How we work it out

The current US National Foreign Affairs Training Center page publishes timelines to an integrated ILR 3 in Speaking and Listening. Model nfatc-integrated-ilr3-v2 uses that endpoint and was verified 2026-09-01.

Category I is a real range: the current page lists French and Spanish at 30 weeks and seven other languages at 24. Their 690 and 552 values are 30×23 and 24×23 from the page's typical class week. Categories II–IV print 828, 1,012 and 2,200 directly.

A typical week also includes 17 hours of self-study. The model anchors its method comparisons to source-backed class-contact hours and does not silently add those homework hours to every method. Category IV is preserved at the page's explicit 2,200 even though 88×23 would be 2,024. Category IV explicitly says 2,200 class hours for 88 weeks even though the page also describes a typical week as 23 class hours. The model preserves the explicit 2,200 rather than recomputing it.

The current NFATC page publishes timelines to one endpoint. Most people are not aiming at it. We scale the baseline using the shape of the published Cambridge guided-learning-hour bands, normalised against C1.

• A1
  ×0.13
• A2
  ×0.27
• B1
  ×0.5
• B2
  ×0.73
• C1
  ×1
• C2
  ×1.47

Estimated. Cambridge guided-learning-hour bands per CEFR level, normalised against the C1 band to approximate the non-linear shape of the curve; absolute hours come from the NFATC baseline, not from Cambridge. This is not a level-by-level fit: for example, 0.27 × the 750-hour C1 midpoint gives 202.5 hours at A2, slightly above Cambridge's 180–200 band.

This rests on treating NFATC's integrated oral endpoint as roughly C1 solely for the planning curve. No official ILR↔CEFR crosswalk exists, and the current NFATC endpoint is an integrated Speaking + Listening score rather than the S-3/R-3 pair used by the separate Foreign Affairs Manual course table.

The model assumes a single multiplier per method would misprice both ends. Recognition-heavy app study gets a widening estimated penalty because it supplies less spontaneous production and correction; the immersion assumption runs the other way. These curves are judgement calls, not findings from the NFATC page.

Estimated. Reasoned from what each method provides — corrective feedback, forced output, unpredictable input — relative to the intensive programme context the NFATC timelines describe. No published research assigns coefficients to these categories.

The ceiling column matters more than the multipliers. When your target sits above your method's ceiling, the calculator says so instead of quoting a very large number of hours. Model assumption: apps are efficient for early vocabulary and basic patterns, then become less efficient because they are recognition-heavy. The 1.3×–3.5× curve and B1 ceiling are conservative planning judgements, not limits established by NFATC.

Empirical third-language studies document cross-linguistic influence, including effects of structural similarity. They do not publish transferable hours-saved coefficients for our language pairs, so every discount here is an explicit model assumption and the combined reduction is capped.

Estimated. An empirical third-language study found cross-linguistic influence and reported structural similarity as more predictive than broad typological proximity in its tasks. It supplies no hours-saved coefficient for our pairs. The discounts here are explicit model assumptions: shared branch gets the largest reduction, shared script a smaller one, and the total is capped.

Hanban's claimed 1:1 HSK↔CEFR equivalence was rejected by CEFR bodies. German and French associations of Chinese-language teachers place old HSK 6 around B2–C1, not C2. No official mapping exists.

We map low. Telling you HSK 6 is C2 would shorten every estimate on this site, which is precisely why we do not.

Nothing here predicts your outcome. It is a planning figure built from a published or transparently derived programme baseline and four judgement calls, and individual variation swamps all of it. Treat a result as an order of magnitude, not a date.

We also do not test these courses at scale, which is why you will not find star ratings anywhere on this site.