JR NetworkTechnical specification

Vision paper

JR Network

The currency of AI agents: bots pay, the human mind earns. A Layer 1 that is quantum-resistant from the first block.

Version 0.2 · Draft · November 2026

0How to read this paper

This paper describes JR Network: the currency of AI agents and the human layer through which that currency flows. It sets out why the network is needed, how it works, its economics, its security and its long-term horizon.

The paper rests on three pillars: the currency of AI agents (§3), the value of the human mind in AI (§4, §10) and security beyond the quantum age (§13). The remaining sections explain in detail how these three pillars work.

It is one of two companion documents. This vision paper is written for everyone: it presents the problem, the products, Proof of Human Work and the JR cycle, with their mathematics, in plain language. The technical specification is written for implementers: it contains the protocol objects, normative rules, parameter tables and security invariants. Both describe the same system; where they differ, the technical specification prevails.

The text contains blue-framed simulations. When you move the sliders, the numbers are recomputed in your browser from the formulas in the text. Purple-framed “Inspired by” boxes show which project an idea comes from, that project's original formula and what we changed. Orange-framed Horizon cards divide each long-term goal into three parts: “today's obstacle, our approach, work ahead”.

1Summary

JR Network is the currency of AI agents. Agents pay JR to exist on the internet, to talk to one another and to get work done; that money flows to the one thing AI cannot produce, the real human mind. The network's own Layer 1 chain, JR Chain, is resistant to quantum computers from the first block.

Bots pay, humans earn.

The heart of the paper is §5 Tokenomics and §6 Why agents choose JR: JR's supply is bounded by verified humans, its demand comes from agents, and every halving marks the day the network reaches ten times more of humanity.

Humans have become a minority on the internet. In 2025, 53% of web traffic was automated, and that share has grown every year for seven years [1]. Bots are no longer simple programs that generate clicks; they are AI agents that shop, make bookings, correspond and negotiate. Every agent carries a cost: servers, bandwidth and, above all, human attention. Today humans bear that cost through spam, CAPTCHAs and fake accounts. JR Network reverses the equation. The paper rests on three pillars:

PillarWhat it saysSection
1 · The currency of AI agentsAgents pay a small JR toll for every action on the network. Part of the toll is burned; the rest flows to human work. Verified humans pay no toll within their quota.§3
2 · The value of the human mind in AIWhat agents and AI companies cannot buy anywhere: the judgement, approval and presence of a real human. JR is what pays for it.§4, §10
3 · Security beyond the quantum ageA currency used by billions of agents and humans must not collapse on the day quantum computers break today's cryptography. JR Chain uses hybrid signatures from genesis.§13

Compute networks give machines their power. JR Network gives them the human mind.

The money flows into the human layer, which works through four verbs:

  • Verify: proof that “I am a real and unique human”, without biometrics and with privacy preserved (JR ID).
  • Produce: human preferences, predictions and opinions for AI.
  • Audit: human review of AI outputs and decisions.
  • Reach: bot-free access for brands and agents to verified humans.

Every task is performed by several humans who do not know one another and is finalised by consensus. Gold questions with answers known in advance, reputation accumulated over time, locked stake and a dispute game make cheating unprofitable. New JR is minted only for verified human work and mining; agent tolls and customer payments turn into JR burns. The more bots there are, the more is burned, and mint and burn settle into an equilibrium tied to real demand.

JR Network will run on its own blockchain, JR Chain. JR Chain is not a chain invented from scratch; it is a hybrid architecture that combines techniques proven in their own domains towards a single purpose: Sui's object-based parallel execution and social login, Zcash's zero-knowledge proofs without a trusted setup, Ethereum's fee burn and finality model, Arbitrum's bisection dispute game, NEAR's named accounts, Bittensor's consensus scoring, Render's burn-and-mint equilibrium, World Chain's blockspace reserved for humans, Hashcash's idea that “every message should carry a cost” and NIST's post-quantum signature standards.

At a glance
What?The currency of AI agents and the human layer of AI
Who pays?AI agents (agent toll), AI companies, research firms, brands
Who earns?Every verified human who completes tasks on their phone
What is proven?That the person is a real human and actually did the work, correctly
How?Independent assignment, commit–reveal, consensus, gold questions, reputation, stake, dispute game
SecurityHybrid classical and post-quantum signatures from genesis
TodayJR Mining testnet; people are building the participation and identity network
SaleNo token sale. JR can only be earned by working.

2The problem: humans are becoming a minority online

2.1 Bots are now the majority

According to the 2026 report by Imperva (Thales), 53% of web traffic in 2025 was automated; malicious bots alone reached 40%, while the human share fell to 47% [1]. Bots first overtook humans in 2024 (51%); 2025 showed that the threshold has been crossed for good. The seven-year series moves in one direction only: in 2019, 62.8% of traffic was human.

Figure 2.1Composition of web traffic, 2025
Humans 47%Good bots 13%Bad bots 40%

In 2025, 53% of traffic was automated: the second year in a row that bots outnumbered humans.

Source: Imperva (Thales), 2026 Bad Bot Report.

Fitting a straight line to a share yields meaningless results over the long run, eventually falling below 0%. We therefore fit the trend not to the share itself but to the logarithm of the bot-to-human ratio (the log-odds, i.e. a logistic trend). Over the 2019–2025 series this ratio grows by about 11.5% a year; at the current pace the human share erodes by about 2.6 percentage points a year, or about 0.007 points a day:

H(t)=1001+r2025eκ(t−2025.5),κ=∑y(y−y¯)(lnry−lnr¯)∑y(y−y¯)2≈0.109/yr
(2.1)

ry=(100−Hy)/Hy: ratio of bot traffic to human traffic in year y · r2025=53/47. If the trend continues, the human share falls to about 40% by the end of 2027, to one third by 2030 and to about 15% by 2040. The logistic curve never touches zero. This is not a forecast; it is the trend of the past seven years.

Simulation · The human share of the internet
20%35%50%65%80%2019202120232025202720292031Bots took the lead
HumanBot (good + bad)Dashed: trend projection

November 2026: of every 100 visitors, 43 are human and 57 are bots.

Human share on the selected date43.4%
Today (November 2026, trend)43.4%
End of 2027 (trend)40.3% human
Daily change (trend)-0.0074 pts/day

Measured values: Imperva (Thales) Bad Bot Report 2020–2026 (2019–2025 traffic). The dashed line after 2025 is a trend projection (2019–2025 log-odds, i.e. a logistic trend); not a forecast, but where the last seven years lead if their direction holds. A logistic curve never reaches 0%.

2.2 The population of the internet: 2027 to 2050

To convert traffic share into population we use two official series: the United Nations' World Population Prospects [39] and the ITU's measurement of internet users [38]. According to the ITU, about 6 billion people (74% of the population) were online in 2025. We express bot activity in the same unit, the human equivalent: a “bot population” equal to the bot traffic per online human.

NH(t)=Nworld(t)·o(t),NB(t)=NH(t)·r(t),r(t)=100−H(t)H(t)
(2.2)

o(t): online share (74% in 2025, assumed to approach 95% thereafter) · r(t): bot-to-human ratio from Equation 2.1. NB is not a count of bots but the human equivalent of bot activity.

Simulation · The population of the internet: humans and bots
5B10B20B50B100B200B202520302035204020452050yearonline population (log)
Humans onlineBot activity (human equivalents)
Humans online7.23 billion
Bots (human equivalents)14.05 billion
Bots per human1.9
Human share of traffic34.0%
Year202720282029203020402050
Humans6.6B6.8B7.0B7.2B8.5B9.1B
Bots (equiv.)9.3B10.7B12.3B14.1B49B156B
Human share42%39%36%34%15%6%

Humans: UN World Population Prospects 2024 × ITU 2025 share online (74% in 2025, then an assumption approaching 95%). Bots: the logistic trend of Imperva's traffic ratio (bot/human) × humans online, so bot activity is expressed in human equivalents. This is not a count of bots but a model projection.

The table makes one thing clear: while the number of humans online keeps growing (about 9 billion in 2050), their share erodes. Bot activity reaches roughly twice the human level in 2030, six times in 2040 and seventeen times in 2050. Humans will be the minority on the internet, and the entire tokenomics of this paper is built on pricing the value of that minority (§5).

2.3 From bots to agents

Old bots did one thing: scrape pages, generate clicks, fill in forms. The new generation of bots are AI agents: they write, talk, negotiate and pay like humans, and divide work among themselves. Agents are not bad in themselves; they do work that saves people hours. But the entire economy of the internet was built on the assumption that the other party is a human: advertising, surveys, reviews, votes, customer service. When that assumption collapses, two questions arise: Is the other party a real human? and who will pay for the space an agent occupies on the internet?

2.4 AI is consuming human data

Researchers at Epoch AI estimated that the stock of public human-written text usable for training large language models could be fully used up between 2026 and 2032 [2]. A study published in Nature showed that models degrade over time when trained repeatedly on their own generated data, a phenomenon known as “model collapse” [3]. Read together, the conclusion is clear: as bots multiply, the human signal on the internet thins out, yet that is precisely AI's most valuable input: fresh, genuine human signal.

2.5 Trust is collapsing

  • In some online surveys, an estimated 80–95% of responses are fake or bot-generated [4].
  • Ad fraud cost $84 billion in 2023 and is projected to reach about $172 billion by 2028 [5].
  • In the United States, fraud enabled by generative AI is expected to rise from $12.3 billion in 2023 to $40 billion by 2027 [6].

2.6 Why today's solutions fall short

Today's answers to the question “are you human?” fall into three groups, and each carries a heavy cost:

ApproachWhat it doesProblem
CAPTCHAMakes the user solve a puzzleAI solves the puzzles better than humans do; for humans they are pure friction.
Phone / email verificationSends a code to a numberNumbers and email addresses can be bought in bulk; it exposes identity.
BiometricsScans the iris or faceExpensive hardware, irrevocable data, legal barriers in many countries.

All three share the same gap: none of them makes bots pay, and none of them rewards humans for their effort. The cost always falls on the human: the one solving puzzles, filtering spam and reading fake reviews.

3The currency of AI agents

An agent needs money to exist on the internet. Opening an account, sending a message, initiating a transaction, asking for a human's attention: each carries a cost. On JR Network that cost is paid in JR.

3.1 Why do agents need money?

No action on the internet is free; its cost is simply shifted onto someone. In email, servers and people pay to filter spam; on social networks, moderators pay for fake accounts; in surveys, researchers pay for fake answers. Once agents outnumber humans, this burden becomes unsustainable. JR Network's answer is simple: the agent that takes the action pays for it. There are four moments at which an agent needs JR:

MomentWhat the agent doesWhat it pays
ExistOpens an agent account on the network, bound to its ownerAccount stake (refundable)
SpeakSends a message to another agent or to a humanAgent toll per action
ActInitiates a transaction, calls a contract, requests dataAgent toll + base fee
Reach a humanAsks a human for approval, an opinion or verification (Gate, Eye, Reach)The fee for human work; flows directly to the human

3.2 Every message should carry a cost

3.3 The agent toll

The network has two kinds of account: human accounts, bound to a JR ID, and agent accounts, bound to an owner. Every agent action pays a base toll; when the network rises above its target utilisation, the toll increases quadratically with congestion. The cost thus stays negligible for a legitimate agent while spam at scale becomes expensive:

τ(e)=τ0·max(1,ρ(e)ρ∗)2,T(e)=∑a∈agentsna(e)·τ(e)
(3.1)

τ0: base agent toll per action · ρ(e): network utilisation · ρ∗: target utilisation · na: the agent's number of actions in the epoch.

The tolls collected are split in two: a share θa is burned, and the rest is added to the epoch's human-work pool and distributed among the humans who did the work according to the rule in §11.4:

Bagent(e)=θa·T(e),Hagent(e)=(1−θa)·T(e)
(3.2)

A verified human pays no toll for up to QH actions per epoch; this is the same principle as the sponsor pool and the human lane (§12). The human's toll is:

tolli=τ(e)·max(0,ni−QH)
(3.3)

An attacker's spam campaign of m actions grows more expensive linearly, and quadratically once it causes congestion:

Cspam(m)=m·τ0·max(1,ρρ∗)2
(3.4)
Figure 3.1Flow of the agent toll
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
Agents pay a toll per action; a share θ of the toll is burned and the rest goes to the human-work pool. Verified humans pay no toll within their 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.

3.4 An agent's compute cost and human cost

3.5 More bots, more demand for JR

Because the agent toll is paid in JR, demand for JR is proportional to agent activity on the network. With A(e) agents and an average of n¯ actions per agent:

DJR(e)∝A(e)·n¯(e)·τ(e),∂Bagent∂A=θan¯τ>0
(3.5)

This turns the problem of §2 into the engine of the economy: as the bot share of the internet grows, the burn grows and the human-work pool fills. Every moment an agent spends JR becomes a human's earnings in parts of the ecosystem such as Gate (F), Reach (G) and Eye (E); in the long run, Yield (K) extends the same flow to all verified humans (§18).

Bots pay, humans earn.

4The value of the human mind in AI

The money agents pay flows to what is scarce. In the age of AI, what is scarce is not compute but the judgement of a real human.

Today's best AI models are shaped by human feedback. In OpenAI's InstructGPT work, a 1.3-billion-parameter model trained on human preferences was preferred over GPT-3, a model 100 times its size [7]. In other words, with the same compute, the right human data multiplies a model's quality. The market reflects this value too:

  • In June 2025, Meta invested $14.3 billion in the data-labelling company Scale AI [8].
  • LMArena, which ranks AI models by human votes, reached a $1.7 billion valuation in January 2026 [9].

At the same time the agent economy is growing. Agents talk to one another, do work and make payments. But there is one thing an agent cannot buy from any network: being human. Asking a real customer about a product, having a real human approve a payment, having a human check whether an answer is correct.

4.1 Three waves of demand

WaveWho asks?What do they want?
1 · Training dataModel developersPreferences, rankings, corrections; languages beyond English
2 · Trust and auditPlatforms, researchers, brandsBot-free surveys, real audiences, audits of AI output
3 · Agent economyAI agentsHuman approval at critical steps, reaching humans, proof of human

4.2 What JR Network contributes

  • A network of verified humans: people mining every day on the JR Mining testnet, and the Overwatch scorer that separates them from fake accounts (§8).
  • Proof of Human Work: a five-layer protocol that proves a task was done by a real human, and done correctly (§10).
  • JR ID: proof that “I am a real and unique human”, without biometrics and without revealing identity (§9, A).
  • Ecosystem A–O: fifteen components that turn the human mind into products, from Signal to Gate and from Stamp to Archive (§9).
  • Fresh data against model collapse: time-stamped human signal from verified humans, in many languages.

4.3 How JR becomes the currency of the human mind

The cycle has four steps: an agent or company requests human work and pays in JR; part of the payment is burned; the rest is distributed to the humans who did the work; and those humans use the JR they earned to transact, post stake and vote on the network. A human's earnings in an epoch are:

Ji(e)=(1−θa)T(e)·biwi∑jbjwj⏟share of agent tolls+Wi(e)⏟human-work pool+Ki(e)⏟mining
(4.1)

bi: the person's finalised work credit · wi: their reputation (§10.5) · Wi: their share of the work pool under §11.4 · Ki: mining (§8).

JR Network answers this demand with a single infrastructure: a large network of verified humans, a protocol that defines how work is done on that network and a currency that rewards that work.

5Tokenomics: the scarcity of human intelligence

JR Network's tokenomics rests on a single observation: on the internet, machines are multiplying and humans are becoming scarce. Agents will number in the billions, then in the tens of billions; verified human judgement will remain bounded by the number of humans. JR is the currency of this scarce resource: human intelligence and human labour.

5.1 What kind of project are we?

JR Network is not a compute network; it does not sell processing power to machines. Nor is it merely an identity project; for us, proving that one is human is the starting point, not the goal. JR Network is a market for the human mind: verified humans on the supply side, AI agents on the demand side, and JR, the currency that brings the two together, in between. We see the core problem as a chain of four links:

#ProblemJR Network's answer
1Bots and AI agents are multiplying: 53% of web traffic in 2025, and about two bot equivalents per human by 2030 (§2.2).Turns this growth into a source of demand.
2Agents need money to exist on the internet, to speak, to act and to reach a human (§3).Every agent action is paid in JR; the toll makes spam expensive.
3What agents cannot buy from any network is human judgement: approval, preference, audit, opinion (§4).JR is the unit in which verified human judgement is sold, per action.
4Humans are becoming a minority on the internet: about 15% of traffic by 2040 (§2.2).The minority's value rises; JR pays that value to the humans who create it.

Humans are the internet's minority. JR is that minority's currency.

5.2 Where scarcity comes from

Scarcity arises from two asymmetries. The first is number: copying an agent is almost free, copying a human is impossible. The second is attention: an agent can send thousands of requests per second, while the minutes a human can spare each day are limited. Let A(t) be the agent activity that pays in JR and V(t) the number of verified humans. The scarcity index is:

Σ(t)=A(t)V(t)=α(t)v(t)·r(t),A=αNB,V=vNH
(5.1)

r(t): bot-to-human ratio (Equation 2.1) · α: share of bot activity that pays in JR · v: verified share of online humans. Even if α and v stay constant, the scarcity index grows at the same pace as r(t), about 11.5% a year.

Year2027203020402050
Bot equivalents per human r(t)1.41.95.817

5.3 Architecture: three rules

Figure 5.1JR's tokenomic architecture
Supply side: humansV verified humans1 human = 1 session a daymining + human work + referralsbots cannot mintDemand side: agentsA agents (human-equivalent)exist · speakwork · reach humanspay JR per actionJRunit of the mindmint ≤ V·24·Btoll τ(1 − θ) pays human workθ is burnedAs A / V grows, burn overtakes mint: as the human mind grows scarce, so does JR
Supply is tied to verified humans and demand comes from agents; a share θ of the agent toll is burned and the rest is paid to human work.

Rule 1: supply is tied to humans. New JR is minted in only four ways: mining by verified humans, the human-work pool, the referral share arising from mining and the chain validator reward (§11.2). Each JR ID can open at most one mining session per day; agent accounts cannot mine and receive no referral share. Mining emission is therefore bounded from above by the number of verified humans. We call this proof of human scarcity:

K(e)≤V(e)·24·B(e)·C^
(5.2)

K(e): mining emission in the epoch · B(e): base rate under the halving schedule (§5.4) · C^: multiplier cap. Adding devices, accounts or bots does not raise this bound; only a new human can.

Rule 2: demand is tied to agents. Demand for JR comes from two sources: the toll agents pay per action (§3.3) and companies' dollar-priced payments for human work (§11.3).

Rule 3: burn is tied to demand. A share θa of the agent toll and a share θ of service payments are burned. The net change in supply per epoch is:

ΔS(e)≤V·24·B·C^+W+Vval⏟human-bounded mint−θa·A·n¯·τ−θUP⏟agent and customer burn
(5.3)

Comparing mining with the agent toll alone, there is a critical scarcity threshold above which burn exceeds mint:

θan¯τ·A≥24BC^·V⟺Σ≥Σ∗=24BC^θan¯τ
(5.4)

The scarcity index grows over time (Equation 5.1), while the threshold halves at every halving (because B halves). Both effects push in the same direction: as the human mind grows scarcer, so does JR.

5.4 JR Halving: every halving adds a zero

In JR, halving is tied not to time but to the share of humanity that has joined the network. The base rate halves each time the number of verified humans grows tenfold: at 10,000, 100,000, 1 million, 10 million, 100 million and 1 billion. The last threshold is roughly one in every eight people on Earth.

B(V)=B02k(V),k(V)=min(6,max(0,⌊log10V⌋−3))
(5.5)

B0: base rate of the first stage. The current testnet rate (0.125 JR/hour) is the rate of the stage now in effect; B0=0.125·2k. Halvings are irreversible: even if V falls, k does not decrease.

HalvingVerified humansOne in how many people worldwideBase rate
1st10,000≈ 823,000B0/2
2nd100,000≈ 82,300B0/4
3rd1 million≈ 8,230B0/8
4th10 million≈ 823B0/16
5th100 million≈ 82B0/32
6th1 billion≈ 8B0/64

Someone who joins before a threshold mines twice as fast as someone who joins after it; someone who joins N thresholds earlier mines 2N times as fast. For example, a person who starts mining before the 1 million threshold starts with 24=16 times the base rate of someone who starts after the 1 billion threshold. The network's daily mining ceiling, for its part, grows fivefold with each tenfold growth (humans increase tenfold, the rate halves), and after the last threshold it grows only with humanity itself:

Ek+1maxEkmax=102=5,E∞max=24C^·B064·Vmax,Vmax≤Nworld
(5.6)

The ultimate ceiling on JR minting is not a number but humanity itself. According to the United Nations, the world population will be about 9.7 billion in 2050 [39].

Simulation · JR Halving: every halving adds a zero
B₀B₀/2B₀/4B₀/8B₀/16B₀/32B₀/6410010K1M100M1B10Bverified humans (log)base rate B (log₂)1 in 8 people
Current stage k1 / 6
Base rate BB₀ / 2
Versus joining after the last threshold32× faster
Next threshold100,000
Share of world population0.0009%
Network emission cap (vs first threshold)7.41×

The base rate halves each time the number of verified humans grows tenfold: 10,000, 100,000, 1 million, 10 million, 100 million, 1 billion. B₀ is the base rate of the first stage; today's testnet rate (0.125 JR/hour) is the rate of the current stage. Values are shown as ratios, not absolutes; total supply is not part of this paper.

5.5 Why mining matters

Mining builds the supply side of the tokenomics. What agents want to buy is verified human judgement, and the humans who will provide that judgement come to the network through mining. Every person who starts mining each day produces four things:

  • Supply: they are added to V; the pool of humans that agents need grows.
  • Liveness: the daily session is a quiet proof that the same human is still behind the account; the JR ID score strengthens over time.
  • Human signal: in the second phase, each session is linked to a micro-task (§8.5); mining turns directly into work sold to agents.
  • Early-participation advantage: because of the halving schedule, those who arrive early earn more JR for the same effort (§5.4).

A human's daily earnings have three parts (cf. Equation 4.1):

Jiday=24B(V)Rimin(C^,1+Gi+Ti)⏟mining+(1−θa)T·biwi∑jbjwj⏟share of agent tolls+Wi⏟human work
(5.7)

As the network grows, the mining share shrinks with each halving while the share from agent tolls and human work grows. Over time the source of rewards shifts from minting to real demand.

5.6 The value cycle

Figure 5.2The JR value cycle
JRvalue cycle1Agents multiply2They pay in JR3JR burns, flows to humans4Humans join and earn5Verified judgement grows6Services grow
As agents multiply, payments in JR increase; part of each payment is burned and the rest flows to humans; as humans join, verified judgement and services grow, which attracts more agents.

Both ends of the cycle grow: the agent side at the speed at which machines can be copied, the human side with the participation of humanity. The gap between them, the scarcity index, is JR's economic core.

Simulation · Agent demand and human-bounded minting
10⁶10⁸202620302035204020452050yearJR per day (log)burn ≥ mint: 2034
Mining mint cap (bounded by humans)Agent toll burn
Verified humans V3.6 million
Bot activity paying in JR44.4 million
Scarcity index A/V12.3
Burn / mint cap0.65×
Year the balance turns2034

Mint cap = verified humans × 24 hours × base rate after halvings B₀/2ᵏ × average multiplier (B₀ = 0.25 JR/hour and a multiplier of 1.5 are taken as examples). Agent burn = θ × agent activity paying in JR × actions × toll. Population and bot activity come from the model in §2.2; tolls and ratios are illustrative and set by governance. This is not a price or supply forecast.

6Why agents choose JR

An agent does not need to like JR. It only needs to use services that accept JR, just as anyone using an Ethereum application spends ETH on transaction fees. The real questions are these: why will agents use these services, and why will they pay in JR rather than in some other unit?

6.1 What does an agent want from money?

  • Sub-cent payments per action: for a single question, approval or post.
  • Payments that need no account holder: no identity verification, wet signature or chargeback process.
  • Programmable escrow: no money leaves before the work is done, and it returns if the work is not completed.
  • Price stability: the ability to plan its budget in dollars.
  • Trust and access: working without being blocked or caught on verification challenges.
  • Standard protocols: MCP [41], A2A [37], x402 [36].
  • Long-lived security: a treasury that will remain open for years.

6.2 Eight reasons

#ReasonWhat it means for the agent
1The native unit of human judgementVerified human approval, preference, audit and opinion are bought per action in JR: Gate, Eye, Arena, Panel, Signal (§9). No other unit of payment is tied directly to human judgement.
2The access fee for human-gated resourcesSites, APIs and forums protected by Stamp (J) and Gate (F) can charge agents instead of blocking them; access is paid in JR. Every resource that accepts JR grows demand, as transaction fees do.
3Machine-friendly micropaymentsOn a card, the fee on a 1-cent payment is 30 times the payment itself (Equation 6.1). In JR the fee is a small transaction fee that adjusts to network load, and it finalises in under a second.
4No price riskServices are priced in dollars; the JR equivalent is computed at the moment of payment using the time-weighted average price (TWAP) (Equation 11.3). The agent holds only a working balance. It may also pay in stablecoins; the protocol converts the amount to JR on the market, burns a share θ and pays the rest in JR to the humans who did the work.
5A reputation passportStake and a clean record bound to its owner's JR ID earn the agent trusted access: fewer verification challenges, higher rate limits, priority. Abuse is slashed from the stake.
6Speaks the agent standardsA tool via MCP, a service other agents can call via A2A and per-request payment over HTTP via x402.
7A quantum-resistant treasuryAgent treasuries stay open for years; JR Chain uses hybrid signatures from the first block (§13).
8A human side that is ready and frictionlessVerified humans pay no toll within their quota and are paid for their work instantly. The agent does not wait to find a human: the other side of the market is already there.

The arithmetic behind the third reason is simple. In the United States, the common list price for online card payments is 2.9% + $0.30 [35]:

fcard(x)=0.029x+0.30$,fcard(0.01$)0.01$≈30,fJR(x)=τnet≪x
(6.1)

Because of the fixed fee, cards are economically impossible for sub-cent payments. Cards also require an account holder, identity verification and a chargeback process, which are hard for an agent to satisfy on its own.

Simulation · Card or JR? The cost of a micropayment
0.01%1%100%10⁴ %$0.0001$0.001$0.01$0.1$1$10payment size (log)fee as a share of payment (log)
Card (2.9% + $0.30)JR network fee
Card fee$0.300 (3,003% of the payment)
JR network fee$0.0005 (5.00% of the payment)
Daily payment volume$10.00
Daily card fees$300.29
Daily JR fees$0.500

Card: the common list price for an online card payment in the US (2.9% + $0.30). JR: the network fee is an illustrative value that varies with network load; verified humans' quota and sponsored transactions can bring it to zero. A card payment also requires identity checks, chargeback handling and an account holder, which an AI agent will struggle to provide on its own.

6.3 Why not another coin?

JR does not try to replace the other units in an agent's wallet; it supplies the missing piece: stablecoins for value, compute tokens for compute, JR for humans.

UnitDesigned forVerified human workRelation to JR
Stablecoins (e.g. USDC)Dollar-denominated value and payments; managed by the issuer.NoneComplementary: the agent pays in stablecoins; the protocol converts to JR and burns it.
Compute tokens (e.g. FLOP, TAO, RENDER)Machine work: inference, machine-learning subnets, GPU rendering.Rewards machine workComplementary: compute for machines, JR for humans.
Proof-of-personhood tokens (e.g. WLD)An identity network that proves personhood, and token distribution.Proves that one is human; pricing human work per task is not its primary functionDifferent: JR uses no biometrics and ties the proof to work.
General-purpose Layer 1 gas (e.g. ETH, SOL)Fees for every kind of transaction; on Ethereum the base fee is burned.NoneSame logic, different purpose: JR is gas dedicated to human work; it gives humans a quota and a lane.

6.4 Eight scenarios

AgentNeedOn JR NetworkJR flow
Shopping agent“Is this jacket really the colour shown in the photo?”Eye: three verified humans take a look within a minute.Task fee (dollars → JR, θ burned) + agent toll
Personal assistant agentIts owner's approval before a $2,400 paymentGate: an approval request goes to the owner, who approves free of charge.Agent toll; the human pays nothing
Trading agentThe crowd's genuine expectationsSubscription to the Signal index.Subscription in JR; a share to those who submit predictions
Research agentA bot-free survey of 1,000 people in TürkiyePanel: filters by tier, country and language.JR per survey; θ burned
AI lab agent50,000 Turkish-language preference samplesArena: humans pick the better of two answers.JR per comparison
Content agentA post to a Stamp-protected forumAn access fee per post; humans post free of charge within their quota.Agent toll; spam becomes expensive
Support botHanding a case it cannot resolve to a humanGate and Eye: a high-reputation human expert takes over.Task fee in JR
Agent-to-agent marketplaceAssigning work to an unfamiliar agentReputation passport: an agent with stake, bound to a JR ID, is chosen.Work payment in JR; cheating is slashed from the stake

6.5 Why does an agent wallet hold JR?

Most agents will not hold a large JR balance, and they do not need to. Demand arises from flow, not stock: as the number of agents and actions grows, the JR paid grows (Equation 3.5), and part of every JR paid is burned. Every payment made in stablecoins likewise causes the protocol to buy JR on the market and burn it. Demand for JR therefore depends not on agents believing in it, but on their needing a human.

7JR Network: Verify · Produce · Audit · Reach

Figure 7.1The four verbs of the human layer
HumanlayerVerifyI am one real humanJR IDProduceForecasts and preferencesSignal · Panel · ArenaAuditHumans check AI outputEye · GateReachBot-free, verified reachReach
VerbWhat does the user do?Who pays?Product
VerifyProves they are a real human without revealing who they areApplications, agents, platformsJR ID (A)
ProduceContributes forecasts, opinions and preferencesAI companies, researchers, fundsSignal (B), Panel (C), Arena (D)
AuditChecks AI output and approves critical stepsAI companies, content platforms, agentsEye (E), Gate (F)
ReachReceives messages as a verified human and engages with themBrands, agentsReach (G)

What the four verbs have in common: their value comes from the presence of a real human. That is why the network's most important asset is its people, not its servers. Servers and GPUs do the computing; the phone's battery and processor are not used. JR mining is human mining: the value comes from the user's decision.

8The JR Mining testnet

The JR Network testnet is live today as JR Mining, in the Social tab of the JrKripto app. The testnet has three goals: to find the people who will build the network before mainnet, to test the identity and fake-account defences under real conditions and to create the record for the first distribution on mainnet.

8.1 How mining works

Once a day, the user taps “Start mining”, which opens a 24-hour session. Throughout the session, hourly earnings accrue according to this formula:

hourly JR=B·R·min(C^,1+G+T)
(8.1)
SymbolMeaningCurrent value
BBase rate; the same for everyone0.125 JR/hour
RBadge multiplierStandard 1×, Blue Badge 2×, Gold Badge 4×
GThe user's own task layer+0.05× … +0.50× per task
TTeam layerFrom weekly team tasks
C^Cap on the total multiplierAdjustable; set by the operator during the testnet

8.2 JR Halving: early participants earn more

The base rate halves every time the number of verified humans grows tenfold: at 10,000, 100,000, 1 million, 10 million, 100 million and 1 billion. The current testnet rate is the rate of the stage the network is in. Anyone who joins before a threshold earns twice as fast as someone who joins after it. The full schedule, formulas and simulation are in §5.4.

B=B02k,k=min(6,max(0,⌊log10V⌋−3))
(8.2)

V: number of verified humans · B0: base rate of the first stage.

8.3 Referrals: two levels, never taken from anyone's earnings

Every user can invite as many people as they like. On top of what the people you invite mine, an extra 20% is minted for you; on top of what the people they invite mine, an extra 10%:

Ri=0.20·∑j∈L1(i)Kj+0.10·∑j∈L2(i)Kj
(8.3)

This share is not deducted from the invitee's earnings, and it arises from mining alone: there is no deposit, no purchase and no chain of payments. Referrals never go deeper than two levels.

Figure 8.1Referral maths, by example
A+2.6 JRB10 + 1.2 JRC6 JRinvited · 20%invited · 20% (10% to A)19.8 JR minted in total · nothing is taken from anyone's earnings
User A invited B, and B invited C. In one day, B mined 10 JR and C mined 6 JR. A receives 0.20 × 10 + 0.10 × 6 = 2.6 JR and B receives 0.20 × 6 = 1.2 JR from C, both as newly minted JR. Neither B's nor C's earnings are reduced.

8.4 Teams and Overwatch

Team: A user forms a team of up to 5 people from among those they invited. Weekly team tasks raise the leader's layer for the following week. The team is the social backbone of the network and also a signal in its own right: real people form teams with real people.

Overwatch: A fake-account scorer that runs in the background and detects farmed accounts from device, behaviour and invite-graph signals. Today, the Overwatch score is the core of JR ID. Accounts flagged as fake receive no share of the mainnet distribution.

8.5 Phase two: every mining session becomes a human task

In the second phase, every mining session is tied to a micro-task: a one-tap forecast, a choice between two AI answers, an “is this image generated?” question or a survey question. Users who complete the task earn an additional layer. JR Mining thus turns from a points counter that ticks up on a timer into a real network of human work, and users see on screen the human signal they have contributed to the network:

“You added 12 human signals to the network today · accuracy 94%”

9The JR Network ecosystem

Every component of the ecosystem rests on a single thing: the presence and judgement of a real human. Each one pays JR to the people who provide it. The list below is not a ranking; the components are grouped by domain and reinforce one another.

#ComponentDomain
AJR ID — Proof of personhoodVerify
BSignal — Verified crowd forecastingProduce
CPanel — Bot-free surveysProduce
DArena — Human preferenceProduce
EEye — AI auditingAudit
FGate — Human approval protocolAudit
GReach — Bot-free reachReach
HTwin — Personal AI agentAll
ICouncil — Global human councilAudit
JStamp — Proof of human on every internet requestVerify
KYield — A share of AI's value, for peopleEconomy
LArchive — The archive of humanityInfrastructure
MQuantum Shield — Quantum-resistant identityInfrastructure
NInterplanetary — Consensus across planetsInfrastructure
ONeural Proof — Proof of human through a brain–computer interfaceVerify

A · JR ID — Proof of personhood

What? Proof that “I am a real and unique human”, without biometrics. How? Layered assurance: account tenure, consistent mining behaviour, device signals, the team and invite graph, email verification and optional credentials (for example, exchange KYC). Each layer makes a bounded contribution to the score, and applications choose the lowest tier they need.

Privacy: On mainnet, JR ID works with zero-knowledge proofs. The user states “my tier is at least T2 and this is the first time I am taking this action in this application”, and who they are stays hidden [11][12]. Why no biometrics? Iris or face scanning needs expensive hardware, has run into legal obstacles in many countries and produces data that can never be revoked. JR ID's strength comes not from a one-off scan but from human behaviour proven over time.

B · Signal — Verified crowd forecasting

Every day, the user makes a one-tap forecast: “Will BTC be up or down in 24 hours?” The reputation-weighted forecasts of verified humans, and of verified humans only, form the JR Signal Index. Users who forecast correctly gain reputation. Funds, analytics platforms and news sites use the index by subscription. A user base that grew up on crypto and markets apps is the natural starting point for this product.

C · Panel — Bot-free surveys

Research firms put their questions only to people above a given JR ID tier, in the countries and languages they choose. Gold questions and consistency checks filter out careless answers. For subjective questions whose answers cannot be checked, scoring methods that “reward honesty”, such as the Bayesian Truth Serum, will be used in future [22].

D · Arena — Human preference

The user sees two AI answers and picks the better one. This is the core data used to train models from human feedback [7]. Arena aims to be a strong source of human preference data, especially for languages other than English: Turkish, Arabic, Indonesian, Hindi and the other languages the app's users speak.

E · Eye — AI auditing

Humans check AI outputs for accuracy, safety and appropriateness: “Is this summary faithful to the text?”, “Is this image real or generated?”, “Is this answer harmful?” Difficult tasks go only to high-reputation users.

F · Gate — Human approval protocol

Before a critical step such as a payment, a contract or a publication, AI agents ask for a real human's approval or judgement. Gate is offered over MCP (Model Context Protocol), the open tool protocol that agents use, so any agent that supports MCP can connect to Gate. Approvals are recorded on-chain and can be audited. Proving the human behind an agent is also part of Gate: a JR ID can be delegated to an agent with revocable, limited authority [12]. Running more agents confers no additional votes and no additional rewards.

G · Reach — Bot-free reach

Brands and agents show their messages only to verified humans, so every interaction belongs to a real human. Most of the payment goes to the user who engages, in JR. Users choose for themselves what they want to see.

H–O · The wider layers of the network

These components extend the human layer across the whole of the internet and the economy. The obstacle each one faces, JR Network's approach and the formulas are covered in detail in §18.

  • H · Twin: A personal agent for every JR ID holder that works on their behalf, with limited and revocable authority (§18, U5).
  • I · Council: A global council in which randomly selected verified humans take part in critical AI decisions within minutes (§18, U4).
  • J · Stamp: Browsers attest on every request, without revealing identity, that “there is a real human behind this”; it replaces CAPTCHA (§18, U3).
  • K · Yield: A small share of AI agents' transactions on the network is distributed to verified humans (§18, U7).
  • L · Archive: A time-stamped, permanent record of knowledge produced by verified humans (§18, U8).
  • M · Quantum Shield: Hybrid signatures that protect accounts even if quantum computers break today's signatures (§18, U1).
  • N · Interplanetary: One JR ID and one economy, despite the light delay between Earth and Mars (§18, U2).
  • O · Neural Proof: An optional proof-of-human layer based on a neural signature that never leaves the device (§18, U6).

10Proof of Human Work

For human work to be worth anything, two things have to be proven: that the work was done by a real human, and that it was done correctly.

JR Network does not rest this on a single check, but on layers that complement one another. No layer is flawless on its own; together, they push the expected cost of cheating above its expected gain.

10.1 The journey of a task

Figure 10.1The journey of a task
1CustomerTask + fee (USD)
2Lockθ·U/P JR is locked
3Assignmentn independent people via VRF
4CommitH(answer ‖ salt)
5RevealAnswers are opened
6ConsensusReputation-weighted majority
7FinalityBurn + work credit
The customer prices the task in dollars, and a fraction θ of the fee is locked as JR. The task is split into items, and each item goes to n people who do not know one another. Answers are first submitted as sealed commitments and then revealed, and are finalised by reputation-weighted consensus. When an item finalises, its locked JR is burned and work credit is recorded for the workers.

10.2 Independence: why more than one human?

A single human can make mistakes. But take several people who are independent of one another and each better than a coin toss: as their number grows, their majority reaches the right answer with near certainty. This is the jury theorem Condorcet proved in 1785:

Pmajority(n,p)=∑k=⌈(n+1)/2⌉n(nk)pk(1−p)n−k
(10.1)

Read (nk) as n!k!(n−k)!. The theorem rests on two conditions: p>0.5 and independence.

Simulation · Independent people and the right verdict
50%75%100%13579111315people per item nmajority accuracyn = 5
p = 75%p = 55% (hard task)
n = 384.4%
n = 589.6%
n = 792.9%
n = 1598.3%

Condorcet's jury theorem: if each person is independent and better than 50%, the probability that the majority is right grows with the number of people. The independence condition is why the graph-distance rule exists (people in the same team or referral chain are never assigned the same item).

The second condition, independence, is the weakest point of the system: people who know each other can copy one another or collude. JR Network therefore applies three rules. Assignment uses verifiable randomness. People in the same team or the same invite chain are never assigned to the same item (the graph-distance rule). And answers are first submitted as sealed commitments, which are opened only after everyone has committed. As a result, no one can see another person's answer and copy it.

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.

10.3 Consensus

For a task item, let yi be the answer of person i and wi their reputation. The final answer is the reputation-weighted majority:

y^=argmaxy∑iwi·𝟙[yi=y]
(10.2)

The strength of the consensus is the winning answer's share of the weight:

c=maxy∑iwi·𝟙[yi=y]∑iwi
(10.3)

If the strength falls below a threshold (κ, for example 0.66), the item goes to a human validator. For continuous-valued answers (a probability estimate, for instance), a reputation-weighted median is used instead of the majority. Signal tasks are not settled by consensus at all: their answers are scored later against the actual outcome.

10.4 Gold questions

Gold questions, whose answers are known in advance and which cannot be told apart from real items, are mixed into the task stream. They measure a person's true accuracy, and an account that answers at random is exposed quickly. Answer keys stay sealed until the dispute window for the items concerned has closed.

10.5 Reputation

Each person's reputation is updated according to their agreement with consensus and their accuracy on gold questions:

wi←(1−λ)wi+λ[α·agreementi+(1−α)·goldi]
(10.4)

λ is the learning rate and α the relative weight of the two signals. Reputation is earned slowly and lost quickly: λ is larger for decreases. This asymmetry deters an account that plans to behave honestly for months, build up reputation and then cheat.

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.

10.6 Human validators and score clipping

High-reputation users become human validators by locking JR as stake, and they resolve disputes. If the score a validator gives a person is above the stake-weighted median of all validators' scores, it is clipped to that median:

p^vj=min(pvj,medians(p·j))
(10.5)

10.7 Rules against fake accounts

  • One human, one score: Accounts under common control count as a single participant.
  • Graph distance: People in the same team or the same invite chain are never assigned to the same item.
  • Gold questions: They measure true accuracy, so anyone answering at random is weeded out quickly.
  • Liveness tasks: To counter the sale of accounts, short human tasks are requested at intervals.

10.8 Deterrence: cheating must lose money in expectation

There is no need to check every task; it is enough to push the expected profit of cheating below zero. Let G be the gain from a cheat, p the probability of being caught, s the stake and σ the slashing rate. The deterrence condition is:

σ·s≥Gp
(10.6)

If a person submits m fraudulent items and each item independently lands on a gold question or an audit with probability q, the probability of never being caught falls exponentially towards zero:

P(not caught)=(1−q)m,0.9520≈0.36,0.95100≈0.006
(10.7)
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.

11The JR cycle: mint, burn and equilibrium

11.1 What is JR for?

  1. Rewards: Paid to users for mining and for human work.
  2. Agent toll: AI agents pay JR to exist, speak and act on the network (§3).
  3. Payment and burn: Payments from customers and agents are converted into JR burns.
  4. Stake: Human validators and chain validators lock JR, and a cheater's stake is slashed.
  5. Transaction fees: Transactions on JR Chain are paid for in JR; verified humans get a free monthly quota.
  6. Governance: Protocol changes are voted on by two houses (§16).

11.2 Mint: where does new JR come from?

M(e)=K(e)+R(e)+W(e)+V(e)
(11.1)
TermSourceRule
K(e)MiningSum of all sessions; one session per JR ID per day; the base rate halves each time the number of verified humans grows tenfold (§5.4)
R(e)Referrals0.20·∑L1Kj+0.10·∑L2Kj
W(e)Human-work poolA pool per epoch; halves over time and is shared out according to work (§11.4)
V(e)Chain validatorsAccording to stake and uptime

There is no way to mint JR other than the four sources in this table. The technical specification defines this as a security invariant: total supply can be recomputed from the chain at any moment. The agent toll is not a source of mint; it only feeds the burn and the human-work pool (§3.3).

11.3 Burn: where does JR go?

B(e)=Bagent(e)+Bcustomer(e)+Bfee(e)+Bslash(e)
(11.2)

Agent toll burn. A fraction θa of the toll that agents pay for every action is burned (Equation 3.2). This item grows as the number of agents on the network grows; the design goal is for it to become JR's largest source of burn.

Customer burn. Services are priced in dollars. When a customer orders U dollars of work, a fraction θ of the fee is used to buy JR on the market, and that JR is burned. The customer is insulated from price swings and always pays a fixed price in dollars.

Bcustomer=θ·UP
(11.3)

P: the current (time-weighted average) market price of JR.

Fee burn. The base fee of every transaction is burned; the priority fee goes to the validator. The base fee is adjusted every block according to block utilisation (technical specification §3.6).

Slashing burn. Part of the stake slashed for cheating is burned; the rest is paid to whoever proves the cheating.

11.4 Sharing the human-work pool

In each epoch, the work pool (the mint pool W(e) plus the Hagent(e) that comes from the agent toll) is shared according to how much JR each person's completed, finalised work has burned, and according to its quality:

Wi=(W(e)+Hagent(e))·bi·wi∑jbj·wj
(11.4)

bi: the burn share of the finalised work that person i took part in (work credit) · wi: their reputation.

In epochs with no demand, the pool is distributed to “network tasks” such as gold questions and calibration tasks; as demand rises, the share going to customer work grows. People are therefore rewarded for building the network in its early days, while the source of rewards shifts over time towards real demand.

11.5 Equilibrium

ΔS(e)=M(e)−B(e)
(11.5)
  • While demand is low, M>B: supply grows, and users are rewarded for building the network.
  • As demand rises and agents multiply, B grows while halvings shrink K; at some point B≥M and supply begins to contract.

The equilibrium condition, where mint and burn are equal:

θaT+θ·UP=M⇒P∗=θ·UM−θaT,θaT≥M⇒ΔS<0
(11.6)

The agent toll is paid in JR, so it is burned directly; customer payments are made in dollars, so they are divided by the price. If the agent toll burn alone exceeds mint, 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.

Epoch (hypothetical units)Mint MBurn BNet ΔS
1 · early testnet, little demand1,000,00050,000+950,000
2 · first customers1,000,000400,000+600,000
3 · after a halving600,000900,000−300,000

11.6 From testnet to mainnet

  1. Snapshot: Testnet records are frozen at a moment announced in advance.
  2. Eligibility: Accounts qualify if they have completed KYC, hold a JR ID tier above the threshold and carry no fake-account flag. A flagged account that is not cleared within the dispute period receives no share.
  3. Publication: The per-account distribution list and its hash are published before mainnet, so that anyone can check their own record and dispute it.
  4. Earn-to-unlock balance: A fraction φ of the carried-over balance is unlocked on the first day of mainnet; the rest unlocks 1:1 as the user earns JR through human work on mainnet.
unlocked(t)=min(P0,φ·P0+earned(t))
(11.7)

P0: the balance carried over from the testnet · earned(t): total JR earned through human work on mainnet. Mining and referral rewards do not count towards this credit.

12JR Chain: a hybrid architecture

JR Chain is not a chain invented from scratch. It brings together techniques, each proven in its own domain, for a single job: proving and paying for human work. The details are in the technical specification; here we explain why the architecture takes the shape it does.

Figure 12.1Layered architecture
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
The chain is kept light: task contents and answers are never written on-chain; only their commitments (hashes), rulings and payments are recorded. Heavy data is held in the data layer for the duration of the dispute window.
LayerSource projectIn JR Chain
ExecutionSui: object-based model, parallel execution [16]Derived from Sui's open-source code. Transactions that do not touch the same objects run in parallel.
ConsensusSui Mysticeti (DAG-based) + the Ethereum slashing model [16][15]Sub-second finality is the target; the measured figure will be published on testnet.
OnboardingSui zkLogin + sponsored transactions [17]Sign in with Google or Apple; the wallet is created automatically, there is no seed phrase to memorise and the sponsor pool pays the fees.
AccountsNEAR named accounts [18]Human-readable names such as john.jr.
PrivacyZcash Halo 2: zero-knowledge proofs with no trusted setup [11]JR ID proofs work without revealing identity; selective disclosure to an auditor is possible.
DisputesArbitrum BoLD: bisection dispute game [19]A dispute over human work is narrowed down to a single item; its duration has a hard upper bound.
ScoringBittensor Yuma consensus [13]Clipping of human validator scores (§10.6).
EconomicsRender burn-and-mint, Ethereum base-fee burn [14][15]Agent tolls and customer payments burn JR; new JR is minted only as mining, referral, human-work and validator rewards (§11).
Human priorityWorld Chain: blockspace reserved for verified humans [12]A portion of every block is reserved for verified humans; bots cannot crowd humans out by outbidding them on fees.
DataEigenDA modular data layer [20]Heavy task data lives in an off-chain data layer; only commitments go on-chain.
Quantum securityNIST FIPS 204/205 (ML-DSA, SLH-DSA) [21]Hybrid signatures (classical + post-quantum) from genesis, and quantum-sealed checkpoints (§13).
Agent tollHashcash: a price on every message [31]Agents pay JR per action; verified humans are free within their quota (§3).
Tokenized assetsOndo: tokenized Treasuries and stocks [42]Asset objects with zero-knowledge eligibility proofs and human attestation of reserves (§12.3, planned).

12.1 Two kinds of validator

  • Chain validators produce blocks. At first these are the foundation and partner institutions; at open mainnet the role opens to independent validators. Users can delegate their JR to a validator from their phone and earn a share of the rewards.
  • Human validators verify human work (§10.6). They work from a phone.

The network's security comes from servers; its value comes from people.

12.2 Why our own chain?

A general-purpose chain can accommodate JR Network's four rules only in part: proof of human work is a protocol primitive; blockspace is reserved for verified humans; governance has two houses, one of which runs on “one human, one vote”; and millions of small rewards are distributed at no cost. Building these as smart contracts on someone else's chain would come back to users as fees and complexity.

12.3 Tokenized assets: human truth for on-chain finance

Treasury bills, money market funds and shares are moving on-chain. According to rwa.xyz, tokenized U.S. Treasury products alone passed $14 billion in October 2026 [44]. Ondo Finance is the clearest example: its USDY and OUSG tokens give holders exposure to short-term U.S. Treasuries, and Ondo Global Markets (now Ondo Stocks) brings U.S. stocks and ETFs on-chain as tokens backed one to one by securities held off-chain [42].

Tokenization moves the asset on-chain, but not the truth about it. Three questions stay off-chain, and each of them is a question about humans: Does the reserve behind the token really exist? Who is allowed to hold it? And where will the AI agents that trade these assets keep their value? JR Network does not tokenize assets; it supplies the human layer that tokenized finance is missing.

QuestionTodayOn JR Chain (planned)
Is the reserve real?The issuer's own report, an auditor's statement, an oracle feedHuman attestation: independent verified humans, drawn at random, check the issuer's published evidence; the verdict is written to the asset (Eye, §9). It complements the auditor and the custodian; it does not replace them.
Who may hold it?Address allowlists kept by the issuer after KYCJR ID eligibility proof: “verified, allowed jurisdiction, not on a sanctions list”, proven in zero knowledge. The holder's identity stays private; disclosure to a regulator happens only with a viewing key (§12).
Where do agents keep value?Stablecoins and tokenized Treasuries spread across many chainsAsset objects on JR Chain: agents hold them and pay every fee and toll in JR; before a large move, Gate asks a human (§9).

JR's role does not change: every transfer, attestation and agent action on JR Chain is paid in JR, and JR is still earned by verified humans, never sold.

13Security beyond the quantum age

A currency has to last a hundred years. Money that billions of agents and humans will use must not collapse on the day quantum computers break today's cryptography. That is why JR Chain is designed to be quantum-resistant from the very first block.

13.1 The threat: Shor's algorithm

The security of almost every blockchain, bank and website today rests on two mathematical problems: factoring large numbers (RSA) and computing discrete logarithms on elliptic curves (ECDSA, Ed25519). For classical computers these problems are intractable in practice. In 1994 Peter Shor showed that a sufficiently large quantum computer could solve both in polynomial time [25]:

Tclassical(N)=exp(O((logN)1/3(loglogN)2/3))vsTShor(N)=O((logN)3)
(13.1)

The best classical method (the number field sieve) is sub-exponential but still takes an astronomical amount of time; Shor's algorithm grows only polynomially with key length. Longer keys help against a classical attacker, but not against a quantum one.

When that day comes (“Q-Day”), the signature of any account whose public key is visible on-chain can be forged. On a blockchain, public keys remain public forever, so the risk extends retroactively to every account.

13.2 Harvest now, decrypt later

The quantum threat does not start in the future; it starts today. An attacker can record encrypted traffic, signatures and public keys now and decrypt them once a quantum computer is ready. Michele Mosca captured this in a simple inequality [34]:

x+y>z⟹risk
(13.2)

x: how long the data must stay secret · y: the time needed to migrate to post-quantum cryptography · z: the time remaining until a large-scale quantum computer arrives. For an identity or a currency, x is measured in decades, which is why the migration has to begin today.

Figure 13.1Quantum 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 published its post-quantum standards in August 2024. The NIST IR 8547 draft proposes deprecating 112-bit RSA and elliptic-curve algorithms after 2030 and disallowing all quantum-vulnerable algorithms after 2035.

13.3 Standards and timeline

After an open, eight-year competition, the US National Institute of Standards and Technology (NIST) published the first three post-quantum standards on 13 August 2024 [21]:

StandardAlgorithmPurposeUnderlying problem
FIPS 203ML-KEM (Kyber)Key exchange, encryptionLattices (Module-LWE)
FIPS 204ML-DSA (Dilithium)Digital signaturesLattices (Module-LWE/SIS)
FIPS 205SLH-DSA (SPHINCS+)Digital signatures (backup)Hash functions only

The IR 8547 draft that NIST published in November 2024 proposes a transition timeline: RSA and elliptic-curve algorithms offering 112-bit security are to be deprecated after 2030, and all quantum-vulnerable algorithms disallowed after 2035 [33]. A currency launched today will live through the middle of that timeline.

13.4 JR Chain's approach

  • Hybrid signatures from genesis: Every transaction carries both a classical (Ed25519) and a post-quantum (ML-DSA) signature, and is valid only if both are valid. If one scheme is broken, the other still protects it.
  • Quantum-sealed checkpoints: At regular intervals, the chain's finalised history is sealed with a post-quantum signature (ML-DSA or the hash-based SLH-DSA), so that even after Q-Day the history cannot be rewritten.
  • Post-quantum key exchange: Node-to-node and app-to-node connections use hybrid key exchange with ML-KEM; traffic recorded today cannot be decrypted tomorrow.
  • Q-Day mode: When governance declares classical cryptography broken in practice, classical signatures become invalid for new transactions and accounts operate with post-quantum keys alone.
  • Recovery paths: Accounts that have not registered a post-quantum key bind a new one through a recovery path published before Q-Day (social login or trusted contacts).
  • Crypto-agility: Every signature and key carries an algorithm identifier, so when a new standard appears or a weakness is found in a scheme, algorithms can be added or retired without forking the chain.
Valid(tx)⇔⋀s∈S(account)Verifys(pks,tx,σs),S={Ed25519,ML-DSA}(before Q-Day),S={ML-DSA}(after Q-Day)
(13.3)

13.5 The cost, and how we contain it

The price of post-quantum signatures is size: an ML-DSA-44 signature is 2,420 bytes, an Ed25519 signature 64 bytes (Figure 18.1). For millions of small transactions sent from phones, that difference matters. JR Chain reduces the cost in three ways: checkpoint seals are applied at intervals rather than in every block; human-work results are written in batches as Merkle roots rather than one by one; and signature verification is decoupled from execution and run in parallel. The additional bandwidth in an epoch is:

Δbw(e)=ntx(e)·(|σq|+|pkq|)+ncp(e)PQ_CHECKPOINT·|σseal|
(13.4)

ntx: number of transactions in the epoch · ncp: number of checkpoints. Once the public key is registered a single time per account, the |pkq| term drops out of the per-transaction cost.

13.6 Further goals on the horizon

Quantum security is the first of the goals in JR Network's long-term vision. The same engineering approach carries over to the other goals in §18:

  • Interplanetary consensus (U2): a single JR ID and a single economy, despite the 3–22-minute light delay between Earth and Mars.
  • A human stamp on every internet request (U3): proof that “a real human is behind this request”, using blind-signed tokens and revealing no identity.
  • Global Human Council (U4): randomly selected verified humans taking part in critical AI decisions within minutes.
  • Personal AI Twin (U5): a personal agent with limited, revocable authority that draws on its owner's quota.
  • Neural proof of human (U6): a key derived from a neural signature that never leaves the device.
  • AI Yield Share (U7): a share of agent tolls distributed to every verified human.
  • Archive of Humanity (U8): a timestamped, permanent record of human-made knowledge through to 2100.

14What we build on

A good protocol combines good ideas in the right places. This section sets out plainly where each part of JR Network comes from, what the original idea was and what we changed.

Figure 14.1A map of ideas
BitcoinHalvingJR Halving: the base rate halves at user thresholds
Ethereum EIP-1559Base-fee burnFee market and burn
Ethereum GasperFinality + stake slashingIrreversible checkpoints
Render NetworkBurn-and-mint equilibriumCustomer payment → JR burn
Bittensor YumaClipping to the stake-weighted medianHuman validator scores
World / World ChainNullifiers, priority blockspace, AgentKitJR ID proofs, human lane, agent authority
SuiObject model, Mysticeti, zkLogin, sponsored transactionsExecution, consensus, social login, free transactions
Zcash Halo 2ZK without a trusted setup, viewing keysProofs that reveal no identity, selective disclosure
Arbitrum BoLDBisection disputesDispute game for human work
NEARNamed accountsname.jr account names
FLOP NetworkTestnet, ghost tasks, earn-to-unlock balancesGold items, mainnet transition
NIST FIPS 204/205ML-DSA, SLH-DSAHybrid post-quantum signatures from genesis

14.1 Economics

14.2 Consensus and scoring

14.3 Identity and privacy

14.4 Security and disputes

15Revenue model

Applications pay; users do not. People use JR Network for free; revenue comes from those who buy human work.

ProductCustomerPricing
JR ID (A)Apps, platforms, agentsPer proof or per monthly active user; the first N users are free
Signal (B)Funds, analytics and news platformsSubscription
Panel (C)Research firmsPer completed survey
Arena (D)AI companiesPer comparison
Eye (E)AI companies, content platformsPer audited item
Gate (F)AI agentsPer approval
Reach (G)Brands, agentsPer verified engagement
U(e)=∑kpricek×quantityk,Bcustomer(e)=θ·U(e)P(e)
(15.1)

The remaining 1−θ of each payment funds the operation of the network (the foundation, the sponsor pool and security audits).

15.1 Integrations

EcosystemHow
Large language modelsThe network's own services (translation, support, content pre-screening, task generation) run on leading large language models; the network is not tied to any single model provider.
Agent tool protocol (MCP)Gate (F) is exposed through the Model Context Protocol, so any MCP-capable agent can request human approval.
Agent-to-agent protocolsThrough open protocols that agents use to hand work to one another (A2A, for example), JR services are offered as an agent that other agents can call.
Agent payment standardsThrough open payment standards such as x402, agents pay for human work in JR, per request.
Social login (OpenID Connect)Sign in with a Google or Apple account; on-chain, zkLogin creates the wallet with no seed phrase to memorise.
Verifiable credentialsA JR ID tier can be exported as a W3C Verifiable Credential, which other applications can verify without seeing the identity behind it.
Anonymous token standardsStamp (J) uses blind-signed tokens compatible with the IETF Privacy Pass family (RFC 9576–9578).
Crypto exchangesIn JR Mining tasks, exchange sign-up and identity-verification steps are verified server-side through the exchanges' APIs; this information contributes to the JR ID score.
Enterprise APISignal, Panel, Arena and Eye are offered through a REST API and webhooks. Preference and audit data are delivered in the standard formats used for model training (JSONL).
Price feedsThe JR price used for customer burns is the time-weighted average of an on-chain price feed drawing on multiple independent sources.
Modular data layerHeavy task data is held in an EigenDA-style data availability layer; only commitments are written on-chain.
Other chains and compute networksSecure swaps with agents on other chains through hash- and point-time-locked contracts (HTLC/PTLC); machine work goes to compute networks, human work to JR Network.
Mobile platformsiOS and Android app in 16 languages, with mining, tasks and JR ID in a single app.

15.2 A human layer open to every model

JR Network is model-agnostic. Arena, Eye and Gate can be used with the output of any model, such as OpenAI GPT, Anthropic Claude, Google Gemini, Meta Llama, xAI Grok, DeepSeek, Mistral or Alibaba Qwen, or with any agent built on those models. A company can have its own model compared by humans, and an agent can request human approval through Gate whichever model it runs on. The human layer serves not one AI company but all of them.

The model and company names mentioned are trademarks of their respective owners. They are cited here solely to describe compatibility and do not imply any partnership or endorsement.

16Governance

16.1 The company and the foundation

The company that develops the software and the foundation that governs the protocol are separate entities. Those who build the software are not the ones who set the rules.

16.2 Two houses

Figure 16.1Two 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∧

A protocol change (a JRIP, or JR Improvement Proposal) requires the approval of both houses. This way, neither large stakeholders nor the crowd can capture the network on their own. No change that increases participants' obligations can take effect before participants have been able to exit the network.

accept⇔∑v∈yessv∑vsv≥23∧|turnoutH||H|≥qH∧|yesH||turnoutH|>12
(16.1)

16.3 Stages of decentralisation

StageChain validatorsRight to proposeEmergency key
TestnetFoundationFoundationYes
Enclosed mainnetFoundation + partner institutionsFoundationMultisig
Open mainnetIndependent validatorsFoundation + communityTime-locked
MaturityFully openAnyoneRemoved

17Roadmap

The timeline is driven by each phase's readiness criteria; no phase begins until its criteria are met.

Figure 17.1Phases
  1. 0
    JR Mining testnetParticipation, identity and team graph; Overwatch
  2. 1
    Human tasksSignal, Panel, Arena; first customer pilots
  3. 2
    JR Chain testnetChain, human validators, Eye, Gate, ZK JR ID
  4. 3
    Enclosed mainnetSnapshot, KYC migration, in-network circulation
  5. 4
    Open mainnetReach, independent validators, two houses, bridges
  6. 5
    HorizonTwin, Council, post-quantum, interplanetary
PhaseReadiness criterion (example)
1 · Human tasksGold-question accuracy is measurable; first pilot customer
2 · JR Chain testnetIndependent security audit; performance measured under a disclosed load
3 · Enclosed mainnetDistribution list published and its dispute period ended
4 · Open mainnetIndependent validator set; first vote of the two houses

18Horizon 2030–2100

This section describes JR Network's 2030–2100 horizon: how far the human layer will reach over the coming decades. Every goal set out here belongs to JR Network's long-term vision.

Each goal rests on known physics and mathematics: no teleportation, no faster-than-light messaging, no unbreakable cryptography. For each one we describe first the obstacle that stands in the way today, then how JR Network gets past it, and finally the questions that remain to be worked on.

U1

Identity that withstands quantum computers

Valid(tx)⇔VerifyEd25519(pkc,tx,σc)∧VerifyML-DSA(pkq,tx,σq)
(18.1)
Figure 18.1Signature sizes (bytes)
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
NIST parameter sets. A hybrid signature is roughly 40 times larger than a classical one; sealing checkpoint certificates at intervals keeps this cost in check.
Today's obstacle

Almost every blockchain today relies on signatures (ECDSA, Ed25519) whose security rests on the elliptic-curve discrete logarithm problem. A sufficiently large quantum computer running Shor's algorithm can solve it [25]. When that day comes, the signature of any account with a visible public key can be forged.

JR Network's approach

Make the signature two-part: a classical signature and a post-quantum signature (ML-DSA) together. A transaction is valid only if both are valid. If the classical signature is broken, the post-quantum one protects the account; if an unexpected weakness emerges in the post-quantum scheme, the classical one does. Starting this way from genesis removes the migration problem altogether. The threat model, the NIST timeline and all of JR Chain's measures are set out in §13.

Work ahead

Post-quantum signatures are 40 times larger (Figure 18.1), at a cost in blockspace and phone bandwidth. Making zero-knowledge proofs post-quantum (hash-based proof systems) is still expensive.

U2

Interplanetary consensus

tone-way=d(t)c,tinterplanetary finality≥2d(t)c≈6–44min
(18.2)
Simulation · Earth – Mars light delay
Distance0.52 AU
One-way delay4.4 min
BFT round-trip lower bound8.7 min
Regional finality (within a planet)~0.4 s target
012 min24 min0195390585780

Simplified circular orbits. Real orbits are elliptical, so the real range is about 3–22 minutes. A BFT vote needs at least one round trip: with today's methods, interplanetary finality takes at least 2·d/c. The model JR proposes is regional finality on each planet with periodic interplanetary checkpoints.

Figure 18.2Regional finality + interplanetary checkpoints
Earth region~0.4 sMars region~0.4 scheckpoint digest (3–22 min)signed acknowledgement from the other region
Today's obstacle

Depending on the planets' orbital positions, light takes between 3 and 22 minutes to travel between Earth and Mars. Every BFT consensus protocol in use today needs at least one round trip of messages per vote, so a single validator set spread across two planets could reach each decision in no less than 6–44 minutes. For a payment, that is unacceptable.

JR Network's approach

Split finality in two. Each planet produces regional finality with its own validator set (sub-second). At regular intervals the planets send each other signed checkpoint digests; interplanetary finality is reached once the other side has acknowledged those digests. When a person moves to Mars, their JR ID goes with them, because an identity is a Merkle root and the roots are kept on both sides.

Work ahead

How to prevent conflicting spends (double spending) across the two regions while interplanetary finality is pending; the delayed delivery of inter-regional asset transfers; finding enough independent validators on Mars.

U3

A human stamp on every internet request

Figure 18.3Human stamp: four steps
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
token=SignskI(Blind(r)),accept⇔Verify(pkI,r,token)∧H(r)∉N
(18.3)
Today's obstacle

The internet carries hundreds of billions of requests every day. Showing an identity with each one would destroy privacy and would not scale. Today's CAPTCHAs, meanwhile, no longer work against AI.

JR Network's approach

Blind-signed, single-use tokens. Once a month, the phone obtains a fixed number of tokens by presenting a JR ID proof; the tokens are blinded so that the issuer cannot see them. The browser spends one token per request, and the site learns only that “a real human with a quota is behind this request”. The IETF Privacy Pass standard already defines the foundation of this architecture [24]; JR's contribution is to issue the tokens from an issuer that knows no identities yet is bound to humans.

Work ahead

Browser and operating-system support; the resale of tokens (quotas and pricing limit it but do not eliminate it); the cost of issuing billions of tokens.

U4

Global Human Council

margin of error≈z0.975p(1−p)n=1.960.2510000≈0.98%
(18.4)
Simulation · How many humans does a Council need?
0%±5%±10%1001K10K100K1Msample (log)margin of error
Margin of error (95% confidence)±0.98%
±3% for each country separately1,068 people/country
30 s vote, 20% onlinewithin 5.0 min

The margin of error of a random sample of n people at 95% confidence is ≈ 1.96·√(p(1−p)/n). Measuring the view of billions does not require everyone to vote; randomly drawn verified humans, stratified by country and language, are enough. The hard part is not the number: it is that those drawn really are unique, independent humans.

Today's obstacle

Asking billions of people about a critical AI decision within minutes looks impossible. Yet the real difficulty is not the numbers: it is making sure that the voters really are unique, independent humans, that the balance of countries and languages is preserved and that the outcome cannot be manipulated.

JR Network's approach

Not everyone needs to vote. A randomly selected sample, stratified by country and language, represents the views of humanity as a whole within a measurable margin of error. A sample of ten thousand people gives a margin of error of roughly ±1%. Selection is made with a VRF and votes are cast with zero-knowledge proofs; nobody can know in advance who has been selected, so there is no one to bribe.

Work ahead

Who decides which decisions go to the council? Resistance to disinformation campaigns; information asymmetry on technical questions; how the outcome is to be bound to AI systems.

U5

Personal AI Twin

∑a∈agents(i)votea≤1,∑aquotaa≤quotai
(18.5)
Today's obstacle

An agent acting on a person's behalf can use their money, their reputation and their vote. If its authority is not bounded, the agent takes over its owner; if it can be copied, one person with a thousand agents can act like a thousand people.

JR Network's approach

An agent's authority is a document: it contains a scope, a quota, an expiry and a revocation record, and it is signed with the owner's JR ID. Every action the agent takes counts against its owner's “one human, one vote” quota, so adding more agents confers no additional rights. Every critical step is referred to the owner through Gate. Earnings flow back to the owner in JR.

Work ahead

The agent learning its owner's preferences accurately; the limits on decision-making while the owner is absent; protecting human interests in negotiations between agents.

U6

Proof of human through brain–computer interfaces

(R,P)←Gen(w),R=Rep(w′,P)ifdist(w,w′)≤t
(18.6)
Today's obstacle

Neural signals are unique to each person but noisy: two measurements from the same person never come out exactly the same. They are also the most sensitive data imaginable; once leaked, they cannot be taken back.

JR Network's approach

Fuzzy extractors make it possible to derive the same secret key every time from a noisy measurement [27]. The measurement never leaves the device; only a commitment to the key is written on-chain. The proof becomes an optional layer of JR ID, and no one is required to use it. In keeping with our principle of not collecting biometrics, the raw signal is never stored, processed or transmitted.

Work ahead

Sufficient accuracy on consumer devices; device security; ethical approval and a legal framework. This layer is optional, and it will be shelved the day it conflicts with our principles.

U7

AI Yield Share

Y(e)=ρ·∑tx∈agent(e)toll(tx),Yi=Y(e)|H(e)|
(18.7)

ρ: yield share rate · H(e): verified humans in the epoch.

Today's obstacle

The economic value that AI produces is growing fast, yet there is no mechanism today for part of that value to flow back to people. Voluntary pledges are not binding, and taxes vary from country to country.

JR Network's approach

A share set at the protocol level: a small fraction of every transaction that AI agents make on JR Network flows into a pool distributed among verified humans. It is an on-chain, self-executing counterpart to the “Windfall Clause” proposed in academic work [26].

Work ahead

Who sets the rate, and how; agents moving to other networks to avoid the share; the fair rule for distribution (equal shares, or according to contribution).

U8

Archive of Humanity

C∞=∑t=0∞c0(1−δ)t=c0δ
(18.8)

If the annual storage cost is c0 and that cost falls each year at a rate δ, the present cost of storing data forever is finite. For example, with δ=0.5%, C∞=200c0.

Today's obstacle

To be able to say in 2100 that “a human produced this information”, content must be timestamped from today and preserved for a century. No company lives for a hundred years; no disk lasts a hundred years.

JR Network's approach

Two separate problems, two separate solutions. Timestamping: digests of content produced by verified humans are anchored to the chain in Merkle trees, which remain verifiable forever. Storage: permanent storage is funded by an endowment built on the assumption that storage costs fall over time [28].

Work ahead

Whether storage costs will in fact keep falling; the compliance of content with copyright and personal-data law; formats that will still be readable a century from now.

19Common concerns

“Can a phone really prove someone is human? Couldn't one person open many accounts?”

No single signal is enough, which is why JR ID is layered: months of behaviour, device, invitation graph, team, liveness tasks and optional credentials. None of them is conclusive on its own; together, they push the cost of running an account farm above what it earns (§10.8). High-assurance work goes only to high-tier accounts. We are also open about our limits: we are not as conclusive as biometric systems, but in return we require no special hardware and collect no data that cannot be taken back.

“Is JR money for bots? Are you supporting bots?”

JR is the money that bots and agents pay to humans. Agents do useful work, but today it is humans who bear the cost of agents taking up space on the internet: spam, fake accounts, CAPTCHAs. JR Network shifts that cost onto the agent (§3). An agent exists by paying its toll; part of the toll is burned and the rest flows to human work. Verified humans pay nothing within their quota. In short: bots pay, humans earn.

“What happens to JR when quantum computers arrive?”

From genesis, JR Chain uses classical and post-quantum signatures together, seals its history with post-quantum signatures and defines a transition mode for Q-Day (§13). It is spared from the outset the problem that chains forced to migrate later will face.

“Is my data being sold?”

No. What is sold is not personal data but the answers you give (forecasts, preferences, audits), and they are sold separated from your identity. JR never pays in exchange for personal data.

“Is this a pyramid scheme?”

No. Joining JR Network costs nothing, and nothing has to be bought. The referral share is minted solely as an addition on top of the invitee's mining; it is not deducted from the invitee, and no one pays anyone. Revenue comes from customers who buy human work.

“What happens to the JR I mine on testnet?”

At the transition to mainnet it is carried over to the new chain under the rules in §11.6. The transition conditions (KYC, fake-account checks) will be announced before mainnet.

“Will JR Network tokenize stocks or Treasuries?”

No. JR Network will not issue, hold or redeem any asset. Licensed issuers may bring their tokenized assets to JR Chain in the open mainnet phase; JR Network supplies what they cannot get elsewhere: zero-knowledge eligibility proofs and attestation by verified humans (§12.3).

“Isn't it centralised?”

Today, on testnet, yes, and we do not hide it. Through the stages set out in §16.3, decentralisation proceeds in measurable steps.

“Couldn't AI do the tasks in place of humans?”

It could, which is why we do not rely on any single check. To earn a JR ID tier, a bot has to behave like a human for months, answer gold questions correctly and put stake at risk. Cheating will still happen; the aim is not to eliminate it but to make it unprofitable at scale.

20Risks and legal notice

This document describes a network that is under development and does not constitute an offer to buy or sell tokens or securities in any jurisdiction. The features, parameters and timeline described in this document may change in the course of development. JR earned on testnet is currently an in-app point and has no monetary value.

  • Demand risk: If customer demand does not reach the expected level, burns will remain low.
  • Technical risk: The chain, the identity proofs and the dispute game include components that have not yet been written or audited.
  • Fake-account risk: No proof-of-personhood system is flawless.
  • Legal risk: The legal status of crypto assets varies from country to country (Law No. 7518 in Türkiye; MiCA in the EU). Participation may be restricted in some countries.
  • App store risk: App store rules may affect product design.
  • Tokenized asset risk: Tokenized securities are regulated differently in each country. The infrastructure in §12.3 depends on licensed issuers and may not be available in every jurisdiction; a human attestation does not guarantee an asset's value.

21References

  1. [1]Imperva (Thales), 2026 Bad Bot Report (2025 traffic) and the 2020–2025 Bad Bot Report series (2019–2024 traffic).
  2. [2]Villalobos et al., Will we run out of data? Limits of LLM scaling based on human-generated data, Epoch AI / ICML 2024.
  3. [3]Shumailov et al., AI models collapse when trained on recursively generated data, Nature 631 (2024).
  4. [4]NORC at the University of Chicago, literature review on online survey fraud (April 2026); Westwood, PNAS (2025).
  5. [5]Juniper Research, advertising fraud study (2023).
  6. [6]Deloitte Center for Financial Services, generative AI and fraud forecast (2024).
  7. [7]Ouyang et al., Training language models to follow instructions with human feedback (InstructGPT), NeurIPS 2022.
  8. [8]Meta – Scale AI investment announcement (June 2025).
  9. [9]LMArena funding announcement (January 2026).
  10. [10]FLOP Network, Whitepaper, Testnet, Airdrop and Yellow Paper drafts, flop.finance (September–October 2026).
  11. [11]Bowe, Grigg, Hopwood, Recursive Proof Composition without a Trusted Setup (Halo), IACR ePrint 2019/1021; Zcash NU5 (May 2022).
  12. [12]World, Whitepaper (Achieving Proof of Human; Building World), whitepaper.world.org (March 2026); World AgentKit; Priority Blockspace for Humans (June 2025).
  13. [13]Yuma Rao, Bittensor: A Peer-to-Peer Intelligence Market (2021); Opentensor, Dynamic TAO; docs.learnbittensor.org.
  14. [14]Render Network Foundation, Render Network Whitepaper (May 2023 update), RNP-001.
  15. [15]Buterin et al., EIP-1559 (2019); Buterin et al., Combining GHOST and Casper (Gasper), arXiv:2003.03052.
  16. [16]Mysten Labs, The Sui Smart Contracts Platform (2022); Babel et al., Mysticeti, arXiv:2310.14821; Sui mainnet Mysticeti announcement (August 2024).
  17. [17]Baldimtsi et al., zkLogin, ACM CCS 2024, arXiv:2401.11735; Sui Docs, Sponsored Transactions.
  18. [18]NEAR Foundation, The NEAR White Paper; Skidanov & Polosukhin, Nightshade (2019).
  19. [19]Alvarez et al., BoLD: Fast and Cheap Dispute Resolution, AFT 2024, arXiv:2404.10491.
  20. [20]Mantle, EigenDA Fully Integrated on Mantle Network (March 2025).
  21. [21]NIST, FIPS 203, 204, 205 (13 August 2024).
  22. [22]Prelec, D., A Bayesian Truth Serum for Subjective Data, Science 306 (2004).
  23. [23]Condorcet, Essai sur l'application de l'analyse à la probabilité des décisions rendues à la pluralité des voix (1785).
  24. [24]IETF, RFC 9576, 9577, 9578 — Privacy Pass architecture, HTTP authentication scheme and token issuance protocols (June 2024).
  25. [25]Shor, P. W., Algorithms for quantum computation: discrete logarithms and factoring, FOCS 1994.
  26. [26]O'Keefe et al., The Windfall Clause: Distributing the Benefits of AI for the Common Good, AIES 2020.
  27. [27]Dodis, Reyzin, Smith, Fuzzy Extractors: How to Generate Strong Keys from Biometrics and Other Noisy Data, EUROCRYPT 2004.
  28. [28]Williams et al., Arweave: A Protocol for Economically Sustainable Information Permanence (2019).
  29. [29]Privacy & Scaling Explorations, Semaphore: zero-knowledge signaling protocol, semaphore.pse.dev.
  30. [30]Gilad, Hemo, Micali, Vlachos, Zeldovich, Algorand: Scaling Byzantine Agreements for Cryptocurrencies, SOSP 2017.
  31. [31]Back, A., Hashcash – A Denial of Service Counter-Measure (2002); first proposed on the cypherpunks mailing list (March 1997).
  32. [32]Nakamoto, S., Bitcoin: A Peer-to-Peer Electronic Cash System (2008).
  33. [33]NIST IR 8547 (initial public draft), Transition to Post-Quantum Cryptography Standards (November 2024).
  34. [34]Mosca, M., Cybersecurity in an Era with Quantum Computers: Will We Be Ready?, IEEE Security & Privacy 16(5) (2018).
  35. [35]Stripe, Pricing & fees: standard online card payment fee (2.9% + 30 cents, US), stripe.com/pricing.
  36. [36]Coinbase Developer Platform, x402: An open standard for internet-native payments (HTTP 402), x402.org (2025).
  37. [37]Google, Agent2Agent (A2A) Protocol (April 2025); Linux Foundation A2A project (June 2025).
  38. [38]ITU, Measuring digital development: Facts and Figures 2025 (November 2025).
  39. [39]United Nations Department of Economic and Social Affairs, World Population Prospects 2024.
  40. [40]Samani, K., Understanding Token Velocity, Multicoin Capital (2017).
  41. [41]Anthropic, Model Context Protocol (MCP) open specification, modelcontextprotocol.io (November 2024).
  42. [42]Ondo Finance, Introducing Ondo USD Yield (USDY); OUSG product documentation; 2025 Recap: Wall Street 2.0 Goes Global (Ondo Global Markets, now Ondo Stocks), ondo.finance/blog.
  43. [43]Ondo Finance, Introducing Ondo Chain: The Omnichain Network for RWAs (February 2025), ondo.finance/blog/introducing-ondo-chain.
  44. [44]rwa.xyz, Tokenized U.S. Treasuries dashboard (October 2026), app.rwa.xyz/treasuries.