Paper-to-dataset engine

Literature witness corpus

Every published reference behind the ChipletOS BEM calibration. Each record is source-typed — measurement means VNA / probe / lab data; simulation means an HFSS-coaxial or FEM extraction paraphrased from a published figure or table.

Honest framing — buyer DD

0/0 Z₀ entries in this corpus are source_type:simulation (HFSS-coaxial extractions). The ChipletOS BEM headline 3.53 % MAE is therefore a simulation-vs-simulation calibration, not a VNA-validated measurement. The queued ~$200–500 K coupon-pack VNA campaign in deliverables/EXPERIMENT_SOWS/ is the next unlock.

Run BEM-vs-paper comparison →Validation hub

By metric × source_type

Metricmeasurementsimulationtotal

All records

Sorted by metric → source_type → year. Click DOI to retrieve the source paper. The dashboard provides a Compare BEM button per record for live BEM-vs-paper comparison.

Corpus could not be loaded at build time. Visit the dashboard to run live queries.

What this engine does NOT do

  • Replace VNA measurement. The ~$200–500 K coupon-pack queued in deliverables/EXPERIMENT_SOWS/ is the next unlock.
  • Re-paraphrase tables. Each record's measured_value is the published number; tolerance bands reflect the model-vs-reference systematic (BEM multi-conductor vs HFSS-coaxial, +25–55 %).
  • Auto-extract new papers. PDF ingestion + plot digitization is a manual-loop pipeline; human review is required before a record enters the corpus.