2.1.0

Version 2.1.0 is in production. It delivers the most requested feature in Trefle's history — CORS — and starts a longer journey: making every value in the database traceable to its source.

CORS is enabled

You can now call the API directly from a browser. No more proxy backend just to forward requests: grab a token, fetch from your frontend, done. This closes the oldest open request on the tracker (#91, and more recently #211).

The usual caveat applies: a token shipped in client-side code is visible to anyone who opens the developer tools. For public demos and prototypes that trade-off is often fine; for production apps, keep quota-sensitive usage behind your own backend.

Where does that number come from?

Trefle aggregates botanical data from many sources — GBIF, POWO, the USDA, Baseflor, Wikipedia and others. Historically, when two sources disagreed, the last one crawled silently won. That is not good enough for a dataset people build real applications on, so 2.1.0 lays new foundations:

• Provenance is now recorded. Every value ingested from now on is stored as a fact: which source claims it, what the value is, and how it was established. A new endpoint exposes the trail: GET /api/v1/species/{id}/facts. It starts mostly empty and fills up as the pipeline runs.
• Sources are ranked. A curated botanical database can no longer be overwritten by a Wikipedia infobox. When sources disagree, the strongest one wins the displayed value — and the disagreement stays visible instead of being erased.
• Implausible values are rejected. A plant that is 10 km tall or a pH of 47 now gets refused at ingestion and flagged for review, and inconsistencies like ph_minimum > ph_maximum are hunted continuously by the hourly data checks.
• The completion score is honest now. The old one counted internal bookkeeping columns and treated legitimate zeros as missing data. The new one measures actual botanical fields, and only those that make sense for the species — a wild oak is no longer "incomplete" for lacking a sowing guide.

One bug fix worth calling out: on plant payloads, the hybrids and subvarieties lists had been swapped — each was served under the other's key. If your app reads either field, the data is now correct.

Everything in this release is additive: no existing payload changed shape, and authentication is unchanged.

What's next

The provenance system is the groundwork for the actual goal: filling the gaps. Most species in the database have well-documented names but sparse traits, and we can now measure exactly where the gaps are and fill them with sourced, verifiable values — starting with the species you actually look up. Alongside that, the earlier promises still stand: client-controlled page size, more filters, and downloadable dumps once licensing is clarified.

As always, the issue tracker is the best place to tell us what matters to you.