AR6

Inquiries · open · since 30 Sept 2026 · last tended 16 Oct 2026

What do ecological measurements really contain?

In ecological data, rows are plentiful; independent targets are the scarce resource, and many “measured” values were never measured.i

  1. det. 16 Oct · correctedCorrected: this page said that sixteen trait databases had been audited and that the experiment would take four of them further. The sixteen were listed by a research model, which could not determine their imputed values, and none of them is among the four databases Syntype v0.1 uses. The experiment concerns those four.
  2. det. 30 Sept · interpretedThe question moves from “which model wins” to “what do the tables contain”: how many trait values are derived or estimated, and how much they bias comparisons.
  3. det. 30 Sept · correctedA review finds a survival task that mostly predicts harvest, modelled tree heights, and species reduced to their epithet. Forest results invalid, trait results preliminary.
  4. det. 30 Sept · measuredFirst round of Syntype: the foundation model wins on species traits, not on forest inventories.

Opened by Testing a tabular foundation model on the Mac, then a public benchmark corrected the same day. Roots: Plant identification (Pl@ntNet), herbarium publishing and biodiversity platforms, since 2009.

Position

Never turn an absence of information into a value; publish raw numbers and failures; correct in public.i

What it refuses to claim

  • A random split does not say how well a model predicts for a new plot, site or species.
  • A benchmark score is not a finding about ecology.

Experiments

0 done · 0 running · 4 proposed m

  1. proposed

    Syntype v0.1.1. Fix the four blocking issues and rerun every task. Announced on 30 September; no change since.

  2. proposed

    Derived values in trait databases. In the four trait databases used by Syntype v0.1, how many “measured” values are derived or estimated, and how do they reorder the models?

  3. proposed

    Name only. Does what a language model knows from a species name predict a trait as well as a tabular model?

  4. proposed

    Model, rules or human correction. To annotate what a column means, which helps most: a more accurate model, explicit rules, or a correction that is quick to make?

What it could lead to

  • already feeding Syntype, an open benchmark of tabular models on ecological data. · Syntypei
  • sought Layon, an ecological data tool that does not exist yet: which of four problems is a real pain is still to find out, through interviews not yet started. · Layoni

Reading

The four trait databases and seven forest inventories of Syntype v0.1, all open data; sixteen more trait databases listed by a research model, not yet checked.

From the Syntype repository and notes. The sixteen extra databases come from a review written by a research model on 6 October: it could not determine imputation for them, and two downloads failed.

Crossings

In the register

  1. correction Syntype

    A second model reviewed Syntype v0.1 and found four blocking issues. Forest results declared invalid, trait results preliminary, the same evening. m

    1. det. 30 Sept · correctedA second model’s review finds four blocking issues: forest results invalid, trait results preliminary.
    2. det. 30 Sept · measuredv0.1 published: the foundation model wins on species traits, not on forest inventories.