Minwoo Kim, Founder & Chief AI Officer, Calamophyton Industries · November 3, 2026
When a decision goes wrong, people ask who made it. That's the wrong question. The only one that matters: what was it standing on?
Most organizations can't answer. The conclusion survives. The ground under it is gone. We believe this is the most common and least discussed reason high-consequence decisions fail.
We did what no one asked for
We set out to build a skeleton for an early review. The system had already built something far deeper. The evidence kept coming.
We didn't see excess. We saw the minimum a decision should rest on. The response was different: "Who looks this deep before deciding anymore? You don't need all this."
We sat with that for a long time. Then we understood. The problem wasn't depth. It was a habit of deciding without ground. Where that habit rules, ground looks like waste.
So we don't search less. We surface only the layer the current decision needs. The rest stays below, intact, for the day someone asks what that decision was standing on.
Same evidence, different judgments
Give two organizations the same material and they will see different things. That is legitimate. But when finding evidence and interpreting it happen in the same layer, you lose track of whose judgment it was and what it rested on.
We separated them. The layer that finds evidence does not judge. Judgment belongs to people, and to the institutions they answer to.
A summary is not an answer
At first we defined a source as a link. We were wrong. Most evidence in the field has no link: a scanned page, meeting minutes, a handwritten note. Accept only what links, and you lose what matters most.
Someone told us then: "So it's a document summarizer that saves reading time."
Fair. A summary makes the conclusion shorter. What we had to build was the ground beneath it.
So we set one rule
To count as evidence, an item carries three things. Where it came from. As of when. Which version it is.
Miss one, and it does not move forward. Nor is it discarded. It is marked missing, and we follow what's missing to the end. A source is not only a link. A page number, a timestamp, the place an original sits — all of it is source.
This is what HARBINGER does. It doesn't decide. It builds the ground a decision can stand on.
Conclusions can change, as long as the ground remains.