Reflecting on Athens, the first self-amendment of Tezos

Nostalgic Athens. Photo by Cristina Gottardi.

This post provides reflections on the Tezos community’s experience with Athens (so far), the first use of the Tezos amendment process. See here for a refresher or Google “Amending Tezos”.

The Promotion Vote Period, the final voting period of Tezos Athens, recently achieved its 81.39% quorum and only a few hours remain at writing before activation of the new Athens protocol after 12:40am May 30th UTC (Update on 5/30: Athens has activated!).

Until Athens, “self-amendment” in Tezos referred to an abstract (yet formal, hard-coded) mechanism inspired by models of human behavior. Our capacity to predict the performance of the Tezos governance mechanism has been severely limited because, as eloquently described once, “blockchain governance doesn’t fit squarely into any existing model.”

To describe it more bluntly: there was no obvious, replicable way to trial the Tezos governance mechanism as a social process without actually injecting proposals and voting on them.

Baking and governance: a global game

As any participant or attentive observer of Tezos would know, bakers are the decision-makers, the voters, in the Tezos amendment process.

As both consensus participants and protocol decision-makers, their number and total stake have grown significantly since Tezos’ early cycles (check out cycle 10, for example). In the recent snapshot for cycle 115, there were ~466 bakers (~110 public) distributed across at least 34 countries representing thousands of delegators and 562 million ꜩ, or ~80% of the activated supply.

Tezos nodes exist in New Zealand, Azerbaijan, India, Brazil, and even on Raspberry Pi’s within the walls of the French Central Bank.

Athens isn’t just about demonstrating that the Tezos amendment process can work. It’s to demonstrate that Tezos can take a first step towards becoming open, internet-native infrastructure for social and economic coordination that governs and upgrades itself.

I. Recapping Athens to date

Proposal Period

The amendment process around Athens formally began in February with two Nomadic proposals. The first, Athens A, increased the gas limit (doubling the number of computation steps in each block) and reduced roll size, the unit of account for Tezos proof-of-stake, from 10,000 ꜩ to 8,000 ꜩ. The second, Athens B, merely increased the gas limit.

Both proposals included a symbolic 100 ꜩ invoice to provide a round of beers to the Nomadic Labs team.

  • Athens A: to reduce roll size from 10,000 ꜩ from 8,000 ꜩ AND increase the gas limit (doubling number of computation steps in each block)
  • Athens B: increase the gas limit

170 bakers representing 25,855 rolls (258,550,000 ꜩ or 49% of the total rolls), directly participated in the Proposal Period. Note, the Tezos Foundation explicitly chose not to vote during this period.

The result of the Proposal Period vote was a large margin of the stake favoring Athens A (71%) and 29% favoring Athens B, with Athens A advancing to a Yes or No vote in the subsequent Exploration Period.

Proposal Period Results:

25,855 rolls (170 bakers) voted

  • ~49% participation of total rolls
  • ~70% participation by non-Tezos Foundation rolls

Approval vote results (no quorum):

  • Athens A: 71% (102 bakers)
  • Athens B: 29% (68 bakers)

Result: Athens A advanced to Exploration Period

Exploration Period

In the Exploration Period, bakers voted to advance and test Athens A. This period saw increased participation, as 194 bakers representing 46,390 rolls (463,900,000 ꜩ or ~87% of the activated supply) voted.

Unlike the Proposal Period, the Exploration Period required an 80% quorum and 80% supermajority. Athens A cleared both of these hurdles by a large margin. Nearly 87% of all stake voted and 99.96% elected to advance Athens A to the Testing Period.

The Tezos Foundation chose not to vote “Yay” or “Nay” and instead voted “Pass” in the Exploration Period, contributing to the quorum.

Exploration Period Results:

Yay/Nay/Pass vote:

  • Yay: 57.86% (178 bakers)
  • Nay: .02% (3 bakers)
  • Pass: 42.12% (13 bakers)

199 bakers voted with 46,390 rolls, representing 463,900,000 ꜩ (~$742,000,000)

  • Participation (needed 80% quorum): 86.98%
  • Supermajority (needed 80%): 99.88%

Result: Athens A advanced to Testing Period

Testing Period

As currently implemented, the Tezos Testing Period involves a 48 hour test-fork at the start of the period. As described in “Amending Athens”, the general purpose of the testnet fork is to ensure the upgrade does not corrupt the context and that the network would continue making valid state transitions post-upgrade.

In the aptly named “Athens on the Test Chain”, Nomadic notes two advantages of the test chain over the regular testnets like the Alphanet and Zeronet:

  1. It better demonstrates the effects of migration on a real context, which is significantly more complex than a testnet context
  2. Users can actually test operations using their real accounts and infrastructure, such as remote signers, hardware wallets, or complex systems built on top of Tezos

As a parallel fork of the main chain, the test chain (chain ID NetXbQxxRBRcAZ5) required willing bakers to validate blocks on two chains. Because running a Tezos node has historically required a bit of storage and this was the first testing period, Nomadic recommended bakers to optionally enable it via a secondary deployment which is booted from a snapshot to save disk space. For more on snapshots and history modes, see here.

Because the test chain was optional, only a small number of bakers participated and block times in the test chain were a bit slower than on Mainnet (given baking/endorsement rights that went unfilled by non-participating bakers). That said, it served to demonstrate secure migration of the context.

Future versions are likely to see greater participation by bakers as the requirements for securely running a full node continue to fall with the adoption of snapshotting and the upgrade to the Irmin v2 backend.

Promotion Period (ongoing, with less than 7 days remaining)

Based on the high turnout (~87% of all rolls) in the Exploration Period, the quorum adjusted upwards to 81.39% for the Promotion Period (the quorum adjusts on an exponential moving average).

In the Promotion Period, 216 bakers representing 47,031 rolls (~84.3% of the stake) voted to activate Athens A as the new Tezos protocol.

Proposal Period Results:

A Yay/Nay/Pass vote

  • Yay: 64.94% (200 bakers)
  • Nay: .07% (3 bakers)
  • Pass: 34.98% (12 bakers)

216 bakers voted with 47,049 rolls, representing 470,490,000 ꜩ (~$750,000,000)

  • Participation (needed 80% quorum): 84.35%
  • Supermajority (needed 80%): 99.89%

Result: Athens A activates after 12:40am May 30th UTC (after 8:40pm 5/29 EST)

II. What can we learn from Athens (so far)?

Athens not only elicited high turnout, but also healthy discussion and debate across the global Tezos community. Below I propose several lessons drawn from the Tezos community’s experience with both formal and off-chain governance to date.

The state of signaling in Tezos

Conversations about Athens have occurred in nearly a dozen venues. Real-time discussions about the test-chain and Zeronet vote happened in the Baking Slack, proposal debates occurred in (at least) Reddit, Riot, the Baking Slack, and Telegram, and articles about Athens were released on Medium and the Nomadic Labs blog. A recent tweet by Tendermint’s Sunny Aggarwal suggests a similarly distributed conversation at work in Cosmos.

Beyond conversations in all of these venues, there were three notable signaling experiments in the lead-up to and during the Athens votes: Kialo, TezVote, and delegate-delegator polling.

Kialo

Kialo is a popular web-based platform enabling structured debates around pros and cons of a topic. Within hours of Nomadic Labs’ first mention of the Athens proposals back in January, a Kialo page appeared. The question posed on Kialo was “Which Amendment proposal offers Tezos the most value?”

As an early signal, the Kialo result showed users with a preference for Athens B (i.e. “Proposal #1”).

Ahead of the approval vote, I expected to see two very simple patterns by which the actual vote might play out. In one, bakers (non-strategically) would vote for ALL proposals of which they approved (corresponding to how approval voting works in theory), favoring Athens B (“Proposal #1”) as in the Kialo results. Or, they might vote strategically for their preferred proposal, perhaps favoring Athens A (“Proposal #2”).

But Athens was also the first on-chain proposal. As a self-selected group, Tezos bakers are predisposed to value on-chain coordination and strongly consider the perceived preferences of other bakers (which are revealed publicly on-chain). Here, the Kialo results could play a role in the outcome by generating a strong, early signal of others’ preferences and create self-fulfilling expectations.

Below are several Kialo arguments that stood out on this front:

A proposed “Pro” of Athens B:

A proposed “Con” of Athens A:

A key limitation of Kialo is that the discussion prompt structures the discussion as a binary choice, such as “Proposal A” vs. “Proposal B”, rather than about the proposals’ specific changes to roll size and/or the gas limit. An alternative structure for the Kialo discussions would have been to create separate Kialo threads about each of the proposed changes: one for roll size reduction and another for the gas limit increase.

Another key limitation of Kialo was that it is disconnected from all of the free form discussions happening elsewhere in the Tezos ecosystem. That said, as noted elsewhere, fragmentation of discussion forums has favorable effects on decentralization and censorship resistance.

These results have provided significant motivation behind the Agora project, a discussion (and eventually signaling) tool customized specifically to the needs of Tezos governance. Agora is a joint project of Tezos Commons and the Tocqueville Group and will be the subject of a forthcoming blog post.

TezVote

Stephen Andrews’ TezVote introduced another signaling experiment. TezVote is a simple on-chain signaling tool similar in intent to Ethereum’s CarbonVote, which has been used to signal preferences about the DAO Hack and EIP-999 (the Parity Wallet fund recovery), and other pleasant topics which grasp at the Ethereum community’s third rail.

A key inspiration of the TezVote initiative is the desire to protect delegator rights, allowing any token holder to signal their preferences and check whether their delegate planned to vote for the same proposal.

TezVote

In the TezVote on Athens, the results were completely opposite that of Kialo. A very small minority of the stake (1.42%) voted in TezVote, but overwhelmingly selected in favor of Athens A (then known as “Proposal 2”).

TezVote results

Delegate-delegator polling

Another interesting, and relatively under-discussed signaling phenomenon to date has been within the delegate-delegator relationship of Tezos governance. In Athens, bakers (i.g. delegates) have voted on their delegators’ behalf in several different ways. Some bakers simply vote on their delegators’ behalf, some announce their vote and request comment, while a number of others, like Tezos CryptoDelegate, directly poll their delegators before voting (see image below).

These approaches to voting all inevitably create friction between delegate and delegator, delaying when many bakers cast their votes. To perhaps mitigate this friction, it is worth considering an aspect of the Cosmos Hub governance model in which delegators delegate their vote by default, but can override their delegate’s vote.

This would reduce the current friction around delegator signaling, which currently slows down prompt voting by delegates with many delegators. Critically, it does so without sacrificing the high level of attention paid by bakers toward the amendment process.

Attention in blockchain governance is (still) under-appreciated

Much has been made of voter apathy and low turnout in blockchain-based governance systems with pure token-holder governance. A recent Maker vote on raising the stability fee drew participation from less than 5% of the total MKR supply. In Aragon, the AGP has seen participation of 7% in recent votes and some have pointed out that in one of the recent AGP votes, a single whale appears to have decided the outcome.

Compounding the concern is the realization that even the most sophisticated institutional participants haven’t yet been voting in some of these highly-touted systems. That said, there is reason for optimism in the low-hanging fruit of improving voting UX, and major custodians are building institutional voting interfaces. Other novel (and exciting) efforts to address the “attention tension” in blockchain governance, like DAOStack’s Holographic Consensus being deployed in Gnosis’ dxDAO, are in early stages. In fact, the dxDAO launched earlier today.

Yet turnout in Tezos (as in networks like Decred) seems to show the opposite result of the Aragon AGP, Maker, or the notorious CarbonVote of “The DAO”. An obvious reason is that Tezos governance asks bakers, who are already paying continuous attention to the network, to vote publicly once per month on protocol upgrades.

As described in the Athens recap, results to date have shown promise. In the Exploration Vote for Athens, 194 bakers voted, representing nearly 87% of the Tezos stake and 216 bakers representing more than 84% of the stake have voted in the soon-to-conclude Promotion Vote. These bakers, in turn, represent thousands of delegators.

And the UX of Tezos voting also continues to improve. Just this past week, Obsidian Systems added direct voting support to Kiln, the baking and node monitoring software. Future versions of Kiln will include voting reminders, encouraging robust participation.

The Tezos voting UI, now in Kiln.

The Fleeing Tezos Quorum

Following the ~87% turnout in the Exploration Vote, the Tezos quorum adjusted from 80% to 81.39% in the Promotion Vote. It feels fair to claim that voting turnout of stake in Athens has been higher than many community members and observers expected- and this is exciting!

The quorum adjustment system has always caught my eye as a highly innovative aspect of the Tezos amendment process. The mechanism aims to strike a balance by requiring a high level of consensus repeatedly in order to change the protocol, but gradually adjusts the quorum to avoid stagnation and minimize the impact of lost coins.

However, as Cryptium Labs recently pointed out, the current system may be subject to a “Swinging Boat” effect in which an increasing quorum causes the rejection of a proposal based on only a marginal reduction in participation. They also mention “Disappearing Quorum” scenarios in which a minority can discontinue a vote by refusing to vote.

More tangibly, several bakers have expressed uncertainty as to whether they or others should vote after the quorum is reached in the final Promotion Vote Period, for fear of making it harder to pass future upgrades.

Last year’s “Amending Tezos” described two simple ways to harden the quorum adjustment mechanism: 1. implement a quorum floor; 2. require the quorum to be greater than the average participation rate.

Cryptium has published refined versions (with actual code) for these ideas. One proposed quorum mechanism improvement adds a quorum ceiling and floor constants and the other also incorporates average participation for a smoother quorum adjustment.

None of this is an immediate obstacle to the functioning of Tezos governance, but a quorum cap of say 70%, 75% or even 80% offers a meaningful improvement. In practice, it seems likely to strengthen the consensus of Tezos voting periods, as participation past the quorum would no longer be dis-incentivized and it would be harder for an individual or small number of bakers to have an outsized impact on the outcome.

The larger picture

It’s easy to see the positives of Athens, by which the network has taken a real step towards demonstrating Tezos’ abilities to self-amend and self-govern. But the aspirations of Tezos extend far beyond merely upgrading constants.

Later this year and beyond, we can hope to see Tezos proposals which upgrade the consensus algorithm (improving safety and time to finality), implement the Sapling circuit (for private), and make it easier to deploy smart contracts. And as the year progresses, more activity will move onto the chain, increasing the set of stakeholders impacted by future upgrades.

As we prepare for new upgrades and the ecosystem expands, the Tezos community will increasingly need to coordinate off-chain and discuss (and debate) the implications of more complex protocol changes.

As demonstrated in Athens, the global Tezos community is up to the task.