What it means when a calculator tells you no

Ask many language calculators for C1 Japanese by app and you will get a large number. This model instead applies a conservative app-only ceiling. Neither choice is neutral: one assumes hours can be extrapolated indefinitely, while the other assumes the method category stops at B1.

Ours returns a refusal instead, and that decision is worth explaining because it looks like a missing feature.

Where the refusals come from

Each study method has a ceiling — the highest level the model treats it as capable of producing on its own:

Ask for a level above the ceiling and the model does not present the arithmetic as a prediction. It says the target crosses its assumed method cap.

The same logic runs through the tools. The daily-time tool returns no figure when a target crosses the selected method's modelled ceiling, because dividing the pre-ceiling arithmetic by days would hide that assumption. The deadline tool distinguishes running out of time from crossing a modelled ceiling because they call for different plan changes.

Why a big number is worse than no number

A learner who is told "1,486 hours to B2 by app" may reasonably read it as a prediction. Here it is the pre-ceiling arithmetic, not an observed outcome.

The two look identical when both are printed as a number, and they call for completely different decisions:

• Long project → keep going, adjust the schedule
• Crosses this model's ceiling → change the method or challenge the assumption

Presenting the second as the first sends people to spend years on the wrong plan. The number is not even wrong, exactly — 1,486 is what the arithmetic gives. It is that arithmetic is the wrong tool once the question has become qualitative.

This is also why the method chart on every language page hatches a bar above the method's assumed ceiling rather than presenting it as an ordinary long estimate. The label now says the model caps the method, so a screenshot cannot turn the assumption into a universal claim.

Where the ceilings come from, honestly

They are estimates. The cited NFATC timeline publishes no method ceilings, and we found no published controlled comparison that supports the specific app-only ceilings in this model.

Their rationale is structural rather than experimental: upper levels demand more open-ended production and corrective feedback, while the model's app-only category assumes neither is sustained. A product or routine that supplies both no longer matches that input. The ceilings encode this argument and are marked as estimates in the code and every result caveat.

If you think a ceiling is wrong, the specific claim to argue with is that structural one, not the number.

The other refusals

The pattern runs through the site.

Ask what a language you already speak is worth for an unrelated pair and it returns zero, rather than a small encouraging discount. Ask for a language the current NFATC page does not name — sign language, Latin — and there is no current baseline, with pages explaining why instead.

The rule underneath all of it: an answer that looks like an answer, and is not, is worse than an admission. It is also the whole reason to build a tool like this, because the pages competing for the same searches are full of the first kind.