How we choose the best Edge Device to run AI Tasks?

• We don’t want one machine to do all the jobs.

• We do want fast, reliable, and trustworthy machines to handle important stuff.

1️⃣ Step 1: Finding the Best Machines via “Weighted Scoring”

• Stake: How much $TOPS a machine has staked (higher = more trusted).
• Uptime: How often it’s online and ready to work.
• Latency: How fast it responds (lower latency = better).

Why normalize?

Machines have very different numbers (like stake could be 1000, uptime could be 97%, latency could be 80 ms). Normalization squashes all those numbers into a 0–1 scale so they’re fair to compare. And for latency, we do (1 – normalized_latency) so lower latency = higher score.

2️⃣ Step 2: Selection Strategy: Top K + Probabilistic Sampling

1. Pick the Top K machines with the best scores.
2. Give them a chance based on their score. (Higher score = higher chance.)
3. Pick one at random using those chances.

3️⃣ Step 3: Scoring Machines by Their Work

• More complex tasks are worth more points.
• Only successful tasks get counted.
• Your bonus can’t go above +1.0, no matter how good you are (so it’s fair for everyone).

AI_TASK_SCORE = 1 + min(1.0, (Σ(task_weight_i × success_i) / TASK_WEIGHT_NORM))

• task_weight: how important the task is.
• success: 1 if done correctly, 0 if failed.
• TASK_WEIGHT_NORM: a cap to stop scores from getting too big.

• Task A (weight 2) → ✅
• Task B (weight 3) → ✅
• Task C (weight 3) → ❌
• Task D (weight 1) → ✅

→ Total points = (2×1) + (3×1) + (3×0) + (1×1) = 6

→ AI_TASK_SCORE = 1 + min(1.0, 6 / 10) = 1 + 0.6 = 1.6

We score machines on both performance and capabilities, then select from the top performers with a bit of randomness. This keeps the network fast, fair, and healthy.