The screen vs. four laboratories — the EASI-STRESS benchmark arch (316L)

Run 2026-08-10, updated same day with exact geometry · NOT a blind run, and labeled so · files here: our reconstructed STL, the unedited report JSON, both CC-BY data workbooks, the stress-vs-depth figure.

The artifact

The EU EASI-STRESS project's residual-stress benchmark: an LPBF 316L arch, 20 × 20 × 10 mm — a central 8 mm tunnel with vertical sides and a 4 mm-radius crown whose apex sits 2 mm below the top surface, leaving a 2 mm ligament. Built on two machines (MetalFAB1 and EOS M290), measured as-built and after 700 °C / 2 h stress relief by four diffraction instruments across three facilities: neutron diffraction at ILL (SALSA) and synchrotron X-ray diffraction at DESY (P07, P61A) and ESRF (ID15A). Inter-instrument agreement ~50 MPa; the source paper's own words: "near-yield level residual stresses."

• Data: Zenodo doi 10.5281/zenodo.13890911 and 10.5281/zenodo.13890709, CC-BY-4.0 — both workbooks in this folder, credited per their terms.
• Geometry: reconstructed from Laurence et al., Strain 2025, 61:e70005 (doi 10.1111/str.70005, CC-BY), Fig. 1 — our STL in this folder.

Protocol — what this receipt is and isn't

Not blind: the measurement workbooks were opened before the screen ran (they were needed to establish the measurement-line coordinate system). The screen itself takes only geometry and has no stress input — but the honest claim class here is localization comparison, not blind prediction. The blind receipts are the two NIST bridges.

What the screen said vs. what four labs measured

The labs measure from the top surface downward; their 0–2 mm measurement line is the ligament. The screen, from geometry alone:

• overhang watch at z = 6.85–8.0 mm — the crown band where the 4 mm arc passes the 45° self-support angle on its way to horizontal — directly beneath the ligament where all four instruments measured peak in-plane stress: ~350–415 MPa as-built.
• thin_wall finds the ligament at 2.0 mm exactly — and correctly does not flag it (it's above the process floor).
• Verdict "printable with cautions" — matching reality: four labs printed this part; the concern is stress, not build survival.
• The heat-treatment comparison in the data quantifies the standing advice in every report ("stress relieve before cut-off"): 700 °C/2 h drops peak in-plane stress roughly 40–75% (machine-dependent).

Figure: — measured Sxx vs. depth, three instruments, as-built beside heat-treated, with the screen's flagged crown boundary marked.

A retraction, kept visible

Before the source paper was in hand, we reconstructed the arch from database nominals and guessed a flat-roofed opening. That wrong shape produced sharp 90° internal corners, which the screen flagged — and we initially matched that finding to the measured stress. The real crown is a smooth 4 mm radius: no corners exist, and on the exact geometry the screen reports zero severe concave edges. ~~The re-entrant corner finding~~ is withdrawn; the stress-localization claim rests on the overhang/downskin band alone, where it belonged. Lesson recorded: bracketing an unknown dimension cannot rescue a wrong shape family — get the drawing.

The boundary

Localization, not magnitudes: nothing here claims 400 MPa was predicted. The claim is that the geometric screen pointed at the band where four independent instruments found the stress — and that when our own reconstruction was wrong, we said so in writing.