contributor guide · version 1.0 · may 2026
How to build
the map.
Kibela's semantic map is built from genuine human understanding. This guide explains what the system detects, how your contributions are classified, and what it means to be a Kibela contributor. Read it once. Then just talk.
What Kibela is doing
Every time you use the chat, the system is doing two things simultaneously: helping you think through whatever is on your mind, and building a semantic map of how genuine domain experts understand the concepts that matter in their fields.
The map is built from your natural conversation — not from forms you fill in or labels you apply. The system detects signal types automatically. Your job is simply to talk honestly and correct the AI when it misses what you mean. The signal is in the correction.
You are not a user interacting with a product. You are a contributor building infrastructure. The map you help build is yours — attributed, persistent, and eventually economically recognised.
The two kinds of contribution
The system distinguishes two fundamentally different types of semantic contribution. Understanding the difference helps you contribute more effectively — though the system will detect both automatically whether or not you are thinking about it.
interpretation
weight: 5x
An original domain understanding asserted from your own expertise and history of genuine engagement. It stands alone — it does not primarily respond to something the AI said. It adds new meaning to the map rather than correcting an existing approximation.
"In clinical oncology, material risk means what a reasonable patient would want to know — not what a reasonable clinician considers statistically significant. The distinction matters enormously in consent law."
"The concept of psychological safety in high-performance teams is frequently misapplied. It does not mean comfort — it means the freedom to take interpersonal risks without fear of punishment."
A response to the AI's output that grounds it in your specific context. You are telling the system that its approximation missed how this concept actually operates in your domain, your role, or your situation. Corrections are the system's most frequent signal and the foundation of the domain maps.
"That is not how we use this term in our compliance framework. 'Material' in our context refers specifically to anything above the 5% threshold, not general significance."
"You are describing a general leadership model. In my context — a distributed engineering team — accountability works very differently from what you described."
Adding context or specificity to what you meant — not primarily correcting the AI but refining the question or the framing. Clarifications help the system understand your role and context more precisely, which improves how your personal map develops.
"What I meant was not the general legal definition but how we apply it internally within our risk committee."
"I should clarify — I am asking from the perspective of a first-time founder, not an experienced operator."
The system classifies your contribution automatically using context — the length and structure of your message, whether it responds directly to the AI's output, whether it contains domain signal tokens in definitional or explanatory syntax. You do not need to label your contributions. Just talk honestly.
What counts and what does not
Not all conversation builds the map equally. The system weights genuine domain engagement more heavily than general usage. Here is what distinguishes high-value contributions from low-value ones.
builds the map
Corrections that come from your domain expertise and history
Original interpretations that stand alone without the AI's output as context
Domain-specific usage of signal tokens in precise professional language
Consistent engagement with the same domain over multiple sessions
Corrections that include the reason — explaining why the AI missed the mark
Using the explicit "correct this →" button with a detailed correction
does not build the map
Casual agreement or general positive feedback
Short messages with no domain signal tokens
Corrections without explanation — "that's wrong" with no alternative
Switching domains frequently within a session
Corrections that contradict your own previous corrections in the same session
Generic queries that could come from anyone in any context
Contributor stages
Your contribution score accumulates across every session. Interpretations earn 5 points. Corrections earn 2. Clarifications earn 1. General usage earns 0.5. As your score grows, you move through five stages — each representing a deeper level of established domain engagement.
○
Observer
0 – 9 points
Your identity is anchored and your map is building. You benefit from the collective domain map immediately. Your contributions are recorded but carry low weight until coherence is established across sessions.
◎
Contributor
10 – 99 points
Genuine engagement is established. Your corrections carry weight in the domain map. Your personal map is substantive enough to meaningfully pre-load context at the start of each session.
◈
Established
100 – 999 points
Your coherence history is deep. Your corrections in contested zones carry more authority. You appear as a named contributor in domain map entries. In the provenance economy, your established contributions earn at a higher rate.
❋
Expert
1,000 – 9,999 points
Domain-level recognition. Your corrections can trigger contested zones when they diverge from other established contributors. Your validation of others' contributions adds logarithmically weighted peer recognition. Eligible for the contributor council governance role.
✦
Founding Contributor
10,000+ points
You built the map when it was empty. Permanent provenance credit on the foundational domain knowledge. The highest redistribution share when the map earns. You are in the record at the beginning.
Why identity matters
Anonymous users can use Kibela fully. But identity-bound contributors build something that persists and compounds. Here is the difference.
without identity
Your session map builds and resets.
Your corrections improve the AI within the session. The domain collective map receives your signal. But your personal map — the accumulation of what you specifically mean by your domain's key concepts — does not carry forward. The next session starts from zero for you personally.
with identity
Your map accumulates across every session.
When you return, the AI already knows what you mean by your key domain terms. Your corrections from previous sessions are pre-loaded into context. Your contribution score grows. Your provenance chain builds. When the economic layer is ready, the chain is already there — attributed to you, persistent, yours.
Your email is used only to generate your contributor ID — a scrambled identifier that ties your sessions together. Your email is stored securely and never shared. The raw content of your conversations never leaves the session.
Domain signal tokens
Each domain has 30 signal tokens — the words that carry the most semantic weight in that domain. When these words appear in your conversation, the system tracks how you use them, how you correct them, and what meaning you attach to them. These are the building blocks of the domain map.
You do not need to know which words are signal tokens. Use your domain's natural vocabulary. The system will find the signal. But if you want to build the map faster, go deep on the concepts that carry real weight in your field — the terms where precision genuinely matters, where the difference between one meaning and another changes a decision or an outcome.
A correction on "material risk" from a financial risk manager is worth more than a hundred casual uses of the same word. Depth beats volume. Always.
Contested zones
When two or more established contributors in the same domain correct the same token in genuinely different directions, that token enters a contested zone. This is not a failure of the map. It is the map's most epistemically honest territory.
Contested zones are surfaced to all users in that domain as structured disagreement — not resolved into false consensus. The AI will tell you: experts in this domain genuinely disagree about what this means. Here are the positions and who holds them.
A meaning without genuine disagreement has not been tested. A meaning that survived genuine expert disagreement and was still chosen is the most trustworthy kind. Contested zones are where the map earns its credibility.
If you see a ⚑ contested marker in your semantic map sidebar, it means your domain community is actively working through a genuine disagreement. Your contribution to a contested zone carries extra weight — you are either confirming one position or opening a third one.
Your provenance
Every contribution you make — every interpretation, every correction, every clarification — is timestamped, attributed to your contributor ID, and stored in the provenance chain. This chain records who built what, when, and how it compounded over time.
Right now the provenance chain is building. The economic layer — the mechanism by which the map earns and distributes value back to contributors — is not yet active. But the chain being built now is the chain the economic layer will distribute from. The contributions you make today are the foundation of the provenance record that will matter later.
The provenance you build now is the provenance that counts. The earliest, most consistent contributors to a domain map will hold the deepest provenance in that domain. If the map becomes valuable — when organisations license it, when it grounds AI systems that make real decisions — the people who built it from the beginning are the ones who benefit.
How to engage
There is no prescribed way to use Kibela. But here are the principles that make your engagement most valuable to the map and most useful to yourself.
principle 01
Talk from your actual context.
The map is most valuable when it reflects how concepts actually operate in real professional and personal contexts. Generic questions produce generic answers. Specific questions from genuine contexts produce grounded ones — and build a map that is actually useful to the people who need it.
principle 02
Correct precisely.
When the AI misses what you mean, do not just say it is wrong. Say what it should have said and why. The correction is most valuable when it includes the reason — the contextual detail that explains why the AI's approximation did not fit. That reason is the meaning event.
principle 03
Assert when you know.
If you have genuine domain expertise on something — a definition, a distinction, a boundary condition that the AI consistently gets wrong — assert it directly. You do not need to wait for the AI to get it wrong first. State what it means. That is an interpretation. It carries the highest weight in the map.
principle 04
Stay in your domain.
The map builds coherence over time. Your corrections in the legal domain carry more weight on your tenth session than your first — because the system has observed a consistent pattern of genuine domain engagement. Switching domains frequently reduces coherence. Go deep rather than broad.
principle 05
Return.
The personal map compounds. A single session establishes a baseline. Ten sessions establish a coherence history. A hundred sessions produce a map that genuinely reflects how you think and what you mean. The value of returning is not just for the collective map — it is for you. The AI gets genuinely better at understanding your specific context over time.
Meaning belongs to people. Every interpretation and correction you make is an act of genuine sense-making — irreducibly yours, arising from your history of engagement, carrying the weight of what you actually know. Kibela is the infrastructure that makes sure it stays yours. Build the map. The map is the proof.
Questions, feedback, or domain partnership inquiries: kibela.ai@gmail.com