DIIICE
Six types. Every DOT you'll ever write fits one of them.
DIIICE is the universal taxonomy for human knowledge. Every insight, instruction, idea, piece of context, data point, or example — anything you'd want an AI to know about you — belongs to exactly one of six DIIICE types.
When To Use This
DIIICEis for you if…
- Your second brain has 100+ notes and search has started to lie to you.
- You've invented the same tag three different ways and can't remember which one you used last.
- You want AI to reason over your knowledge, but every DOT looks identical to the retriever.
- You're tired of knowledge graveyards. You want a system that gets cleaner as it grows, not messier.
The Parts
What's Inside DIIICE
Raw facts. Numbers, records, measurements. The ground truth.
E.g.Q3 revenue: $847K. Churn: 2.1%. Pipeline: 142 deals.
Synthesized insights. What the data means. Patterns, lessons, conclusions.
E.g.Enterprise deals close 3× faster when the champion already uses the free tier.
How to do something. Runbooks, procedures, step-by-step recipes.
E.g.To onboard a new agent: run `cc agent init`, attach HUMAN.md, verify with a dry run.
Unproven thoughts. Hypotheses, proposals, creative sparks that haven't been tested yet.
E.g.What if the pricing page showed a live counter of DOTs created in the last hour?
The situation around a fact. History, relationships, constraints, people.
E.g.We chose Supabase because Railway + Postgres had latency issues in us-west-2 during our pilot.
Demonstrated patterns. Samples of what good looks like. Few-shot material for AI.
E.g.Here are five cold emails that landed demo calls — use the pattern, not the exact words.
Why Six Types
Before DIIICE, knowledge management was a graveyard of tags. Everyone made up their own labels, and search devolved into guessing which synonym someone used two years ago.
DIIICE fixes that by being complete and mutually exclusive. Every piece of knowledge fits exactly one of the six types — and together they cover the full surface of what an AI needs to know about you.
It's the periodic table for human context. Not cute. Not aspirational. Exhaustive.
How To Classify In Under Five Seconds
Ask the questions in order. Stop at the first yes.
1. Is it a raw fact? → Data
2. Is it a conclusion drawn from data? → Intelligence
3. Does it tell someone how to do something? → Instructions
4. Is it untested — a proposal? → Ideas
5. Does it describe the situation around a fact? → Context
6. Is it a sample of what good looks like? → Examples
If you can't pick in five seconds, the DOT is too big. Split it.
The AI Side Of DIIICE
When a model retrieves context, knowing the type changes how it uses the content.
- Data goes straight into the answer without reinterpretation.
- Intelligence becomes the model's prior — it shapes every downstream reasoning step.
- Instructions get followed literally.
- Ideas get flagged as speculative so the model doesn't treat them as truth.
- Context scopes the answer — "given this situation, here's what applies."
- Examples seed few-shot behavior without needing the model to generalize first.
Without DIIICE, all context looks the same to a retrieval system. With DIIICE, the model knows which pieces to trust, which to follow, and which to doubt.
The DOT That Started It All
DIIICE came out of one observation: every framework I'd tried for organizing my second brain collapsed within a month.
Tags proliferated. Folders got stale. "Just search it" fell apart once I had 10,000 notes and the same idea existed under four different phrasings.
I needed something a) small enough to remember, b) complete enough to cover everything, c) rigid enough to keep me honest. Six letters. No exceptions. No drift.
That's DIIICE. And once it existed, the Vault could grow to 43,000+ DOTs without getting harder to navigate.
The Traps
Common Mistakes With DIIICE
Tagging everything as Context
Context has gravity — it feels right for almost anything. That's the trap. Context is the situation *around* a fact, not the fact itself. When you're unsure, it's almost always Data or Intelligence. Use the five-second decision tree, in order.
Skipping classification to save time
An unclassified DOT is a DOT you'll never find again. The five seconds you save now becomes thirty minutes of searching later — or, worse, a retrieval system that can't tell a fact from a hypothesis. Classification isn't overhead. It's the whole point.
Hiding two DOTs inside one
If a note contains a fact AND the conclusion drawn from that fact, you have two DOTs glued together. Data and Intelligence are siblings, not twins. Split them. The graph is denser with ten small DOTs than one sprawling one.
Objections Answered
But What About…
Tags multiply. Tags drift. Inside a year you'll have invented 'todo', 'to-do', and 'task' for the same concept. DIIICE has six types and six is the ceiling — forever. The rigidity is the feature.