What is DS-NEXUS — and why does it exist?

The problem

When you talk to an LLM without any particular setup, it defaults to a communication style optimized for the average user: moderate density, some social warmth, step-by-step explanations, a trailing summary. This works for nobody in particular and adequately for almost everyone.

That's fine if "adequate" is what you're after. But if you've ever found yourself skimming through paragraphs of preamble to find the one sentence that matters, or feeling like an explanation was talking down to you, or wishing the AI would just get to the point — you've felt the friction that DS-NEXUS is designed to eliminate.

The insight

People process information differently — not better or worse, differently. Some people need the conclusion first and find all preamble irritating. Some need the hook before the detail or the concept doesn't stick. Some think spatially and compress complex systems into 3D structures in their heads. Some need the social context to feel safe enough to explore unconventional ideas.

These aren't preferences you can easily articulate in a single system prompt like "be more concise" or "explain things simply." They're structural — they describe how your working memory loads information, how you retrieve it, how you validate it.

DS-NEXUS maps this structure. Not as a personality test, but as a functional communication calibration: if your input is processed this way, then output should be structured this way.

How it works

The protocol has three components:

1. A calibration questionnaire. 12 situational questions — not abstract ("do you prefer visuals?") but behavioral ("when you're stuck on a problem, you tend to..."). Situational questions are harder to game and more predictive.

2. Five reference profiles. Each profile is a complete system prompt that configures an LLM across six cognitive dimensions: logic, density, spatial, specificity, social context, and depth. You paste it into any LLM — Claude, ChatGPT, Gemini — and the calibration kicks in immediately.

3. A scoring algorithm. The calibrator maps your questionnaire answers to a 6-dimensional vector and computes Euclidean distance to each reference profile. The closest one becomes your starting point.

The five profiles

Each profile specifies not just tone but structure: the density target in bits/word, which analogy level to use (surface, structural, or isomorphic), how to handle ambiguity, what operation modes to activate under different triggers, and which anti-patterns to avoid. It's a complete behavioral specification, not a vibe.

Does it actually work?

DS-NEXUS has been applied to four real projects. The pattern is consistent:

The consistent variable across all four: 5–10× faster execution compared to uncalibrated AI collaboration. The mechanism is straightforward — when the AI's output matches your processing architecture, less time is spent re-reading, re-asking, and filtering noise. The cognitive load drops, and throughput goes up.

What it isn't

DS-NEXUS is not a personality framework. It doesn't claim to describe who you are — only how you process information in a specific context. You can switch profiles mid-project. You can use NEXUS-E for executive decisions and NEXUS-L to learn a new domain and NEXUS-D for technical deep dives. The protocol is a tool, not an identity.

It's also not magic. A well-calibrated system prompt doesn't make a bad LLM good. It makes a good LLM significantly more useful for your specific way of working — by eliminating the friction between how the model communicates and how you receive information.

Where it's going

v3.0 extends the protocol to all five reference profiles, pairs them with a calibrator that maps you to your closest starting point, and adds a feedback loop so the calibration improves with use. The framework is open source and open to community contribution. The goal: every person who uses AI regularly should have a protocol that's calibrated to how they actually think.

The calibrator is already live. The profiles are open source. The rest is iteration.