Human judgment should stay visible.
Understanding a person isn't the same as scoring them.
Libelle is being built to help people reason with better evidence and context without quietly transferring responsibility to an algorithm.
A few things should remain true even as Libelle changes.
These principles guide the current experiment at The Chamber of Us.
People aren't scores.
A person's usefulness, potential, or place in the organization should not collapse into a numerical fit score.
Possibilities aren't decisions.
Something may look worth exploring without becoming an assignment, commitment, or conclusion.
AI can assist. People remain responsible.
Generated interpretation can help a reviewer think. It does not get authority to accept, reject, rank, or assign volunteers.
“Not now” doesn't mean “not useful.”
The right person and the right work do not always appear at the same time. What matters is remembering what we learned instead of starting over later.
Evidence and interpretation stay different.
What someone actually told us should remain distinguishable from anything a system derives or generates later.
Context has boundaries.
Understanding someone better does not mean collecting everything available about them.
What someone said should not quietly become what the system believes.
Libelle keeps different kinds of information conceptually separate so a reviewer can understand where something came from.
Form answers, résumé or CV, submitted links, interests, availability, and other intentional evidence.
For example, information parsed from a résumé. Useful, but still traceable back to its source.
Things worth noticing, possible directions to explore, and important unknowns. Helpful — not authoritative.
The reviewer's own conclusion remains separate from the generated interpretation.
Reach out, wait, revisit, close the loop, or begin exploring something together.
The model can suggest. It doesn't get the final word.
Libelle's use of AI is intentionally narrow. It can help someone understand evidence and think through possibilities. Consequential actions remain human.
A generated possibility does not trigger outreach by itself.
AI-generated interpretation cannot silently become the reviewer's conclusion.
A possibility begins a conversation rather than ending one.
Generated output is assistance, not institutional truth.
Understanding someone better doesn't require knowing everything about them.
The current design favors relevant, intentionally provided information over unrestricted collection. More data is not automatically better context.
In the picture
Not automatically in the picture
People and useful work don't always arrive at the same time.
There are people with capabilities. There are problems that need capabilities. But the timing rarely lines up neatly.
Things change. Don't start over.
People's skills change. Projects change. New opportunities appear. Someone who has no clear next step today may make sense later. Libelle should help us remember what we learned, what we decided, and when it may be worth looking again.
These are commitments, not a claim that we're finished.
Libelle is an open-source project being built and tested at The Chamber of Us. The product is still changing. The point of the experiment is to learn which parts genuinely improve the experience and which ones do not.
We don't want to rescue an unnecessary system with more AI, more automation, or more complexity.
Start with what you want to share.
You don't need to know the perfect role. Tell us about yourself and let a human take it from there.