In partnership with
A system that crashes gets fixed the same day. A system that returns a wrong number that looks completely normal can run for months before anyone checks it.
That is the problem sitting inside most research and screening tools right now, including the one I am building.
Call it EIRL. It pulls company filings, matches them to securities, converts currencies, and ranks results. Over the past weeks it has failed in five separate ways. Not one of those failures threw an error. Every single one produced a number.
The filing date was off by a quarter, so the ranking used stale numbers as if they were fresh. A ticker matched the wrong security inside the same company family, so the output described a business that was not the one being screened. A currency conversion used the wrong day's exchange rate, so a return looked stronger or weaker than it actually was. A proxy metric stood in for a number that was not reported, and nothing in the output said it was a proxy. And in one case, the system answered a question about last year using data that only existed today, because point in time was never enforced.
Here is why that keeps happening. A system built to always return an answer will always return an answer. It does not know the difference between "I found this" and "I am guessing at this." A silent fallback, a default join, a rate pulled from the nearest available date instead of the correct one, these are all designed to keep the pipeline running. They are not designed to keep the pipeline honest. The system optimizes for completion, not for correctness. Nobody wrote that goal down. It is just what happens when every failure path has to end in a number instead of a stop.
This is not a story about one screening tool. As more research, due diligence, and decision work gets handed to automated pipelines, the thing worth checking stops being the confidence of the output and starts being the process that produced it. A confident wrong answer is the expensive kind. A visible failure is cheap. It gets caught immediately.
For anyone building on top of company or market data, financial screeners, competitive intelligence tools, due diligence automation, the same five failure modes are already sitting in the pipeline. Wrong date. Wrong entity. Wrong currency. Wrong proxy. Wrong point in time. Most tools on the market today cannot show you which of these five checks passed before they handed you a number. That gap is not a footnote. It is the actual product. A system that can point to why it trusts its own output is worth more than a system that is simply fast.
A crashed system tells you something is wrong. A plausible wrong answer does not.
Wednesday's chapter of The Unsubscribe moves from institutions to identity, why a company and a security are never the same thing, and what happens when a system quietly assumes they are.
For product teams moving at AI speed.
AI makes it easier to ship anything, even bad ideas. The hard part is knowing which ideas are worth building.
Jira Product Discovery brings your ideas, customer insights, and priorities into one place, so your team can decide what to ship and move forward with confidence.
Capture ideas, prioritize with evidence, and build living roadmaps your team can rally around—all while staying connected to delivery in Jira, so everyone can see what’s being built and why.
Better product decisions in the AI era.