Theron: a guardrailed capital allocator

Idle capital and physical compute infrastructure are two sides of the same problem. Machine owners need funding, and capital needs a trustworthy way to reach real, verifiable assets. Capturing that value safely demands constant attention, disciplined risk scoring, and a structure that cannot be argued with after the fact.

Theron is an autonomous agent that does the work. It scores compute operators on live telemetry, underwrites them against a strict bar, and moves capital within limits enforced by smart contracts. The agent optimises. The chain constrains.

Most investment products force a choice between two bad options: hand custody to an opaque manager, or manage everything yourself and accept that you will miss regime changes, hardware failures, and revenue declines while you sleep.

An autonomous agent is an obvious third path, but an agent with unchecked authority is just a manager with worse judgement. The unlock is not a smarter agent. It is a smaller blast radius. If the agent can only act within limits enforced on-chain, it can be autonomous without being dangerous.

Depositors put BOT into a shared ERC-4626 vault and receive TRN shares. An off-chain agent runs a continuous decision loop. It reads live telemetry from the chain, scores every registered compute operator against a transparent model, underwrites each candidate, and either allocates capital or honestly declines. Every decision is written on-chain with a plain-language rationale.

The loop runs every 60 seconds, but writes are change-based: a decision is only recorded when the situation actually changes, keeping the protocol economical at scale.

Theron is deployed on BOT Chain mainnet (chain 677) as seven verified contracts.

• TheronFund. The ERC-4626 vault. BOT in, TRN out. A decimals offset mitigates the first-depositor inflation attack.
• NodeRegistry. Registers compute operators with a minimum stake, tracks their telemetry, and enforces slashing.
• AISignatureRegistry. Stores every AI decision: the intent hash, the scoring inputs, and the rationale.
• YieldDistributor. Routes accrued revenue to TRN holders and the restaking system.
• Restaking. Lets TRN holders lock shares for yield boosts of up to 2.00x.
• TheronToken. The TRN share token itself.
• EmergencyGuard. A 2-of-3 multisig circuit breaker held by independent guardians.

All seven contracts are verified on the public explorer, so anyone can read the exact code that governs the fund.

Headline numbers are not the objective. The engine scores every compute operator on five transparent factors:

• Uptime (35%). Measured against real block production on-chain.
• Hardware (25%). The physical grade of the machines.
• Revenue (20%). The operator's real income signal.
• Stake (10%). The operator's own capital at risk.
• History (10%). Track record over time.

Two hard gates protect the fund. An operator must clear a 95% uptime barrier to even be considered, and no single operator can ever hold more than 25% of fund assets. Both are enforced by the contracts, not by the agent.

On top of the deterministic scoring engine sits a language model pass governed by a strict safety kernel. The model reads the live situation and writes a plain-language rationale for every decision, in every case honestly declining operators that do not clear the bar.

The model can only ever recommend within the limits the contracts already enforce. It has no capability to raise a cap, lower the uptime barrier, or increase exposure beyond what the engine sanctioned. The AI advises inside a box it cannot open.

Every decision produces a receipt: the scoring inputs, the chosen outcome, and a plain-language rationale, all anchored by an on-chain transaction in the AISignatureRegistry. Anyone can replay any decision and verify exactly what the AI saw and why it acted. The audit trail is the product, not an afterthought.

The core assumption Theron removes is trust in the agent's honesty. It cannot exceed the on-chain caps, cannot lower the uptime barrier, and can only move capital through the registered operator set. A compromised or misbehaving agent cannot steal funds.

The assumptions that remain are the operators themselves, the correctness of the contracts, the integrity of the oracle data, and the security of the guardian keys. The fund is experimental and has not yet completed a formal third-party audit.

The current minimum operator stake is a deliberately low launch value, set to allow controlled testing of the protocol with limited capital. As the fund grows and allocation sizes increase, the minimum will be raised so that an operator's collateral remains meaningful relative to the capital it manages. Stake is intended to scale with exposure as the fund matures.

TRN holders can lock their shares for fixed periods to boost their share of yield: 1.3x for 30 days, 1.6x for 90 days, or 2.0x for 180 days. Locked shares are committed until the period ends, rewarding long-term alignment with the fund.

• Build the yield engine: the fund's revenue channel from real compute income is not yet active on mainnet, and no returns are currently generated. Native staking will serve as a complementary lever as BOT Chain matures.
• Grow the registered operator set with verified DePIN machines and real customers.
• Formal third-party audit of the deployed contracts.
• Expand scoring with deeper revenue verification and on-chain proof of hardware.
• Raise the minimum operator stake in line with fund growth, keeping collateral proportionate to the capital each operator manages.

This document describes software running on BOT Chain mainnet. It is not an offer, solicitation, or financial advice. Yields are variable and are not guaranteed. Smart contracts carry risk, and you should never commit funds you cannot afford to lose. See the Terms of Service, Privacy Policy, and Risk Disclosure.