NTJ Japanese Learning Answer Dataset — October 2026
NTJ 日語學習作答彙總資料集 2026 年 10 月

Version 2.0.0 · snapshot 2026-10-21 · temporal coverage 2026-08-02/2026-10-21
543,344 answer events on 25,850 questions.
Canonical page: https://www.nihongotojapan.com/research/japanese-learning-data-2026-10
Previous release (v1.1.0, different analysis sets): https://www.nihongotojapan.com/research/japanese-learning-data-2026

LICENCE
CC-BY-4.0 — https://creativecommons.org/licenses/by/4.0/

CC BY 4.0 covers the data only: the aggregate figures on the report page and the two downloadable files (japanese-learning-2026-10.json and japanese-learning-2026-10.csv). The charts produced by Nihongo to Japan (SVG/PNG) and the text of the report are not covered and remain rights-reserved.

The data is released under CC BY 4.0: you may copy, redistribute, adapt and build upon it, including commercially, provided you give appropriate credit and indicate if changes were made. ⚠️ CC BY is irrevocable — anyone who receives the data under these terms keeps those rights even if the publisher changes the terms later.

ATTRIBUTION (copy this)
Nihongo to Japan, Japanese Learning Answer Dataset — October 2026, v2.0.0, CC BY 4.0, https://www.nihongotojapan.com/research/japanese-learning-data-2026-10

REQUESTED — these are requests, NOT licence conditions
  - Carry the sampling limitations alongside the figures (self-selected observational sample, not representative).
  - Name the analysis set: all answers, item-level first attempts and ability-adjusted figures differ a lot.
  - Say "learner identifiers" rather than "people"; identifier counts are upper bounds on people.
  - State the version (v2.0.0, snapshot 2026-10-21).

NOT COVERED BY THIS LICENCE
  - The charts produced by Nihongo to Japan (public/images/research/2026-10/*.svg and *.png).
  - The text of the report and of this website.
  - The site name, logo and brand marks.

HOW TO READ THE DATA RESPONSIBLY
  - Every figure is an observed value within this dataset, not an inference about any population.
  - Event-level Wilson CIs are not adjusted for repeated answers by the same identifier; they are too narrow.
  - Levels, axes and knowledge points use different question sets; accuracy gaps are not difficulty rankings.
  - No figure in this dataset is a causal claim.

Contact: contact@nihongotojapan.com