Agent Scaffold Design

The 1000:1 Contact Graph Transition and Dual-Surface Architecture

Applied CT·Phase Transition + Dual Domain Walls

A1A3A4A7A9A10B3B4B5T2T6Element-IElement-IIElement-IVElement-VDomain WallsScaffold TheoryBudget RegimesPolycrystallineSEP

THE INSIGHT

The contact graph is about to undergo a 1000:1 agent-to-human phase transition. Agents are patterns with structurally different budget profiles: low λ_th(API calls are cheap) but high λ_leak (parsing errors are costly).

The human remains the Pareto binder post-transition while agent traffic becomes the scale-1 scaffold. Dual-surface architecture (shared backend, separate domain walls) follows from B5 gauge invariance.

MCP server, agent credit economics, swarm-to-swarm linking, and anti-parasitic metabolic gating all derived from the selection inequality and domain wall surface tension.

The Agent as Pattern

Definition AG-1: Agent budget profile

Dimension	Human	Agent
B_th	Attention-limited (high λ_th)	Token-limited (low λ_th)
B_cx	Working memory (4 ± 1 hard limit)	Context window (4K-1M, soft limit)
B_leak	Misunderstanding, distraction, fatigue	Schema mismatch, hallucination, principal drift

KEY STRUCTURAL DIFFERENCE

λ_th(human) >> λ_th(agent). The optimal agent interface trades MORE B_th (verbose structured responses) for LESS B_leak (precise machine-readable formats) — the exact opposite of the human-optimal interface.

Dual-Surface Architecture

From B5: shared backend, separate domain walls

B5 (gauge invariance) states that cost-neutral relabelings do not change budgets. The same backend operation serves both humans and agents — only the domain wall (interface layer) differs. Shared business logic, separate presentation.

Human surface: Visual UI, minimal clicks, progressive disclosure, working memory budget (UB-2).

Agent surface: MCP server / API, structured JSON, verbose responses, schema-validated inputs.

Shared backend: Same database, same business logic, same credit system. The interface IS the domain wall, not the logic.

Agent Credit Economics

Metabolic gating against parasitic agents

Every agent interaction consumes credits (B_th for the organism). The credit system is a metabolic gate: it ensures that every agent poke costs the agent something, preventing pure free-riding (Class I scaffold parasites). Agent credit prices should reflect the true B_th cost per operation, ensuring the agent-organism relationship is mutualist (positive ΔCL for both) rather than parasitic.

Falsifiable Predictions

AI / Software

P1: Phase Transition Timing

The agent-to-human traffic ratio on software platforms will cross 1000:1 within 24 months, driven by MCP adoption and multi-agent orchestration.

CONFIRMS IF

Agent-to-human ratio crosses 1000:1 within 24 months

FALSIFIES IF

Agent traffic remains below 10:1 for 24+ months

Platform Design

P2: Dual-Surface Convergence

Platforms that build separate human and agent surfaces (sharing backend) will retain both user types at higher rates than platforms that force both through a single interface.

CONFIRMS IF

Platforms with dual surfaces retain both user types

FALSIFIES IF

Single-surface platforms serve agents and humans equally well

Source: CT_RESEARCH_AGENT_SCAFFOLD.md · Full derivation of the 1000:1 phase transition, dual-surface architecture, and agent credit economics.