# Extended 3D Dataset — NeurIPS Closure Challenge additions

Three 3D cases added to extend the closure challenge beyond the existing 2D set:

| Folder | Case | Source | Required metrics |
|--------|------|--------|------------------|
| `ERCOFTAC_AhmedBody25/` | Ahmed Body 25° slant | ERCOFTAC case082 (Lienhart-Becker-Stoots 2003) | LDA velocity + Reynolds stresses on canonical x-z and y-z planes; rear surface Cp |
| `NASA_FaithHill/` | Faith Hill 3-D smooth-body separation | NASA TMR (Bell et al. 2012) | PIV centerline U/V/W + Reynolds stresses; PSP centerline Cp; FISF surface Cf |
| `ERCOFTAC_WingBodyJunction/` | Wing-body junction (NACA0020) | ERCOFTAC kbwiki DNS 1-6 (Bassi et al. 2023) | 10 symmetry-plane profiles upstream of root; bottom-wall + wing-root Cp lines |

## Per-case layout
```
<CASE>/
├── README.md                     case description, refs, submission format
├── 0/, <converged>/, constant/,
│   system/, postProcessing/      OpenFOAM case (rsync from OF_Baseline_Cases)
├── baseline_komegasst/           k-ω SST RANS extracted at experimental locations (.csv)
├── highfidelity/                 LDA / PIV / DNS reference data (.csv + raw .dat where applicable)
└── plots/
    ├── plot_profiles.py          set MODEL_DIR env var to overlay submitted model
    └── *.pdf                     pre-rendered baseline-vs-experiment figures
```

## Tools at this level
- `commented_dat_to_csv.py` — converts simple `#`-commented headed .dat files (Ahmed format) to flat CSV
- `faith_hill_to_csv.py` — Faith Hill ingester; merges 14 single-scalar Tecplot POINT files per PIV dataset into one CSV; converts PSP and FISF too
- `extract_wbj_surface_lines.py` — Python+vtk script that reads the WingBody surface `.pvtu+_NNNN.vtu` partitioned files and extracts bottom-wall centerline (z/T=0) and wing-root chord (y/T~0.05) Cp strips

Re-run any of these by passing the source dir/file as arg.

## Cases skipped or excluded
- **Ahmed Body 35°**: 35° configuration data is in the same archive but excluded from this packaging — only the canonical 25° case (the famous one bracketing the critical 30° angle) is shipped.
- **Faith Hill Cobra probe**: 4-hole probe with 40° acceptance cone — cannot measure inside reverse-flow regions. Including it would mislead participants. PIV is the velocity reference; Cobra is intentionally not provided.
- **DNS 1-6 volume data**: 60–70 GB per scalar field (averaged_pressure, averaged_velocity_{x,y,z}, Reynolds_stress_{xx,xy,...,zz}, Taylor/Kolmogorov scales). Far too large to ship. The 10 pre-curated profile CSVs + 2 surface line CSVs (~5 MB total) are the curated equivalent.

## Decisions log (2026-04-29)
- Case naming: mixed prefixes — `NASA_FaithHill` (NASA TMR), `ERCOFTAC_AhmedBody25` and `ERCOFTAC_WingBodyJunction` (ERCOFTAC sources). Reflects source authority.
- Faith Hill OF: case is non-dimensional (U_inlet=1.0); sampled p maps to Cp directly. Documented in case README.
- Wing-body 5000/ extra fields (`Pk, PkDelta, Prodk, Rall, nutSwitch`) kept as-is (matches "keep as is" preference).
- Wing-body surface data: ship BOTH the extracted CSV centerline lines (primary) AND the raw partitioned `.pvtu+_NNNN.vtu` archives in `highfidelity/raw_surface/` (~298 MB) for full self-containment.
- Wing-body OF: WBJ provides constant/system/postProcessing/log; **time dirs 0 and 5000 are taken from `revised_inflow/`** (Inlet pressure changed from `zeroGradient` to `fixedValue uniform 0` — better convergence). 90+14 Windows `:Zone.Identifier` files skipped on copy.

## TODO
- `baseline_komegasst/` content per case — needs OpenFOAM post-hoc sampling at experimental locations. See `*/run_baseline_extraction.sh` (per-case script).