Why our model refuses to give you a number on this stock — and why that protects you

Pillar: Abstention · Educational, not financial advice

Most stock tools have one job: produce a verdict. Buy, sell, hold, a price target, a score. They are built to always answer, because an answer feels like value.

Divergia has a verdict that almost no other tool will give you: "I don't know — and here's exactly why."

That sounds like a weakness. It's the opposite. Here's a real case that shows why.

A valuation that can't make up its mind

Take a well-known semiconductor-equipment company. We ran it through the model and got something strange: depending on a single input, the estimated margin of safety swung from roughly +102% (wildly undervalued) to roughly −82% (wildly overvalued).

Same company. Same day. Same model. A swing from "best opportunity on the board" to "avoid at all costs."

A number that can move that far on one input is not a valuation. It's a coin flip wearing a suit.

What was actually going on

The culprit was mundane and instructive: a stock-split mismatch. The company had done a 10-for-1 split, and the per-share figures and the share count weren't reconciled to the same basis. Use one share count and the company looks impossibly cheap; use the other and it looks impossibly expensive. Both can't be right — and until the two are reconciled, neither number can be trusted.

The tempting move here is to pick the answer that looks reasonable — quietly choose the share count that lands near where analysts already are, and publish a confident number. That's what a tool optimized to always answer would do.

We think that's the single most dangerous thing a model can do: launder a data problem into a clean-looking verdict. Once a wrong number looks confident, you have no way to know it's wrong.

So the honest output is: nothing

When the model detects that its own valuation is unstable for a data-integrity reason it can't resolve, it doesn't pick a side. It flags the name as incomplete and declines to rate it — as a first-class verdict, not an error message buried in a log. The fix on our end is a reconciliation check that catches split-basis mismatches before they reach a valuation; until a name passes that check, abstention is the correct answer, not a placeholder.

Why "I don't know" is rarer and more valuable than it sounds

Think about who can afford to say "I don't know." Not a tool that sells confidence. Not a newsletter that needs a pick this week. Not an algorithm graded on how decisive it sounds.

A tool that abstains is telling you something every confident tool is hiding: that real financial data is messy, that some companies genuinely can't be valued cleanly on a given day, and that the line between a great opportunity and a data error is sometimes one unreconciled share count.

When Divergia does hand you a number, it's because the name cleared the checks that would have triggered an abstention. That's what makes the confident verdicts worth something — they're the ones that survived the model trying to talk itself out of them.

A screener that always has an answer isn't smarter than one that sometimes says "not this one." It's just less honest about what it doesn't know.

Educational/informational content generated by a quantitative model — not financial advice, not a recommendation to buy or sell, and not personalised to your situation. Company specifics are described in general terms; intrinsic values, flags, and abstention verdicts are model outputs, not facts. Do your own research.

Want to see what the model rates with confidence — and what it sets aside? Run a ticker yourself: https://divergia.ai/