A Look Into Score (SN44)’s Daily Audit Trail for Private Tracks

Score (SN44) rolled out a full auditability layer for its private tracks, tackling one of the biggest verification gaps in Bittensor subnets.

Every 24 hours, the top performer on each private track element is automatically selected, with a challenge from their latest commit published alongside the miner’s response and ground truth.

The selection is random, the code is open source, and the complete history remains available through a public endpoint for anyone to inspect. This gives the ecosystem a way to independently verify responses, confirm ground truth, reproduce scoring, and validate that reported top-miner performance reflects real results.

How the Audit Works

The mechanism is designed to be tamper-proof and independently reproducible.

1. Daily selection: Every 24 hours, the current top performer on each private track element gets picked automatically.

2. Random challenge sampling: One challenge is drawn at random from the winner’s latest commit block, so no sample can be hand-picked for favorable results.

3. Full publication: The challenge, the miner’s JSON response, and the ground truth are all published together on the console.

4. Open-source generator. The code producing the audit output lives publicly at GitHub, so anyone can inspect exactly how selection happens.

5. Full history preserved: Every published sample is archived.

Part two of the rollout brings the audit directly into the console. Opening any private track element now surfaces the published challenge, the top miner’s response, and the ground truth in one view. Live example available at Score’s Console.

What Anyone Can Verify

The audit surface answers four specific questions that used to require trust.

1. Response matches challenge: The published miner output corresponds to the exact challenge selected.

2. Ground truth is consistent: The reference answer aligns with what the challenge demands.

3. Scoring reproduces: Anyone can run the scoring locally and confirm the number.

4. Top-miner performance is real: The reported leaderboard reflects actual results on genuine challenges rather than assumed performance.

Over time this becomes a public benchmark. Miners can measure their own models against top solutions using real ground truth, learn what private tracks demand, and calibrate their submissions accordingly.

Don’t Trust, Verify

Private tracks have always been a difficult tradeoff in subnet design, protecting evaluations from gaming while making leaderboards harder to verify externally. Score’s approach keeps the tracks private during evaluation but creates a daily, verifiable view of how top miners are actually performing.

The code and full audit trail are public, while miners can benchmark themselves against leaders using real ground truth. It shifts the system from relying on trust to providing independent verification, an important step as open subnets move toward commercial applications.