umarsqur/cld-fcc-ee-fullsim-rec

Parnassus CLD full simulation and reconstruction dataset: $e^{+}{e}^{-}\to Z\to q\bar{q},s\bar{s},c\bar{c},b\bar{b}e^+e^-\; \backslash to\; Z\; \backslash to\; q\backslash bar\{q\},\backslash ,s\backslash bar\{s\},\backslash ,c\backslash bar\{c\},\backslash ,b\backslash bar\{b\}$

Full Geant4 detector simulation and PandoraPFA reconstruction of $ZZ$-pole hadronic events for the CLD detector concept at the FCC-ee, in the EDM4hep format. 350 files, $\mathrm{5,000,000}5\{,\}000\{,\}000$ events (5.04 TB). The files are hosted here as-is; read them with uproot or podio (see below). Per-file MD5 checksums are in md5sums.txt and the full size manifest in manifest_sizes.csv.

Dataset at a glance

Process	$e^{+}{e}^{-}\to Z\to q\bar{q}e^+e^-\; \backslash to\; Z\; \backslash to\; q\backslash bar\{q\}$ $(ud)(ud)$, $s\bar{s}s\backslash bar\{s\}$, $c\bar{c}c\backslash bar\{c\}$, $b\bar{b}b\backslash bar\{b\}$ at $\sqrt{s}=91.2\mathrm{G}\mathrm{e}\mathrm{V}\backslash sqrt\{s\}\; =\; 91.2\backslash ,\backslash mathrm\{GeV\}$
Detector	CLD, CLD_o2_v07 (FCCee/CLD compact geometry, via K4GEO)
Simulation	ddsim (DD4hep/DDG4), field stepper ClassicalRK4
Reconstruction	Key4hep CLDReconstruction (PandoraPFA particle flow)
Format	EDM4hep (podio, ROOT), tree name events
Files	$350\times 350\; \backslash times$ ee_Z_<flavor>_<job>_REC.edm4hep.root
Events	$\mathrm{5,000,000}5\{,\}000\{,\}000$ total
Volume	5.04 TB

Per-flavor breakdown:

Flavor	Files	Events/file	Events	Size
$q\bar{q}q\backslash bar\{q\}$ $(ud)(ud)$	50	25,000	1,250,000	1.28 TB
$s\bar{s}s\backslash bar\{s\}$	100	12,500	1,250,000	1.26 TB
$c\bar{c}c\backslash bar\{c\}$	100	12,500	1,250,000	1.26 TB
$b\bar{b}b\backslash bar\{b\}$	100	12,500	1,250,000	1.23 TB

Production chain

1. Generation: Event generation (including initial-state radiation) with Whizard v3.1.5 (arXiv:0708.4233, hep-ph/0102195), interfaced with Pythia v6.427 (hep-ph/0603175) for parton showering and hadronization; $Z\to q\bar{q}Z\; \backslash to\; q\backslash bar\{q\}$ at $\sqrt{s}=91.2\mathrm{G}\mathrm{e}\mathrm{V}\backslash sqrt\{s\}\; =\; 91.2\backslash ,\backslash mathrm\{GeV\}$.
2. Detector simulation: ddsim with cld_steer.py on CLD_o2_v07 ($K4GEO/FCCee/CLD/compact/CLD_o2_v07/CLD_o2_v07.xml).
3. Digitization and reconstruction: CLDReconstruction.py / CLDReconstruction.xml (tracking, calorimeter digitization, PandoraPFA particle flow, jet/vertex reco).
4. Output: one *_REC.edm4hep.root per job. The SIM step output is not shared since the REC files carry the hit collections, reconstructed collections and the MC-truth links needed for analysis.

Software stack: Key4hep nightly release 2026-08-26 (ddsim, PandoraPFA, K4GEO). The exact steering files used are included here (config/cld_steer.py, config/CLDReconstruction.py, config/CLDReconstruction.xml, config/collections_rec_level.txt).

File contents (key EDM4hep collections)

Reconstructed: PandoraPFOs (particle-flow objects), tracks and track states, calorimeter clusters, and the MC-truth MCParticles with reco $\leftrightarrow \backslash leftrightarrow$ truth association collections. See config/collections_rec_level.txt for the full list written at reco level.

Download

from huggingface_hub import snapshot_download, hf_hub_download

# everything (~5 TB) into ./cld_rec
snapshot_download("umarsqur/cld-fcc-ee-fullsim-rec", repo_type="dataset",
                  local_dir="cld_rec")

# or a single file
f = hf_hub_download("umarsqur/cld-fcc-ee-fullsim-rec",
                    "ee_Z_bb_10_REC.edm4hep.root", repo_type="dataset")

or with the CLI:

hf download umarsqur/cld-fcc-ee-fullsim-rec --repo-type dataset --local-dir cld_rec
# a subset:
hf download umarsqur/cld-fcc-ee-fullsim-rec --repo-type dataset \
    --include "ee_Z_bb_*_REC.edm4hep.root" --local-dir cld_rec_bb

Reading the files

# uproot (no Key4hep install needed)
import uproot
ev = uproot.open("ee_Z_bb_10_REC.edm4hep.root")["events"]
print(ev.num_entries)

# or the full EDM4hep object interface with podio
from podio.root_io import Reader
reader = Reader("ee_Z_bb_10_REC.edm4hep.root")
for event in reader.get("events"):
    pfos = event.get("PandoraPFOs")

Integrity

Verify downloads against md5sums.txt:

md5sum -c md5sums.txt

Citation & related resources

• Paper: Umar Sohail Qureshi, Benjamin Nachman, and Caterina Vernieri, Parnassus for the CLD Detector: A Generative Machine-Learning Surrogate for Detector Simulation and Reconstruction at the FCC-ee, arXiv:2609.30775 (2026).
• Parnassus source code: https://github.com/parnassus-hep/parnassus
• CLDConfig (reconstruction configuration): https://github.com/key4hep/CLDConfig
• Key4hep: https://key4hep.github.io/key4hep-doc/
• EDM4hep: https://github.com/key4hep/EDM4hep

If you use this dataset, please cite it via its Hugging Face DOI (from the "Cite this dataset" sidebar, generated in the repo Settings) and the paper above.

@article{Qureshi:2026oks,
    author = "Qureshi, Umar Sohail and Nachman, Benjamin and Vernieri, Caterina",
    title = "{Parnassus for the CLD Detector: A Generative Machine-Learning Surrogate for Detector Simulation and Reconstruction at the FCC-ee}",
    eprint = "2609.30775",
    archivePrefix = "arXiv",
    primaryClass = "physics.ins-det",
    month = "9",
    year = "2026"
}

License

Creative Commons Attribution 4.0 International (CC-BY-4.0).