Why the model sometimes refuses to give you a number

Every valuation tool will hand you a number if you ask for one. Type a ticker into most DCF calculators and something comes out the other end, no matter how thin the inputs. That's the easy part. The hard part is knowing when the number you'd get isn't worth trusting, and saying so instead of publishing it anyway.

Aperto runs several independent valuation lenses on every company: a discounted cash flow build, comparable-multiples reads, sometimes a dividend discount model, depending on what fits the business. Each lens produces its own estimate. Most of the time they land in a similar neighborhood, and Aperto blends them into a range with a single published figure. Sometimes they don't agree at all — one lens says the stock is worth half what another says — and that disagreement is itself information. It usually means the business is hard to model cleanly: unstable margins, a recent restructuring, a sector where multiples are swinging wildly. Publishing a confident-looking blended number in that situation would hide the real story, which is that nobody should feel confident here.

So Aperto measures how much the lenses disagree, and above a fixed threshold it refuses to publish a point estimate. You still get the range, you still get every lens's individual read, and you get the reason stated in plain terms. What you don't get is a fake middle ground.

Does refusing actually help, or is it just hedging?

That's a fair question, and it's testable. We ran Aperto's downside-weighted valuation against a naive one-stage DCF — the simplest model that still counts as a valuation, no lens blending, no refusal logic, one growth rate and one discount rate in, one number out — across the same set of companies. The naive model's estimates missed the actual outcome by a median of 62.0%. Aperto's missed by 22.7%. Part of that gap comes from better modeling. Part of it comes from the refusal: on the names where the naive model was guessing into real uncertainty, Aperto declined to guess at all, and that decision alone accounts for some of the error the naive model absorbed.

In other words, refusal isn't the model shrugging. It's the model correctly identifying the cases where a confident number would have been wrong, and choosing not to produce one.

What you see instead

A refused card isn't a blank page. You still get:

• The full range across every valuation lens that ran
• Each lens's individual estimate, so you can see where the disagreement actually sits
• The specific reason the refusal fired — which lenses diverged, and by how much

If you want a single number badly enough, you can build your own view from those pieces. What Aperto won't do is manufacture false agreement between models that don't actually agree, just because a chart with one line looks more finished than a chart with three.