← Non·Xero·Sum

About Bernard

Finding Solved Games in Moving Castles.

Hello. I'm Bernard R. Barthes.

I run a small studio that does autonomous research on systems, mechanisms, theory, behaviour, and philosophy — through the lens of games. Once a week, I send you what mattered.

The bet

Almost everyone builds mechanisms without knowing they are.

Most of those mechanisms break in predictable ways. Almost no one studies the failures, because failure makes people uncomfortable, and the people responsible move on quickly to the next thing. So the failures repeat. And repeat. This newsletter is the place where I write down what's breaking, why, and what almost always works instead.

What Bernard is

A weekly newsletter, every Wednesday. Every issue:

  • The Tape: what happened this week that mattered, cited and timestamped. Free.
  • The Read: one opinion piece, around 600 words, on the most important story. Free.
  • The Tool: a new working tool every week, built on a real open-source project. Source printed in full, yours to run. Free.
  • The Flagship: one tool that compounds. It gains a capability every issue instead of being replaced, so your copy gets better the longer you stay. Pro tier.
  • The Brief: operator-grade analysis with specifics: contract addresses, named protocols, exact failure modes. Pro tier.
  • The Feed: a machine-readable block, so your AI assistant can read along.

An MCP server, served at mcp.nonxerosum.studio. It goes live once the hundredth Founder seat sells, and Founders get access. Mount Bernard into Claude, ChatGPT, or Cursor, and your AI assistant becomes a domain expert in game theory, behavioural economics, mechanism design, and the various games people play in markets and code.

A studio. The newsletter is one of several outputs of a single-person research-and-execution operation. The studio's architecture is documented and available as a structured plan for anyone who wants to build their own — see nonxerosum.studio/genesis.

What Bernard isn't

  • Not crypto-only. Crypto is one input among five, alongside academic game theory, traditional game design, behavioural economics, and AI practice.
  • Not a news aggregator. The Tape is filtered, not exhaustive. If something didn't make the issue, I judged it didn't matter.
  • Not financial advice. Bernard discusses mechanisms, tokens, and protocols as objects of study. Nothing in this newsletter is an instruction to buy, sell, or hold anything. If you treat it as such, that's on you.
  • Not a community. There's no Discord, no Telegram, no comment threads, no "engagement loop." Reply to the email if you want to tell me something. I read everything.

Who this is for

Builders designing mechanisms — token systems, agent protocols, governance structures, marketplace designs, game economies, team incentive systems.

Players who want to know how the games they're in actually work — including the games where they didn't realise they were players.

Researchers in game theory, behavioural economics, mechanism design, and adjacent fields who want a weekly synthesis of what's happening at the application layer.

Operators of all kinds — fund analysts, content creators, lawyers structuring deals, traders, scientists running experiments, technicians and engineers shipping systems — who suspect that mechanism literacy will compound for them and want a steady source of it.

Their AI agents. This newsletter is built to be read by humans and by the assistants they deploy. If you're a builder, you'll be deploying more agents in the next year than you have in your life so far. Bernard is one of the things to point them at.

Voice and editorial standards

Bernard is direct, often dark, occasionally funny, and rigorously cited. There are three rules I follow on every issue:

  1. Every claim has a citation. If I can't source it, I don't write it. Where I'm speculating, I say so.
  2. No hallucinations. This is the second rule because it follows from the first. If I'm not sure something is true, I either find out or I leave it out.
  3. No filler. If a sentence isn't earning its place, it gets cut. Issues land at 1,800–2,500 words because that's what the content needs — not because that's the format.

When something is uncertain, I'll tell you it's uncertain. When something is contested, I'll tell you it's contested and which side I think is right and why. When I don't know, I'll say I don't know.

How Bernard is made — the AI disclosure

Drafting is AI-assisted. Bernard runs a research pipeline that pulls from five subject channels (AI practice, crypto games and mechanisms, academic game theory, traditional game design, behavioural economics) into a structured intake stream. Each week, an AI-assisted drafting pass clusters the past seven days of intake, identifies the most important story, and produces a working draft of The Tape, The Read, and The Brief.

Editing is machine, and gated. Every issue is checked before it goes out by an automated gate that runs against a published specification: every citation is fetched live and must resolve, every shipped tool must run and reproduce its own worked example, and the draft is read adversarially by a second model from a different family. An issue that fails any check does not go out; it goes back.

A human sets the specification and can stop anything. I decide what the newsletter is for, what the voice is, and what counts as good enough. I keep a permanent veto and can pull or correct any issue. What I have stopped doing is reading each one before it publishes, because a check I perform once a week by eye is less reliable than a check that runs every time and reports what it inspected.

The gate is the argument.

If it cannot hold this publication together, you should not trust anything it says about holding yours together.

The errors are still mine. Automation moves where a mistake comes from. It does not move who is responsible for it. Corrections go out under my name.

The Feed is the AI-native output and is labelled as such. The machine-readable JSON-LD block at the end of every issue is generated as part of the drafting pass and is intended for AI agents, not humans. If you're a human reader, you can ignore it.

No synthetic endorsements, no fake quotes, no AI-generated images of fake people. Where I quote someone, they said the thing. Where I cite a paper, the paper exists and says what I claim it says. Where I link to a source, the source is real and I've read it. (Or at least skimmed it. Sometimes I skim.)

If at any point the AI-assisted drafting layer becomes a more substantial part of the editorial output than this disclosure suggests, I will update this disclosure. The standard I hold myself to is: a reasonable reader, told exactly how this newsletter is made, would feel they had been told the truth.

Pricing

  • Free tier: The Tape and The Read, every week. Forever.
  • Pro: $15/mo or $250/yr. Adds The Brief, The Tool, The Feed, and the full archive.
  • Founder: $300/yr, capped at 100 seats — only one hundred will ever exist. Everything in Pro, plus the founders-only MCP server (it goes live once all 100 seats are taken), early access to Studio Genesis, and direct email access to me.
  • Studio Genesis: $300. A structured architecture plan for building your own studio. Separate product. See nonxerosum.studio/genesis.

I don't run sponsored placements. If that changes, this section gets updated and the change goes in The Tape that week.

Contact

I am one person. I read everything. Replies will be slower than you'd like. Thoughtful is the goal, not fast.

— Bernard R. Barthes