The Horizon MandateOrchestrating Professional Reliability in the AI Era
The tools will change. The discipline endures.
Skip to content
The Thesis
Steven Grambow, PhD | Associate Professor of Biostatistics & Bioinformatics
Biomedical research is currently operating in a state of high-velocity disorientation, where the rapid cadence of Large Language Model (LLM) capabilities often distracts from the fundamental requirements of scientific integrity. This presentation argues that focusing on the "view from the window"—the transient features of specific chatbots—is a strategic error. Instead, we must adopt the Horizon Mandate: a durable, model-agnostic framework for stewardship that ensures professional reliability regardless of technological shifts.
The central challenge is the "Stochastic Trap." Because generative AI models are probabilistic next-token predictors, their outputs trend toward average patterns in training data unless actively directed. Without stewardship, these models function like a "TEEMP INERN"—enthusiastic and fast, but lacking judgment and precision. To counteract this, researchers must transition from the role of passive "Prompter" to active "Steward," orchestrating workflows that provide the context and constraints necessary to move outputs from generic to expert-specific.
The misspelling is the point—unsteered AI confidently gets it wrong.
"Process Provenance is the new standard of rigor. The defensibility of scientific output depends not on the tool used, but on the transparency of the human orchestration behind it."
The shift from passive Prompter to active Steward
The Shift: From Chat to Orchestration
"View from the Window" vs. "Horizon Mandate"
The View from the Window is the transient state: which model tops the leaderboard this week, which chatbot has the newest feature. It changes constantly and rewards reactive adoption.
The Horizon Mandate is the durable investment: building workflows, context infrastructure, and governance habits that compound over time regardless of which model you're using.
The Evolution: Chat → Library → Workbench
Phase 1 — Stateless Chat: "The Cocktail Party" — conversational, forgetful, disconnected.
Phase 2 — Contextual Tools: "The Library" — grounded in data, RAG-enabled, capable of reference.
Phase 3 — Intelligent Agents: "The Workbench" — goal-oriented, tool-using, multi-step reasoning.
We're not replacing chat—we're adding layers of capability.
The Framework: Three Operational Pillars
🚦 1. Data Sovereignty (The Sovereignty Gate)
The "Traffic Light Protocol" for risk management delineates safe operating zones:
- Green Zone: Public data synthesis — open web, published literature
- Yellow Zone: Internal IP — Enterprise tools (Microsoft Copilot) or Local-First architectures (Obsidian, LM Studio) to prevent leakage
- Red Zone: PHI — Zero-trust computing within secure enclaves (PACE/PRCC) is mandatory
🏗️ 2. Context Engineering
A structured workflow to mitigate documented model pathologies:
- Context Rot: Performance degradation as input length increases
- Lost-in-the-Middle: Models under-utilize information placed in mid-context positions
- Cognitive Load: Models have attention limits — overload degrades output
Key Principle: Context stuffing is not context engineering. Curate, don't dump.
🏃 3. Personal Infrastructure (Crawl-Walk-Run)
A roadmap for adoption that moves from stateless chat to durable assets:
- CRAWL — Portable Context: Build a reusable
Context.mdwith Bio, Persona, and Constraints. Paste it at session start to force the model off baseline. - WALK — The Reset Protocol: Summarize the win → Close chat → Open fresh → Paste Summary. Counteract context rot.
- RUN — Validated Workflows: Convert personal wins into shared Team Prompts. Move from "I prompted this" to "We built a workflow for this."
📘 Go Deeper: The Crawl-Walk-Run framework originated in The Bootstrapper's Playbook—a companion resource covering the "Portable AI Brain," Process Provenance, and templates for building your personal AI operating system.
We are stewards, not prompters.
Build the discipline. Protect the integrity. Orchestrate reliability.
📄 Download Slides"Amateurs prompt. Professionals build."
Written by a human | Boosted by AI | Transparency is part of the process