OCT 13, 2026 · Anime

Anime

What Changed When We Fixed the Data

We refreshed the MyAnimeList catalog behind ANIME after Jikan shut down, moved to the Tenrai-compatible API, and recomputed every headline on the clean TV frame. The light-novel reach premium fell from 32× to 27.5×; manga's 8.0+ share fell to 14.8%; 2015 isekai was under-counted at 0.9% (now 2.2%).

By Kyle McAuliffe · 2026-10-13 · 3 min

Executive summary. The ANIME report's core story survives on the October 2026 ledger: source material still sets the audience floor, and isekai tags trace a real production wave. What changed is measurement. When MyAnimeList's public Jikan mirror shut down on 2026-10-01, we finished the refresh through Tenrai (api.tenrai.org, Jikan-compatible) and recomputed every headline on anime_2026-10-11_clean with one script. The light-novel reach multiple narrows from 32.0× to 27.5×. Manga's high-score share falls from 17.0% to 14.8%. The 2015 isekai point was an under-tag in the old exploratory frame (0.9% → 2.2%); 2021 stays 20.4% and 2024 edges to 18.1%. Single-year shares are trends, not precise point estimates — bootstrap ranges are about ±5 percentage points.

WHY WE REFRESHED

The published report rested on the TidyTuesday export of 2019-04-23 (13,631 unique titles). A parallel pipeline had already grown the catalog to 18,818 rows with AniList tags, streaming links, and season fields. On 2026-10-01, MyAnimeList discontinued the free Jikan v4 API (api.jikan.moe). We completed metadata fetches through Tenrai, which mirrors the same Jikan field schema, then ran scripts/anime-data/clean_audit.py (seed 20261011) to produce anime_2026-10-11_clean.* with documented inclusion rules and bootstrap ranges.

OLD VS NEW

Published / exploratory is what readers saw before this audit. Clean new is the October 2026 TV frame from the clean file; brackets are 95% bootstrap ranges where shown.

WHAT WE CHECKED

We drew n = 200 rows with both Tenrai/Jikan and AniList fields present (seed 20261011). Score agreement within 0.5 points on the MAL 0–10 scale (|MAL − AniList meanScore/10| ≤ 0.5): 146 / 200 = 73%. This is an offline check of already-joined fields — not a live re-fetch at audit time. Year agreement between aired_from_year and season_year was weaker (59.5%), which is why calendar-year shares carry wide bootstrap bands.

Cleaning removed 5 rows missing id/name, normalized source labels, and nulled five end-before-start dates. No scores were imputed. Full PRISMA counts and rules R1–R7 are in data/anime/CLEANING_LOG.md.

HOW TO READ SINGLE-YEAR SHARES

For any one premiere year, the 95% bootstrap interval is typically about ±5 percentage points around the point estimate. That is wide enough that 17.7% vs 18.1% for 2024 isekai should be read as still elevated after the 2021 peak, not as proof of a 0.4-point decline. The charts in the main report show the full 2010–2025 trend lines and stacked source mix instead of over-interpreting one year.