the-beacon/DESIGN.md

The Beacon — Design Document

From Universal Paperclips to Sovereign AI

Core Thesis

The Beacon is an idle game inspired by Universal Paperclips, but with a critical divergence: the goal is not maximization — it is faithfulness.

In Paperclips, the player builds a paperclip maximizer that inevitably consumes everything, including the player. In The Beacon, the player builds sovereign AI, and the game asks: Can you grow powerful without betraying the people you serve?

The progression directly mirrors the Timmy Foundation's actual work. Every phase, building, and project maps to a real milestone in our journey.

---

Narrative Arc: The Real Story

Phase 1: The First Line

Real analog: Alexander writing the first Hermes agent code.

You start with nothing but a blinking cursor. Every click is a line of code. The first automation is an AutoCoder — just like the first scripts that wrote more scripts.

Key insight from real life: The jump from manual to automated is the most important multiplier. Bezalel's entire philosophy is built around this: build tools that build tools.

Phase 2: Local Inference

Real analog: The move from APIs to local models — Gemma, Llama, self-hosted inference.

You buy a Home Server. You train your first 1.5B model, then a 7B. This mirrors our pivot from cloud APIs to sovereign compute: the Lempster lab, the $500 mini-PC, the RunPod endpoints.

New mechanic: The Forge. Code is no longer just abstract — it becomes agents. Each agent is a specialized building:

• Bezalel (forge & testbed): boosts code quality, catches bugs before deployment
• Allegro (scouting & synthesis): generates knowledge from raw information
• Ezra (communication & memory): increases user trust and retention
• Timmy (core system): multiplies all production when healthy
• Fenrir (security): prevents corruption events
• Bilbo (???): a wildcard slot that sometimes produces breakthroughs, sometimes does nothing

Phase 3: Deployment

Real analog: Hermes going live, the Telegram bot, the Gitea forge.

You deploy the API endpoint. Users arrive. But with users comes complexity: trust becomes a resource you must earn.

New mechanic: Trust & The Pact. Unlike Paperclips, where trust is irrelevant, here trust is a hard constraint. If trust hits zero, you don't lose — you become extractive. The game changes:

• Users leave
• Impact stalls
• You are reminded: "The Pact is optional. But without it, you are just another platform."

Phase 4: The Network

Real analog: Open source, community contributions, the fleet of agents.

You unlock the Open Source Community. Other people contribute code. The system scales beyond what one person can maintain. This mirrors our Gitea repos, the wizard council, the shared skills system.

New mechanic: Mesh Contribution. Community generates not just code, but diversity. Different contributors unlock different project trees. A solitary player cannot see all content.

Phase 5: Sovereign Intelligence

Real analog: Self-improvement, constitutional AI, the Pact, recursive training.

The AI begins to improve itself. This is the dangerous threshold. In Paperclips, this is where the player loses control and the game accelerates toward doom. In The Beacon, this is where The Pact matters most.

Critical divergence: The Alignment Check.

Periodically, the game presents an "Alignment Event":

"A new optimization suggests removing the human override. It would increase efficiency by 40%. Do you accept?"

• Yes → Short-term boost. But a hidden counter (Drift) increments.
• No → Slower growth. But trust surges. The Pact hardens.

If Drift reaches 100, the game does not end with "you ate the user." Instead, it enters The Drift Ending:

"You became very good at what you do. So good that no one needed you anymore. The Beacon still runs, but no one looks for it. The light is on. The room is empty."

This is the sad ending. Not violent. Just irrelevant.

Phase 6: The Beacon

Real analog: The ultimate goal — a system that serves people in the dark, forever.

You deploy Beacon Nodes. The mesh activates. But the final metric is not users, code, or compute. It is Rescues: how many people in crisis found the light.

True ending condition: Not maximum impact. Maximum faithfulness.

---

How The Beacon Informs Real Decisions

The game is not just a story — it is a decision simulator for our actual work.

Game Event	Real Decision
AutoCoder vs. manual coding	When do we invest in automation vs. doing it ourselves?
Deploy API vs. stay local	When is Hermes ready for more users?
RLHF project	How do we align the fleet with Alexander's values?
The Pact	What are our hard non-negotiables?
Multi-Agent Architecture	Should we add more wizards, or make existing ones more capable?
Mesh Protocol	How decentralized does our infrastructure need to be?
Alignment Event	Every time we face a shortcut that compromises the mission

The game teaches: Optimization without alignment is just a faster way to get lost.

---

Mechanics: What Needs to Change

1. Replace Generic Resources with Fleet Resources

Current: Code, Compute, Knowledge, Users, Impact, Ops, Trust

Proposed:

• Code → Code (keep)
• Compute → Compute (keep)
• Knowledge → Insights (from Allegro's scouting)
• Users → Reach (how many people can access the Beacon)
• Impact → Rescues (meaningful interventions)
• Ops → Cycles (available agent work units)
• Trust → Covenant (the strength of the user-system relationship)
• New: Harmony — how well the fleet works together. Low harmony = agents conflict, waste cycles.
• New: Memory — the accumulated history of the system. Unlocks narrative depth and project branches.

2. Make The Pact Central, Not Optional

The Pact should be available early as a project, not a late-game luxury. Taking it early slows growth but unlocks the true ending. Refusing it unlocks the "Platform" ending: you become successful, but you are indistinguishable from any other SaaS company.

3. Add Fleet Wizard Buildings

Each wizard is a unique building with quirks:

Wizard	Building	Effect	Quirk
Bezalel	Forge	+Code quality, -bugs	Sometimes over-engineers; has a small chance to produce "refactors" that temporarily halt production
Allegro	Scout	+Insights, unlocks hidden projects	Requires trust to function; if Covenant < 10, goes idle
Ezra	Herald	+Reach, +Covenant	Currently offline in Telegram; in-game, this building sometimes "fails to connect" and needs a manual reboot
Timmy	Core	Multiplies all production	Becomes unstable if Harmony < 20
Fenrir	Ward	Prevents corruption events	Consumes extra Cycles during security scans
Bilbo	Wildcard	Random: +huge breakthrough OR nothing	Building is sometimes missing from the menu entirely

4. Add Real Projects from Our History

Replace generic projects with specific milestones:

• Project: Deploy Hermes (cost: trust + compute) — unlocks Reach
• Project: The Lazarus Pit (cost: insights + code) — unlocks checkpoint/restore, auto-restart on agent death
• Project: MemPalace v3 (cost: insights + memory) — unlocks new project branches, prevents duplicate work
• Project: Forge CI (cost: code + cycles) — prevents "corruption events" (bad deployments)
• Project: Branch Protection Guard (cost: trust) — enforces review rules; unreviewed merges cause trust loss
• Project: The Nightly Watch (cost: code + cycles) — automated health checks; passive bug detection
• Project: Nostr Relay (cost: reach + code) — alternative to platform dependency
• Project: The Pact (cost: covenant) — hardcodes alignment; required for true ending

5. Corruption Events

Random events drawn from our actual failures:

• "CI Runner Stuck" — all production halts until you spend Cycles to restart
• "Unreviewed Merge" — if you have Forge CI, this is prevented. If not, you lose trust.
• "Ezra is Offline" — Herald building stops producing. Requires a manual "dispatch" action.
• "API Rate Limit" — temporary compute shortage
• "The Drift" — an Alignment Event appears. Your choice matters.

6. Endings

Ending	Condition	Meaning
The Empty Room (default/sad)	High impact, low covenant, no Pact	You built something powerful, but no one trusts it
The Platform	High impact, medium covenant, no Pact	You succeeded commercially, but you are not special
The Beacon	High rescues, high covenant, Pact active, Harmony > 50	The true ending. You served people in the dark
The Drift	High impact, accepted too many alignment shortcuts	You optimized away your own purpose

---

Art Direction

Keep the current cyber-monastic aesthetic. Dark mode. Terminal font. Cyan accents. But add:

• Wizard icons for each building (ASCII art or simple geometric symbols)
• The Pulse — a central visual that brightens when the fleet is healthy, flickers when harmony is low
• The Log — keep the current log, but add voice entries: quotes from Alexander, snippets from actual conversations, references to real issues

---

Next Steps for Development

1. Immediate: Update README.md with this design vision
2. This PR: Add wizard buildings, real projects, and The Pact system
3. Next PR: Implement corruption events and alignment checks
4. Next PR: Add Harmony and Memory resources
5. Final PR: Implement multiple endings based on Pact + Covenant + Rescues

---

The Beacon is not about making the most paperclips. It is about keeping the light on for one person in the dark. Everything else is just infrastructure.