Preliminary Single-Rater Replication: Applying the ATRS-HS46 Reconstructability Rubric to the Dutch Algoritmeregister

Status: Preliminary, single-rater replication attempt — not claim-grade. This deposit applies the same six-field (F1-F6) strict event-level reconstructability rubric used in the ATRS-HS46 audit (Hon Bor So, 10.5281/zenodo.20181288) to a 46-record high-stakes selection from the Dutch national Algoritmeregister (algoritmes.overheid.nl, live total_count=1517 at time of snapshot). Relationship to ATRS-HS46: This is an independent, parallel replication attempt using the same instrument in a different jurisdiction. It is not part of the core ATRS-HS46 empirical study and should not be cited as extending that corpus. Any structural similarity between the two corpora (see summary statistics below) is reported as a preliminary observation, not a validated cross-jurisdictional finding. This record references the ATRS-HS46 pack as instrument and source-method anchor; it is not derived data. Coverage and verification status: All 46 records were scored on the strict F1-F6 rubric by a single rater. 8 of 46 records (17%) underwent a second-pass spot-check, covering 48 field scores. Four of the 48 field scores showed disagreement on initial pass. All 4 were formally adjudicated against ATRS-HS46 calibration anchors (see 03_spotcheck_adjudication/ADJUDICATION_4_DELTAS_v0_1_1.md and .json inside the archive); one field score was changed as a result (Microsoft CoPilot record, F6: 2 to 1). The remaining 38 records have not been independently re-verified. This is a single-rater audit. No second human rater has yet reviewed this corpus. It is not inter-rater validated and should not be treated as claim-grade evidence. Summary statistics (post-adjudication, N=46): mean PLAN strict total = 6.87/12; median = 7.0; range 2-9. Full event-level actor identification (F1=2): 0/46. Full event-level uncertainty disclosure (F5=2): 0/46; F5 absent (F5=0): 27/46. Event-level oversight gate (F6=2): 8/46; any oversight mention (F6>0): 43/46. Copyright note: author copyright and CC-BY licensing apply to the selection logic, scoring, scripts, method notes, and adjudication materials only. Source Algoritmeregister disclosure text reused within corpus_full.json remains attributable to the relevant Dutch public bodies and register operators; reuse here does not imply endorsement by the Dutch government or register operators. Files: this deposit is packaged as a single archive, NL_AR_HS46_Preliminary_Replication_Evidence_Pack_v0_1_1.zip, containing: 01_scoring/ (the full 46-record scoring table with per-field justifications, plus a scope readme); 02_source_corpus/ (corpus_full.json and corpus_full.sha256.txt, the hash-anchored snapshot of the source records used for scoring, plus per-record JSON files and a scoring-evidence file); 03_spotcheck_adjudication/ (the 8-record spot-check checklist and the adjudication log for the 4 disagreed field scores, in both .md and .json form); 04_scripts/ (the Python script used to recompute the strict PLAN metrics from the scoring table); 05_metrics/ (the recomputed metrics output, domain and field summary tables); plus a top-level scope README, method note, and a FILE_MANIFEST_SHA256SUMS.txt covering every file in the archive for independent verification. Instrument reference: F1-F6 definitions follow the ATRS-application canonical lock (Hon Bor So, Boundary Phenomena in Sociotechnical Decision Systems, Canonical Instruction Section 16).