Data From 2025-06-02 To 2025-06-08

In this week’s issue, in addition to reviewing individual miners’ performance and modeling strategies via forecast summary statistics, we also examine axon-level performance among subnet participants to assess how groups of miners perform collectively.

Important Update: Starting on June 4, 2025, at 4:00 PM UTC, the parameters used for calculating EMA scores were updated:

• Window length increased from 7 to 10 days

• Half-life increased from 3.5 to 5 days

• Beta parameter decreased from -0.0475 to -0.1

The axon analysis offers an initial overview of the impact of these changes on the rewards distribution.

Figure 1 shows the hourly log returns and volatility observed throughout the week. It’s clear that ETH exhibited higher volatility than BTC, and that the two assets are strongly correlated, with volatility spikes occurring at similar times.

This week’s analysis focuses on six miners:

• The top three at the start of the week: Miners 32, 234, and 200

• The top three at the end of the week: Miners 170, 221, and 90

As usual, we evaluate their performance using two key metrics:

• CRPS (Continuous Ranked Probability Score) for both BTC and ETH forecasts

• Leaderboard Scores

Lower CRPS values indicate better predictive accuracy.

The competition was extremely tight this week, as reflected in the CRPS values for both BTC and ETH. The scores were very close, making it hard to clearly distinguish between miners.

Leaderboard scores are an exponentially weighted average of past CRPS values over a 10-day window (half-life: 5 days). Lower scores indicate better performance.

The leaderboard also reflected the week’s close competition, with frequent leadership shifts until the ranking began stabilizing from Friday, June 6, through the weekend. Notably, the six miners finished the week within a narrow range of smoothed scores, between 91 and 97.

This section analyzes summary statistics—volatility, kurtosis, and skewness—to detect potential strategic differences among top miners for both BTC and ETH.

Summary Statistic Definitions:

• Volatility: Variability in forecasted returns

• Kurtosis: Measures the "tailedness" of the distribution—how likely extreme events are

• Skewness: Indicates the asymmetry of return distributions (0 = symmetric)

Figure 4 shows these distributions across the week for the six miners.

The modeling strategies appear to be quite similar, with no significant differences in weekly volatility, kurtosis, or skewness. This helps explain the tight CRPS values.

All miners who submitted forecasts were ranked by total weekly rewards. Figure 5 shows how summary statistics vary across different ranking tiers.

Compared to the previous week, lower-ranked miners struggled more to keep up with top miners, particularly regarding BTC volatility modeling. As in past weeks, top miners showed a clear edge in modeling ETH volatility and kurtosis for both assets.

To evaluate miners’ performance in groupings (axons), we analyzed weekly data on:

• Axon size over time

• Average and total rewards per axon

• Axon-level summary statistics (for BTC forecasts)

Figure 6 shows axon size evolution throughout the week, with a vertical red dashed line indicating when the EMA parameter update occurred.

Notably, axon 116.202.53.142 and axon 95.216.99.113 had the most miners (49 and 33 at the end of the week, respectively). But does a larger size guarantee more rewards?

Figure 7 explores this by showing average and total daily rewards for the top 10 axons by average rewards. The EMA parameter change is again marked by a vertical line.

1. The EMA update steepened the rewards curve (as intended), making performance differences more pronounced (as noticeable from the top graph)

2. Some large axons (e.g., 116.202.53.142 and 95.216.99.113) saw declining total rewards post-update, despite their size

3. Smaller axons with better modeling strategies (e.g., axon 35.77.6.189) significantly outperformed larger ones in total rewards

Figure 8 compares the top 10 axons (by average rewards) on their BTC modeling characteristics—intraday average volatility and its variability, as well as intraday average kurtosis.

There is a clear distinction in modeling quality between top- and lower-ranked axons. High-reward axons were able to model BTC volatility effectively and adjust it throughout the week. Similarly, strong performance in modeling kurtosis contributed to their rewards. By contrast, some axons appear to neglect one or both of these aspects altogether.

This was a highly competitive week, with tight CRPS values and similar modeling strategies across top miners. This is a positive sign for the subnet, suggesting that miners are refining their models and focusing on nuanced aspects of BTC and ETH distributions.

The axon-level analysis reveals that the EMA parameter update improved reward fairness: axons producing higher-quality models were more strongly rewarded, while large axons relying solely on size saw diminished returns. This encourages skill-based clustering rather than strategy based purely on scale.