# Synth Subnet - Inside Synth’s Accuracy Surge

By [Synth](https://paragraph.com/@synthdata) · 2025-08-04

---

Introduction
------------

This report provides a focused assessment of recent progress in the Synth subnet, where miners forecast BTC’s return distribution every day. Over the past month, the subnet has seen a major shift: miners prioritizing volatility accuracy have rapidly climbed the leaderboards, outperforming more complex models that previously held top positions.

We analyze the performance of these top miners across multiple metrics, volatility, intraday variation, and kurtosis, using data from June 15 to July 27, 2025. While recent gains in volatility prediction mark a significant milestone for Synth, they also reveal a new competitive phase: can miners now combine precision in volatility with deeper modeling of the return distribution? This report highlights where things stand, what’s improved, and where the next edge may lie.

Analysis Setup
--------------

The primary focus of this analysis is on **volatility forecasts**, which are the core of the subnet’s design. We evaluate forecasts produced by the **Top 10 miners by leaderboard** as well as the **Top 10 miners by meta-leaderboard** (i.e., total rewards accumulated over the past two weeks). Since the composition of these top miners changes over time, the aim is to track how forecast quality has evolved week by week.

We concentrate on **24-hour volatility** forecasts (computed using all 5-minute return predictions), comparing them against realized BTC volatility over the same period. We also analyze shorter-term volatility forecasts (1-hour, 3-hour, 6-hour, 12-hour, and 18-hour) to assess consistency across time horizons.

Recognizing that volatility is not constant throughout the day, we also explore how miners vary their short-term volatility estimates relative to the 24-hour horizon. Additionally, we assess **kurtosis**, a metric reflecting the probability miners assign to extreme or unexpected outcomes.

To provide a more detailed picture, we supplement aggregated Top 10 statistics with **per-axon data**, helping identify which axons may be driving changes in the overall forecast quality.

The analysis covers the period from **June 15th, 2025 to July 27th, 2025**.

Volatility Analysis By Top Miners
---------------------------------

**Figure 1** shows the 24-hour volatility forecasts from the Top 10 miners by both leaderboard and meta-leaderboard, alongside realized BTC volatility.

![](https://storage.googleapis.com/papyrus_images/26b7662771eeed87bf69f17fc75aeebf4c4e96747b4c1c273f988eaf9c1e6d86.jpg)

From the figure, we observe that the gap between forecasted and realized BTC volatility has **narrowed over time, with an almost perfect match in the final weeks of the period**. Additionally, whereas the two groups of miners diverged in late June, they now appear to be in strong agreement.

**Figures 2 and 3** report the absolute and relative differences between forecasted and realized volatility, along with 7-day moving averages.

![](https://storage.googleapis.com/papyrus_images/c3dcd0f31e844a9ddb0461c1e1587464d6e73a2968d238f6f23eb563629f667c.png)

![](https://storage.googleapis.com/papyrus_images/03f5ed43cfc2e2693d68ab9a9881be09d4a8582fd5610edf1deec557125c71ec.png)

These figures confirm the trend: After a temporary increase in errors from late June to early July, both absolute and relative differences have dropped sharply. In the last two weeks of the sample, the relative difference fell below 20%—to **18.47%** (leaderboard) and **17.04%** (meta-leaderboard)—more than halving the values from the prior two weeks (**43.13%** and **46.48%**, respectively). This improvement represents a major milestone for the Synth subnet.

**Table 1** breaks down volatility forecast performance across multiple horizons and periods.

![](https://storage.googleapis.com/papyrus_images/001385e9fc342c1a35b6fe464929473e81cdabab03f8608402b7ea1861dd7265.png)

Key takeaways from the table:

*   Shorter-term volatility forecasts tend to be less accurate, with relative differences decreasing as the time horizon increases.
    
*   Both top miner groups showed significant improvements across all horizons in the last two weeks, following a dip in late June and early July.
    

Volatility Analysis By Axons
----------------------------

To better understand these improvements, we analyzed the individual axons to which the top miners belong. **Figures 4 and 5** show how miners from each axon moved in and out of the Top 10 (by total rewards and by leaderboard).

![](https://storage.googleapis.com/papyrus_images/21996e4629d12bdf7557826eccfaf13fa93d8023622ef08438a6a7f70b5f7ba6.png)

![](https://storage.googleapis.com/papyrus_images/f8b99153514751b1cdfae513a962d5c9942d9c65a93898fb840597115590ab7c.png)

We identify two main groups of axons:

*   **Previously leading axons (through end of June)**: 35.77.6.189, 18.183.47.137, 3.112.97.164, 186.233.184.223
    
*   **Newly leading axons (late July)**: 116.202.53.142, 95.216.99.113
    

**Figure 6** shows the average forecasted volatility from each group.

![](https://storage.googleapis.com/papyrus_images/f558198d837afd50753c787829ec2a72dcfe887a4f3c73c53a50e2e8b351e51c.png)

A clear pattern emerges: Previously leading axons consistently forecasted **higher volatility**, which initially seemed advantageous. However, as BTC volatility declined, this strategy failed to adapt, causing these axons to fall in the rankings.

By contrast, miners from the newly leading axons forecasted **lower volatility**, more aligned with realized BTC movements. Their apparent strategy of focusing on accurate volatility estimation has clearly paid off in leaderboard performance.

Intraday Volatility Analysis
----------------------------

Intraday volatility variation is a critical aspect in probabilistic forecasting, particularly in financial markets where volatility often fluctuates between day and night.

To evaluate whether the new top miners are adapting to **intraday dynamics**, or simply improving **average 24-hour volatility forecasts**—we conducted the following:

*   Calculated normalized short-term volatilities (1h, 3h, 6h, 12h, 18h) and averaged them.
    
*   Compared these to the normalized 24-hour volatility.
    
*   Computed the ratio of the two metrics: A ratio near 1 suggests **constant volatility**; a ratio far from 1 suggests **adaptive intraday modeling**.
    

**Figure 7** shows this ratio over time.

![](https://storage.googleapis.com/papyrus_images/c032ec69c6ab56bfcabb3e43a533d3989958fe43cc3ca75b0b9dfaf5d9278a2d.png)

Interestingly, while miners initially varied their intraday volatilities effectively, this behavior diminished over time. Ratios approached 1 in late July, indicating a shift toward constant-volatility paths.

**Figure 8** examines these ratios by axon group.

![](https://storage.googleapis.com/papyrus_images/48fa238743c5b1965fe875d1dd5e68ef30571282c784131fae7932cd10a5067b.png)

The pattern is revealing: **Previously dominant axons** made significant efforts to model intraday variation. In contrast, **newer top axons** appear to be submitting forecasts with constant volatility across time. While this simpler strategy has improved long-term accuracy, it reveals a potential **weakness**, namely lack of intraday adaptability, which could be exploited by competitors in the future.

Kurtosis Analysis
-----------------

Finally, we examine the **kurtosis** of forecast distributions, considered an important measure of tail risk and sensitivity to rare events.

For each miner, we averaged the kurtosis across 100 simulated paths. We omit BTC realized kurtosis from the plots, as it is a single-path estimate and not directly comparable to the multi-path averages from simulations. A comparison with BTC will be the subject of a future report.

**Figure 9** displays average kurtosis over time for top miners by both leaderboards.

![](https://storage.googleapis.com/papyrus_images/7c6880afda41dc26de149b1a8beb965881ae6b92bf199da60c5450a9bd409c12.png)

Similar to the intraday volatility findings, kurtosis estimates were higher and more dynamic in early weeks, but have since flattened. By the end of the sample, top miners' kurtosis estimates dropped near zero, particularly for those by leaderboard.

**Figure 10** shows the same analysis split by axon.

![](https://storage.googleapis.com/papyrus_images/a107ee4d6a0971035a871934bedb77ed8168202ee26ed0862b8cf9555b578d8c.png)

Again, we find that **newer leading axons** produce flatter distributions with little to no kurtosis. **Previously leading axons**, despite declining in rankings, still make active efforts to estimate and adapt kurtosis.

Takeaways and Future Challenges
-------------------------------

The past month on Synth has seen a notable **shift in leaderboard dynamics**. A new group of miners—likely relying on **simpler models** narrowly focused on volatility—has climbed to the top by steadily improving their accuracy on this one key metric. Their forecasts are now consistently closer to BTC’s realized volatility, especially over the past couple of weeks, **with errors dropped below 20%—18.47% (leaderboard) and 17.04% (meta-leaderboard), less than half of the previous period’s values, marking a major milestone for Synth.**

Meanwhile, the **previously dominant miners**, who likely used **more complex models**, initially succeeded due to their broader probabilistic modeling, capturing volatility, intraday variation, and kurtosis. However, their failure to adapt to recent moves in BTC volatility caused them to fall behind.

This shift suggests that the Synth competition structure is driving real adaptation: simpler strategies have temporarily outpaced more complex models that were previously dominating. But the big questions are now coming into focus:

*   Will the more sophisticated miners rise again, inspired to improve their models and regain the top spot?
    
*   Will current leaders evolve to include better modeling of intraday variation and tail risk?
    
*   Or will a **new axon** emerge that combines the strengths of both approaches—accurate volatility forecasts **and** rich distribution modeling?
    

The coming weeks will be an important test for the subnet. There’s still clear room for improvement, and the miners who manage to balance precision in volatility with attention to deeper characteristics of BTC’s return distribution are likely to be the ones who rise next. Whether it’s the current leaders who evolve, previous top performers making a comeback, or a new group altogether, the competition is far from over.

In short: the subnet is improving, but there’s still plenty of value to be captured, both in leaderboard positions and in rewards. That makes now a very good time for skilled new quants to enter the game.

---

*Originally published on [Synth](https://paragraph.com/@synthdata/synth-subnet-inside-synth-s-accuracy-surge)*