Skip to content

Gallery overview

Thirteen prototypes, one gallery (index.html, live at met-al-lab.pages.dev). Most share the verification math in lib/met-stats.mjs (and, since Round 4, the lib/met-data-source.mjs funnel) and ship as offline single-file builds; the Real-Data Demo and Scorecard stream from the network, and WebGPU FSS needs a WebGPU browser.

Each card carries honesty badges: a data-provenance chip (real·streamed / real·baked / real·load-a-file / synthetic) and MET-component chips naming which part of the MET ecosystem it reimagines. The one genuinely novel app (07, the metric cube) carries no component badge — nothing in MET does that.

# App Theme Data Reimagines
01 Client-side Wrapper Client-side compute synthetic MET · METcalcpy
02 Stat Explorer Modern interaction synthetic METviewer · METcalcpy
03 Novel Plotting Novel visuals synthetic METplotpy
04 Guided Journey Modernization synthetic METplotpy · METviewer
05 Spatial Maps Novel visuals real · baked MET · METplotpy
06 Ensemble Views Novel visuals synthetic¹ MET · METplotpy
07 Metric Cube (3D) Novel visuals synthetic — (novel)
08 STAT Ingest Real-data bridge real · load-a-file METdataio
09 MET Pipeline Explorer MET in the browser real · streamed MET · METplus
10 Real-Data Demo Real data over the wire real · streamed METdataio · METviewer
11 WebGPU FSS GPU compute real · baked MET
12 Scorecard · Taylor · ROC Report views real · streamed METviewer · METplotpy
13 MODE Lab Objects real · baked MET (MODE)

¹ Unavoidably synthetic: the real archive contains no ensemble line types (ECNT/ORANK/RHIST).

  • MET core — deep: verification math recomputed and validated against MET’s own .stat (counts bit-identical), plus the FSS neighborhood operator on the GPU.
  • METplotpy — broad: performance diagram, reliability + Brier, rank histogram, spread-skill, CRPS, threshold scrubber, spatial fields.
  • METdataio + METviewer — serverless, with the signature views: DuckDB-WASM + Parquet + Zarr replace the database stack, and card 12 adds the scorecard (paired-bootstrap significance), Taylor diagram, and ROC.
  • METcalcpy — covered: aggregation/derivation, true bootstrap CIs (percentile + BCa in the shared lib), and event equalization (the scorecard’s common-cycle pairing) all shipped.
  • METplus / METexpress — thematic: the pipeline explorer reconstructs the workflow narratively rather than reimplementing wrappers.