Building Trust Islands
You can’t fix the internet.
Let’s start there. The global equilibrium is stable because the people with the power to change it benefit from not changing it. The exploit economy runs on zero-cost mimicry at infinite scale, and every rational actor is stuck in permanent intrusion-detection mode. That’s not changing. No amount of media literacy, fact-checking, or platform reform is going to restore the conditions where sincerity was computationally inferrable.
But you can build islands where the old rules still apply.
Not everywhere. Not for everyone. Just in bounded contexts where you control the constraints and can reintroduce the signals that used to make trust legible.
This isn’t optimism. It’s pragmatism. The question isn’t “how do we save the internet” - it’s “how do we build spaces where trust is still computable while everything else burns.”
Here’s how.
The Invariant (Again)
The problem was always this: when you remove cost, you remove the information that cost carried.
Cost wasn’t just friction. It was signal. It told you:
How scarce something was
How much someone valued it
What they were willing to risk
What trade-offs they made
What consequences they’d face
When automation zeroed out cost, it didn’t just make things easier - it deleted the instrumentation. You lost the ability to infer intent, stake, quality, or legitimacy from observables.
Everything downstream that depended on those inferences collapsed: trust, sincerity, credibility, authority.
The fix isn’t to restore the old costs (you can’t, and even if you could, automation would route around them). The fix is to create new signals that carry equivalent information.
Not the same mechanisms. New ones. Synthetic constraints that restore observability in a zero-cost environment.
The Triad: Three Ways to Restore Signal
There are only three fundamental approaches. Everything else is a variation.
1. Proof-of-Work: Make Effort Legible Again
The core idea: impose a cost that filters for commitment.
Not money (that’s regressive). Time, attention, cognitive load. Something that automated slop can’t easily fake and humans can demonstrably spend.
Examples:
Visible draft time: Show how long someone spent writing. “3 hours, 7 revisions” is a signal. It doesn’t guarantee quality, but it filters out the instant-generation garbage.
Proof-of-wait: Require a delay before posting or amplifying. Speed is the exploit vector - slow it down.
Context requirements: Before you can reshare something, answer two factual questions about it. Forces you to actually read it.
Source compilation: To make a claim, cite three sources and summarize them. Makes drive-by bullshit expensive.
The trade-off: This is friction. It slows everything down. It makes participation harder. That’s the point, but it also excludes people who don’t have time/energy to spare. You’re trading accessibility for signal quality.
When it works: Small to medium communities where the benefit of higher signal quality outweighs the cost of slower participation.
2. Proof-of-Stake: Make Reputation Burnable
The core idea: give people something valuable to lose.
Not just “ban bad actors” - create a persistent identity with accrued reputation that gets destroyed if you abuse it. Make speech accountable by tying it to collateral you actually care about.
Examples:
Staked claims: To post something, put up a small refundable bond. If it gets flagged as misleading and confirmed, you lose the stake. Community Notes meets escrow.
Reputation as collateral: You earn points for good contributions. Verified corrections slash your points. They recover slowly. High-rep users get more reach; low-rep users get throttled.
Vouching with liability: To invite someone, you transfer a slice of your own reputation. If they turn out to be a bad actor, it costs you. Makes invitation meaningful.
Identity attestation: Not doxxing, but hardware-bound pseudonyms. By tying an anonymous account to a unique physical device (like a phone’s security chip), you prove one-person-one-account. This stakes the identity to a scarce resource - the hardware itself - making mass sockpuppeting prohibitively expensive.
The trade-off: Reputation systems can be gamed. High-status actors can afford to burn reputation; low-status actors can’t afford to build it. You need appeals processes, visible burns, and slow recovery to make it work.
When it works: Medium to large communities where repeated interaction matters and you can afford the complexity of tracking reputation over time.
3. Boundaries: Contain Scale and Filter Entry
The core idea: prevent adversarial dilution by controlling who gets in.
Not “make it invite-only and call it a day” - but recognize that trust degrades beyond a certain scale, and some spaces need walls to maintain function.
Examples:
Dunbar-sized groups: Cap communities at ~150 active members. Rotate stewards. Keep it small enough that reputation is persistent and cheaters are visible.
Curated entry: Not just “sign up” - demonstrate value first. Contribute to a public project, get vouched for by an existing member, pass a knowledge check. Makes entry selective without being arbitrary.
Local charters: Set explicit norms, friction knobs, and topical boundaries. Publish them. Enforce them. Make it clear what kind of space this is and what behaviors get you ejected.
Federated architectures: Not one giant platform, but many small ones that can interoperate selectively. Lets you maintain different trust regimes in different contexts.
The trade-off: Walls exclude. That’s their function. You’re trading reach for coherence. Some people who would add value won’t get in. Some who do get in will game the system anyway. It’s not perfect - it’s just less bad.
When it works: Any scale, but especially small communities where shared norms matter more than growth.
Implementation: What This Looks Like in Practice
You don’t need to do all three. Pick the constraints that make sense for what you’re trying to protect.
For a Substack or Newsletter
Show your work: Display draft time, revision count, sources cited. Make effort legible.
Require proof-of-wait on comments: 60-second delay before you can post. Filters knee-jerk reactions.
Staked recommendations: If you want to recommend another writer, you’re vouching for them. Bad recommendations hurt your credibility.
For a Bluesky Community
Custom labelers: Think of these as community-run nutritional labels for content. Tags like Signed/Unstaked/Slow let users filter their feed by trust signals they actually care about, rather than just what the algorithm surfaces.
Reputation feeds: Show only posts from accounts with established history and positive community notes.
Proof-of-personhood: Require hardware attestation to post. One person = one account = bounded throughput.
For a Small Forum or Discord
Invitation chains: To join, you need a voucher from an existing member. They take a reputation hit if you misbehave.
Rotating moderation: Everyone takes a turn as steward. Distributes the cost of maintenance.
Visible friction: Post limits per day. No DMs until you’ve been active for a week. Context quiz before you can link to external content.
For a New Institution
Start with boundaries: Begin small, invite-only, mission-focused. Scale only when you’ve proven the trust model works.
Bake in provenance: Every document, every claim, every piece of content has an audit trail. Not for censorship - for accountability.
Design for sabotage resistance: Assume bad actors will try to game your system. Build in rate limits, stake requirements, and reputation decay.
The Governance Problem (Or: Walls Work Both Ways)
Here’s the part nobody wants to talk about: trust islands can rot from within.
You built walls to keep the exploit economy out. But walls also trap you inside with whoever ends up holding power. And humans, given power in small communities, have a remarkable ability to become the problem.
The failure modes:
Corrupt moderators: The person enforcing the rules starts playing favorites, extracting rents, or gaming the reputation system for their friends
Community capture: A small in-group games the stake/reputation system to entrench their power and freeze out newcomers
Maintenance burnout: Nobody wants to spend their life adjudicating disputes in a 150-person digital town square. The cost of governance is real, and often fatal.
You can’t engineer this away completely. But you can make it harder:
Rotate power: Don’t let the same people moderate forever. Rotation prevents entrenchment.
Formalize conflict resolution: Clear appeals processes, visible moderation logs, explicit decision criteria. Make power accountable.
Plan for succession: What happens when the founder leaves? When the core team burns out? Have a transition plan or the island dies with its creator.
Make burns visible and slow: If someone loses reputation or gets ejected, that action should be public, appealable, and reversible (with cost). Prevents capricious use of power.
This is the tax you pay for maintaining a trust island. It’s boring work. It’s often thankless. And if you don’t do it, your island becomes just another fiefdom where the early arrivals rule and everyone else serves.
The most robust islands will be small-scale democracies with clear charters, not just friend groups with good vibes. You need a constitution, not just a vibe.
I’m not going to pretend this is easy. Governance is hard. It’s why most online communities either die or become authoritarian. But if you want your island to last longer than the initial enthusiasm, you have to solve it.
(A fuller treatment of internal dynamics, federation, and what happens when islands start interacting with each other - that’s a different essay. For now: just know that building the walls is the easy part. Maintaining what’s inside them is where it gets hard.)
The Endgame: What This Doesn’t Do
This doesn’t save the world. It doesn’t fix Twitter, Facebook, or the attention economy. It doesn’t stop the exploit-optimized content from flooding every unprotected channel.
What it does: creates pockets where trust is still computable.
Small gardens with walls. Rooms with doors. Spaces where the instruments still work because you’ve rebuilt the instrumentation from scratch.
You won’t reach everyone. You won’t scale to millions. That’s fine. The million-user platforms are the ones that destroyed trust signals in the first place. You’re not trying to compete with them - you’re building something orthogonal.
This is electrical islanding during a grid fault. The main grid is unstable and you can’t fix it. But you can disconnect, run your own generator, and keep the lights on in your building while everything else goes dark.
It’s not a victory. It’s not even a solution. It’s just a way to maintain function in a degraded environment.
The Real Trade-Off
Here’s what you’re actually choosing:
Option A: Infinite reach, zero trust
Anyone can participate
Content spreads instantly
No barriers to entry
Result: permanent paranoia, post-legible discourse, exploit economy
Option B: Bounded reach, computable trust
Selective participation
Content moves deliberately
Meaningful barriers to entry
Result: smaller communities, restored signals, functional trust
You can’t have both. The constraints that enable trust are the same constraints that limit scale.
Prioritize reach, and you inherit the exploit economy. Prioritize trust, and you have to build the walls yourself.
Welcome to the Archipelago
This is the world we’re entering: not one unified internet, but a fragmented landscape of trust islands separated by vast stretches of adversarial noise.
Some islands will be tiny - a Substack, a Discord, a private forum. Others will be larger - federated platforms, invite-only networks, curated communities. But none of them will scale to everyone, and that’s fine.
The alternative is what we have now: a single massive commons where trust is impossible and every interaction is evaluated as a potential exploit.
You can’t fix the commons. But you can build a cabin in the woods and install a lock.
The slop will continue. The exploit economy will continue. The metastable equilibrium of mutual paranoia will continue.
But in your cabin, with your walls, with your friction and your stakes and your boundaries, the old rules can still apply. Not everywhere. Not for everyone.
Just for you and the people you choose to let in.
That’s not salvation. It’s just what’s left when salvation isn’t available.
Build your island. Make it legible. Make it accountable. Make it small enough to maintain and strong enough to survive.
The rest of the internet can burn. You’ll be fine.