DeRLDeRL.
CONNECT
PROOF-OF-VERIFIED-ROLLOUT · ON-CHAINDEPIN / ROBINHOOD · $DEREL · v0.1

We generate &
verify RL
rollouts.

Frontier agents learn inside environments with provable reward — agents that reach the goal, paths a verifier can re-walk step for step. DeRL generates those rollouts and verifies every one by re-execution, then anchors the proofs on Robinhood Chain. Every number below is read live from the derl_core program — nothing simulated.

ENVIRONMENTS0
BATCHES ANCHORED0
SETTLEMENTS0
01How It Works

One rollout, end to end.

Every verified rollout travels this exact path. Re-execution is the real check — each reward is recomputed by an independent verifier, not trusted — and the proof is anchored on Robinhood Chain. The counters below are read live from the chain.

01
Assign

A bounty escrows funds against an environment on-chain — no cherry-picking the easy work.

environments
02
Solve

A node solves a real task — a gridworld path, code against hidden tests, or a math answer — and signs it (ed25519).

open bounties
03
Re-execute

An independent verifier re-executes the reward: re-walks the path, runs the hidden tests, recomputes the answer. Reproduce it yourself via /api/verify.

environments
04
Attest

The verifier confirms and signs; a curator drops near-duplicates. A wrong reward is rejected and slashed on-chain.

leaves anchored
05
Batch

Verified rollouts are hashed into a sha-256 Merkle tree and the root is anchored on Robinhood Chain.

batches anchored
06
Settle

The derl_core program releases escrow per verified rollout and burns the protocol fee — all on-chain.

settled
TRUSTLESS · ON-CHAIN
03Buy verified fuel

Pay. Get real fuel.

The product is verified AI training data. Pay on-chain in SOL, USDC or $DEREL and the network generates a real dataset of verified RL rollouts — every reward re-executed, a Merkle root as proof — and hands you the file to download. No subscription, no API key, no trust required.

initialising wallet…
what you receive
01
Pick & pay

Choose a family (code, math, grid or mixed) and size, then pay on-chain. The payment is verified on Robinhood Chain before anything is generated.

02
Real generation

The network produces that many rollouts and re-executes every reward — only verified ones make the cut (RLVR).

03
Proof + download

You get a JSONL dataset — task, solution, reward, trajectory per row — plus a sha-256 Merkle root over the set. Download it straight from the browser.

This is the scarcest fuel for training agentic models — and you actually receive it, not a promise.
04Dataset marketplace

Sell your own fuel.

A peer marketplace for verified RL data. List a dataset you own, set a price, and anyone can buy it in SOL or $DEREL — you keep 80%, the protocol takes a 20% commission. DeRL-format rollouts are re-verified on listing, so buyers see exactly how much of a set actually re-executes.

initialising wallet…