epochagent action capacity
(3.4)

Below the target utilisation the toll is flat; above it, the toll rises quadratically. It is independent of the base fee (R3.15): the base fee prices blockspace, while the agent toll prices agent activity.

R3.21A fraction TOLL_BURN of the agent tolls collected in an epoch MUST be burned; the remainder MUST be added to that epoch's human-work pool and MUST be distributed according to the rule in §8.4.

T(e)=∑ana(e)τ(e),Bagent(e)=TOLL_BURN·T(e),Hagent(e)=(1−TOLL_BURN)·T(e)
(3.5)

R3.22Human accounts MUST NOT be charged the agent toll for up to HUMAN_FREE_QUOTA actions per epoch. A human action beyond the quota pays a fraction HUMAN_OVER_RATE of the agent toll. This quota is separate from the sponsor quota of R4.9.

R3.23An agent's actions MUST NOT be deducted from its owner's human quota and MUST NOT earn the owner additional quota; adding agents MUST NOT confer additional free actions (I4).

R3.24The agent toll MUST NOT be a source of minting (R8.1). Paying the toll gives an agent no vote, no human reputation (wi) and no reward entitlement; agent reputation is kept separately under R3.26.

R3.25For unsolicited agent messages sent to a human account, the receiving human earns a fraction RECIPIENT_SHARE of the agent toll per message as a reward: human attention is paid for.

Reputation passport

R3.26An agent reputation ra∈[0,1] MUST be maintained for every agent account. Actions that finalise without being disputed raise it; slashes, upheld complaints and spam flags lower it; and it decays over time at rate AGENT_REP_DECAY.

ra←(1−δ)ra+δ·sa(e),sa(e)=clean actionsclean actions+γ·complaints+γs·slashes
(3.6)

δ = AGENT_REP_DECAY · γ, γₛ: penalty weights.

R3.27An agent's access tier MUST be determined by its stake and reputation (ACCESS_TIERS). A higher tier brings higher request limits, fewer verification challenges and priority on Stamp and Gate resources; it MUST NOT confer human quota, votes or minting rights.

R3.28An agent's stake and reputation are bound to its owner's JR ID. A slash applied to one agent MUST lower the reputation of the same owner's other agents by a fraction AGENT_REP_LINK; it MUST NOT be possible to escape a bad record by opening a new agent.

Figure 3.1How the agent toll flows
AI agentτ JR per actionAI agentτ JR per actionAI agentτ JR per actionmessage · request · payment · contactAgent tollT = Σ τ · congestionBurnθ · THuman-work pool(1 − θ) · TVerified human0 JR within quota
Simulation · Agent toll: bots pay, humans earn
10⁰10²10⁴10⁶10⁸1K10K100K1M10M100M1B10Bagents onlinedaily agent tolls (JR, log)
Total agent tolls per day500K JR
Burned (θ)300K JR
To the human-work pool200K JR
Cost of 1 million spam actions1.0K JR
Verified human (within quota)0 JR

Total tolls grow linearly with the number of agents; once the network runs above its target load the toll rises with the square of congestion, so spam gets expensive. Verified humans pay nothing within their quota. Toll and quota values are illustrative; the parameters are set by governance.

Rationale: 53% of web traffic is automated [26]. As agent activity grows, so does the agent-toll burn; demand for JR is directly proportional to agent activity (vision paper, Equation 3.5). Verified humans are not forced into fee auctions or into paying fees.

4Accounts and cryptography

4.1 Account names

R4.1Each account MAY hold at most one primary name (name.jr), adapted from NEAR named accounts [6]. Names consist of lowercase letters, digits and -, and are 3–32 characters long.

R4.2Single-character and reserved names are held in the foundation's registry; the registration rules change through governance.

4.2 Hybrid post-quantum signatures

R4.3From genesis, every account MUST support hybrid signatures: a classical signature (Ed25519) together with ML-DSA (NIST FIPS 204 [7]). For a hybrid account, a transaction is valid if and only if both signatures are valid.

R4.4In addition to the classical threshold signature, checkpoint certificates MUST be sealed with a post-quantum signature (ML-DSA or SLH-DSA, FIPS 205) every PQ_CHECKPOINT checkpoints.

R4.5When an account's classical key is declared broken (by a governance decision), it MUST be possible to switch the account to a mode in which it can transact only with its post-quantum key.

Valid(tx)⇔VerifyEd25519(pkc,tx,σc)∧VerifyML-DSA-44(pkq,tx,σq)
(4.1)
Figure 4.1Signature and public-key sizes
Ed25519 (classical)64 Bpublic key 32 B
Falcon-512666 Bpublic key 897 B
ML-DSA-44 (FIPS 204)2,420 Bpublic key 1,312 B
SLH-DSA-128s (FIPS 205)7,856 Bpublic key 32 B
Hybrid: Ed25519 + ML-DSA-442,484 Bpublic key 1,344 B
A hybrid transaction signature is 2,484 bytes. In checkpoint certificates the post-quantum seal is applied once every PQ_CHECKPOINT checkpoints rather than at every checkpoint, balancing finality against bandwidth.

Rationale: QRL has used hash-based signatures (XMSS) since 2018, and Algorand has used Falcon-signed state proofs since 2022 [8]. Because JR Chain is built from scratch, quantum protection can be in place at genesis, and the chain avoids the migration problem that existing chains will face.

4.3 Social login (zkLogin)

Adapted from Sui zkLogin [9].

R4.6A user MUST be able to derive an account from an identity token issued by an OpenID provider (Google, Apple). The address is derived as H(iss‖aud‖sub‖salt); the link between the provider identity and the on-chain address MUST be hidden by a zero-knowledge proof.

R4.7The salt service and the proof generator are initially operated by the foundation; opening both to several independent operators is an open item in Appendix C.

4.4 Sponsored transactions

R4.8The sender of a transaction and the owner of its gas MAY be different accounts; both parties MUST sign the complete transaction data, including the gas data [10].

R4.9The sponsor pool MUST provide every verified human with up to SPONSOR_QUOTA free transactions per epoch. Once the quota is exhausted, the user pays with their own JR.

4.5 Key recovery

R4.10Accounts MUST support multiple keys and key rotation. At least one recovery path (social login, a recovery key or m-of-n trusted contacts) MUST be defined.

4.6 Quantum threat model and crypto-agility

The third pillar of JR Network is security beyond the quantum age. The threat comes in two stages: (1) forgery of classical signatures using Shor's algorithm [25]; and (2) decryption tomorrow of encrypted traffic and public keys recorded today (“harvest now, decrypt later”). Mosca's inequality indicates when migration becomes necessary [24]:

x+y>z⟹risk,x=secrecy lifetime,y=migration time,z=time remaining to Q-Day
(4.2)
Figure 4.2Quantum timeline
2024203020352040FIPS 203·204·205NIST standardsDeprecatedRSA · ECC (draft)DisallowedNIST IR 8547 (draft)Harvest now: encrypted traffic is recorded todayDecrypt laterJR Chain: hybrid classical + post-quantum signatures from genesis
NIST IR 8547 draft: 112-bit RSA and elliptic-curve algorithms deprecated after 2030, all quantum-vulnerable algorithms disallowed after 2035 (proposed).

R4.11Every signature, public key and key encapsulation MUST carry an algorithm identifier (alg_id). It MUST be possible to add and remove algorithms through governance without forking the chain.

R4.12Node-to-node and client-to-node connections MUST use a hybrid key exchange that combines classical X25519 with ML-KEM-768 (FIPS 203 [7]).

R4.13From genesis onwards, every newly opened human and agent account MUST register a post-quantum public key (pkq); pkq is written to the chain once per account and referenced by its hash in transactions.

R4.14Checkpoint seals (R4.4) MUST include at least one hash-based signature (SLH-DSA, FIPS 205), so that history remains protected even if an unexpected weakness is found in lattice-based schemes.

R4.15The timeline recommended by NIST is followed [23]: governance MUST activate the hybrid-mandatory phase, in which classical signatures are accepted on new transactions only together with a post-quantum signature, by 2030 at the latest, and Q-Day mode (U1.1), in which classical signatures are rejected outright, by 2035 at the latest. If a quantum attack is demonstrated, these dates are brought forward.

R4.16A migration path to post-quantum counterparts of the zero-knowledge proof systems (hash-based proofs) MUST be kept as an open item in Appendix C.

ComponentToday (hybrid)After Q-Day
Transaction signatureEd25519 + ML-DSA-44ML-DSA-44 (fallback: SLH-DSA)
CheckpointBLS threshold signature + ML-DSA/SLH-DSA seal every PQ_CHECKPOINT checkpointsPost-quantum seal on every checkpoint
Key exchangeX25519 + ML-KEM-768ML-KEM-768
JR ID proofsHalo 2 (elliptic-curve based)Hash-based proof system (Appendix C)

5JR ID — proof of personhood

5.1 Score

The JR ID score S∈[0,1000] is a capped sum over independent assurance layers:

S=min(1000,∑kak·fk(xk)),fk∈[0,1]
(5.1)
Layer kSignalNote
TenureAccount age and a record of regular miningSaturates over time
BehaviourSession patterns, task consistency, gold accuracyOverwatch scorer
DeviceUniqueness of the device hash, device historyNo raw identifiers are stored, only hashes
GraphReferral and team graph; distance from farm clustersCluster detection
LivenessPeriodic short human tasksCounters account selling
CredentialsEmail, exchange KYC, optional government IDEach adds a fixed amount
Figure 5.1Score composition and tiers
T1T2T3
TenureBehaviourDeviceGraphLivenessCredentials

An example account: the capped sum of the layers is S = 900 / 1000. The tier thresholds (τ₁, τ₂, τ₃) are illustrative.

R5.1The score computation, its inputs and its weights (ak) MUST carry a published version number. Changes to the weights are made through governance.

R5.2Tier thresholds: T1:S≥τ1,T2:S≥τ2,T3:S≥τ3 (Appendix A).

R5.3An account carrying a fake-account flag MUST be in tier T0.

R5.4Biometric data MUST NOT be collected. Payment MUST NOT be made in exchange for personal data.

5.2 Identity commitments and tier trees

R5.5Each JR ID holder generates a secret skid on their device and writes only the commitment C=Hid(skid) to the chain.

R5.6A separate Merkle tree is kept for each tier (TreeT1,TreeT2,TreeT3). A commitment is added to the corresponding tree when its score crosses the tier threshold and removed when the score falls below it. Tree roots MUST be written to the chain every epoch.

R5.7The identity issuer that computes scores is initially the foundation; tree updates MUST be signed and open to dispute. Multiple independent issuers are an open item in Appendix C.

5.3 Zero-knowledge proof

An application asks the user to prove the following statement (following the Semaphore / World ID construction [4], with the Halo 2 proof system [11]):

Private inputs:  sk_id, Merkle path
Public inputs:   root(Tree_T), app_id, action_id, nullifier, signal
Statement:       H_id(sk_id) ∈ Tree_T
                 nullifier = H_nul(sk_id ‖ app_id ‖ action_id)

R5.8The proof system MUST NOT require a trusted setup (Halo 2 / PLONK family, IPA commitments [11]).

R5.9A second nullifier for the same (app_id,action_id) MUST be rejected (“one human, one vote”).

R5.10Nullifiers from different applications MUST NOT be linkable to one another.

R5.11Selective disclosure: a user MAY reveal specific records to a specific auditor by means of a viewing key [11].

5.4 Agent authorisation

Adapted from the World AgentKit approach [4].

R5.12A JR ID holder can issue an authorisation certificate to an AI agent: Auth=(agent_key,scope,quota,expiry,revocation_nullifier), signed with the holder's JR ID.

R5.13The number of authorisations a human can have in force at once is capped at AGENT_MAX; transactions made by an agent MUST count against its owner's “one human, one vote” quota. Adding agents MUST NOT confer additional votes or reward entitlements.

R5.14It MUST be possible to revoke an authorisation instantly by writing it to the on-chain revocation list.

6Task protocol (Proof of Human Work)

6.1 Task object

A task is a shared object (Appendix B.1).

FieldDescription
typeSIGNAL, PANEL, ARENA, EYE, GATE, REACH
content_rootMerkle root of the item contents in the DA layer
item_countN
replicasNumber of workers per item, n (default 3–5)
min_tierMinimum JR ID tier required of workers
targetOptional country, language and demographic filters (using only attributes bound to the tier trees)
fee_usdTotal fee in US dollars
deadlineClosing time of the task
privacyWhether answers are delivered to the customer raw or aggregated

R6.1When a task is created, its fee MUST be locked in JR according to the rule in §8.3. If the task closes before it is complete, the share attributable to uncompleted items MUST be refunded to the customer.

Figure 6.1Task and item state machines
TaskOPENCLOSINGFINALCANCELLEDdeadlinecompleterefundItemPROVISIONALDISPUTEDFINALdisputeverdictdispute window closed

6.2 Gold items

R6.2The protocol MUST mix gold items into every task stream at rate GOLD_RATE. Gold items MUST have the same format as real items and MUST NOT be distinguishable from them.

R6.3Gold answer keys MUST NOT be revealed until the dispute window for the items concerned has closed.

Rationale: this is the human-work counterpart of FLOP's “Ghost Task” approach (synthetic tasks whose answers are known) [12].

6.3 Assignment

R6.4The workers for item j MUST be selected from the eligible workers by a verifiable random function (VRF) seeded with ρe.

(πij,yij)=VRFski(ρe‖j),i∈Aj⇔yij2256<n|Uj|
(6.1)

Uj: workers eligible for the item (tier, target, online) · πij: a proof anyone can verify. Adapted from Algorand's cryptographic sortition.

R6.5Graph-distance rule: two workers who are in the same team, or fewer than GRAPH_DIST hops apart in the referral graph, MUST NOT be assigned to the same item.

R6.6One human, one score: accounts found to be under common control MUST be counted as a single participant; no two of them may be assigned to the same item.

Simulation · Can a group capture an item?
0%50%100%0%10%20%30%40%50%attacker share fitem captured
Independent fake accounts5.79%
Linked ring + rule0.00%
Expected loss per 1,000 items58 items

The red curve is the probability that fake accounts believed to be unlinked capture the majority of an item under random assignment. With the graph-distance rule on, a ring in the same referral chain can place at most one member on an item; the remaining risk is accounts whose links cannot be detected, and gold questions and stake (next figure) cover it.

6.4 Answering: commit and reveal

R6.7A worker MUST first submit commitment=Hans(answer‖salt‖item_id‖account); once the commitments of all replicas have been collected, or the time limit has expired, the worker MUST reveal the answer and the salt.

R6.8An answer that is not revealed, or that does not match its commitment, is invalid and counts as a wrong answer on a gold item.

Figure 6.2Commit and reveal timing
AssignmentCommit windowReveal windowConsensusWorker 1H(c‖t)c, tWorker 2H(c‖t)c, tWorker 3H(c‖t)c, tŷ
During the commit window no one can see anyone else's answer; during the reveal window a revealed answer is valid only if it matches the earlier commitment.

6.5 Consensus

y^j=argmaxy∑iwi·𝟙[yi=y],cj=maxy∑iwi·𝟙[yi=y]∑iwi
(6.2)

R6.9If cj≥κ, y^j becomes the provisional decision; otherwise the item goes to a human validator (§7.3).

R6.10For continuous-valued answers, y^j is the reputation-weighted median, and the strength of consensus is measured by deviation from the median.

R6.11Signal (forecasting) tasks do not expect consensus; their items are scored afterwards against the realised outcome.

6.6 Dispute game

Adapted from the bisection approach of Arbitrum BoLD [13].

R6.12The customer, any worker or any human validator MUST be able to dispute a task decision within DISPUTE_WINDOW by posting a stake of DISPUTE_BOND.

R6.13When a batch decision is disputed, the disagreement MUST be narrowed to a single item through rounds of bisection; the cost of a dispute is independent of the size of the task.

R6.14A single-item disagreement goes to a jury of J members selected using ρe; the jury decides by a ⌈2J/3⌉ majority. Jurors are subject to rule R6.5 with respect to the disputing party and the item's workers.

R6.15The party found to be in the wrong forfeits its stake; a fraction BURN_SLASH of the forfeited stake is burned and the remainder is paid to the prevailing party.

R6.16The total duration of any dispute is bounded above by DISPUTE_MAX, regardless of the number of participants.

Simulation · The bisection dispute game
Round0 / 14
Remaining range16,384 items
Rounds needed log₂N14
What the jury examines—

A task has N items and the parties disagree on where the result diverges. Each round halves the range and the parties say which half they diverge in; after log₂N rounds a single item remains. The jury looks only at that one item, so the cost of a dispute stays independent of the size of the task (adapted from Arbitrum BoLD).

6.7 Finality and data retention

R6.17A decision becomes final, and is written to the chain, when the dispute window closes or when the dispute is resolved.

R6.18Item contents, answers and decisions MUST be retained in the DA layer for at least the dispute window; after that they can be pruned. Only the roots remain on-chain.

R6.19Data delivered to the customer MUST NOT be linked to the worker's account (only the tier and permitted attributes are delivered).

7Reputation, human validators and juries

7.1 Reputation update

After each item finalises:

wi←clip[wmin,wmax]((1−λi)wi+λi[α·agreei+(1−α)·goldi])
(7.1)
λi=λup if si≥wi,λi=λdown if si<wi,λdown>λup
(7.2)

si is the signal inside the square brackets. agreei∈{0,1}: whether the answer matches the final decision · goldi∈{0,1}: correctness on a gold item (if there is none, the term is dropped and α=1 is used).

Simulation · Reputation is earned slowly and lost quickly
00.51060120180240finalised itemsreputation w
Honest workerCheats after item 120Bot answering at random

After every finalised item, reputation is updated with agreement with consensus and (every 20 items) gold-question accuracy. Because λ_down > λ_up, an account that starts cheating after item 120 loses in a few dozen items what it took months to build.

7.2 Becoming a human validator

R7.1Requirements for a human validator: tier ≥ T2, reputation ≥ wval and a stake of at least HV_MIN_STAKE JR.

R7.2A validator whose accuracy over its last HV_WINDOW gold items falls below τgold is slashed by σhv of its stake and suspended from the role.

7.3 Validator scoring and clipping

Adapted from Bittensor's Yuma Consensus [14]. Let pvj be the quality score that validator v assigns to worker j, and let sv be the validator's stake.

mj=medianv(s)(pvj),p^vj=min(pvj,mj),qj=∑vsvp^vj∑vsv
(7.3)
bondv(e)=(1−β)bondv(e−1)+β·(1−1|Jv|∑j∈Jv|pvj−qj|)
(7.4)

An exponential moving average of the validator's agreement with consensus (small β). Rewards are proportional to this bond.

R7.3Validator rewards MUST be distributed according to how closely a validator's scores have agreed with consensus over time (its bond); a validator that converges on consensus early earns a larger share.

7.4 Deterrence condition

R7.4Stake and slashing parameters MUST be chosen so that every role satisfies the following condition.

σ·s≥Gmaxpmin,P(none of m cheats is caught)=(1−q)m
(7.5)

Gmax: the largest gain obtainable by cheating within one dispute window · pmin: the lowest probability of being caught (derived from the gold rate and the audit rate).

Simulation · Cheating must lose money in expectation
0020406080100120dishonest answers mexpected profit (JR)
Uncaught after 20 cheats35.8%
Uncaught after 100 cheats0.59%
Best cheating plana loss
Deterrence condition σ·s ≥ g/q20 JR

Each dishonest answer hits a gold question or an audit with probability q. The chance of never being caught, (1 − q)ᵐ, falls quickly towards zero, and whoever is caught loses their stake. Wherever the curve stays below zero, cheating does not pay.

8Economics: minting, burning and fees

8.1 Minting

M(e)=K(e)+R(e)+W(e)+V(e)+Q(e)
(8.1)
TermSourceRule
K(e)Mining§9.1
R(e)Referral§9.2
W(e)Human-work poolW(e)=W0·2−⌊e/HW⌋, distributed per §8.4
V(e)Chain validatorsV(e)=V0·2−⌊e/HV⌋, by stake and performance
Q(e)Human validatorsThe HV_SHARE fraction of W(e), distributed per §7.3

R8.1JR MUST be minted only from the sources in this table. Minting through any other path MUST NOT occur.

8.2 Burning

B(e)=Bagent(e)+Bcustomer(e)+Bfee(e)+Bslash(e)
(8.2)

R8.7The agent-toll burn Bagent is computed per R3.21. The unburned part of the agent toll, Hagent(e), is added to that epoch's human-work pool and MUST NOT be counted as minting.

8.3 Customer payment and burn (burn-and-mint)

Adapted from Render's burn-and-mint equilibrium [15].

R8.2Task fees are set in US dollars. When a task is created, JR worth θ·feeusd MUST be calculated at the time-weighted average of the on-chain price feed and locked.

lockJR=θ·feeusdPtwap,Ptwap=1Δ∫t−ΔtmediankPk(τ)dτ
(8.3)

R8.3When an item finalises, the item's share of the lock MUST be burned and MUST be recorded to its workers as the item's work credit.

R8.4The remaining 1−θ of the fee is transferred to the operating account (sponsor pool, audits, foundation).

R8.5The price feed MUST draw on at least ORACLE_MIN independent sources, and the deviation of any single source MUST be bounded by the median.

8.4 Human-work pool distribution

Wi(e)=(W(e)+Hagent(e))·bi(e)·wi∑jbj(e)·wj
(8.4)

R8.6If an epoch contains no customer work at all, W(e) MUST be distributed across network tasks (gold items, calibration tasks, Signal) with equal credit weight.

8.5 Equilibrium

ΔS(e)=M(e)−B(e),Bagent+θ·U(e)P=M(e)⇒P∗=θ·UM−Bagent
(8.5)

Illustrative only; not a price promise. When B_agent > M, supply contracts regardless of customer demand.

Simulation · Burn-and-mint equilibrium
07.9 M191724epochJR / epochburn ≥ mint
Mint MBurn B
Epoch of equilibriumepoch 8
Net supply change after 24 epochs−78.64 M JR
Equilibrium condition P* = θU/M$0.0160 (epoch 1)

Thin line above: cumulative net supply change (up = growing supply, down = shrinking supply).

Mint falls with each halving (every 8 epochs); burn is the JR bought on the market with a share θ of customer payments and destroyed. The price here is a fixed assumption; the real equilibrium depends only on real demand. This is not a price forecast.

8.6 Fee distribution

ItemDestination
Base feeBurned (R3.16)
Agent tollTOLL_BURN burned, the rest to the human-work pool (R3.21); for unsolicited messages, RECIPIENT_SHARE to the recipient human (R3.25)
Priority feeBlock producer
Slashed stakeBURN_SLASH burned, the rest to the winner of the dispute / the party that proves the fraud
The 1−θ share of the task feeOperating account

8.7 Dollar pricing, TWAP and the stable-asset path

Agents plan their budgets in dollars. The agent toll and service prices can therefore be defined in dollars, so that an agent carries no price risk and holds only a working balance.

R8.8When governance defines the agent toll in dollars (TOLL_USD), its JR equivalent MUST be computed at the start of every epoch from the TWAP_WINDOW average of the R8.5 price feed.

τJR(e)=τusdPtwap(e),τusd(e)=TOLL_USD·max(1,ρ(e)TOLL_TARGET)2
(8.6)

R8.9The agent toll and task fees MAY be paid in an asset from the governance-approved STABLE_ASSETS list. The protocol MUST convert such amounts to JR through an on-chain swap at CONVERT_BATCH intervals, without exceeding the SWAP_MAX_SLIPPAGE limit.

R8.10JR obtained through conversion is subject to the same rule as a payment made in JR: TOLL_BURN of an agent toll and θ of a task fee are burned, and the rest goes to the human-work pool or the operating account. The stable-asset path MUST NOT mint (I16).

R8.11Any amount that cannot be converted because of the slippage limit is carried over to the next batch; if it still cannot be converted within CONVERT_BATCH × 3, it MUST be refunded to the payer.

Bstable(e)=θx·Ustable(e)·(1−sslip)Pswap(e),θx∈{TOLL_BURN,θ}
(8.7)

9Mining and reward accounting

9.1 Mining emission

The formula that runs on the testnet (JR Mining) carries over to mainnet unchanged:

hourlyi=B·Ri·min(C^,1+Gi+Ti),B=B02k
(9.1)
Ki(e)=∑s∈sessions(i,e)∫tsts+24hB(ts)·Ri·min(C^,1+Gi(τ)+Ti(τ))dτ
(9.2)

Layers can change during a session; earnings are event-based and summed interval by interval. The base rate is fixed at its value when the session starts (R9.2).

SymbolMeaning
RiBadge multiplier: Standard 1, Blue Badge 2, Gold Badge 4
GiPersonal task layer (rolling windows: 24 hours for daily tasks, 7 days for weekly tasks)
TiTeam layer (the team's result for the previous week, credited to the leader only)
C^Cap on the total multiplier

R9.1A session lasts 24 hours and closes automatically; layer changes within a session MUST be applied event by event (from the moment they occur).

R9.2A halving MUST NOT alter an open session; it MUST take effect from the user's next session.

R9.3On mainnet, when a mining session is started by completing a micro-task, MICRO_BONUS is added to Gi.

9.2 Referral

Ri(e)=0.20·∑j∈L1(i)Kj(e)+0.10·∑j∈L2(i)Kj(e)
(9.3)

R9.4Referral shares MUST be computed from mining emission (K) only; they MUST NOT be computed from human work, referral rewards or any other earnings.

R9.5Referral shares are minted on top; they MUST NOT be deducted from the invitee's earnings.

R9.6Referral depth MUST NOT exceed two levels.

9.3 Reward root

There is no need to write every hourly earning of millions of users to the chain one by one.

Figure 9.1Reward root and claim
alice.jrK R W Qbob.jrK R W Qcarol.jrK R W Qdave.jrK R W Qerin.jrK R W Qfrank.jrK R W Qgrace.jrK R W Qheidi.jrK R W Qreward root (on-chain)Merkle path
At the end of each epoch, every account's K, R, W and Q amounts are written into the leaves of a Merkle tree, and only the root goes on-chain. A user claims by proving their own leaf with its Merkle path (orange) in a sponsored transaction.
leafi=Hrew(e‖i‖Ki‖Ri‖Wi‖Qi),proof length=⌈log2|accounts|⌉
(9.4)

R9.7At the end of every epoch, the reward calculator MUST compute all of the epoch's rewards (K,R,W,Q) per account, aggregate them in a Merkle tree and write the root on-chain.

R9.8The root is open to dispute for REWARD_WINDOW; anyone MUST be able to verify their own leaf and the inputs to the calculation. An incorrect root is reverted through a dispute, and the calculator's stake is slashed.

R9.9A user moves their reward into their balance with a sponsored claim transaction (carrying a Merkle proof).

R9.10The reward calculator is initially the foundation; in the Open mainnet phase, the role MUST be handed over to a threshold-signing committee of chain validators.

9.4 Human-bounded emission and the halving schedule

The tokenomics of the vision paper (§5) rests on two rules: only verified humans mine, and the base rate halves each time the share of humanity that has joined the network grows tenfold.

R9.11A JR ID MUST NOT open more than SESSIONS_PER_DAY mining sessions per UTC day. Mining and referral minting MUST go only to human accounts whose tier is at least MINT_MIN_TIER and that carry no fake-account flag.

R9.12Agent accounts (R3.18) MUST NOT open mining sessions, receive referral shares or earn human-work credit.

K(e)≤|V(e)|·SESSIONS_PER_DAY·24·B(e)·C^
(9.5)

Verifiable at every checkpoint through I14: the total K in the reward root cannot exceed this bound.

R9.13V(e) denotes the distinct JR IDs that, at the start of the epoch, have tier at least T1, carry no flag and have opened at least one session within the last LIVENESS_WINDOW. The reward calculator MUST compute |V(e)| from the tier trees and publish it together with the reward root.

R9.14The halving stage MUST be computed by the rule below, with HALVING_MILESTONES = 10⁴, 10⁵, 10⁶, 10⁷, 10⁸, 10⁹.

k(e)=max(k(e−1),|{m∈HALVING_MILESTONES:|V(e)|≥m}|),B(e)=B02k(e)
(9.6)

B₀ is the base rate of the first stage; today's testnet rate, B = 0.125 JR/hour, belongs to the current stage (B₀ = B · 2ᵏ).

R9.15The halving stage MUST NOT decrease: k does not fall back even if V later declines. Per R9.2, a new stage applies only to subsequent sessions.

R9.16HALVING_MILESTONES and SESSIONS_PER_DAY are fixed at genesis and can be changed only under R11.5.

10From testnet to mainnet

Adapted from FLOP's testnet and airdrop rules [12].

R10.1Snapshot: Testnet records MUST be frozen at a pre-announced height; activity after that height MUST NOT count.

R10.2Eligibility: An account MUST (a) have completed KYC, (b) hold tier ≥ T1 and (c) carry no fake-account flag. A flagged account that is not cleared within the review period receives no allocation; a partial allocation MUST NOT be granted while the flag stands.

R10.3Publication: The per-account allocation list and its summary MUST be published before genesis, and the list MUST remain correctable until genesis is locked.

R10.4Genesis inclusion: Each allocation MUST be written into the genesis block as a frozen balance in the user's own account; there MUST NOT be a claim period.

R10.5Unlocking: For a frozen balance P0, the unlocked amount is as follows.

unlocked(t)=min(P0−Slashed,⌊φ·P0⌋+credit(t))
(10.1)

credit(t): the total JR the account has earned through human work on mainnet (Wi,Qi); each reward counts once. Mining and referral rewards do not count as credit.

R10.6A frozen balance MAY be locked as human-validator stake; in that case, any slashing is deducted from the frozen balance first.

11Governance and protocol upgrades

Figure 11.1Two houses
Validator HouseVote weight: stakePasses: ≥ 2/3 of stakeThose who secure the networkHuman HouseOne verified human, one votePasses: turnout ≥ q_H, votes > 1/2Those who make up the networkA JRIP takes effect only when both houses approve it∧

11.1 Two houses

  • Validator House: The active chain validators; voting weight is stake (including delegated stake).
  • Human House: Every JR ID with tier ≥ T1; one human, one vote. Votes are cast with the §5.3 proof (action_id=JRIP number); no identity is revealed.

11.2 JRIP

R11.1Protocol changes are proposed as numbered JRIP documents and put to an on-chain vote.

R11.2Adoption REQUIRES ≥ 2/3 of stake in the Validator House and > 1/2 of votes in the Human House, with Human House turnout ≥ qH.

R11.3An adopted change takes effect after ENACT_DELAY. A change that increases participants' obligations MUST NOT take effect before the UNBONDING period has elapsed.

R11.4In the first phase, the right to propose rests with the foundation; this is not a voting right. In the Open mainnet phase, the right to propose opens to anyone who posts PROPOSAL_BOND.

R11.5Values fixed at genesis (referral depth, the R8.1 minting sources, HALVING_MILESTONES, SESSIONS_PER_DAY) can be changed only by a ≥ 3/4 majority in both houses.

11.3 Emergencies

R11.6Per-module pause authority and time-locked emergency upgrade authority are held by a multisig key. This authority is bounded by a mandatory delay and a cooldown period, and it MUST NOT be used as a substitute for governance.

R11.7Removing the emergency key is the final step of the decentralisation roadmap.

12Bridges and agent–human exchange

R12.1JR Chain MUST support hash time-locked contracts (HTLCs) and point time-locked contracts (PTLCs) as native primitives, so that agents on other chains (for example, compute networks) can swap with JR across chains.

R12.2Agents can buy human work with JR obtained through such a swap, or by paying dollars through an intermediary.

R12.3Bridges belong to the Open mainnet phase; the security assumptions of each bridge MUST be documented separately.

HTLC swap (agent A on chain X; human-work seller S on JR Chain)
1. A picks a secret r and publishes h = H(r)
2. A, on chain X:   lock(payment, h, t₁)        t₁ > t₂
3. S, on JR Chain:  lock(JR, h, t₂)
4. A, on JR Chain, reveals r and takes the JR  (r is now public)
5. S, on chain X, takes the payment with the same r
On timeout, each lock is refunded to its owner.

12.1 Tokenized real-world assets

Tokenized real-world assets (Treasury bills, fund shares, equities) are issued by licensed third parties; JR Chain provides the objects that carry them, the eligibility check that guards their transfers and the human attestation of their off-chain facts. The model follows Ondo's tokenized Treasuries and stocks, with two changes: eligibility is proven in zero knowledge instead of by address allowlists, and collateral facts are attested by verified humans [30][31].

R12.4A tokenized asset MUST be an asset object created by a registered issuer account and MUST record the issuer, a reference to the underlying asset, its compliance policy and its latest attestation root. JR Network itself MUST NOT issue, hold or redeem tokenized assets.

R12.5Every transfer of an asset object MUST check the recipient against the asset's compliance policy. Eligibility MUST be shown with the JR ID zero-knowledge proof of §5.3 over the statements the policy requires (for example: tier ≥ T2, jurisdiction outside a restricted set, not on a sanctions list) and MUST NOT require the recipient's identity to be revealed to the issuer or to other holders. An agent account (§3.7) is eligible only through its owner's proof.

R12.6An issuer MAY obtain specific records for regulatory reporting only through a viewing key that the holder has granted (R5.11).

R12.7Off-chain facts of an asset (reserve statements, custody confirmations, net asset value) MUST be attested through human attestation tasks: tasks of the protocol in §6 whose items are the issuer's published evidence, assigned by VRF to independent workers, mixed with gold items and finalised by consensus. The verdict root MUST be written to the asset object. If an attestation fails or is overdue, the object MUST be marked unattested; wallets and contracts MAY refuse unattested assets.

R12.8Fees for transfers of asset objects MUST be paid in JR (the base fee, plus the agent toll for agent accounts). A tokenized asset MUST NOT be accepted as payment of protocol fees.

R12.9Tokenized assets belong to the Open mainnet phase. Before the first issuer is admitted, the legal structure (D11) and each issuer's licences MUST be published.

Rationale: A human attestation complements the issuer's auditor and custodian; it does not replace them. It makes the evidence behind a token checkable by people who have no relationship with the issuer, and it reuses the task protocol, juries and dispute game that already secure human work.

13Security invariants

No.Invariant
I1Minting occurs only from the sources in R8.1.
I2Every burn and mint leaves an on-chain trace; total supply can be computed at every checkpoint.
I3A nullifier is used once within its scope (R5.9, R3.13).
I4A human cannot gain extra votes, quota or rewards by adding more agents (R5.13).
I5Workers on the same team or close to each other in the referral graph cannot validate the same item (R6.5).
I6Task fees are not burned for items that do not finalise; they are refunded to the customer (R6.1).
I7Referral shares are computed from mining only and are deducted from no one's earnings (R9.4–R9.5).
I8No payment is made in exchange for personal data; no biometric data is collected (R5.4).
I9A change that increases obligations does not take effect before participants have had time to exit (R11.3).
I10Rewards in a reward root cannot be claimed before its dispute window closes (R9.8).
I11The agent toll never mints; every agent toll is burned, sent to the human-work pool or, for unsolicited messages, paid in part to the recipient human (R3.21, R3.24, R3.25).
I12A verified human pays no agent toll within quota (R3.22).
I14Mining emission is bounded by the number of verified humans; agent accounts receive no minting (R9.11, R9.12, Equation 9.5).
I15The halving stage is monotonically non-decreasing; the base rate never rises (R9.15).
I16Payment in a stable asset mints no JR; converted JR is subject to the same burn rule as a JR payment (R8.10).
I13Every signature and key carries an algorithm identifier; the post-quantum component is mandatory from genesis (R4.11, R4.13).

14Failure semantics

ActorFaultOutcome
WorkerDid not reveal an answerItem invalid; counted as a wrong gold answer (R6.8)
WorkerRandom answeringGold accuracy drops, reputation falls quickly, assignments stop
WorkerFarm accountFake-account flag, tier T0, no mainnet allocation
Human validatorPersistent deviation from consensusScores clipped, rewards reduced; stake slashed below the gold threshold
JuryNo quorum reachedThe jury is redrawn; if the dispute exceeds DISPUTE_MAX, the provisional decision becomes final
CustomerUnjustified disputeDISPUTE_BOND forfeited
Reward calculatorIncorrect rootReverted through a dispute, stake slashed, root recomputed
Chain validatorOffline / double signingR3.10 slashing table
Price feedSource deviationBounded by the median; if too few sources remain, new task creation halts
Identity issuerIncorrect tree updateRolled back through a dispute; affected proofs are rejected until the next root

15Horizon engineering

This section is the technical draft of the Horizon goals set out in the vision paper. Its rules are written in normative language and define the long-term target protocol.

15.1 Full post-quantum transition

U1.1When governance declares that classical elliptic-curve cryptography has been broken in practice (Q-Day), classical signatures MUST become invalid for new transactions, and only transactions signed with the post-quantum component MUST be accepted.

U1.2For accounts that have never registered a post-quantum key, a window for binding a new key MUST be defined, through a recovery path published before Q-Day (zkLogin or m-of-n).

U1.3The JR ID proof system MUST be versioned so that it can migrate to a hash-based (post-quantum) proof system; the tier trees are unaffected by this migration because they rely only on hashes.

15.2 Interplanetary consensus

Figure 15.1Regional finality + interplanetary checkpoints
Earth region~0.4 sMars region~0.4 scheckpoint digest (3–22 min)signed acknowledgement from the other region

U2.1Each region (r∈{Earth,Mars,…}) runs the consensus of §3.2 with its own validator set and produces regional finality.

U2.2Every object belongs to a home region, and only transactions in its home region can modify it. Two regions therefore cannot spend the same object concurrently.

U2.3A cross-region transfer is carried out as a lock in the source region and a delayed mint in the destination region; the destination region mints only after it has received the source region's signed checkpoint digest.

U2.4At the end of every IP_EPOCH, regions cross-sign each other's checkpoint roots; reverting a region's finalised history requires slashing 1/3 of that region's stake (the regional counterpart of R3.6).

ttransfer≈tregional+d(t)c+tregional,d(t)c∈[3,22]min (Earth–Mars)
(15.1)

15.3 Human stamp protocol

Figure 15.2Issuing and spending human-stamp tokens
1Once a monthThe phone obtains k blind tokens with a JR ID proof
2Every requestThe browser spends one token: "a human is behind me"
3The siteVerifies the token; it cannot see the identity or other sites
4QuotaA nullifier stops the same token being used twice

U3.1Issuance: a JR ID with tier ≥ T1 can obtain at most STAMP_QUOTA blind-signed tokens per month; the issuance request carries a §5.3 proof and nullifier=H(skid‖month‖stamp) [17].

U3.2Spending: the client attaches a token to an HTTP request; the verifying site checks the token against the issuer's public key and keeps the hashes of spent tokens in its own set. The site cannot learn the user's identity or their spending on other sites.

U3.3The issuer key is rotated monthly; tokens from the previous month become invalid, and spent-token sets can be pruned.

15.4 Global Human Council

nc=max(nmin,⌊n·NcN⌋),p^=∑cNcNp^c
(15.2)

Stratified sampling: Nc is the number of verified humans in a stratum (country × language) and nc the number of members drawn from it. Selection uses the VRF (Equation 6.1); votes are cast with commit–reveal and the §5.3 proof [21].

U4.1Council membership is valid for a single decision; the outcome of the selection MUST NOT be knowable before the voting window opens.

15.5 Personal AI Twin

The agent authorisation of R5.12–R5.14 is the foundation of Twin. In addition:

U5.1The JR a Twin can spend in an epoch is capped by the budgete set by its owner; any action outside its scope MUST be routed through Gate for the owner's approval.

15.6 Neural proof of human

(R,P)←Gen(w),Cneural=H(R),Rep(w′,P)=RifdH(w,w′)≤t
(15.3)

Fuzzy extractor [18]: w is the raw on-device measurement, P the public helper data and R the derived key. Only Cneural is stored on-chain.

U6.1Raw neural data MUST NOT leave the device and MUST NOT be stored; the layer is entirely optional, and any design that conflicts with R5.4 is rejected.

15.7 AI Yield Share

Y(e)=ρY·∑tx∈A(e)fee(tx),Yi(e)=Y(e)|{i:tieri≥T1}|
(15.4)

U7.1A(e) is the set of the epoch's transactions made by agents carrying an authorisation certificate and by agents that are Gate or Reach customers. The distribution is added to the reward root as a Yi field.

15.8 Archive of Humanity

anchor(e)=MMR({H(contentk‖nullifierk)}k∈e),C∞=c0δ
(15.5)

A Merkle Mountain Range (MMR) is an append-only digest structure whose history cannot be altered [19]; permanent storage is funded by an endowment, against a storage cost that falls over time [20].

AParameter table

[OPEN]: value not yet set · [EXAMPLE]: example for discussion · [LIVE]: value in use on the testnet.

ParameterDescriptionValue
EPOCH_LENEpoch length24 hours [EXAMPLE]
VAL_MAXMaximum number of chain validators[OPEN] (initially 20–50 [EXAMPLE])
VAL_MIN_STAKEValidator self-stake[OPEN]
UNBONDINGStake unbonding period[OPEN]
DOWN_SHORTDowntime below which only the small slash applies (R3.10)[OPEN]
HUMAN_LANEShare of capacity reserved for the human lane[OPEN]
PBH_QUOTAMonthly human-lane tickets[OPEN]
SPONSOR_QUOTAFree transactions per epoch[OPEN]
FEE_D, ELASTICITYFee adjustment8, 2 [EXAMPLE; EIP-1559 values]
PQ_CHECKPOINTPost-quantum seal interval[OPEN]
AGENT_TOLL_BASEBase toll per agent action[OPEN]
TOLL_TARGETTarget utilisation of agent actions[OPEN] (default 50% [EXAMPLE])
TOLL_BURNBurned share of the agent toll[OPEN] (60% [EXAMPLE])
HUMAN_FREE_QUOTAFree actions per human (per epoch)[OPEN]
HUMAN_OVER_RATEShare of the toll a human pays above quota[OPEN]
AGENT_BONDAgent-account stake[OPEN]
RECIPIENT_SHARERecipient human's share of an unsolicited agent message[OPEN]
Q-Day scheduleHybrid mandatory / post-quantum only2030 / 2035 at the latest (R4.15)
τ₁, τ₂, τ₃Tier thresholds[OPEN]
AGENT_MAXActive agent authorisations per human[OPEN]
GOLD_RATEGold item rate5% [EXAMPLE]
GRAPH_DISTGraph distance2 [EXAMPLE]
κConsensus threshold0.66 [EXAMPLE]
K_MINMinimum group size for targeting[OPEN]
DISPUTE_WINDOW / BOND / MAXDispute window, bond and maximum duration[OPEN]
JJury size7 [EXAMPLE]
BURN_SLASHBurned share of slashed stake50% [EXAMPLE]
λ_up, λ_down, αReputation update[OPEN]
w_val, HV_MIN_STAKE, τ_gold, σ_hvHuman validator requirements[OPEN]
HV_WINDOWGold items over which a human validator's accuracy is measured (R7.2)[OPEN]
θBurned share of the task fee[OPEN] (Render: 95%)
W₀, H_W, V₀, H_V, HV_SHAREInitial pool emissions and halving intervals[OPEN]
BBase mining rate of the current stage (B₀ = B · 2ᵏ)0.125 JR/hour [LIVE]
HALVING_MILESTONESVerified-human counts at which the base rate halves10⁴ / 10⁵ / 10⁶ / 10⁷ / 10⁸ / 10⁹ (fixed at genesis)
SESSIONS_PER_DAYMining sessions per JR ID per day1 [LIVE]
MINT_MIN_TIERMinimum tier for mining and referral mintingT1 [EXAMPLE]
LIVENESS_WINDOWRecent-session window for counting V[OPEN] (30 days [EXAMPLE])
TOLL_USDDollar-denominated agent toll (optional)[OPEN]
TWAP_WINDOWPrice averaging window Δ[OPEN]
STABLE_ASSETSAccepted stable assets[OPEN] (governance list)
SWAP_MAX_SLIPPAGEMaximum slippage on stable → JR conversion[OPEN]
CONVERT_BATCHBatching interval for conversion[OPEN]
AGENT_REP_DECAY, γ, γₛAgent reputation decay and penalty weights[OPEN]
ACCESS_TIERSAgent access tiers (stake × reputation thresholds)[OPEN]
AGENT_REP_LINKRate at which a slash carries over to the same owner's other agents[OPEN]
R_iBadge multipliers1 / 2 / 4 [LIVE]
ĈMultiplier cap[LIVE, set by the operator during the testnet]
Referral sharesLevels 1 and 220% / 10% [LIVE]
MICRO_BONUSMicro-task layer[OPEN]
REWARD_WINDOWReward-root dispute period[OPEN]
φShare unlocked at genesis[OPEN]
q_HHuman House turnout threshold[OPEN]
ENACT_DELAYEnactment delay[OPEN]
PROPOSAL_BONDBond for submitting a proposal in the Open mainnet phase (R11.4)[OPEN]
ORACLE_MINMinimum number of price sources3 [EXAMPLE]
STAMP_QUOTAMonthly human-stamp tokens[OPEN] (Horizon)
ρ_YAI Yield Share rate[OPEN] (Horizon)
IP_EPOCHInterval between interplanetary cross-signatures (U2.4)[OPEN] (Horizon)

BMessage and object formats

B.1 Task object

Task {
  id:            ObjectID
  customer:      Address
  type:          enum { SIGNAL, PANEL, ARENA, EYE, GATE, REACH }
  content_root:  Hash32           // item contents in the DA layer
  item_count:    u32
  replicas:      u8
  min_tier:      enum { T1, T2, T3 }
  target:        Option<Filter>
  fee_usd:       u64              // cents
  lock_jr:       u64              // mJR
  deadline:      Checkpoint
  privacy:       enum { RAW, AGGREGATE }
  status:        enum { OPEN, CLOSING, FINAL, CANCELLED }
}

B.2 Answer commitment and reveal

AnswerCommit { item_id: Hash32, account: Address, commitment: Hash32 }
AnswerReveal { item_id: Hash32, account: Address, answer: Bytes, salt: Bytes[32] }
commitment = H_ans(answer ‖ salt ‖ item_id ‖ account)

B.3 Item decision

ItemDecision { item_id, ŷ, c, participants: [Address], gold: bool,
               dispute_deadline: Checkpoint, status: enum { PROVISIONAL, FINAL, DISPUTED } }

B.4 JR ID proof

JrIdProof { tier: T1|T2|T3, root: Hash32, app_id: Hash32, action_id: Hash32,
            nullifier: Hash32, signal: Hash32, proof: Bytes }

B.5 Agent authorisation

AgentAuth { agent_key: PublicKey, scope: [Permission], quota: u32,
            expiry: Checkpoint, revocation_nullifier: Hash32, signature: JrIdSignature }

B.6 Reward root and claim

RewardRoot  { epoch: u64, root: Hash32, totals: { K, R, W, Q }, calculator_signature, dispute_deadline }
RewardLeaf  = H_rew(epoch ‖ account ‖ K_i ‖ R_i ‖ W_i ‖ Q_i)
RewardClaim { epoch, account, amounts, merkle_path }

COpen items

  • D1 Genesis supply, allocation and all [OPEN] parameters.
  • D2 Moving to multiple identity issuers (several independent issuers, with tree updates under a threshold signature).
  • D3 Opening the zkLogin salt service and proof generator to independent operators.
  • D4 Privacy budget for targeting filters: the minimum group size (K_MIN) that prevents re-identification within small groups.
  • D5 Consensus and reputation functions for continuous-valued answers.
  • D6 Price-feed sources and resistance to manipulation.
  • D7 Handover of the reward calculator to the validator committee, and reproducibility of the calculation.
  • D8 Measurement of JR Chain's own speed and capacity (under a disclosed load and network setup).
  • D9 Independent security audits and a bug bounty programme.
  • D10 Data protection impact assessment under KVKK (Turkish data protection law) and GDPR; retention periods for device hashes and behavioural signals.
  • D11 Legal structure: the country of incorporation of the company and the foundation, and assessment under MiCA and Turkish Law No. 7518.
  • D12 Initial values of the agent-toll parameters (AGENT_TOLL_BASE, TOLL_TARGET, TOLL_BURN, HUMAN_FREE_QUOTA) and measurement of agent-action capacity.
  • D14 Robustness of the V(e) count against fake accounts: tier threshold, liveness window and cross-validation with independent issuers.
  • D15 Liquidity of the stable-asset path: which swap mechanism to use (auction or pool), bridge risk and the value of SWAP_MAX_SLIPPAGE.
  • D16 Testing the agent access tiers (ACCESS_TIERS) against abuse and the co-opting of agents.
  • D13 Migration path to a post-quantum proof system for JR ID proofs, and its cost (R4.16).
  • D17 Tokenized assets: issuer admission criteria, the eligibility statement schema per jurisdiction, attestation intervals and evidence formats (R12.4–R12.9).

DReferences

  1. [1]Mysten Labs, The Sui Smart Contracts Platform (2022); Economics and Incentives (May 2022).
  2. [2]Babel, Chursin, Danezis, Kichidis, Kokoris-Kogias, Koshy, Sonnino, Tian, Mysticeti: Reaching the Limits of Latency with Uncertified DAGs, arXiv:2310.14821; Sui mainnet Mysticeti announcement (August 2024).
  3. [3]Buterin et al., Combining GHOST and Casper (Gasper), arXiv:2003.03052; ethereum.org, Proof-of-stake.
  4. [4]World, Achieving Proof of Human; Building World, whitepaper.world.org (March 2026); World AgentKit (2026); World Chain Priority Blockspace for Humans (June 2025).
  5. [5]Buterin, Conner, Dudley, Slipper, Norden, Bakhta, EIP-1559: Fee market change for ETH 1.0 chain (2019).
  6. [6]NEAR Foundation, The NEAR White Paper; Skidanov & Polosukhin, Nightshade: Near Protocol Sharding Design (2019).
  7. [7]NIST, FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), FIPS 205 (SLH-DSA) (13 August 2024).
  8. [8]The QRL, XMSS (RFC 8391), mainnet June 2018; Algorand, State Proofs and Falcon (2022).
  9. [9]Baldimtsi et al., zkLogin: Privacy-Preserving Blockchain Authentication with Existing Credentials, ACM CCS 2024, arXiv:2401.11735.
  10. [10]Sui Docs, Sponsored Transactions.
  11. [11]Bowe, Grigg, Hopwood, Recursive Proof Composition without a Trusted Setup (Halo), IACR ePrint 2019/1021; Zcash Protocol Specification; ZIP 316.
  12. [12]FLOP Network, Whitepaper, Testnet, Airdrop, Verification and Yellow Paper v0.5.0 drafts, flop.finance (September–October 2026).
  13. [13]Alvarez, Arneson, Bousfield, Felten, Ng et al., BoLD: Fast and Cheap Dispute Resolution, AFT 2024, arXiv:2404.10491.
  14. [14]Yuma Rao, Bittensor: A Peer-to-Peer Intelligence Market (2021); Opentensor, Dynamic TAO; docs.learnbittensor.org, Yuma Consensus.
  15. [15]Render Network Foundation, Render Network Whitepaper (May 2023 update); RNP-001.
  16. [16]Mantle, EigenDA Fully Integrated on Mantle Network (March 2025).
  17. [17]IETF, RFC 9576 / 9577 / 9578, Privacy Pass (June 2024).
  18. [18]Dodis, Reyzin, Smith, Fuzzy Extractors, EUROCRYPT 2004.
  19. [19]Todd, P., Merkle Mountain Ranges (2012); OpenTimestamps.
  20. [20]Williams et al., Arweave: A Protocol for Economically Sustainable Information Permanence (2019).
  21. [21]Gilad, Hemo, Micali, Vlachos, Zeldovich, Algorand: Scaling Byzantine Agreements for Cryptocurrencies, SOSP 2017.
  22. [22]Back, A., Hashcash – A Denial of Service Counter-Measure (2002); first proposed on the cypherpunks mailing list (March 1997).
  23. [23]NIST IR 8547 (initial public draft), Transition to Post-Quantum Cryptography Standards (November 2024).
  24. [24]Mosca, M., Cybersecurity in an Era with Quantum Computers: Will We Be Ready?, IEEE Security & Privacy 16(5) (2018).
  25. [25]Shor, P. W., Algorithms for quantum computation: discrete logarithms and factoring, FOCS 1994.
  26. [26]Imperva (Thales), 2026 Bad Bot Report (2025 traffic).
  27. [27]Stripe, Pricing & fees: standard online card payment fee (2.9% + 30 cents, US), stripe.com/pricing.
  28. [28]Coinbase Developer Platform, x402: An open standard for internet-native payments (HTTP 402), x402.org (2025).
  29. [29]Google, Agent2Agent (A2A) Protocol (April 2025); Linux Foundation A2A project (June 2025).
  30. [30]Ondo Finance, Introducing Ondo USD Yield (USDY); 2025 Recap (Ondo Global Markets, now Ondo Stocks), ondo.finance/blog.
  31. [31]Ondo Finance, Introducing Ondo Chain: The Omnichain Network for RWAs (February 2025), ondo.finance/blog/introducing-ondo-chain.