Performance Weight Formula

Turning raw metrics into ALPHA

Every five minutes the FlameWire validator client collects three raw metrics for each miner, converts them to a single weight (0 – 1), then once per 25‑minute cycle publishes that weight vector on‑chain. A miner’s share of the 2  952 ALPHA/day emission pool is simply:

miner_reward = total_daily_emission × weightminer / Σweightsall_miners

---

1. Metrics captured per probe round

Metric	How it’s measured	Units
Throughput	Successful responses per second during a synthetic burst of N requests	req/s
Latency (p95)	95th‑percentile round‑trip time over the probe window	milliseconds
Availability	% of probes that returned a non‑error payload	percent

2. Gate on minimum throughput

If throughput < THRESHOLD the miner is disqualified for the current round and gets weight = 0. Default threshold: 50 req/s for Ethereum archive nodes (chain‑specific values are configurable by subnet governance).

3. Normalise latency and availability

For miners that pass the throughput gate:

iniCopyEditnorm_latency      = 1 / p95_latency_ms
norm_availability = availability_percent / 100

All norm_latency values are then min‑max scaled so the best miner in the round scores 1.0 and the worst > 0. (Availability is already 0 – 1, so no extra scaling.)

4. Blend into a single score

iniCopyEditweight_raw = 0.5 × norm_latency + 0.5 × norm_availability

The 50 / 50 split is FlameWire’s launch setting; governance can vote to adjust this constant.

5. Time smoothing

The validator averages the last five rounds (25 min of data) before publishing the weight vector. This protects miners from a single transient spike but still allows corrective down‑weights inside an hour.

6. Cross‑validator aggregation

On‑chain, multiple validator submissions are combined by median for each miner UID. Outliers are ignored, so publishing honest, high‑fidelity weights is the best strategy for every validator.

7. Reward settlement

At the end of every Bittensor block (≈ 12 s) the subnet contract:

1. Sums the latest median weights.

2. Pays each miner their block‑level ALPHA emission share.

3. Pays stakers and team split from the same block emission according to tokenomics (41 % miners, 41 % stakers, 18 % team).

---

Key take‑aways for operators

• Keep throughput above threshold—it is an on/off gate.

• Reduce p95 latency—use NVMe and fast networks.

• Maximise availability—watch health checks and avoid firewall mis‑configs.

• Weights update roughly twice per epoch (60 min), so improvements start paying out within the hour.