# TRUE: Product and Technology Overview

**True Labs Limited · product and technology · figures current to the September 2026 close**

Financials in this document are unaudited. The authoritative machine-readable numbers are in
`kpis.json` on this domain; where the two differ, `kpis.json` wins.

Four people. Eight and a half million raised, running at an eight million annual run rate. A live
zero-knowledge perpetuals exchange settling on Solana, spot and wallet across eleven chains,
a consumer app on web, iOS and Android, an AI automation layer, and an agentic interface no
other venue in this market has shipped.

This document covers the product and the engineering. Full financials are provided
separately.

---

## 1. What TRUE is

Most companies in this market build one of three things: an exchange, a trading app, or
an AI layer. We built all three, and they run as one system.

**The venue.** TRUE operates its own perpetual futures exchange, built as a zero-knowledge
validity rollup that executes off-chain and settles on Solana. Matching runs in a Rust
engine we wrote. Every block is proven in the SP1 zkVM, compressed to a Groth16 proof and
verified by a Solana program that custodies all collateral. Live on mainnet since August
2026, with $1.20B of notional traded since launch.

**The automation.** AI agents that run strategies continuously, copy trading that mirrors
other traders trade for trade, and wallet mirroring that follows proven on-chain wallets.
A deterministic exit engine manages every position independently of the model that opened
it, so a strategy can never talk itself out of a stop.

**The client.** A web application, native iOS and Android apps, and a conversational
interface. Spot and swaps route across eleven chains. Everything is non-custodial: users
hold their keys and sign their own transactions.

**The distribution.** A performance-based revenue share engine built into the same
codebase, attributing every swap, perp and prediction fee to the referring link in real
time, with a full analytics suite, an earnings simulator, six revenue-share tiers to 25%
and weekly USDC payouts settled on chain. On web and on mobile.

**And one thing nobody else has.** TRUE is addressable by machines. A Model Context
Protocol server exposes 37 tools with OAuth access from Claude, ChatGPT, Perplexity,
Cursor, Grok and Gemini. A documented venue API carries REST, WebSocket and SSE for
anyone who wants to run a bot or make markets. A second, 257-tool administrative MCP is
how we run the company.

---

## 2. The venue: a ZK rollup for perpetuals on Solana

We are the only perpetuals venue running a zero-knowledge validity rollup that settles on
Solana. Every other ZK perp venue settles to Ethereum. Every other Solana perp venue runs
without proofs. That position is ours.

### Architecture

| Component | Language | Role |
|---|---|---|
| Sequencer | Rust, axum, openraft, RocksDB | Matching engine, Raft replication, block production |
| Prover guest | Rust on SP1 zkVM 6.3 | Replays each block inside the zkVM, commits both state roots |
| Proof orchestrator | Rust, tokio | Block cadence, proof requests, settlement handoff |
| Settler | Rust, tokio, axum | Oracles, deposits, liquidation, proof submission, withdrawals |
| Verifier | Rust / Anchor on Solana | Custodies USDC, verifies the proof, advances canonical state |
| History service | Rust | Read models, trade history, fee rollups |
| Infrastructure | Terraform, Flux, Helm, AWS EKS | Deployment, observability, disaster recovery |

Settlement is Solana mainnet and collateral is canonical USDC.

### Who can change anything

Every privileged action on the venue requires its own key. The verifier key, the settler
key, the admin key, the sub-admin key and the custody key are separate, and no one of them
can change the system on its own. Pausing the venue, deciding a withdrawal, moving
collateral, altering market configuration and upgrading the on-chain program are each gated
behind a different credential. The keys are held by the three founders.

This is the honest answer to the question a four-person company invites, which is what happens if one
of them is wrong or unavailable. The answer is that no single key reaches the money, and the
administrative MCP records every privileged call with the identity that made it.

### The engineering decision that sets this apart

Our matching engine exists once. The same Rust crate compiles into the sequencer that runs
production and into the RISC-V guest that runs inside the zkVM.

Every other proving system in this category writes the exchange logic twice: once in the
engine, once in the circuit. Those two implementations drift, and the drift is where the
failures live. Because we share one crate, a proof is a mathematical statement about the
exact code that executed the trade. Our CI proves it on every commit: the pipeline
generates block inputs from the sequencer and replays them through the prover.

It is the hardest thing in this codebase to replicate and the easiest for an engineer to
verify.

### Measured in production

| Metric | Measured |
|---|---|
| Matching engine latency, order received to matched | **p50 0.18 ms, p99 10.8 ms** |
| Proof latency | p50 42 s, p99 47 s per block |
| On-chain verification | 284,491 compute units |
| Groth16 proof size | 356 bytes |
| Blocks verified | 255,748 and counting |
| Raft apply lag | 1 entry, quorum ack age 0 s |

These are instrumented from production, not design targets, and the measurement boundary
is stated so anyone can reproduce them: matching latency is the time from the sequencer
receiving an order to the match being written, not an end-to-end round trip from a user's
device. We publish it that way deliberately. Venue latency numbers in this sector are
quoted against different boundaries and are rarely comparable, which is why the one
independent measurement worth citing is Glassnode's: 884 ms median end-to-end on
Hyperliquid, with a 200 ms geographic advantage for traders sitting in Tokyo.

**Verification is cheap enough to do on every block.** 284,491 compute units on Solana, for
a 356-byte proof. That is the argument for settling here rather than on Ethereum. A validity
rollup is only as good as how often it actually proves, and proving every block is an
economic decision before it is an engineering one. 255,748 blocks verified so far, with no
batching of the verification step to save cost.

Market data reaches the apps over a server-sent-events and WebSocket fan-out with one
shared subscriber per market, so venue load scales with markets rather than users. The
venue also publishes every book, trade, quote and mark event on a public WebSocket
gateway with engine-side timestamps, and the fleet is moving onto it now.

### Run like infrastructure

Five Raft nodes on dedicated NVMe. Terraform Cloud per environment, Flux GitOps, Helm
charts in a container registry. Twenty-eight named alerts covering consensus, settlement,
proving lag, prover budget, custody outflow and contract state. Six Grafana dashboards
provisioned as code. Written runbooks for stop-the-world and voter resync.

Around 175,000 lines of Rust across six production services. The sequencer alone is 93,798
lines with 2,533 test assertions and 26 tagged releases.

---

## 3. Automation: agents, copy trading and wallet mirroring

### Autopilot

A user allocates capital, picks a risk profile, and an agent trades a published strategy
from their own non-custodial wallet, continuously.

Two decision engines sit behind it, deliberately separated. **Spot** builds a candidate
universe from trending data, an internal gem radar, technical detectors and social
discovery, runs a pre-trade safety screen covering impersonation, holder concentration,
mint and freeze authority, transfer restrictions and deployer reputation, then makes one
model call that returns structured JSON. **Perps** trades our own venue, with every model
output strictly validated before execution: asset allowlist, leverage range, mandatory
take-profit and stop-loss, correct side, collateral bounds, risk-reward floor and stop
distance.

**Exits are deterministic.** A separate engine runs its own 20-second loop and enforces
stop-loss, partial take-profit with a move to breakeven, trailing stops and maximum hold.
The model opens positions; it does not get a vote on closing them.

**Risk rails fail closed.** Concurrent position caps, a daily-loss circuit breaker,
per-asset cooldowns, a per-execution permit ceiling, hourly outflow caps and fleet-wide
exposure caps. If the system cannot verify the day's loss total, it blocks the open rather
than allowing it. Exits are never gated.

Strategies are authored in a validated YAML schema with three rules worth knowing: a
strategy may only make risk tighter than its base profile, free-text prompt fragments are
screened for injection before reaching a model, and arbitrary code hooks are forbidden.

Claude Haiku 4.5 runs the standard spot tick and escalates to Claude Sonnet 5 on high
conviction or a material move. Claude Sonnet 5 runs every perp tick. Cost is controlled
with per-workload API keys and hard provider-side caps, prompt caching on a shared system
prefix, model tiering and cadence suppression on ticks that would change nothing.

### Copy trading

Leaders place an order and the engine fans it out to followers in bounded parallel.
Follower size is a percentage of the leader's margin, floored, capped against both the
leader's size and the follower's own ceiling, then downscaled to what the follower can
actually afford with fees.

Three independent reconcilers run every five minutes comparing the ledger against venue
state in both directions. Partial fills rewrite follower size pro-rata. Fan-out is
idempotent under unique database indexes, so a retry cannot double-open or double-charge.

The marketplace works. Our top leader is running a hundred percent win rate across 39
active copiers with $15,742 of capital copying him, and the copiers are in profit after
fees. We charge 1 USDC per copied trade and pay 0.70 of it to the leader.

### Wallet mirroring

Users follow proven external on-chain wallets across five chains. These fire from a
blockchain webhook, trade spot, size as a fixed ticket and apply a protective bracket
whether or not the wallet being followed ever sells. Liquidity floors, pool-depth caps,
minimum token age and same-asset cooldowns gate every mirror. Sandwich legs are discarded
rather than copied.

Nobody else in this market ships this.

---

## 4. Liquidity: our own market maker

We run our own market maker against our own venue, in Singapore, isolated so that no
unrelated deployment can disturb the quoting engine.

It quotes post-only ladders up to 25 levels per side with configurable spacing, a depth
growth curve, per-level jitter and reprice hysteresis. Around it sit fast oracle marks,
book and tape recording, an AIMD depth controller, a treasury rebalancer, a bounded taker
and a volatility guard that widens spreads automatically when the mark moves.

Live sending is armed deliberately, behind four independent conditions and a single-writer
lease, with a dashboard kill switch that disarms every market and cancels resting orders,
and a dead-man sweep on process termination. An external watchdog reads the same public
book a browser reads and shares no code or state with the quoting engine.

Owning the maker means we control our own depth and our own uptime rather than waiting for
third parties to show up.

**And it is not where the volume comes from.** Any venue that runs its own maker should
expect to be asked how much of its volume is itself, so here is the split, taken from the
venue's own position ledger on 8 October 2026.

| Closed-position notional | Client | Market maker | Client share |
|---|---|---|---|
| Since launch, 4 Aug 2026 | **$1,064,033,246** | $140,135,908 | **88%** |
| Trailing 30 days | **$330,488,886** | $21,827,095 | **94%** |

577 distinct client accounts, 154,755 client positions. The maker is on more positions than
the clients are and a tenth of the notional, which is what a maker is supposed to look like:
present on most of the book, small on each fill. Both sides of a matched trade are counted
once each, so the market maker row is the maker side of fills it took and nothing else.

---

## 5. Clients

**Web.** React 19, Vite, TypeScript, Tailwind. Roughly 725,000 lines across the
application, with an end-to-end Playwright suite, Lighthouse budgets, bundle-size limits
and ten pre-build guard scripts that block a deploy on money-path invariants and ledger
correctness. Nineteen locales.

**Mobile.** Native iOS and Android, built on React Native with the new architecture and
compiled through EAS with eight custom native modules. 3,971 commits, 28 CI workflows
covering build, submission, over-the-air updates and localisation parity. Live in the App
Store and on Google Play in nineteen languages.

**Admin.** A 71-tab operations console, roughly 83,000 lines, controlling users, support,
translations, releases, points, revenue share, deploys, perps, DEX risk, fee and points
matrices, autopilot, strategy authoring and market-maker configuration. This is why three
people can run a derivatives venue, a consumer app and a growth programme at once.

---

## 6. Multi-chain by default

Eleven chains: Solana, Ethereum, BNB Chain, Polygon, Arbitrum, Base, Tron, Avalanche,
Optimism, Linea and Robinhood Chain. Deposits from six networks, with USDT on Tron
credited as USDC.

Jupiter is fully integrated for Solana, server-proxied with fee injection, DEX exclusion
rules, a circuit breaker and nine dedicated test suites. Relay is the primary cross-chain
rail and deBridge the fallback, both server-quoted and client-signed. LI.FI and 0x run
server-side as route and price adapters.

Deposit by QR, withdraw with code verification and server-side review, transfer between
venue sub-accounts, and convert any supported token on any supported chain to the user's
own wallet. Fiat in with card, Google Pay, Revolut Pay and bank transfer, across three
on-ramp providers, landing directly in the user's own wallet. The server holds a
trade-only delegate and cannot sweep anything.

### Fully Web3 ready, with no forced signup

A user who already has a wallet never has to create an account with us. They connect and
they trade.

The login surface carries email one-time code, phone, passkey, Google, Apple, X and
Telegram, plus direct wallet connect across **158 wallets**, with wallets already
installed on the device detected and promoted to the top of the list. Dynamic provides
wallets and authentication on web and mobile with MPC-backed embedded wallets, and Privy
runs in parallel behind a facade. Phantom, Solflare and Backpack connect by deep link, and
WalletConnect v2 is supported on mobile.

One account model therefore serves two completely different users. Someone arriving from a
consumer ad signs in with an email code and never sees a seed phrase. Someone arriving from
crypto Twitter connects Phantom and is trading in one click, with no second signup, no
email handover and no custody of their keys by us.

---

## 7. Distribution: the partner engine

Every venue in this sector ends up paying for users. We built the machine that does it, and
it is the same grade of software as the exchange itself.

The revenue share programme is fully performance oriented and fully traceable.
There is no flat bounty and no opaque monthly statement. A partner earns a share of the
fees their clients actually generate, it is paid every week, and they can see exactly where
each cent came from.

**Real-time attribution on every event.** Swaps, perpetual trades, prediction markets and
every other chargeable action on web and on mobile are attributed to the referring link at
the moment they settle and appear in the partner's dashboard immediately. Today, yesterday,
seven days, thirty days, ninety days, broken out by product with volume, fees and the
partner's own cut on each. Nothing has to be taken on trust.

**A full analytics suite, not a referral counter.** The portal reports the whole funnel:
link clicks, signup rate, activation rate, the count of clients past the active threshold,
per-client earnings and a ranked earner board. One live link in the system currently shows
2,463 clicks, a 9% signup rate, 223 signups, 13% of those going active, 28 active clients
and $65,287 of partner earnings at a 25% share. Professional traffic buyers do not scale
spend without that resolution, and they have it from the first day they join.

**An earnings simulator.** Partners model their own economics before they commit. Sliders
for active referrals, new referrals per week, average monthly volume per referral and
product mix produce a projection per week, per month and across twelve weeks, with a
week-by-week growth curve and a per-referral yield. It is the strongest conversion tool in
the programme because it answers the only question a professional partner actually asks,
which is what this is worth to them.

**Six levels, 10% to 25%.** The revenue share scales with performance, and every tier is
visible in the product, so a partner always knows what the next level pays and what it
takes to get there. It is paid on every product a referred client touches, perps, swaps and
prediction markets alike, not on perpetual taker fees alone. Headline percentages in this
sector are not comparable until you ask what they are a percentage of: dYdX publishes an
affiliate tier reaching 50% of qualifying taker fees, which is a larger share of a narrower
base.

**Weekly USDC, settled on chain.** Payouts run weekly in USDC, with the transaction
exposed in the partner's own ledger so they can verify payment independently of us.

**Revenue share and rebates** sit on top: partners earn an agreed share of the fees their
referred traders actually generate, and traders earn rebates against the fees they have
actually paid. Nothing is paid up front and nothing is paid out of capital. All of it is configurable from the admin console without a deploy.

**Team structures** are tracked as a tree, so agencies and sub-partner networks onboard
without a separate reporting layer.

The whole portal ships in the production app on web, iOS and Android, with the same
dashboard, the same simulator and the same payout ledger on every surface. It scales with
the venue for free, because it reads the same fee events the venue already writes.

---

## 8. Agentic finance, and the venue API

TRUE is a first-class client for other AI systems and for other people's trading
infrastructure, not just for humans in a browser. Three surfaces, all shipped, all
documented at docs.truefinance.ai.

**User MCP: 37 tools** over streamable HTTP and SSE covering balances, positions, prices,
charts, market mood, news, watchlists, asset and memecoin search, holder and security
data, agent lifecycle, perp markets, trade execution, limit orders and trader following.
OAuth with scoped access, with Claude, ChatGPT and Perplexity seeded as directory clients.
Execution is human-in-the-loop: the server never signs, it returns an approval request,
the user confirms in their app, and the app signs.

**Admin MCP: 257 tools** on a single authenticated endpoint requiring super-admin
credentials, with per-tool scope gating applied at registration and again at execution.
Read-only SQL, venue pause and kill switches, withdrawal review, account freezes, market
flags and leverage caps, solvency runs, forced closes, autopilot controls, copy
reconciliation, market-maker configuration, payouts, revenue share credits, push, metrics and
logs.

This is not a demo. It is how the company is operated. Incident response, support triage,
reconciliation, release verification and reporting all run through it, with AI agents
doing the work and humans approving anything that moves money. It is the single biggest
reason our output per person looks the way it does.

### The trading API

The venue is open to anyone who wants to run their own execution against it. Three
transports, all live:

- **Venue REST** at `dex-prod.truefinance.ai` for placing, modifying and cancelling
  orders, accounts, positions and transfers, authenticated by API key with an Ed25519
  signature on every write.
- **WebSocket** for sequence-numbered order book deltas, best bid and offer, oracle marks
  and trades.
- **SSE** streams for the same market data where a socket is not wanted.

Users create and revoke their own API keys in the product, scoped to a delegated signer so
a bot never holds the owner key, with per-key rate, burst, socket, topic and SSE ceilings
and an IP allowlist. Keys are visible, revocable and auditable from the app.

The documentation is written for developers and for AI agents, with `llms.txt` and
`llms-full.txt` served alongside it, a CLI (`npm install -g trueai`), and a one-click
connector flow for Claude, ChatGPT, Perplexity, Cursor, Grok and Gemini.

**This is the standard institutional market makers and liquidity providers expect**:
signed REST for order entry, sequence-numbered WebSocket deltas for the book, post-only
quoting, and an explicit convention that every displayed price is the oracle mark rather
than the last fill. We built that surface before we needed it. **No external market maker
or liquidity provider is onboarded yet.** The connectivity is finished, the documentation
is published, and that programme starts now, which means third-party depth is upside that
has not been drawn on.

**In the product**, AI also runs the narrator that explains every agent tick in the user's
language, portfolio and technical analysis, the signals pipeline, the news and trends
desk, withdrawal risk verdicts, support triage through Kira, and translation across
nineteen locales. Kira resolves support herself with a 100% reply rate, a sub-minute
average response and 100% resolved same day.

---

## 9. The integration surface

A list of the third-party systems TRUE runs on, as of 7 October 2026. It is here because
the breadth of it is part of the point: this is the integration load a three-person company
is carrying in production.

**Wallets, authentication and identity.** Dynamic is the primary wallet and authentication
provider across web and mobile, with MPC-backed embedded wallets, an admin API, JWKS
verification and provisioning webhooks. Privy runs in parallel behind a facade as a second
lane. WalletConnect v2 covers mobile. Phantom, Solflare and Backpack connect by deep link.
The connector set reaches 158 wallets in total.

**Settlement and proving.** Solana mainnet for settlement, the Succinct SP1 zkVM for block
proving, and an Anchor program on Solana for Groth16 verification.

**RPC, nodes and explorers.** Helius is the primary Solana provider, including LaserStream
gRPC and webhooks. QuickNode and Alchemy supply additional webhook and notification rails.
Public Solana RPC, PublicNode for Ethereum, the Base RPC and the Robinhood Chain RPC cover
the rest. Solscan, Etherscan, Basescan, Tronscan and Blockscout for explorer links and
verification.

**Swaps, bridging and routing.** Jupiter is fully integrated for Solana across Ultra, Swap,
Price, Trigger and Recurring, with the token registry and a referral account. Relay is the
primary cross-chain rail and deBridge DLN the fallback. LI.FI and 0x run server-side as
route and price adapters. Kamino, Marinade, Sanctum, Meteora, Raydium and Orca are
integrated for staking and DeFi routes.

**Fiat on-ramps.** Banxa, Onramper and MoonPay, carrying card, Apple Pay, Google Pay,
Revolut Pay and bank transfer, with signed sessions and webhook receipts reconciled against
our own ledger.

**Market data and pricing.** Pyth through both Hermes and Lazer for oracle marks. CoinGecko
Pro, Birdeye, DEX Screener, GeckoTerminal, CoinGlass, Bitquery, Binance public market data,
Polymarket data and CLOB endpoints, Finnhub and EODHD for traditional markets, LunarCrush
and CryptoPanic for sentiment and news.

**Research and on-chain intelligence.** Nansen, Glassnode, Moralis and CoinMarketCap, served
to our agents through an internal data MCP so every model call goes through one rate-limited,
audited gateway rather than scattered keys.

**AI.** Anthropic is the primary model provider, with separately scoped keys per surface:
spot agents, perp agents, chat, Kira support, Kira operations, content, translation,
intelligence and social. OpenAI supplies Whisper transcription. ElevenLabs runs the voice
agent.

**Growth, messaging and attribution.** Customer.io for lifecycle email, push messaging and
segmentation. Resend for transactional mail. AppsFlyer with OneLink for mobile attribution,
including server-to-server product events and raw report ingestion. Segment and Mixpanel for
analytics. Expo push for mobile notifications. The Telegram Bot API runs login, referral,
perps, community, status and crash-alert bots.

**Compliance.** Chainalysis for address screening.

**Infrastructure and operations.** AWS as the primary cloud, with EKS, S3, ECR and Secrets
Manager. Terraform for provisioning, FluxCD with Helm and Kustomize for GitOps delivery, and
Vault for secrets. Cloudflare for edge and R2 object storage. MongoDB and Redis for state and
caching. Railway for a set of auxiliary services. Observability is Prometheus, Grafana, Loki
and Alertmanager, with Sentry on frontend, backend, agents and chat, and PagerDuty and Slack
for paging.

---

## 10. How we compare

Nine venues, including the two the first version of this document left out. Drift and GMX
are in because leaving out the closest Solana competitor would be the first thing a
diligent reader noticed.

### Architecture and settlement

| Venue | Architecture | Settles on | Proof system | Order book | Live since |
|---|---|---|---|---|---|
| **TRUE** | **ZK validity rollup** | **Solana** | **SP1 to Groth16, verified on chain** | **CLOB, proven in zk** | Aug 2026 |
| Hyperliquid | Own L1 (HyperBFT) | Own L1 | None | On-chain CLOB | Aug 2023 |
| Lighter | ZK validity rollup | Ethereum | SNARKs, undisclosed | zk-CLOB | Oct 2025 |
| dYdX v4 | Cosmos appchain | Own chain | None | Validator book, unproven | Nov 2023 |
| Jupiter Perps | Solana programs | Solana | None | Oracle-priced pool | 2024 |
| edgeX | StarkEx rollup | Ethereum | STARK (StarkEx) | Off-chain CLOB | 2024 |
| Paradex | Starknet appchain | Starknet | STARKs | Off-chain CLOB, private | Feb 2024 |
| Drift | Solana programs | Solana | None | DLOB, JIT auctions, AMM | 2021 |
| GMX | Smart contracts | Arbitrum, Avalanche | None | Oracle-priced pool | 2021 |

Read the first three columns together, because that is where the position is. Lighter, edgeX
and Paradex prove their books, and all three settle to Ethereum. Jupiter, Drift and TRUE
settle on Solana, and only TRUE proves. A 200-venue public survey of perpetual DEXs turns up
no other Solana venue settling with zero-knowledge proofs. That is a narrow lane and it is
ours, and it is a narrower claim than "the only ZK perps venue", which would not be true.

### Product and distribution

First-party only. Third-party bots, copy tools and ecosystem automation exist around most of
these venues; this table is about what the venue itself ships and operates.

| Venue | Copy trading | AI agents | MCP / LLM | Consumer login | Mobile | Partner programme |
|---|---|---|---|---|---|---|
| **TRUE** | **Native** | **Native** | **37 tools** | **Wallet + email, phone, passkey, social** | **iOS + Android, 19 locales** | **10 to 25%, fully attributed** |
| Hyperliquid | Vaults | No | No | Wallet | Android MVP, no iOS | Referral rewards |
| Lighter | No | No | No | Wallet | iOS + Android | Referral |
| dYdX v4 | No | No | No | Wallet | iOS + Android | Affiliate, to 50% of taker fees |
| Jupiter | No | No | No | Wallet | Jupiter Mobile | Referral |
| edgeX | No | No | No | Wallet | iOS + Android | Referral |
| Paradex | Vaults | No | No | Wallet | Not established | Referral |
| Drift | No | No | No | Wallet | Ecosystem apps | Referral |
| GMX | No | No | No | Wallet | Mobile web | Affiliate, tiered |

Two honest notes on this table. Several of these venues can be reached with an embedded
wallet through a third-party frontend or an SDK, so "wallet" describes the venue's own front
door, not the limit of what is possible around it. And most of them pay affiliates; the
question is not whether a programme exists but whether a partner can see the fee events
behind their own payout. That is the distinction section 7 is about.

### Capital and team

Only the venues with a clean public number are listed. Token market capitalisation,
ecosystem treasury and parent-company funding are not the same thing as capital raised, and
we would rather show four comparable rows than nine misleading ones.

| Venue | Capital raised | Team |
|---|---|---|
| **TRUE** | **~$8.5M equity** | **4** |
| Hyperliquid | No conventional VC | 11 reported, 2026 |
| Lighter | $68M round, Nov 2025, at $1.5B | Not disclosed |
| dYdX | VC-funded, above $80M historically | Not disclosed |

Our equity valuation was priced in June 2026 at $150M. The business has since reached an
annual run rate of approximately $8M with paid acquisition not yet switched on. For
reference rather than comparison, token market capitalisation and fully diluted valuation
from CoinGecko on 7 October 2026: Hyperliquid $19.6B / $84.4B, Lighter $880M / $3.52B,
Jupiter $1.05B / $2.16B, edgeX $142M / $405M, dYdX $112M / $127M, Paradex $28M / $42M.

The three is the current operating team, not the total resource that built this.

### Where we win

**The ZK-on-Solana lane is ours alone.** Lighter, edgeX, Paradex and Extended all settle
to Ethereum. Every Solana perp venue runs without proofs. A 200-venue public survey
confirms it: no Solana venue settles with zero-knowledge proofs. We are the only one.

**The automation layer is vacant at the first-party level.** None of the nine venues ships
its own trading agents, strategy automation or LLM interface. Hyperliquid's native copy
product is a vault taking a 10% profit share. Everything else in this category across the
set is a third-party bolt-on, which matters because a bolt-on cannot see the book, cannot
size against the user's real collateral and cannot be held to the venue's own risk
controls. We ship agents, copy trading, wallet mirroring and an MCP interface inside the
product, against our own matching engine. For a market whose next generation of traders
will delegate to software, that is the whole game.

**Mobile is an open goal at the top.** Hyperliquid, the largest venue in the world by a
wide margin, shipped its first native app in April 2026, Android only, limited to
notifications, and still has no iOS app at all. We ship both stores in nineteen languages.

**We are the only venue that is also a consumer product.** Perps on Solana, spot and wallet
across eleven chains, prediction markets, staking, signals, a social layer and an AI
assistant in one account. Everyone else in these tables sells a trading screen.

**Both ends of distribution are ours, and both are measured.** Every venue here has a front
door built for someone who already holds crypto, and most of them pay affiliates. The
difference is instrumentation. We take a wallet in one click and we also take an email code,
a phone number, a passkey, Google, Apple, X or Telegram, so paid acquisition converts rather
than bouncing at a wallet prompt. At the other end, a partner sees the fee events behind
their own payout: every swap, perp and prediction attributed to their link in real time, the
whole funnel from click to active client, an earnings simulator, and weekly USDC settled on
chain that they can verify without trusting our accounting. A referral code that pays out
monthly from a spreadsheet is not the same product, whatever the headline percentage says.
Owning both ends, in the same codebase as the exchange, is what lets four people buy users
profitably.

### On the published numbers in this sector

We have left volume and user rows out of this table on purpose, because the public figures
do not hold up. edgeX reports $34.1B over thirty days and $556M over twenty-four hours, a
run rate under half its own average. Aster runs at 62% of its average, Veranta at seven
times. Lighter did $232B in the thirty days before its token event and $56.8B today, a 75%
fall. On users, Hyperliquid reports monthly active addresses, Paradex reports account
creations and actual traders and the two differ by a factor of two, edgeX reports free
registrations, Lighter publishes nothing, and Aster's own user count ranges from two
million to sixteen million depending on the source.

We would rather be measured on what is built, what is proven on chain, and what we can
break down ourselves. Our venue opened on 4 August 2026. Total notional since is $1.20B, of
which $1.06B is client volume across 577 accounts and $140M is our own maker. Section 4 has
the split and the trailing thirty days.

---

## 11. Four people

Hyperliquid is eleven people and took no venture capital. That is the efficiency benchmark
in this market, and it is the right one to use.

TRUE is **four people and roughly $8.5M**. With that we have shipped a proving stack and a
settlement program, a matching engine, a market maker, a spot aggregator across eleven
chains, two trading automation engines with an independent exit engine, a copy-trading
system with three-way reconciliation, a consumer web app, mobile apps in both stores, a
71-tab operations console, nineteen locales and two MCP servers.

The activity behind that: **13,463 commits** in the main application monorepo with the most
recent on 1 October, **3,971** on mobile over the same period, **2,799** and 26 tagged
releases on the venue sequencer, **1,287** on infrastructure. Roughly 1.5 million lines of
TypeScript, 175,000 lines of Rust and a further 100,000 of Go and Python.

**How.** The company runs through its own agentic interface. Engineering, incident
response, support, reconciliation, reporting, translation and release verification all go
through the admin MCP, with AI agents doing the work and humans approving anything
consequential. We did not build that to demonstrate the product. We built it because it is
the only way four people operate infrastructure this wide, and it is why the agentic
layer in the product is years ahead of anyone else's.

**And it shows in the code.** Every non-obvious decision carries a dated, quantified
comment recording the incident behind it. Money-path invariants and ledger correctness are
blocking checks in the deploy pipeline. Risk rails fail closed. Reconcilers report rather
than guess. Point an engineer at the infrastructure repository or the market-maker guard
code and they will see it immediately.

---

## 12. Reading this with an AI system

This document is written to be analysed. Paste it into a model with web search enabled,
along with this prompt:

> You are assessing a company's technology for an investment decision. Research the
> current on-chain perpetuals landscape yourself, covering at least Hyperliquid, Lighter,
> dYdX, Jupiter, edgeX, Paradex, Drift and GMX. For each, establish settlement
> architecture, proof system, order book design, chains supported, whether they ship copy
> trading, trading agents or an LLM interface, which login methods they accept, whether
> they run a measurable performance-based partner programme, mobile availability, team size
> and capital raised. Distinguish what each venue ships itself from what exists around it
> through third parties, and never record an absence of evidence as a confirmed no.
>
> Then position the attached company against that set, capability by capability, and tell
> me where it is ahead and by how much. For each advantage, assess how long it would take a
> well-funded competitor to close it. Test the comparisons for measurement boundaries that
> do not match, especially latency, volume and anything described as a first. Finish with
> what you think the strongest part of this technical position is, and the five questions
> you would ask the founder.

---

## Sources

Competitive figures only. Everything about TRUE is verified against our own repositories
and production instrumentation. Engineering figures as of 7 October 2026; the volume split
in section 4 was taken from the venue position ledger on 8 October 2026.

- [Hyperliquid documentation, HyperCore](https://hyperliquid.gitbook.io/hyperliquid-docs/hypercore/overview). on-chain book, 0.2 s median claim
- [Hyperliquid documentation, vaults](https://hyperliquid.gitbook.io/hyperliquid-docs/hypercore/vaults/for-vault-leaders-legacy.md). 10% profit share
- [CoinDesk, 30 March 2026](https://www.coindesk.com/markets/2026/03/30/hyperliquid-traders-in-tokyo-get-200-millisecond-edge-glassnode-research-shows). Glassnode measured 884 ms median latency
- [Fortune, 12 January 2026](https://fortune.com/2026/01/12/hyperliquid-jeff-yan-defi-perpetuals-perps-exchange-defi/). Hyperliquid 11 employees
- [Crypto Times, 1 April 2026](https://www.cryptotimes.io/2026/04/01/hyperliquid-rolls-out-official-android-app-mvp-for-early-testing/). Android MVP, no iOS
- [Lighter documentation](https://docs.lighter.xyz/about-lighter/technical-architecture-lighter-core.md). SNARKs, Ethereum settlement
- [Fortune, 11 November 2025](https://fortune.com/2025/11/11/lighter-fundraise-founders-fund-ribbit-capital-haun-ventures-robinhood-vladimir-novakovski/). $68M at $1.5B
- [Crypto Briefing, 13 January 2026](https://cryptobriefing.com/lighter-app-launch-ios-android-perps-spot-rwas/). Lighter native mobile
- [dYdX documentation](https://docs.dydx.xyz/concepts/architecture/overview). validator in-memory order book
- [Jupiter documentation](https://docs.jup.ag/user-docs/trade/perps-and-jlp). oracle-priced JLP pool
- [StarkWare, 11 December 2025](https://starkware.co/blog/paradex-s-two-the-fastest-prover-meets-the-fastest-appchain/). Paradex S-two prover
- [awesome-perp-dex survey](https://github.com/buddies2705/awesome-perp-dex). no Solana venue settles with ZK proofs
- [dYdX Foundation, affiliate programme](https://www.dydx.foundation/blog/dydx-affiliate-program) and [BeInCrypto](https://beincrypto.com/dydx-performance-based-affiliate-tier/). performance-based tier to 50% of taker fees
- [CoinDesk, 26 October 2021](https://dev.coindesk.com/zh/business/2021/10/26/solana-based-perpetual-swaps-dex-raises-38m). Drift raise, dates the venue
- [DefiLlama perps dashboard](https://defillama.com/perps). volumes, read 7 October 2026
- CoinGecko coin pages, read 7 October 2026: [Hyperliquid](https://www.coingecko.com/en/coins/hyperliquid), [Lighter](https://www.coingecko.com/en/coins/lighter), [dYdX](https://www.coingecko.com/en/coins/dydx-chain), [Jupiter](https://www.coingecko.com/en/coins/jupiter-exchange-solana), [edgeX](https://www.coingecko.com/en/coins/edgex), [Paradex](https://www.coingecko.com/en/coins/paradex)

---

*True Labs Limited · Incorporated in Hong Kong, Company No. 78786155 · Flat 5, 4/F, Won Hing Building, 74-78 Stanley Street, Central, Hong Kong*