The OptiLearn docs. Camera to settlement, on one page.

Robots learn to manipulate the world by watching it being done, and first-person footage of real work barely exists at scale. OptiLearn posts bounties for the specific data labs are asking for. A person reviews every submission against a published capture standard. Approved work is paid in tokenized stock on Robinhood Chain, and each approved contribution gets an attestation on OP Mainnet so a buyer can check where every file came from.

1. 1
   Connect a wallet or paste an address

   Sign in with a wallet, or with email and we create a wallet for you through Privy. You can also paste any EVM address. That address is where rewards are sent. There is no password to set and nothing to fund.
2. 2
   Pick a bounty and capture

   Choose an open bounty, read its brief, then record in the browser or upload a file you already made. One task per submission, filmed or recorded by you.
3. 3
   Human review

   A reviewer checks the submission against the capture standard and the bounty brief, gives it a score, and sets a reward in US dollars inside the bounty's range. This usually happens inside a day.
4. 4
   Settlement and attestation

   The dollar reward is converted into the stock token you picked at the Chainlink price and sent to your address on Robinhood Chain, after a person checks the payout. The contribution is attested on OP Mainnet. Track every step in Portfolio.

Each bounty has a brief, an accepted length, a pay range in US dollars and a target count. It feeds exactly one dataset in the marketplace, and it closes when the dataset has the approved items it needs. There are 17 bounties open across video, audio, images, text and sensor logs.

Tiers

The same review applies across the range. What changes is how hard the task is to capture well, and what it pays.

Video lengths above are for video bounties. Audio, image, text and sensor bounties carry a tier too, with their own length rules and pay range shown on each bounty page. The exact reward is set at review, inside the range.

From bounty to dataset

Every approved item lands in one dataset. This is the full map as it stands today.

• Household ManipulationVideo, commercial license
  Folding laundry$2–$4Loading the dishwasher$2–$4Surface wipe down$2–$3Kitchen prep$3–$4
• Skilled TradesVideo, exclusive license
  Plumbing install$8–$15Patch and paint walls$8–$12Furniture assembly$8–$14Oil and fluid change$10–$15
• Care and OutdoorsVideo, commercial license
  Pet grooming and feeding$4–$7Gardening$4–$6Car wash and detail$4–$7
• Retail and WarehouseVideo, commercial license
  Shelf restocking$3–$4
• Everyday AudioAudio, research license
  Kitchen soundscapes$1–$3Tool and machine audio$2–$5
• Object ViewsImages, commercial license
  Everyday objects, many angles$1–$3
• Task ProceduresText, research license
  Written procedures$3–$8
• Motion SensorsSensors, research license
  Phone motion logs$2–$6

1. 01Framing
   First person, camera held or worn at eye level, both hands in frame for the whole task.Photos: the object fills most of the frame, on a plain surface.
2. 02Duration
   One task per submission, inside the bounty’s length range. Video runs 5 to 60 seconds.Audio 10 to 120 s. Sensor logs 30 to 300 s. Text 150 to 600 words.
3. 03Resolution
   1080p or better, 30 fps or better.Photos 2000 px on the short edge. Audio 44.1 kHz. Sensors 50 Hz or faster.
4. 04Lighting
   Even enough that hands, tools and the object stay readable from start to finish.Audio: no clipping, no music, and no speech unless the brief asks for it.
5. 05Subjects
   Your own space. No other people, faces, screens or documents in shot.Text: no names, addresses or personal details of anyone, including you.
6. 06Audio
   Optional on video. It is stripped before a clip enters a dataset.On audio bounties the recording is the data, so keep voices out.
7. 07Format
   Straight from the device. No edits, filters, speed changes or overlays.Accepted files and size limits are listed under Data types.
8. 08Review
   Checked by a person against this sheet and the bounty brief.Scored 0 to 100. The score sets where the reward lands in the range.

Sensor logs

Export accelerometer and gyroscope readings from any logging app. Put a header row first, time in milliseconds from the start in the first column, acceleration in m/s² and rotation in rad/s. JSON uploads should be an array of objects with the same keys.

t_ms,ax,ay,az,gx,gy,gz
0,0.118,-9.793,0.342,0.002,-0.011,0.004
10,0.121,-9.801,0.337,0.003,-0.010,0.004
20,0.164,-9.786,0.351,0.019,-0.008,0.006

The file type is checked twice, by extension and by its contents, and every file is fingerprinted with SHA-256 on upload. Resubmitting a file that is already in review or approved is refused as a duplicate.

What a reviewer checks

• The brief. It shows the task the bounty asked for, from start to finish, and only that task.
• The standard. Each of the eight lines in the capture standard, in order.
• Originality. You made it, recently, and it has not been submitted before by you or anyone else.
• Consent. The consent box was ticked for this submission and the file matches what was described.

What gets rejected

• Other people's faces or voices, or anything showing screens, mail, ID cards or addresses.
• Edited, filtered, sped up, stitched or screen-recorded files.
• Footage you did not record yourself, including anything downloaded or AI generated.
• Duplicates and near duplicates of earlier submissions.
• Hands or the object out of frame for most of the task, or too dark or shaky to follow.
• A different task from the one the bounty describes.

Status and turnaround

Review usually happens inside a day. A busy queue can push it further, never silently: the status is always visible in Portfolio.

Scores and rewards

Each approved submission gets a score from 0 to 100, and the score decides where the reward lands inside the bounty's range. A Folding laundry clip that scores 90 in a $2 to $4 bounty earns close to $4. A rejected submission can be recorded again and resubmitted; the reviewer note tells you what to change.

Reward set at review

$3.00T1$6.00T2$12.00T3

Paid in

NVDATSLAAAPLMSFTGOOGLMETAAMZNPLTR

Reward

$6.00

NVDA Chainlink price

You receive

Sent to your address on chain 4663

tokens = reward_usd / price_usd…

• Priced by Chainlink. The price comes from the stock's Chainlink feed on Robinhood Chain when the payout is prepared. When the market is closed the feed holds its last price, and that price is used.
• Checked by a person. A person checks the amount, the token and the address before any payout is signed. Nothing is sent automatically.
• Chosen per submission. You pick the payout stock when you submit. Different submissions can pay out in different stocks.
• Gas is on us. OptiLearn pays the network fee. You need nothing in your wallet to receive.

Payouts go out by hand in batches after approval. Until the transaction is sent, Portfolio shows the submission as approved; once it lands you get the transaction link on the explorer.

Robinhood Chain

Network

Robinhood Chain mainnet, an Arbitrum Orbit L2

Chain ID

4663 (0x1237)

RPC

https://rpc.mainnet.chain.robinhood.com

Explorer

robinhoodchain.blockscout.com

Gas token

ETH

Price source

Chainlink data feeds, read on chain

Payout stocks

Any of these 35 stock and ETF tokens can be chosen as a payout. Each one has a live Chainlink feed on Robinhood Chain, which is what makes a fair conversion possible.

EAS contract

0x4200000000000000000000000000000000000021

Schema registry

0x4200000000000000000000000000000000000020

Explorer

optimism.easscan.org

The schema

Each attestation is made against this schema with no resolver. The recipient is the contributor's address, so every attestation also shows up under that wallet on the explorer. The schema UID is the keccak256 of the schema string, resolver and revocable flag; the Network page shows whether it is registered.

bytes32 submissionId,address contributor,string modality,string bounty,bytes32 contentHash,uint16 reviewScore,string payoutSymbol

What stays off chain

The file itself, your email, your IP hash, the reviewer's notes and the dollar value of the reward are never written to an attestation. Attestations are public and permanent: they cannot be deleted, only revoked, and a revoked attestation stays visible with its revocation time.

How a buyer verifies a file

1. 1
   Find the attestation

   Every item in a licensed dataset carries its attestation UID in the manifest.
2. 2
   Read it on OP Mainnet

   Open it on easscan, or call getAttestation(uid) on the EAS contract directly.
3. 3
   Check who signed it

   The attester must be the OptiLearn attester address listed in your license, and the revocation time must be zero.
4. 4
   Match the fingerprints

   contentHash must equal the item's source_sha256, the hash of the original upload. The file you received hashes to sha256; the two differ only when audio or metadata was stripped before delivery.

import { createPublicClient, http, parseAbi, decodeAbiParameters, parseAbiParameters } from "viem";
import { optimism } from "viem/chains";
import { createHash } from "node:crypto";
import { readFileSync } from "node:fs";

const row = JSON.parse(manifestLine); // one line of manifest.jsonl
const ATTESTER = "0x…"; // OptiLearn attester, listed in your license
const EAS = "0x4200000000000000000000000000000000000021";

const abi = parseAbi([
  "function getAttestation(bytes32 uid) view returns ((bytes32 uid, bytes32 schema, uint64 time, uint64 expirationTime, uint64 revocationTime, bytes32 refUID, address recipient, address attester, bool revocable, bytes data))",
]);
const client = createPublicClient({ chain: optimism, transport: http() });
const att = await client.readContract({ address: EAS, abi, functionName: "getAttestation", args: [row.attestation.uid] });

const [, , , , contentHash] = decodeAbiParameters(
  parseAbiParameters("bytes32, address, string, string, bytes32, uint16, string"),
  att.data,
);
const sha = (f) => "0x" + createHash("sha256").update(readFileSync(f)).digest("hex");

const ok =
  att.revocationTime === 0n &&
  att.attester.toLowerCase() === ATTESTER.toLowerCase() &&
  contentHash === row.source_sha256 && // what was reviewed
  sha(row.file) === row.sha256; // what you received
console.log(ok ? "verified" : "does not match");

• Not how rewards are paid. Contributor rewards are paid in stock tokens. $LEARN is separate.
• Not required. You do not need to hold $LEARN to contribute, get paid or license data.
• No promises. We make no promises about its price, supply or future use. If that changes, it will be written here before it happens.

The same data is available from GET /api/learn. Buying or holding any token carries risk and nothing on this site is investment advice.

Licenses

Each dataset page shows the license it is offered under and a starting price. Volume and exclusivity move the price; the final terms are in the license you sign.

Delivery

• Files. Video as MP4 with the audio track removed, audio as WAV, images in their original format, text as UTF-8 Markdown, sensor logs as CSV.
• Manifest. One JSON line per item with capture details, annotations, the consent record, both file hashes and the attestation UID.
• Transfer. Delivered into your S3 or R2 bucket. Sample records are shared before you sign.
• Refreshes. Datasets that are still collecting ship a new manifest as items clear review. Item IDs never change between refreshes.

{
  "id": "4f1c9a2e-7b3d-4e21-a8c5-1d93e07b6f42",
  "file": "clips/4f1c9a2e.mp4",
  "sha256": "0x3b7e…91c4",
  "source_sha256": "0x9d02…4ae8",
  "contributor": "sha256:5c81e0a2…7f13",
  "task": "fold-laundry",
  "captured_at": "2026-08-14T10:32:00Z",
  "country": "BR",
  "duration_sec": 34.2,
  "resolution": "1920x1080",
  "fps": 30,
  "audio": "stripped",
  "annotations": {
    "task_label": "fold_laundry",
    "hand_visibility": 0.94,
    "segment": [12, 1018],
    "objects": ["t-shirt", "towel", "basket"]
  },
  "consent": { "recorded": true, "version": "2026-06", "at": "2026-08-14T10:32:00Z" },
  "attestation": { "chain_id": 10, "uid": "0x8a4f…e2d0" }
}

• GET/api/tickerChainlink prices for every payout token
• GET/api/learn$LEARN price and contract
• GET/api/configLive site settings
• GET/api/networkRobinhood Chain and OP Mainnet status
• GET/api/statsPlatform totals
• GET/api/submissions?wallet=A wallet's submissions
• POST/api/inquiriesSend a licensing inquiry

• Format. JSON in and out, UTF-8. Responses are not cached unless noted.
• Errors. A non-2xx status with { "error": "bad_request", "message": "…" }. The message is written to be shown to people.
• Limits. Writes are rate limited per network and return 429 with a Retry-After header. Poll reads every 30 seconds or slower; prices do not change faster than that.

curl -s https://optilearn.ai/api/ticker

{
  "quotes": [
    {
      "symbol": "NVDA",
      "name": "NVIDIA",
      "kind": "stock",
      "address": "0xd0601CE157Db5bdC3162BbaC2a2C8aF5320D9EEC",
      "feed": "0x379EC4f7C378F34a1B47E4F3cbeBCbAC3E8E9F15",
      "price": 225.66018707,
      "updatedAt": 1790366165,
      "state": "held"
    }
  ],
  "block": 74069201,
  "gasGwei": 0.020218,
  "at": 1790525258207
}

curl -s https://optilearn.ai/api/learn

{
  "address": null,
  "symbol": "LEARN",
  "priceUsd": null,
  "change24h": null,
  "marketCap": null,
  "liquidityUsd": null,
  "volume24h": null,
  "url": null,
  "source": null,
  "updatedAt": 1790526092927
}

curl -s https://optilearn.ai/api/config

{
  "learnAddress": null,
  "learnPool": null,
  "xHandle": "gkimbwala",
  "updatedAt": null
}

curl -s https://optilearn.ai/api/network

{
  "robinhood": {
    "name": "Robinhood Chain",
    "chainId": 4663,
    "rpc": "https://rpc.mainnet.chain.robinhood.com",
    "explorer": "https://robinhoodchain.blockscout.com",
    "ok": true,
    "block": 74077518,
    "blockTime": 1790526092,
    "blockAgeSec": 1,
    "avgBlockTimeSec": 0.1,
    "gasGwei": 0.020086
  },
  "optimism": {
    "name": "OP Mainnet",
    "chainId": 10,
    "ok": true,
    "block": 157463658,
    "avgBlockTimeSec": 2,
    "gasGwei": 0.001000409,
    "eas": {
      "address": "0x4200000000000000000000000000000000000021",
      "schemaRegistry": "0x4200000000000000000000000000000000000020",
      "schemaUid": null,
      "schemaRegistered": false,
      "attestations": 1318576
    }
  },
  "feeds": { "total": 38, "live": 0, "held": 38 },
  "at": 1790526093796
}

curl -s https://optilearn.ai/api/stats

{
  "submissions": 42,
  "approved": 31,
  "contributors": 17,
  "paidUsd": 118.5,
  "byBounty": {
    "fold-laundry": { "submitted": 12, "approved": 9 }
  }
}

curl -s "https://optilearn.ai/api/submissions?wallet=0x7a3f…c21e"

{
  "submissions": [
    {
      "id": "4f1c9a2e-7b3d-4e21-a8c5-1d93e07b6f42",
      "wallet": "0x7a3f…c21e",
      "bountySlug": "fold-laundry",
      "modality": "video",
      "title": "Folding laundry",
      "fileName": "IMG_4821.MOV",
      "fileSize": 48213771,
      "mime": "video/quicktime",
      "durationSec": 34.2,
      "captureSource": "upload",
      "payoutSymbol": "NVDA",
      "status": "paid",
      "reviewScore": 88,
      "rewardUsd": 3.75,
      "reviewerNote": null,
      "payoutAmount": 0.016618,
      "payoutPriceUsd": 225.66,
      "payoutTx": "0x51be…09aa",
      "attestationUid": "0x8a4f…e2d0",
      "createdAt": "2026-09-26T14:02:11.000Z",
      "reviewedAt": "2026-09-26T19:40:05.000Z",
      "paidAt": "2026-09-27T09:12:48.000Z"
    }
  ]
}

curl -s -X POST https://optilearn.ai/api/inquiries \
  -H "Content-Type: application/json" \
  -d '{"name":"Ada Lovelace","email":"ada@lab.ai","org":"Analytical Robotics","useCase":"Bimanual folding policy, 2k clips to start","datasets":["household-manipulation"],"budget":"10k-50k"}'

{
  "ok": true,
  "id": "b2e0d6c4-3f1a-4c7e-9a52-6e8d1f0c7a31"
}

Still stuck? Contact us.