The preregistration called for median absolute relative error, signed bias, worst overprediction factor, and leave-one-out ranges over at least four Fe-W references. The panel contains at most ONE signed error (alpha-Fe control, +6.7%). All four statistics are therefore undefined for Fe-W intermetallic chemistry.
The empirical calibration is invalid for Fe-W intermetallic chemistry at frozen v1 settings. One control on an endpoint element does not calibrate a compound space; no Fe-W intermetallic reference converged under the frozen protocol (WFe2 C14 failed SCF in both seed arms of the retargeted sensitivity test).
bcc-Fe anchor error: +6.7% (route 2.2967 T vs 2.152 T experimental). Direction: overprediction. Single sample; no distributional claim.
WFe2 two-temperature bound: NOT computable — the revised seed-sensitivity pair did not converge, so the conditional bound contemplated at the checkpoint lapses.
The Fe17W3 route observation of 1.7402 T sits in a chemistry the calibration cannot bound. The only element-level anchor (alpha-Fe) overpredicts by ~7%, and the known failure mode of semilocal DFT in Fe-rich intermetallics is overestimation of itinerant moments, so the plausible envelope tilts DOWNWARD for the 1.7402 T figure — but with a single anchor and zero Fe-W references, no numerical correction factor is defensible.
The candidates-dataset observation of 1.7402 T is NOT replaced with a corrected value. It stands as a route observation with an unbounded, direction-informed uncertainty: treat as an upper-leaning estimate pending a converged multi-reference panel.
At least four converged Fe-W intermetallic references spanning Fe2W, FeW, Fe7W6, and an Fe-rich phase, each with M(H) or M(T) literature anchors, under a convergence-first protocol amendment.