Datasets:
Scoring correction (2026-10-05): band-filling convention fixed, all scores re-derived
All participants please read.
A community bug report (#15, #18, by @Yuseon84 ) identified a factor-of-2 in how the band filling was
derived in pipeline/downfold.py. In a non-spin-polarized pw.x run the per-band occupations are per-spin
fractions in [0,1] and the spin factor sits in the k-point weights, so the filling must be the density
n = 2f with delta = |1 - 2f|. dos_at_ef() already applied the spin factor; active_band() did not.
This is now fixed. We re-derived the full active set at the corrected filling and re-deployed. Effects:
- Every material filling delta (hence A_d and the canonical score) is updated under one rule, at once.
- Materials wrongly marked overdoped (score 0) at the wrong filling now score correctly; for example
CuS2 moves to the top of the canonical ranking. - The active-band selector now targets the band nearest half filling rather than the fullest band, which
resolves the multi-band cases (CHf2, CuI2) that had sat high on the un-corrected filling. Where the SCF
outputs were available we re-ran the selection directly; single-band materials are exact from the fix.
The corrected pipeline/downfold.py and active_challenge.csv are live. Your submitted materials are scored
by the corrected canonical values; no re-submission is needed. Credit for the diagnosis goes to @Yuseon84 .
Thank you for the careful, reproducible reports.