core-lens 0.1.dev124__tar.gz → 0.1.dev129__tar.gz

core_lens-0.1.dev129/.github/ISSUE_TEMPLATE/blank-proposal.yaml

@@ -0,0 +1,33 @@
+name: Blank Proposal
+description: Use this for all proposals.
+title: "[Proposal]: "
+labels: ["proposal"]
+body:
+  - type: textarea
+    id: issue
+    attributes:
+      label: Is the proposal related to an issue you are facing?
+      description: If yes, then please give a clear and concise description of the problem.
+    validations:
+      required: true
+  - type: textarea
+    id: feature
+    attributes:
+      label: Describe the proposal you have in mind.
+      placeholder: Describe the proposal you are in detail.
+    validations:
+      required: true
+  - type: textarea
+    id: alts
+    attributes:
+      label: Have you considered any alternatives?
+      description: If so, please give a clear and concise description.
+    validations:
+      required: false
+  - type: textarea
+    id: additions
+    attributes:
+      label: Any additional context you want to share?
+      placeholder: Add any other context or screenshots about the proposal here.
+    validations:
+      required: false

core_lens-0.1.dev129/.github/ISSUE_TEMPLATE/bug-report.yaml

@@ -0,0 +1,63 @@
+name: Bug Report
+description: File a bug report.
+title: "[Bug]: "
+labels: ["bug"]
+body:
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: Please describe issue you faced.
+      description: Also tell us, what did you expect to happen?
+      placeholder: Describe the issue in detail
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Version
+      description: What is the version of library you are using?
+      placeholder: 0.0.1
+    validations:
+      required: true
+  - type: dropdown
+    id: python
+    attributes:
+      label: What is the python version you are using?
+      multiple: true
+      options:
+        - Python 3.11
+        - Python 3.12
+        - Python 3.13
+        - Python 3.14
+  - type: dropdown
+    id: package-manager
+    attributes:
+      label: Which package manager are you using?
+      description: If you are unsure select pip.
+      options:
+        - uv
+        - poetry
+        - conda
+        - pip
+        - pixi
+        - pipenv
+  - type: input
+    id: OS
+    attributes:
+      label: Which OS are you using?
+      description: OS Name and version
+      placeholder: Ubuntu 22.04
+  - type: textarea
+    id: logs
+    attributes:
+      label: Relevant log output
+      description: Please copy and paste any relevant log output. This will be automatically formatted into code, so no need for backticks.
+      render: shell
+  - type: upload
+    id: screenshots
+    attributes:
+      label: Upload screenshots
+      description: If applicable, add screenshots to help explain your problem.
+    validations:
+      required: false
+      accept: ".png,.jpg,.gif,.log,.txt,.zip"

core_lens-0.1.dev129/.github/ISSUE_TEMPLATE/feature-request.yaml

@@ -0,0 +1,33 @@
+name: Feature Request
+description: File a feature request.
+title: "[Feature]: "
+labels: ["feature"]
+body:
+  - type: textarea
+    id: issue
+    attributes:
+      label: Is the feature related to an issue you are facing?
+      description: If yes, then please give a clear and concise description of the problem.
+    validations:
+      required: true
+  - type: textarea
+    id: feature
+    attributes:
+      label: Describe the solution you have in mind.
+      placeholder: Describe the solution you are proposing in detail
+    validations:
+      required: true
+  - type: textarea
+    id: alts
+    attributes:
+      label: Have you considered any alternatives?
+      description: If so, please give a clear and concise description.
+    validations:
+      required: false
+  - type: textarea
+    id: additions
+    attributes:
+      label: Any additional context you want to share?
+      placeholder: Add any other context or screenshots about the feature request here.
+    validations:
+      required: false

{core_lens-0.1.dev124 → core_lens-0.1.dev129}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: core-lens
-Version: 0.1.dev124
+Version: 0.1.dev129
 Summary: Query, analyse, and visualise CoreStack's microwatershed and Earth science data through a clean, composable Python API.
 Project-URL: Homepage, https://github.com/ApoorvaKashyap/core-lens
 Project-URL: Issues, https://github.com/ApoorvaKashyap/core-lens/issues
@@ -17,11 +17,13 @@ Requires-Python: >=3.13
 Requires-Dist: core-lens[core]
 Requires-Dist: core-lens[spatial]
 Provides-Extra: core
+Requires-Dist: awscrt>=0.23.0; extra == 'core'
 Requires-Dist: loguru>=0.7.3; extra == 'core'
 Requires-Dist: polars<2,>=1.39.0; extra == 'core'
 Requires-Dist: pyarrow<25,>=23.0.0; extra == 'core'
 Requires-Dist: pydantic<3,>=2.13.3; extra == 'core'
 Provides-Extra: full
+Requires-Dist: awscrt>=0.23.0; extra == 'full'
 Requires-Dist: bokeh<4,>=3.6.0; extra == 'full'
 Requires-Dist: duckdb<2,>=1.5.3; extra == 'full'
 Requires-Dist: geoarrow-rust-core>=0.3.0; extra == 'full'
@@ -37,6 +39,7 @@ Requires-Dist: scipy<2,>=1.17.1; extra == 'full'
 Requires-Dist: shapely<3,>=2.1.2; extra == 'full'
 Requires-Dist: statsmodels<1,>=0.14.6; extra == 'full'
 Provides-Extra: full-gpu
+Requires-Dist: awscrt>=0.23.0; extra == 'full-gpu'
 Requires-Dist: bokeh<4,>=3.6.0; extra == 'full-gpu'
 Requires-Dist: cudf-cu13==26.6.*; extra == 'full-gpu'
 Requires-Dist: cudf-polars-cu13==26.6.*; extra == 'full-gpu'
@@ -54,6 +57,7 @@ Requires-Dist: scipy<2,>=1.17.1; extra == 'full-gpu'
 Requires-Dist: shapely<3,>=2.1.2; extra == 'full-gpu'
 Requires-Dist: statsmodels<1,>=0.14.6; extra == 'full-gpu'
 Provides-Extra: gpu
+Requires-Dist: awscrt>=0.23.0; extra == 'gpu'
 Requires-Dist: cudf-cu13==26.6.*; extra == 'gpu'
 Requires-Dist: cudf-polars-cu13==26.6.*; extra == 'gpu'
 Requires-Dist: loguru>=0.7.3; extra == 'gpu'
@@ -61,6 +65,7 @@ Requires-Dist: polars<2,>=1.39.0; extra == 'gpu'
 Requires-Dist: pyarrow<25,>=23.0.0; extra == 'gpu'
 Requires-Dist: pydantic<3,>=2.13.3; extra == 'gpu'
 Provides-Extra: spatial
+Requires-Dist: awscrt>=0.23.0; extra == 'spatial'
 Requires-Dist: duckdb<2,>=1.5.3; extra == 'spatial'
 Requires-Dist: geopandas<2,>=1.1.3; extra == 'spatial'
 Requires-Dist: loguru>=0.7.3; extra == 'spatial'
@@ -70,6 +75,7 @@ Requires-Dist: pydantic<3,>=2.13.3; extra == 'spatial'
 Requires-Dist: pyproj<4,>=3.7.2; extra == 'spatial'
 Requires-Dist: shapely<3,>=2.1.2; extra == 'spatial'
 Provides-Extra: stats
+Requires-Dist: awscrt>=0.23.0; extra == 'stats'
 Requires-Dist: loguru>=0.7.3; extra == 'stats'
 Requires-Dist: polars<2,>=1.39.0; extra == 'stats'
 Requires-Dist: pyarrow<25,>=23.0.0; extra == 'stats'
@@ -77,6 +83,7 @@ Requires-Dist: pydantic<3,>=2.13.3; extra == 'stats'
 Requires-Dist: scipy<2,>=1.17.1; extra == 'stats'
 Requires-Dist: statsmodels<1,>=0.14.6; extra == 'stats'
 Provides-Extra: viz
+Requires-Dist: awscrt>=0.23.0; extra == 'viz'
 Requires-Dist: bokeh<4,>=3.6.0; extra == 'viz'
 Requires-Dist: geoarrow-rust-core>=0.3.0; extra == 'viz'
 Requires-Dist: loguru>=0.7.3; extra == 'viz'

core_lens-0.1.dev129/benchmarks/README.md

@@ -0,0 +1,60 @@
+# core-lens Benchmarks
+
+Focused Scalene profiling scripts — one per module/subsystem.
+
+## Structure
+
+```
+benchmarks/
+├── README.md
+├── run_all.sh                  # Run all benchmarks in sequence
+├── bench_aoi.py                # AoI init, register, named-boundary resolution
+├── bench_spatial.py            # build_bbox_index, bbox_intersects_geometry, exact_spatial_filter
+├── bench_view.py               # View.where, spatial_filter, between, materialise (static/annual/fortnightly)
+├── bench_result.py             # Result.derive, aggregate, with_geometry, df/gdf
+├── bench_season.py             # resolve_time_filter, add_temporal_columns, SeasonConfig.season_for
+├── bench_schema.py             # schema detection (detect, _infer_* helpers)
+├── bench_polars_utils.py       # scan_with_key_filter, collect_lf (CPU path)
+└── bench_export.py             # parquet, json, csv, geoparquet, geojson
+```
+
+## Running
+
+### Single benchmark
+
+```bash
+uv run scalene --html --outfile benchmarks/profiles/bench_spatial.html benchmarks/bench_spatial.py
+```
+
+### All benchmarks (sequential, HTML output)
+
+```bash
+chmod +x benchmarks/run_all.sh
+./benchmarks/run_all.sh
+```
+
+Profiles land in `benchmarks/profiles/`.
+
+### Quick timing-only run (no Scalene overhead)
+
+```bash
+uv run python benchmarks/bench_spatial.py
+```
+
+## What to look for in Scalene output
+
+| Signal | Meaning |
+|---|---|
+| High **CPU %** on line | Compute bottleneck — vectorise or push into Polars expr |
+| High **memory** spike | Possible copy-on-write or large intermediate frame |
+| `[native]` vs `[Python]` | Native = C ext (fast). Python = pure Python (optimise) |
+| **GPU %** column | cudf path hit or not |
+
+## Prerequisites
+
+```bash
+uv sync --group dev
+```
+
+Each script needs `DATA_ROOT` pointing at a directory with the expected entity Parquet files.
+Edit the `DATA_ROOT` constant at the top of each file to match your local data path.

core_lens-0.1.dev129/benchmarks/bench_aoi.py

@@ -0,0 +1,172 @@
+"""
+bench_aoi.py — Benchmark for core_lens.aoi
+===========================================
+Targets:
+  - AoI.register()              entity registration + path probe
+  - AoI.__init__(bbox=)         bbox boundary mode
+  - AoI.__init__(geometry=)     raw Shapely geometry mode
+  - AoI.__init__(**kwargs)      named-boundary resolution (Parquet scan + unary_union)
+  - AoI.__getattr__             lazy View creation + first spatial_filter call
+  - AoI.deregister()            cleanup
+
+Scalene focuses on:
+  - _validate_entity()          schema read at init time
+  - _resolve_named_boundary()   scan_parquet + shapely.from_wkb + unary_union
+
+Usage:
+    uv run scalene run --html --outfile benchmarks/profiles/bench_aoi.html benchmarks/bench_aoi.py
+    uv run python benchmarks/bench_aoi.py   # timing only
+"""
+
+from __future__ import annotations
+
+import time
+
+import shapely.geometry as sgeom
+
+from core_lens import AoI, SeasonConfig
+from core_lens.entities.mws import MWSEntity
+
+# ── Config ────────────────────────────────────────────────────────────────────
+DATA_ROOT = "data/"
+
+# A mws_id known to exist in your dataset.
+MWS_ID = "12_322881"  # update to a valid mws_id in your data
+
+# Bounding box covering all of India (WGS-84).
+INDIA_BBOX: tuple[float, float, float, float] = (68.0, 6.0, 98.0, 38.0)
+
+# A small bbox for the named-boundary AoI sanity check.
+SMALL_BBOX: tuple[float, float, float, float] = (86.0, 21.0, 87.5, 22.0)
+
+
+def _section(title: str) -> None:
+    print(f"\n{'─' * 60}")
+    print(f"  {title}")
+    print("─" * 60)
+
+
+# ── 1. Registration ───────────────────────────────────────────────────────────
+_section("1. AoI.register(MWSEntity)")
+t0 = time.perf_counter()
+
+AoI.register(MWSEntity)
+
+t1 = time.perf_counter()
+print(f"register()       : {(t1 - t0) * 1000:.2f} ms")
+print(f"Registered       : {AoI.registered_entities()}")
+
+
+# ── 2. AoI init — bbox mode ───────────────────────────────────────────────────
+_section("2. AoI.__init__(bbox=INDIA_BBOX)  [validates all entities]")
+t0 = time.perf_counter()
+
+aoi_bbox = AoI(DATA_ROOT, bbox=INDIA_BBOX)
+
+t1 = time.perf_counter()
+print(f"AoI(bbox)        : {(t1 - t0) * 1000:.2f} ms")
+print(f"Geometry type    : {type(aoi_bbox.geometry).__name__}")
+
+
+# ── 3. AoI init — raw geometry mode ──────────────────────────────────────────
+_section("3. AoI.__init__(geometry=shapely_poly)")
+poly = sgeom.box(*SMALL_BBOX)
+t0 = time.perf_counter()
+
+aoi_geom = AoI(DATA_ROOT, geometry=poly)
+
+t1 = time.perf_counter()
+print(f"AoI(geometry)    : {(t1 - t0) * 1000:.2f} ms")
+
+
+# ── 4. AoI init — named boundary (most expensive) ────────────────────────────
+_section(f"4. AoI.__init__(mws_id={MWS_ID!r})")
+t0 = time.perf_counter()
+
+aoi_named = AoI(DATA_ROOT, mws_id=MWS_ID)
+
+t1 = time.perf_counter()
+print(f"AoI(named)       : {(t1 - t0) * 1000:.2f} ms")
+print(f"Geometry type    : {type(aoi_named.geometry).__name__}")
+
+
+# ── 5. AoI init — custom SeasonConfig ────────────────────────────────────────
+_section("5. AoI.__init__ with custom SeasonConfig")
+t0 = time.perf_counter()
+
+aoi_season = AoI(
+    DATA_ROOT,
+    bbox=INDIA_BBOX,
+    seasons=SeasonConfig(
+        kharif=("06-15", "10-15"),
+        rabi=("10-16", "02-28"),
+        zaid=("03-01", "06-14"),
+    ),
+)
+
+t1 = time.perf_counter()
+print(f"AoI(seasons)     : {(t1 - t0) * 1000:.2f} ms")
+print(f"Current season   : {aoi_season.current_season}")
+
+
+# ── 6. View access via __getattr__ ───────────────────────────────────────────
+_section("6. aoi.mws  [first __getattr__ call — runs spatial_filter]")
+t0 = time.perf_counter()
+
+view = aoi_bbox.mws
+
+t1 = time.perf_counter()
+print(f"First access     : {(t1 - t0) * 1000:.2f} ms")
+print(f"Keys shape       : {view.keys.shape}")
+
+# Second access — should hit _scoped cache (near-zero cost).
+t0 = time.perf_counter()
+view2 = aoi_bbox.mws
+t1 = time.perf_counter()
+print(f"Cached access    : {(t1 - t0) * 1000:.4f} ms  (expect ~0 ms)")
+
+
+# ── 7. current_season / current_year (property overhead) ─────────────────────
+_section("7. AoI.current_season / current_year")
+REPS = 10_000
+t0 = time.perf_counter()
+for _ in range(REPS):
+    aoi_bbox.current_season
+t1 = time.perf_counter()
+print(
+    f"current_season ×{REPS}: {(t1 - t0) * 1000:.2f} ms total  "
+    f"({(t1 - t0) / REPS * 1e6:.2f} µs/call)"
+)
+
+t0 = time.perf_counter()
+for _ in range(REPS):
+    aoi_bbox.current_year
+t1 = time.perf_counter()
+print(
+    f"current_year   ×{REPS}: {(t1 - t0) * 1000:.2f} ms total  "
+    f"({(t1 - t0) / REPS * 1e6:.2f} µs/call)"
+)
+
+
+# ── 8. registered_entities() ─────────────────────────────────────────────────
+_section("8. AoI.registered_entities()")
+REPS = 50_000
+t0 = time.perf_counter()
+for _ in range(REPS):
+    AoI.registered_entities()
+t1 = time.perf_counter()
+print(
+    f"registered_entities ×{REPS}: {(t1 - t0) * 1000:.2f} ms total  "
+    f"({(t1 - t0) / REPS * 1e6:.2f} µs/call)"
+)
+
+
+# ── 9. deregister ─────────────────────────────────────────────────────────────
+_section("9. AoI.deregister(MWSEntity)")
+t0 = time.perf_counter()
+AoI.deregister(MWSEntity)
+t1 = time.perf_counter()
+print(f"deregister()     : {(t1 - t0) * 1000:.4f} ms")
+print(f"Remaining        : {AoI.registered_entities()}")
+
+print("\n✓ bench_aoi.py complete")

core_lens-0.1.dev129/benchmarks/bench_export.py

@@ -0,0 +1,212 @@
+"""
+bench_export.py — Benchmark for core_lens.export.formats
+=========================================================
+Targets:
+  - export.parquet()            Polars write_parquet (no geometry)
+  - export.json()               Polars write_json (no geometry)
+  - export.csv()                Polars write_csv (no geometry)
+  - export.geoparquet()         DuckDB spatial COPY → Parquet (WKB → geometry)
+  - export.geojson()            DuckDB spatial COPY → GeoJSON (WKB → GeoJSON)
+  - _get_duckdb_conn()          singleton connection + INSTALL/LOAD spatial
+
+Scalene focuses on:
+  - DuckDB spatial extension load overhead (first call)
+  - write_parquet / write_json / write_csv serialisation speed
+  - DuckDB ST_GeomFromWKB + COPY overhead vs Polars native writers
+  - Memory copies between Polars DataFrame → DuckDB
+
+Usage:
+    uv run scalene run --html --outfile benchmarks/profiles/bench_export.html benchmarks/bench_export.py
+    uv run python benchmarks/bench_export.py
+"""
+
+from __future__ import annotations
+
+import tempfile
+import time
+from pathlib import Path
+
+
+from core_lens import AoI
+from core_lens.entities.mws import MWSEntity
+from core_lens.export import csv, geoparquet, geojson, json, parquet
+from core_lens.export.formats import _get_duckdb_conn
+
+# ── Config ────────────────────────────────────────────────────────────────────
+DATA_ROOT = "data/"
+INDIA_BBOX: tuple[float, float, float, float] = (68.0, 6.0, 98.0, 38.0)
+SMALL_BBOX: tuple[float, float, float, float] = (86.0, 21.0, 87.5, 22.0)
+
+AoI.register(MWSEntity)
+
+# Materialise two result sizes for comparison.
+aoi_all = AoI(DATA_ROOT, bbox=INDIA_BBOX)
+aoi_small = AoI(DATA_ROOT, bbox=SMALL_BBOX)
+
+result_all = aoi_all.mws.static  # full dataset — geometry present
+result_small = aoi_small.mws.static  # small subset
+
+print(f"All   result : {result_all.data.shape}")
+print(f"Small result : {result_small.data.shape}")
+
+
+def _section(title: str) -> None:
+    print(f"\n{'─' * 60}")
+    print(f"  {title}")
+    print("─" * 60)
+
+
+with tempfile.TemporaryDirectory() as tmpdir:
+    tmp = Path(tmpdir)
+
+    # ── 1. DuckDB singleton init (first call installs spatial extension) ──────
+    _section("1. _get_duckdb_conn()  [INSTALL spatial; LOAD spatial — cold]")
+    t0 = time.perf_counter()
+    conn = _get_duckdb_conn()
+    t1 = time.perf_counter()
+    print(f"DuckDB init (cold)  : {(t1 - t0) * 1000:.2f} ms")
+
+    REPS = 10_000
+    t0 = time.perf_counter()
+    for _ in range(REPS):
+        _ = _get_duckdb_conn()
+    t1 = time.perf_counter()
+    print(
+        f"_get_duckdb_conn ×{REPS}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS * 1e6:.3f} µs/call)  (cached)"
+    )
+
+    # ── 2. export.parquet() — no geometry (strip geom col first)
+    _section("2. export.parquet()  [Polars write_parquet, no geometry]")
+
+    # Strip geometry (and nested bbox) to pass has_geometry=False and allow CSV export.
+    geom_col = result_small.entity.geometry_col
+    drop_cols = [geom_col]
+    if "bbox" in result_small.data.columns:
+        drop_cols.append("bbox")
+    df_no_geom = result_small.data.drop(drop_cols)
+    result_no_geom = result_small._replace(data=df_no_geom, has_geometry=False)
+
+    REPS_IO = 5
+    out = tmp / "bench.parquet"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        parquet(result_no_geom, out)
+    t1 = time.perf_counter()
+    sz = out.stat().st_size
+    print(
+        f"parquet (small) ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz // 1024} KB"
+    )
+
+    # All dataset.
+    df_all_no_geom = result_all.data.drop(geom_col)
+    result_all_no_geom = result_all._replace(data=df_all_no_geom, has_geometry=False)
+    out_all = tmp / "bench_all.parquet"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        parquet(result_all_no_geom, out_all)
+    t1 = time.perf_counter()
+    sz_all = out_all.stat().st_size
+    print(
+        f"parquet (all)   ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz_all // 1024} KB"
+    )
+
+    # ── 3. export.json() ──────────────────────────────────────────────────────
+    _section("3. export.json()  [Polars write_json]")
+    out_json = tmp / "bench.json"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        json(result_no_geom, out_json)
+    t1 = time.perf_counter()
+    sz = out_json.stat().st_size
+    print(
+        f"json (small) ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz // 1024} KB"
+    )
+
+    # ── 4. export.csv() ───────────────────────────────────────────────────────
+    _section("4. export.csv()  [Polars write_csv]")
+    out_csv = tmp / "bench.csv"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        csv(result_no_geom, out_csv)
+    t1 = time.perf_counter()
+    sz = out_csv.stat().st_size
+    print(
+        f"csv (small)  ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz // 1024} KB"
+    )
+
+    # ── 5. export.geoparquet() — DuckDB spatial (small) ──────────────────────
+    _section("5. export.geoparquet()  [DuckDB ST_GeomFromWKB + COPY]")
+    out_gpq = tmp / "bench.geoparquet"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        geoparquet(result_small, out_gpq)
+    t1 = time.perf_counter()
+    sz = out_gpq.stat().st_size
+    print(
+        f"geoparquet (small) ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz // 1024} KB"
+    )
+
+    out_gpq_all = tmp / "bench_all.geoparquet"
+    t0 = time.perf_counter()
+    geoparquet(result_all, out_gpq_all)
+    t1 = time.perf_counter()
+    sz_all = out_gpq_all.stat().st_size
+    print(
+        f"geoparquet (all)   ×1: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) * 1000:.2f} ms/call)  → {sz_all // 1024} KB"
+    )
+
+    # ── 6. export.geojson() — DuckDB spatial ──────────────────────────────────
+    _section("6. export.geojson()  [DuckDB ST_GeomFromWKB + COPY GDAL GeoJSON]")
+    out_gj = tmp / "bench.geojson"
+    t0 = time.perf_counter()
+    for _ in range(REPS_IO):
+        geojson(result_small, out_gj)
+    t1 = time.perf_counter()
+    sz = out_gj.stat().st_size
+    print(
+        f"geojson (small) ×{REPS_IO}: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) / REPS_IO * 1000:.2f} ms/call)  → {sz // 1024} KB"
+    )
+
+    out_gj_all = tmp / "bench_all.geojson"
+    t0 = time.perf_counter()
+    geojson(result_all, out_gj_all)
+    t1 = time.perf_counter()
+    sz_all = out_gj_all.stat().st_size
+    print(
+        f"geojson (all)   ×1: {(t1 - t0) * 1000:.2f} ms total  "
+        f"({(t1 - t0) * 1000:.2f} ms/call)  → {sz_all // 1024} KB"
+    )
+
+    # ── 7. Head-to-head: geoparquet vs geojson ────────────────────────────────
+    _section("7. Head-to-head: geoparquet vs geojson (all dataset)")
+    out_a = tmp / "hth.geoparquet"
+    out_b = tmp / "hth.geojson"
+
+    t0 = time.perf_counter()
+    geoparquet(result_all, out_a)
+    t1 = time.perf_counter()
+    t_gpq = (t1 - t0) * 1000
+
+    t0 = time.perf_counter()
+    geojson(result_all, out_b)
+    t1 = time.perf_counter()
+    t_gj = (t1 - t0) * 1000
+
+    print(f"geoparquet : {t_gpq:.2f} ms   ({out_a.stat().st_size // 1024} KB)")
+    print(f"geojson    : {t_gj:.2f} ms   ({out_b.stat().st_size // 1024} KB)")
+    print(
+        f"Ratio      : geojson is {t_gj / t_gpq:.2f}× {'slower' if t_gj > t_gpq else 'faster'}"
+    )
+
+
+# ── Cleanup ───────────────────────────────────────────────────────────────────
+AoI.deregister(MWSEntity)
+print("\n✓ bench_export.py complete")