Published by BEBEL / Decision systems

What should software do when the data is incomplete?

We built a proprietary decision-system prototype that refuses to turn missing evidence into a confident answer.

Evidence
Known · Estimated · Unknown
Proceed or refuse
Owner decision
Correction history

Evidence stays qualified all the way to a decision. A correction changes the current answer without erasing what came before.

Method: Model the evidence states, enforce explicit refusal boundaries, and exercise both accepted and deliberately faulty cases.
Scope: A private decision-system prototype. The public note excludes its product identity, source implementation, interfaces, and private data.
Result: Missing, estimated, observed, and contradictory information remain distinguishable; unsupported outputs can be refused; corrections remain legible.
Limits: The evidence supports this prototype and its tested boundaries. It does not establish fitness for every decision, production availability, or independent certification.

The problem

Software is often rewarded for producing an answer. That incentive becomes dangerous when the available evidence is partial, estimated, stale, or contradictory. A confident interface can make those differences disappear at exactly the point they matter.

The response

Known is not estimated. Observed, estimated, missing, and contradictory information remain distinguishable.
Corrections remain legible. A new answer supersedes an earlier one; it does not rewrite history.
Refusal is a valid result. Some comparisons or recommendations should not be produced when their basis is inadequate.
The owner decides. The system can organize evidence and enforce its boundaries. Final judgment remains with the person accountable for the outcome.

Publication boundary

The proof is public. The product is not.

This is proprietary product R&D—proprietary research and development—and is not publicly available. This explanation contains no private data, source implementation, internal product identity, or claim that the system is appropriate for every decision.

Tell us what’s hard.

Send the problem, the constraints, and what a useful result would change. BEBEL considers qualified conversations without promising availability.

Product or partnership inquiry Selective client-work inquiry

Please do not send confidential, sensitive, or third-party-restricted information in an initial email. Contacting BEBEL does not create a client relationship; confidentiality begins only under a written agreement.

Both links open your email app at info@bebel.io.