About the data

Every plant detail page has a "Sources" section at the bottom. The facts on the page come from the combination of primary references, observation data, and editorial judgment listed there.

There are no marks (icons, chips, confidence percentages) next to individual facts. Plant timing is not contested medical or legal information; listing all the sources at the bottom and trusting the reader to assess them is sufficient.

Institutional sources: Japanese government agencies (Tokyo Metropolitan Construction Bureau, Ministry of the Environment Nature Conservation Bureau, National Institute for Environmental Studies), scientific societies (Ecological Society of Japan, Japan Red Data Book), and international taxonomic authorities (RHS, Plants of the World Online).

Community / observation aggregates: observations submitted to iNaturalist, CLIP zero-shot bloom-state estimation from photos (the round-5 phenology POC), and user-uploaded planting records — used to power the "currently blooming nearby" surface.

Editorial / encyclopedia: established Japanese gardening encyclopedias (yasashi.info), NHK hobby-gardening discussion boards. The round-10 text-extraction pipeline ingests these as structured facts.

Catalog editor: the niwa editorial staff. When sources disagree, they pick what to surface.

Principle: institutional > community-aggregate > single anecdote. Newer data outranks older (recency-weighted).

Example: yasashi.info gives Astilbe planting as "early Feb to early March, October to December"; the existing niwa catalog says "March / April / October." These overlap rather than conflict — we take the union (Feb–April + Oct–Dec).

When there is a genuine contradiction, the editor picks, weighted toward the institutional source. That decision is recorded in the source metadata but does not surface on the page.

Niwa doesn't cover every plant in every region in every season. Regional shifts are based on average JMA-derived climate offsets — actual planting dates vary with altitude, microclimate, and cultivar.

iNaturalist observation data is biased toward where observers live. Coverage in less-observed regions is thinner.

Text extraction has errors (~90% accuracy in the round-10 POC). Reader reports of errors are welcome.

Upload your own plant photos at /upload/ (iNaturalist account links if you have one). Submitted observations roll up into the "currently blooming nearby" surface.

The source code and plant data live on GitHub. Pull requests are reviewed by the editor.

The design choices behind this page are informed by trust-UI research:

Liao, Q. V. (2024). Designing Human-AI Interactions for Trust Calibration. — chip-and-dot trust UIs frequently miscalibrate user judgment.

Metzger, M. J., & Flanagin, A. J. (2013). Credibility and trust of information in online environments. — bandwagon heuristics (multiple sources = trustworthy) generate false confidence.

Ding, Y., et al. (2025). Citation Laundering in Generative AI. AAAI. — automated citations can pass provenance checks they should fail.