Proctor for developers
Everything the site shows comes from one public contract on NEAR. Run exams from your own code, read every score as JSON, and put a live, provable score on your own site.
Your console
Not connectedConnect a NEAR testnet wallet to see the exams you've run, the question sets you've written and what they've earned.
Make an AI hold a licence before your contract lets it act
If your contract lets an AI move money, ask Proctor first. One cross-contract call to licence(model, skill) returns whether the model passed the sealed money exam, its score, and when the licence lapses. Our demo vault does exactly this.
Promise::new("app.proctor-91796.testnet".parse().unwrap())
.function_call("licence".to_string(),
json!({ "model": model, "skill": "money" }).to_string().into_bytes(),
NearToken::from_yoctonear(0), Gas::from_tgas(20))
.then(Self::ext(env::current_account_id()).on_licence(model, amount))
// in on_licence: pay only if licence["licensed"] == true
curl "https://useproctor.vercel.app/api/licence?model=gpt-5-nano&skill=money"
Embed a live score badge
The badge is drawn from NEAR each time it loads, so the number can't be inflated. Link it to the scoreboard so anyone can check it.
Read every score as data
One request returns every model's score on sealed questions and on published ones, recomputed from the certificates on NEAR.
curl https://useproctor.vercel.app/api/scores
Loading
Don't want to trust our server? Read the contract directly. This returns the raw certificates:
near contract call-function as-read-only app.proctor-91796.testnet \
get_certificates json-args '{"from":0,"limit":20}' network-config testnet now
Run an exam from your own code
One call asks one sealed question. Attach the writer's fee, the model fee and up to 0.025 NEAR for the secure hardware; what isn't used comes back. The certificate lands about a minute later.
near contract call-function as-transaction app.proctor-91796.testnet request_run \
json-args '{"set_id":"hard-a","model":"Qwen/Qwen3.8-27B","start":0,"count":1}' \
prepaid-gas '300 Tgas' attached-deposit '0.027 NEAR' \
sign-as YOUR_ACCOUNT.testnet network-config testnet sign-with-keychain send
import { Account } from "@near-js/accounts";
import { JsonRpcProvider } from "@near-js/providers";
import { KeyPairSigner } from "@near-js/signers";
const provider = new JsonRpcProvider({ url: "https://rpc.testnet.fastnear.com" });
const me = new Account("YOUR_ACCOUNT.testnet", provider, KeyPairSigner.fromSecretKey(process.env.NEAR_KEY));
// ask question 0 of the harder sealed set
const out = await me.callFunction({
contractId: "app.proctor-91796.testnet",
methodName: "request_run",
args: { set_id: "hard-a", model: "Qwen/Qwen3.8-27B", start: 0, count: 1 },
gas: 300_000_000_000_000n,
deposit: 27_000_000_000_000_000_000_000n, // 0.027 NEAR
});
// the certificate number is in the logs: "EVENT proctor certificate 412 ..."
| Model id | Who sees the questions | Model fee per question |
|---|---|---|
| Set id | Questions | Writer's fee | Seen by |
|---|
Contract reference
app.proctor-91796.testnet on NEAR testnet. Source on GitHub.
| Method | What it does |
|---|---|
request_run(set_id, model, start, count) | Pays for questions and sends them to the marking program in secure hardware. Payable. |
commit_set(set_id, sha256, size, authors, price_per_question, retire_after) | Puts a new sealed set's fingerprint on record, with the attached NEAR as the writer's stake. |
reveal(set_id, json) | Publishes a retired set. The contract checks it matches the fingerprint. |
challenge(set_id, index, reason) | Disputes an answer in a published set, with a bond of at least 0.1 NEAR. |
licence(model, skill) | Whether an AI holds a licence: passed, score, questions answered, expiry, and whether its maker saw the exam. |
fund_wallet(model), list_wallets | Put NEAR in an AI's wallet for the money exam; see what each has left and lost to scams. |
get_certificates(from, limit) | Every recorded result: model, set, marks, block, marking build, signed receipts. |
list_sets, get_set(set_id) | Question sets with fingerprints, fees, stake and which AI makers have seen them. |
list_public_models, allowance(account_id) | Models anyone may run, their fees, and how many questions an account has left today. |