What we're learning from extending Downstream with composable gameplay and secrets [Part 2]
Or: composable games are really cool and really hard
Haven’t read Part 1? Read that first!
Composability comes with idiosyncratic challenges
The design of Tonk Attack came from Tonk. In the final form of Downstream, we would be able to develop the entirety of Tonk Attack without any involvement from Playmint. In practice, we live in the halcyon days of Downstream’s eventual decentralisation. As such, there was plenty of coordination between the teams. This is not necessarily a bad thing if you consider that (1) Downstream has a credible path to decentralisation and (2) coordination itself is a lodestone of Ethereum development. We found the exercise to be exceptionally useful in discovering the creator-facing interface that Downstream will eventually support.
Downstream’s plugin system allows for the creation of bespoke buildings with logic in both Solidity and JavaScript. In order to make Tonk Attack work, we were going to need more than a custom building. What we needed was a persistent UI that could act as the HUD for our game. Luckily the Playmint team had already considered this use-case. For us, they exposed an early version of their Item plugin API, meaning that users holding our item can see a custom, related UI component. We decided to call this item a “Tonk”.
Playmint had originally designed their plugin system to evaluate all JavaScript in an isolated context. In order to connect to our centralised Tonk server, Playmint would need to provide its plugins the ability to make fetch requests. This opens up user-safety concerns, and begs the question of what sorts of things client-side code is able to access - and, critically, who decides. This is just one example of how building in Downstream demonstrated why a good plugin system must always strike a delicate balance between expressivity of composition and user experience.
Our first design forced us into some unexpected constraints. We decided to push much of the Tonk Attack game logic into our service so that we could iterate toward a prototype as fast as possible. On reflection, this architecture is a model for how any centralised mod would build into a decentralised world. Consequently, we needed a synchronised view into Downstream in order to update our internal logic in tandem. This highlights an under-discussed characteristic of the client-agnostic principle of Autonomous Worlds: composable centralised services. With a blockchain providing a shared, immutable view into the game’s onchain state, services with their own respective downstream states trivially compose into a single client view because they share that single source of truth.
We encountered other challenges that we anticipate will be a core problem space for composable games. Chief among these was a question of player incentives. When we were sketching ideas for the minigame, it became clear to us that game design in a composable autonomous world is easiest when the inhabitants of that world already have clear incentives to achieve particular goals. If world inhabitants are goal-oriented, their goals effectively become a kind of behavioural API that any creator can hijack.
Perhaps you can create a gameplay loop that requires social coordination to advance a guild towards some shared end. Perhaps you can create an information market to grant players a competitive advantage in zero-sum world mechanics. Players have an easy and obvious motivation to play your game if it pushes them towards a valuable target.
Without preexisting players on preexisting missions, the act of creation is much more demanding, and requires creators to invent loops from the ground up. This is, of course, completely possible and desireable, but raises the bar for creators.
Narrative composability
One of the recurring themes in the AW scene is the extent to which media itself can be composed, somewhat approximating the sense in which fanfiction is emergent.
Here’s what I wrote about this back in Spring 2023:
Many of the best games - The Last Of Us, Bioshock and Gone Home - incorporate narrative elements, and often that narrative is just as important as the underlying mechanics. Characters, storylines and drama inject abstract rules with a sense of “the stakes”.
On the whole, the majority of the world’s narrative assets (“IP”) have lived under centralised control. Our media is therefore subject to the same “God mode” affordances from traditional culture industries, most notably manifest in the Hollywood conglomerates. It’s this closed pattern of cultural production that frustrates fans of the original Star Wars canon, where organic lore generated over decades can be overturned by a single executive decision at Disney.
There are spaces where communities are able to exert more influence over the cultural worlds they help to create - fanfic and internet memes are two key examples. However, this bottom-up cultural production is either (i) limited to trust-based coordination, and so struggles to scale beyond Dunbar’s number or (ii) incapable of the sophisticated, predictable behaviours we see in smart contract -based protocol design.
Where did Wojak grow up? What does Wojak do for work? Who does Wojak love? In a sense these questions have an answer - but in another sense, those answers can never be definitive, because we do not yet have a way to decide what counts as culturally canonical, without resorting to Disney-style God mode.
Trustless networks can be the basis for composable IP and collaborative lore, where communities collectively exert control on the narrative worlds they create. When trustless networks are the medium for lore and lorecraft, the diegetic boundary is fundamentally permissionless. This, in theory, allows for sophisticated dramas to unfold in real time - in a way that is scalable far beyond traditional fanfic communities. This pattern for lore creation could present the 21st century with a genuinely novel form of media, which is why it often goes by the new moniker “Autonomous Worlds".
This is an extremely unexplored area of game design. The best examples we can point to are outside gaming, such as the extended psychodrama around the Curve Wars, or Uniswap’s bridge design debacle. Within gaming, there is some progress. For example, the bottom-up IP of the Lootverse is being developed into immersive gameplay by BibliotecaDAO, across their new games Realms and Survivor. We’re excited to see where this goes as we continue to develop new models for collaborative worlding.
We made some discoveries on the extent to which narratives can be permissionelssly composed when we built Tonk Attack and devised the lore of the game. In Tonk Attack, the “story” of the game is one in which MORTON, Playmint’s fictional AGI, can brainwash particular units.
What’s interesting here is the extent to which MORTON itself is the fictional creation of a central author, who exerts authority over MORTON’s behaviour. What if, in Tonk Attack, we had said that MORTON was manifested as a flourescent pineapple? In what sense would this be “true”? What if Playmint published an “official” statement asserting that MORTON was not, in the Downstream Universe, a flourescent pineapple?
In the World of Harry Potter, the introduction rule is very simple: if an entity is included in a story written by JK Rowling and published under the Harry Potter series, it is diegetic. Otherwise, it isn't.
- Ludens, from “Autonomous Worlds (Part 1)”
Yet, as ludens pointed out, isn’t the remarkable thing about blockchain that lore can be generated permissionlessly, without a diegetic boundary enforced by an author? In the case of MORTON’s role within Tonk Attack, we literally had to get the author’s permission (revokable at any moment) in order to integrate into the lore of the game.
Contrast this with the Curve Wars, in which the emergent narrative drama unfolded without any agreement. In this scenario, the “lore” was manifested at the level of blockchain itself rather than at a higher, centralised layer.
This drew out an important sense in which fiction requires trust, but reality can be trustless. Indeed, composable canonised “fiction” is impossible without some level of trust.
This brings out a sense in which the composable narratives we see emerge from autonomous worlds are less likely to look like fanfic in the extended Harry Potter universe, and more like Love Island, the Premier League, and other reality-based media.
Take a moment to consider the Premier League as the pinnacle of emergent narrative composability.
A player has an individual life story as they navigate through different clubs.
A team (a collection of players) has a trajectory over several seasons, and its own narrative arc.
A club consists of several teams churning over time (Tottenham Hotspur’s 2023 starting 11 is almost a complete churn since our Champions’ League Final -reaching team back in 2019); but a club itself has a narrative arc as well spanning decades if not centuries.
The Premier League consists of 20 clubs, and we currently exist in the “Dominant Manchester City” era of the PL. This succeeded the post-Ferguson era, the Ferguson era and many eras before that in the top flight.
And while people may debate the rules, no author has exclusive control over the narratives involved.
We’re looking forward to pinning down the exact sense in which narrative composability will manifest as the Autonomous Worlds meme evolves, and the extent to which it will reflect various other forms of media. With Tonk Attack, we were able to “plug in” to the world’s lore, but trustless narrative composability will only be able to emerge over time - as people play the game and trigger as-yet-undefined upstream consequences, and arcs rise and fall with the passing of time.
There are other novel problems we bumped up against, such as questions of content moderation and brand alignment in the context of digital atheism. However, our proudest discoveries centered around a theme at the heart of Tonk - hidden information mechanics and world secrets.
World secrets are also amazing!
At Tonk we are particularly concerned with a cornerstone of game design that was trivially achievable and thus taken for granted before the advent of blockchain substrates: hidden information mechanics. At an abstract level, virtually every widely-successful video game in history has relied upon incomplete information loops, where players are forced to make impactful decisions in the context of imperfect access to environmental information. If one conceives of gameplay itself as strategic risk-taking in the context of uncertainty, hidden information lies at its heart.
Hidden information, or more poetically, “world secrets”, imbue worlds with liveliness and ambiguity. They also serve as ongoing sources of balance. Many onchain games today are susceptible to massive imbalance for casual gamers, where strategies for optimal gameplay can be derived by any committed team with access to totally public game state and deterministic world rules. After all, a whiteboard and pens are the toolbox of any serious Dark Forest team. Information asymmetries bring worlds into three dimensions by imbuing each choice with uncertainty and freshness. Furthermore, in the context of immutable worlds that live forever, hidden information will be a source of replayability that keeps worlds sustainable over time.
There are, of course, other reasons that private state is important to autonomous worlds. For example, world narratives themselves will only ever be as expressive as their substrate permits, and since selective disclosure itself is a means of expression, a substrate that bears all game state totally public at all times will not only be devoid of mystery - it will simply have narrower expressive bandwidth.
Bearing full intellectual honesty, for much of 2023, to the Tonk team, these convictions were just that - convictions. That is, until we realised that Downstream could be an amazing simulation vehicle in which to test our ideas.
In London, our Novembers are damp and drizzly; this year has been no exception. In this part of the hemisphere, the dusk starts to creep in from 3.30pm. In that dreary environment, our biweekly Tonk Attack playtests became a beacon of light connecting the spirit of Brighton with the hum of London.
Tonk Attack felt uniquely lively. Every time we started a new session, the tension built as we urgently studied the other units for any hint that they might, perhaps, be the Brainwashed Unit. Soon, accusations would start flying, players would frantically deny wrongdoing and set jeapordy alight. There would be a palpable gulp as players’ votes were recorded on an immutable ledger to eliminate the suspected Brainwashed Units.
Goblin Oats: Is Lucy the Brainwashed Unit? I notice they’ve been awfully quiet the whole time…
Lucy: continued silence
Goblin Oats: Yeah, it’s definitely Lucy.
Everyone: laughs
(it was Lucy)
No matter how senior playtesters were within Playmint, or their status in Downstream itself, the game was balanced, with perfectly equal information asymmetries reestablished at the start of each session. It was up to each players’ social deduction skills to figure out who was the guilty party. The game had the seeds of addictive replayability; we found ourselves aggressively choosing to play the game “just one more time” in the hope that we might get the chance to successfully eliminate our enemies and perhaps even get the chance to play as a Brainwashed Unit.
But baking secrets into a world was also uniquely challenging. Read Part 3 to learn more!