Motivation

After reading both h4wk's When do transactions fail dashboard and Kiana's Flow vs L1s dashboard I wanted to understand where the massive increase in failed transactions came from.

As can be seen in Kiana's dashboard, Flow maintains a high success rate well over 90% for some days but sinked to 60% levels on May 13th and plummeted below 20% through a downtrend phase in late May.

h4wk's throughout analysis on the correlation between transaction volume and failure rate on different time frames concludes that the transaction failure is unrelated to transaction volume.

But what is happening then? Why is the Flow drifting apart?

Flow: innovation from crypto cats

The blockchain company Dapper Labs developed Cryptokitties, a NFT cat-collectible game on Ethereum which was one of the first of its kind and a clear use of blockchain technology with a use case not related to digital currency. In 2017, there was a major spike in Ethereum's transactions by Cryptokitties players which caused the network to be clogged. This revealed the need to level Ethereum up to be ready for massive adoption.

But Dapper Team had other plans and after some years researching and partnering with big players in and out of the cryptoindustry announced the launch of a new blockchain designed with mainstream consumers in mind - a network capable of scaling to the needs of games with massive player base and providing the tools for developers to create decentralized apps (dApps) with its own resource-oriented programming language Cadence. With it, upgradeable smart contracts can be created easily.

The blockchain became operational in May 2020 and in June 2020 the first application went live - NBA Top Shot, an officially-license digital collectible marketplace.

Analysis methodology

The flow.core.fact_transactions table is queried to obtain the number of daily transactions and the success rate as successful txs / total txs.

Then, all addresses initiating a transaction (proposer) are separated in two categories: "human" and "bot" addresses. The criteria is to have done more than 200 transactions in at least one day in the time period studied. Cumulative total and failed transactions are calculated for each category as well as the daily success rate.

Currently, Flipside Flow core tables have data from May 9th. Queries are set to refresh daily, therefore graphs will be updated but text mentions will convey the value on June 9th when the dashboard was published.

Results

Total transactions

The graphs below show the number of total transactions and daily transaction success rate. The second graphs highlights days over and under 85% success rate. The downtrend phase happened between May 25th and May 31st, reaching the absolute low of the series on May 29th (around 17% success). Absolute high reached is almost 97% on May 21st.

Bot vs humans

The graph below shows daily success rate by bots and humans. With the current definition, the rate for humans remains relatively constant above 95% while the rate for bots follows the trend for the total transactions, with a similar downtrend phase starting one day before on May 24th and reaching an absolute low of 11% - barely 1 out of 10 transactions from "bots" went through on May 30th.

In terms of cumulative values, bots made 13.2M transactions of which 4.6M failed (average 65% success) and humans made 6.9M transactions of which 83k failed (average 98.8% success).

Between May 9th and June 9th, the Flow blockchain processed more than 20M transactions of which around 4.6M failed (average 23% success)

Conclusions

The massive increase in failed transactions was caused by "bots", addresses with at least once perform more than 200 transactions per day.

"Bots" is just a definition based on the number of transactions an address performed. An analysis on this addresses profiles is necessary to find out what these mean bots are doing with the flow!

Thanks to Kiana's and h4wk's for their awesome submissions!

Analysis done by Kask_CEA powered by Flipside Crypto