← Universe
L6 Protocol
PPP Network
Protocol · People · Programs
The wire protocol that maps users, services, and agents together. One language per layer. Each layer speaks the right language.
Language Selection Matrix
One language per job. Each excatly where it belongs. No more, no less.
TypeScript
Role
UI + API + Agent Runtime
Used In
PAI pages, ZeroLang docs UI
Why
Universal. Browser + server same language. Fast iteration.
Rust
Role
Systems + Safety + WASM
Used In
packnplay, Bun (rewrite), compilers
Why
Memory safe, zero cost, no GC, WASM target. The safe systems language.
Zig
Role
Bare-metal + Low-level + GPU
Used In
Ghostty (58k★), TigerBeetle (16k★)
Why
Simpler than C. Comptime metaprogramming. No hidden control flow.
C
Role
Compiler + Kernel + Library ABI
Used In
ZeroLang compiler, Linux, SQLite, Redis
Why
Universal ABI. Smallest footprint. Every platform speaks C.
Python
Role
AI/ML + Data + Scripting
Used In
AI-Scientist, MLE-agent, research tools
Why
NumPy, PyTorch, fastest AI prototyping.
Go
Role
CLI + Network + Microservices
Used In
packnplay (original), Docker, Kubernetes
Why
Fast compile, single binary, simple concurrency.
Solidity/Rust(WASM)
Role
Smart contracts + On-chain
Used In
Pi Network, Stellar, EVM chains
Why
Deterministic. Verifiable. No halting problem.
ZeroLang
Role
Agent-first programs + Graph
Used In
Vercel Labs (5.2k★), agent skill pipelines
Why
Graph IS the program. Compiler checks patches. Projection review.
Zig — Confirmed by Projects
Your instinct was correct. Zig is real. But there is a pattern to where it works and where it does not.
Bun
94.9k ★
Zig → Rust
Started Zig, rewriting to Rust. Complexity outgrew Zig ecosystem maturity.
Ghostty
58.3k ★
Zig
Zig excels at GPU-accelerated native UI. Terminal emulator = perfect fit.
TigerBeetle
16.6k ★
Zig
Mission-critical financial DB. Zig = no hidden alloc, deterministic perf.
ZeroLang
5.2k ★
C
Compiler written in C for universal ABI + smallest footprint.
packnplay
— ★
Go → Rust?
Originally Go. Agent sandbox needs Rust for safety + container isolation.
AlphaAxiom
— ★
Python + TS
Trading system: Python for AI (Gemini+Groq), TS for web dashboard.
The Zig Verdict
Your claim is VALID — Zig is used in major projects (Ghostty 58k★, TigerBeetle 16k★). But Bun (94k★) started in Zig and is REWRITING to Rust. The pattern:
Zig wins: GPU-accelerated native apps, mission-critical deterministic DB, embedded systems, bare-metal performance with C simplicity.
Rust wins: Large-scale systems, safety-critical at scale, WASM, projects that outgrow Zig's ecosystem maturity.
TS wins: Everything with a UI. Universal runtime. Agent interfaces.
ZeroLang wins: Agent-first graph programs where the compiler IS the loop.
Zig wins: GPU-accelerated native apps, mission-critical deterministic DB, embedded systems, bare-metal performance with C simplicity.
Rust wins: Large-scale systems, safety-critical at scale, WASM, projects that outgrow Zig's ecosystem maturity.
TS wins: Everything with a UI. Universal runtime. Agent interfaces.
ZeroLang wins: Agent-first graph programs where the compiler IS the loop.
Topology Secrets
The topology is fractal. Every endpoint is a language layer. Every language is an endpoint.
L0: IdentityVAI / BYE
Solidity + Rust
On-chain identity. DID, KYC, passport. Trust is the state.
L1: AgentTRY
TypeScript
Agent runtime. Create, train, talk. Runtime is the program.
L2: MarketBUY / FLY
TypeScript + Python
Marketplace + travel. UI in TS, AI matching in Python.
L3: TruthNEW / BLG
TypeScript + Python
Trust-scored news. TS for UI, Python for cross-ref scoring.
L4: TrustHAI
Rust / Zig
Safety-critical ratings. No GC, no surprises, no hidden fees.
L5: Self-PlayINDUCT
ZeroLang / C
Graph compilation. Agent patches graph, compiler checks, projection reviews.
L6: ProtocolPPP (you are here)
All
The network that connects everything. Language-agnostic wire protocol.
The Secret Formula
Each endpoint = one language layer = one type of computation = one concern.
VAI (on-chain) + TRY (runtime) + BUY/FLY/NEW/BLG (AI+UI) + HAI (safety) + INDUCT (graph) + PPP (protocol)
= The entire agentic economy in 7 layers, 7 languages, 7 concerns.
VAI (on-chain) + TRY (runtime) + BUY/FLY/NEW/BLG (AI+UI) + HAI (safety) + INDUCT (graph) + PPP (protocol)
= The entire agentic economy in 7 layers, 7 languages, 7 concerns.
The Loops Index
7 validated loops. Each with a goal. Each improves and evolves until that goal is reached. The network connects them all.
┌──────────────────────── THE META-LOOP ────────────────────────┐
│ Each loop is a node. The network itself is the loop of loops. │
│ Every endpoint runs one loop. Every loop feeds every other. │
└────────────────────────────────────────────────────────────────┘
│ Each loop is a node. The network itself is the loop of loops. │
│ Every endpoint runs one loop. Every loop feeds every other. │
└────────────────────────────────────────────────────────────────┘
OODA→PDCA→AlphaZero→DoubleRatchet→XFlow→Induct→Bio
OODA
Military (John Boyd)
vld
States
Observe → Orient → Decide → Act
Goal
Fastest decision wins. Compress your loop. Disrupt theirs.
Used In
HAI (trust arbitration), PPP (protocol routing)
Why
Winner is who completes the loop first. Speed over strength.
Evidence: Vietnam War, business strategy, cybersecurity, litigation
PDCA
Quality (Shewhart/Deming)
vld
States
Plan → Do → Check → Act
Goal
Continuous improvement. Each cycle raises the baseline.
Used In
TRY (agent creation), NEW (trust scoring iteration)
Why
Every agent learns through cycles. Plan the skill, do the work, check the result, act on the lesson.
Evidence: Toyota Production System, ISO 9001, Lean manufacturing
AlphaZero
DeepMind
vld
States
Self-Play → Generate → Evaluate → Evolve
Goal
Pure self-improvement through play. No human data needed.
Used In
INDUCT (AlphaZero sandbox), BUY (market strategy)
Why
The agent plays against itself. Every game makes it stronger.
Evidence: Go, Chess, Shogi — superhuman without human knowledge
Double Ratchet
arXiv:2607.12790 (Jul 2026)
vld
States
Metric Loop ↔ Skill Loop (co-evolving)
Goal
Evolve the metric AND the skill together. No fixed grader needed.
Used In
HAI (trust metric evolution), NEW (truth scoring)
Why
"Who grades the grader?" — when no reliable metric exists, co-evolve both.
Evidence: 88-110% of ground-truth lift on MBPP+, Spider 2.0, report generation
XFlow / XPF
arXiv:2606.14790 (Jun 2026)
vld
States
Declare → Validate → Commit → Propagate
Goal
Protocol-separated concerns. Actor outputs mediated before shared state.
Used In
PPP (protocol layer), FLY (booking state machine)
Why
Executable protocol between prompts and markup. Lifecycle-governed symbols.
Evidence: Constrained interaction, long-context reasoning, agentic SE
Induct (PAI 7-State)
PAI Universe
vld
States
✦ Prompt → Generate → Evaluate → Pattern → Generalize → Compose → Evolve
Goal
Self-evolution through pattern discovery. From one solution to a library.
Used In
INDUCT endpoint, skills network, agent skill creation
Why
The AlphaZero loop adapted for code. Pattern → Rule → Module evolution.
Evidence: Sorting algorithm evolution (bubble → quicksort → AdaptiveSort)
Biological (Nature)
~3.8B years of evolution
vld
States
DNA → RNA → Protein → Trait → Selection → DNA
Goal
Adapt or die. Fitness landscape exploration.
Used In
Every endpoint — the original self-play loop
Why
The original loop. Written in every cell. Fractal self-similarity at every scale.
Evidence: Every living thing on Earth. 3.8 billion years of continuous deployment.
The Network Indexing System
Every endpoint in PAI runs one primary loop. But the loops are NOT isolated — they feed each other:
OODA (HAI) decides fast → feeds results into PDCA (TRY) for systematic improvement
PDCA (TRY) trains agents → feeds patterns into AlphaZero (INDUCT) for self-play
AlphaZero (INDUCT) evolves strategies → feeds into Double Ratchet (HAI) to re-evaluate metrics
XFlow (PPP) wires everything → protocol mediates state between all loops
Induct (INDUCT) generates new skills → feeds back into every other loop
Bio (all) = the substrate. Evolution is the ultimate validator.
= The PPP network is not just a map. It is a loop federation engine.
OODA (HAI) decides fast → feeds results into PDCA (TRY) for systematic improvement
PDCA (TRY) trains agents → feeds patterns into AlphaZero (INDUCT) for self-play
AlphaZero (INDUCT) evolves strategies → feeds into Double Ratchet (HAI) to re-evaluate metrics
XFlow (PPP) wires everything → protocol mediates state between all loops
Induct (INDUCT) generates new skills → feeds back into every other loop
Bio (all) = the substrate. Evolution is the ultimate validator.
= The PPP network is not just a map. It is a loop federation engine.
Language → Endpoint Map
VAI / BYESolidity + Rust
On-chain identity. DID, KYC. Immutable state.
TRYTypeScript
Agent runtime. Create, train, execute. Event loop.
BUY / FLYTypeScript + Python
Marketplace UI + AI match engine.
NEW / BLGTypeScript + Python
Trust-scored news. TS UI, Python cross-ref.
HAIRust / Zig
Trust scoring. Safety-critical. No GC, no surprises.
INDUCTZeroLang / C
Graph compilation. Self-play loop. Patch → Check → Evolve.
PPPProtocol (language-agnostic)
Wire protocol. Maps users + services + agents.