Subnet 13: Incentive Mechanism
Subnet 13 incentive overview
Last updated
Subnet 13 incentive overview
Last updated
Subnet 13's incentive mechanism rewards miners for providing successfully validated, non-duplicate, fresh data. There are two main parts that make up a miner’s total scaled score: their raw score and their credibility.
To maximise their raw scores, miners either scrape data according to their own preferences (with a value equal to our default scale factor, which is currently set 0.3), or scrape data from dynamically specified labels that validators can submit with the .
Validator voting power is proportional to the amount they have currently staked on subnet 13, and indicates the amount the custom label will be incentivised for. From this incentivised list, miners can choose labels to scrape and a receive significantly higher reward for returning data with said labels. This mechanism allows validators with significant subnet stake to leverage their bandwidth and request large amounts of fresh data for use in analytics, training, and more.
Miners will also be rewarded for their uploads to our data hub, currently set to HuggingFace (HF), which contributes to their raw score. These uploads enable Subnet 13 to have reliable, reputable and publicly accessible open source data. HF rewards, which are programmed , comprise an up-to 10 million score boost based on HF validation results. HF validation works similarly to local validation, in which miners are given a reward based on the percentage of randomly chosen HF posts that are successfully re-scraped and validated. See the current for the latest status on planned changes.
To maximise their credibility, miners must provide data that matches a real-time scrape. This is determined by an exponential moving average based on the percentage of successfully validated bytes. The scaled score is heavily reliant on credibility– raw scores are multiplied by credibility of 2.5, so this is key to miner success on the network. Credibility is such an important mechanism on the network because unreliable data is worthless. Keeping miners accountable for failed validation keeps data on SN13 fresh and trustworthy.
Subnet 13’s reward mechanism is based on the source, type and age of scraped data. This is separated into 2 parts, the scoring system and the credibility system.
Where:
Σ is the sum over all data entity buckets
DSW is the Data Source Weight (weight from DataSourceDesirability)
LSF is the Label Scale Factor (from label_scale_factors or default_scale_factor)
ASF is the Age Scale Factor (calculated based on data age)
SB is the Scorable Bytes
C is the Miner Credibility
The Data Source Weights sum to 1 and currently include Reddit and X. Label Scale Factors determine the weight placed on certain subreddits and hashtags for Reddit and X respectively, ranging anywhere from 0-1. Age Scale Factors depend on content age, with data over 30 days old valued at 0.
This total is then multiplied by the miner’s credibility to the power of 2.5, the credibility is calculated with an exponential moving average. Credibility is scored from 0-1 (worst to best). Good credibility is obtained by providing data that is successfully validated. This means the miner's uploads has the right labels, matching size, etc. to the original, and that the original post hasn’t been deleted or misrepresented.
This is calculated as:
Where:
C_new is New credibility
C_old is Old credibility
α is EMA alpha (set to 0.15 in the code)
V is Validation score (percentage of valid bytes in the latest validation)
For more information, see our , and our .
