Elevon in ForbesMany Sensors, Conflicting Reports, One Ranked Decision
We built an end-to-end crisis response suite for ourselves: fuse incoming reports from many sources, resolve the contradictions between them, and produce a ranked set of options with the reasoning attached.
Client
Elevon (internal)
Industry
Public sector & safety
Solution
Internal reference suite (Elevon platform)
Deployment
Internal build, not client-deployed
Built as an internal reference implementation. No client data, no live deployment, and every design decision written down so it can be rebuilt for a real operator.
The Problem We Set
In any crisis the problem is rarely a lack of information. It is that reports arrive from many places at once, in different formats, at different levels of reliability, and several of them contradict each other.
A person under time pressure then does the reconciliation by hand, which is slow, and does it differently each time, which means two people looking at the same picture reach different conclusions.
We wanted to know what a governed suite would actually have to do to help there, so we built one rather than writing a proposal about it.
How We Designed It
The design principle throughout: the suite may rank and explain, but it never quietly discards a report a human has not seen.
Fusion before interpretation
Incoming reports are normalized into one shape and deduplicated first. Only then does anything try to interpret them, because interpreting the same event three times produces three different situations.
Contradictions are surfaced, not resolved silently
When two sources disagree, the suite does not pick a winner behind the scenes. It marks the conflict, states which source says what, and carries that disagreement into the recommendation.
Options with reasoning, not one answer
The output is a small ranked set of courses of action, each with what supports it and what argues against it. A single confident answer is the one thing a decision maker under pressure cannot check.
“A recommendation you cannot argue with is not a recommendation. It is an instruction, and in a crisis that is the wrong shape of help.”
How the Suite Runs
The suite ingests reports, normalizes and deduplicates them, builds a single current picture, flags where sources conflict, and then generates candidate courses of action ranked with their supporting and opposing evidence.
Everything that can be computed is computed in code rather than generated, so the same set of reports always produces the same picture. Language models are used where judgment is genuinely needed, not for arithmetic or ranking.
What this looks like in practice
The suite is packaged as a self-contained decision package: the current picture, the conflicts, the ranked options and the trail of what produced each one, in a single document a person can read end to end.
What It Produces
This is an internal build. There are no client outcomes to report and none are claimed. What follows is what the suite does.
Reports from many sources normalized and deduplicated into one current picture
Source conflicts flagged explicitly and carried through to the recommendation
A ranked set of courses of action, each with supporting and opposing evidence
Deterministic where it can be, so the same inputs always produce the same picture
One readable decision package rather than a dashboard to interpret
“We built it to find out what the hard parts really are. Fusion and contradiction handling turned out to be the whole job, and the recommendation was the easy end.”
What We Learned
Building it taught us more than specifying it would have. The genuinely hard part is not the recommendation, it is deciding what to do when two trusted sources describe the same event differently.
Keeping computation in code rather than in a model is what makes the output defensible. In a domain where a decision gets reviewed afterwards, a picture that changes between runs is worse than no picture.
The same structure applies well beyond crisis work. Any situation with many partial sources, real time pressure and a decision that has to be justified afterwards has the same shape.
Want a similar transformation in your organization?
Let's talk about how Elevon can help your team too.
Book consultation