B. PROJECTS

RESEARCH NOTE 04 / NOVEMBER 15, 2026 · INTERACTIVE

One event,
many explanations.

In 2.84 tonne-years, the LUX-ZEPLIN detector recorded one nuclear recoil at 248 keV that its known backgrounds explain poorly. About fifty papers followed. This explorer puts their explanations under one set of assumptions and lets you change them.

24PUBLISHED BENCHMARKS 14PAPERS 7INDEPENDENT CHECKS PASSED

THE EXPLORER

Load an explanation, then push on its assumptions.

The coloured area is what LZ would have selected; the dotted line is every recoil, including those above its window. The red band is a high-energy sideband where LZ saw nothing. A good explanation puts its weight at 248 keV and little in the sideband.

Explanation

Particle

10 GeV to 3 TeV, logarithmic

Positive: the particle is kicked up to a heavier state (endothermic). Negative: it drops to a lighter one and releases energy (exothermic). Zero: ordinary elastic scattering.

Normalisation
Dark-matter halo

Detector
selected by LZall recoilssideband, nothing seen
Vertical line: the event at 248 keV, with its ±23 keV statistical uncertainty shaded.
Cross-section for one event–
Sideband events per signal event–
Fit penalty Δχ²–
Rate on June 16 vs. year average–
Spectrum peak (selected)–
Audit value (Python)–

THE WHOLE LANDSCAPE

Every mass and splitting at once.

Colour is the fit penalty for one event at 248 keV with an empty sideband, on the common halo. Dots are the published benchmarks; the ring is your current choice. Click or tap anywhere on the map to move there.

Precomputed on a 40 × 90 grid with the same code as the audit (Δχ², cross-section profiled). Black: no recoils in LZ's window (an endothermic splitting too large to pay, or an exothermic one that throws every recoil above 270 keV). The explorer above recomputes exactly at any point.

WHAT THE AUDIT FOUND

Five things to try for yourself.

  1. The halo moves the answer more than the physics. Load with their cross-section fixed, then switch between "This paper's halo" and "Common". The same particle goes from about 1.4 selected events to about 67.
  2. A factor of 4 in the higgsino cross-section. Two conventions are in print, 7.4 × 10⁻³⁹ and 1.86 × 10⁻³⁹ cm². Retuned to one event, the larger one predicts about 4.5 sideband events, the smaller about 2.1: a 1% versus 12% chance that the sideband stays empty.
  3. The sideband splits the field in two. Splittings near 360–380 keV with electroweak strength predict 1.2–5.8 sideband events per signal event; everything else predicts under 0.5. Raise the escape speed to 600 km/s on and watch that number collapse.
  4. One event barely discriminates. Across all 24 benchmarks the fit penalty runs from 1.7 to 4.9: at most a factor of about 12 in likelihood.
  5. The date counts. The event came on June 16, close to Earth's fastest point through the halo. Endothermic models expect up to 3.9 times their average rate then; exothermic ones barely change. Set the time to June 16 and December 2 to see it.

Limits

Helm nuclear form factor only (its second zero sits at 279 keV, where shell-model responses differ); LZ's efficiency approximated as a piecewise-linear curve; sideband acceptance assumed complete because LZ has not published it; backgrounds neglected (about 0.01 events expected near 248 keV); contact spin-independent couplings only, so boosted dark matter and momentum-dependent interactions are not covered. Benchmarks were read from each paper by an AI reader and checked where a reproduction was possible. See the full write-up.

Sources

LZ Collaboration, arXiv:2609.02823; benchmark papers are linked from the explorer by arXiv number. Reference values in the "Audit value" box come from benchmarks_common.csv.