Taxonomy Design Principles

Every ConvoTag taxonomy is built against the same design framework. These principles were developed while designing Classic and applied to each taxonomy that followed. What they have in common: symbol systems fail in predictable ways, and the failures are easier to avoid once they’re named.

1. Omission is the default state

An untagged conversation is the default. A symbol earns its place in a taxonomy only if its absence changes a decision — if you would triage the conversation differently without it. If a symbol doesn’t change what you’d do next, it doesn’t belong in the set.

Corollary: “Done” cannot be the omission state. Untagged means “undefined” — finished conversations need their own symbol to be distinguishable from conversations that haven’t been triaged yet. Every ConvoTag taxonomy ships with an explicit terminal symbol.

2. Axes, not flat menus

Every taxonomy operates on one or more conceptual axes.

• Classic operates on two: content/action (what the conversation contains or demands) and lifecycle state (where it sits in its arc).
• Kanban, GitHub, and Notion operate on one: workflow state.
• Eisenhower operates on two: urgency and importance.
• Signal operates on one: health status.

Designing by axis prevents two failure modes:

• Flat menus produce redundant options. Without an organizing principle, designers add symbols that overlap in meaning.
• Flat menus produce gaps. Without axes, entire states go unrepresented because no one noticed they were missing.

ConvoTag brings symbolism to naming conventions as seamlessly as possible.

3. Shape carries primary meaning; color modifies

The primary semantic channel is shape. A question mark means “there is a question” regardless of color. Color serves as an intensity or variant modifier.

Classic example: ❓ and ❔ both mean “has questions.” The filled ❓ is urgent; the outlined ❔ is non-urgent. Same meaning, two intensities.

This matters because mixing color and shape semantically creates collisions. If red means “urgent” on some shapes but “done” on others, users get confused when both reds appear near each other.

4. Established conventions beat custom ones

A user who recognizes the mental model before reading the documentation is a user who adopts the taxonomy quickly. Kanban’s columns are widely understood. The Eisenhower 2×2 is widely understood. Traffic light semantics are near-universal.

When a taxonomy can ride on top of an existing framework, it should.

5. The Miller ceiling

No taxonomy ships with more than 9 symbols. Working memory for novel symbol sets breaks down around 7±2. Classic has 8, the practical upper bound. Kanban and GitHub have 5. Eisenhower, Signal, and Notion have 4.

6. Every symbol must change a decision

For each candidate symbol, ask: what does the user do differently when they see this symbol versus any other symbol in the set?

7. Ease that compounds, not ease that erodes

A good taxonomy gets more useful the more you use it. Bad taxonomies optimize for the first session and degrade over time.

Signs of compounding ease:

• Triage speed increases as the system becomes familiar
• Tag patterns surface meaningful signals — which conversations are always 🔥, which follow predictable sequences
• Tags can be trusted to act on without re-reading titles

Signs of erosion:

• Tagging starts to feel like overhead rather than help
• Tags go stale — conversation states drift from what’s labeled
• Re-reading titles happens anyway, defeating the point

Every taxonomy is evaluated on compounding criteria, not first-impression criteria.