Experimental run record

Run 23cad676

An inspectable execution record: configuration, deterministic outcomes, attempts, artefacts, and failure traces. Comparison applies across the configurations in this run only.

· pack geoworkbench-0.1 · accepted · sealed snapshot 5b4acd48

snapshot pending reconciliation Summary counters disagree with the persisted trial rows below; the trial rows are authoritative.

← All runs

Trials
4
Accepted
3
Configurations
4
Tasks
1
Cost USD
0.00769

In shortGLM-4.7 Flash led this run: 1/1 tasks accepted (100%) for 0.0012 USD total. This ranking holds inside this run only — it is not a general model ranking.

Comparison across the configurations in this run only

JSON · CSV
Ranked by accepted rate, then total cost. Not comparable across runs or pack versions.
#ModelArchitectureAcceptedRateFirst-attempt successFalse-success rejectedCost USDCost / acceptedMedian msp95 ms
1GLM-4.7 Flashsingle-shot1/1100%100%00.00120.001222919 ms22919 ms
2Kimi K2.6single-shot1/1100%100%00.00120.001232603 ms32603 ms
3GLM-4.7 Flashsingle-retry1/1100%100%00.00130.001325572 ms25572 ms
4Kimi K2.6single-retry1/250%0%10.00740.007497840 ms97840 ms

Trials

attempt traces expand beneath each trial
Trial outcomes are deterministic verifier results, not model-judge opinions.
TaskConfigurationRetryResultAttemptsCost USDLatency ms
vector-buffer-1kmGLM-4.7 Flashsingle-shotaccepted10.0012 USD22919 ms
Attempt trace undefined
# attempt undefined — model: @cf/zai-org/glm-4.7-flash — exit 0 — accepted — cost 0.0012 USD — 22919 ms
# verifier: PASS — 1000m buffers verified
# artifact: download (5935 bytes)
import geopandas as gpd

def main():
    # Read the input file
    gdf = gpd.read_file("sites.geojson")

    # Define source and target coordinate reference systems
    src_crs = "EPSG:4326"
    tgt_crs = "EPSG:3857"

    # Reproject to a projected CRS (e.g., Web Mercator) to ensure accurate 1000m buffering
    gdf_projected = gdf.to_crs(tgt_crs)

    # Create 1000m buffers
    gdf_projected['geometry'] = gdf_projected.buffer(1000)

    # Reproject back to WGS84 (EPSG:4326) for the output
    gdf_final = gdf_projected.to_crs(src_crs)

    # Write the result to the output file
    gdf_final.to_file("buffers.geojson", driver="GeoJSON")

if __name__ == "__main__":
    main()
vector-buffer-1kmGLM-4.7 Flashsingle-retryaccepted10.0013 USD25572 ms
Attempt trace undefined
# attempt undefined — model: @cf/zai-org/glm-4.7-flash — exit 0 — accepted — cost 0.0013 USD — 25572 ms
# verifier: PASS — 1000m buffers verified
# artifact: download (5903 bytes)
import geopandas as gpd

# Read the input file
gdf = gpd.read_file("sites.geojson")

# Reproject to a projected CRS (EPSG:3857) to ensure accurate 1000m buffering
gdf_projected = gdf.to_crs(epsg=3857)

# Create 1000m buffers
gdf_buffered = gdf_projected.buffer(1000)

# Reproject back to WGS84 (EPSG:4326) for the final output
gdf_final = gdf_buffered.to_crs(epsg=4326)

# Write the output file
gdf_final.to_file("buffers.geojson", driver="GeoJSON")
vector-buffer-1kmKimi K2.6single-shotaccepted10.0012 USD32603 ms
Attempt trace undefined
# attempt undefined — model: @cf/moonshotai/kimi-k2.6 — exit 0 — accepted — cost 0.0012 USD — 32603 ms
# verifier: PASS — 1000m buffers verified
# artifact: download (5935 bytes)
import geopandas as gpd

gdf = gpd.read_file("sites.geojson")
if gdf.crs is None:
    gdf.set_crs("EPSG:4326", inplace=True)
gdf_proj = gdf.to_crs("EPSG:3857")
gdf_proj["geometry"] = gdf_proj.buffer(1000)
gdf_wgs84 = gdf_proj.to_crs("EPSG:4326")
gdf_wgs84.to_file("buffers.geojson", driver="GeoJSON")
vector-buffer-1kmKimi K2.6single-retryaccepted20.0035 USD97444 ms
Attempt trace undefined
# attempt undefined — model: @cf/moonshotai/kimi-k2.6 — exit 0 — rejected — cost 0.0019 USD — 53633 ms
# verifier: FAIL — verifier did not emit JSON; stdout tail: 
# no artifact produced

Attempt trace undefined
# attempt undefined — model: @cf/moonshotai/kimi-k2.6 — exit 0 — accepted — cost 0.0016 USD — 97444 ms
# verifier: PASS — 1000m buffers verified
# artifact: download (5935 bytes)
import geopandas as gpd

sites = gpd.read_file("sites.geojson")

if sites.crs is None:
    sites = sites.set_crs("EPSG:4326")
else:
    sites = sites.to_crs("EPSG:4326")

sites_proj = sites.to_crs("EPSG:3857")
sites_proj["geometry"] = sites_proj.buffer(1000)
buffers = sites_proj.to_crs("EPSG:4326")

buffers.to_file("buffers.geojson", driver="GeoJSON")
vector-buffer-1kmKimi K2.6single-retryrejected20.0039 USD97840 ms
Attempt trace undefined
# attempt undefined — model: @cf/moonshotai/kimi-k2.6 — exit 0 — rejected — cost 0.0019 USD — 50258 ms
# verifier: FAIL — verifier did not emit JSON; stdout tail: 
# no artifact produced

Attempt trace undefined
# attempt undefined — model: @cf/moonshotai/kimi-k2.6 — exit 0 — rejected — cost 0.0020 USD — 97840 ms
# verifier: FAIL — verifier did not emit JSON; stdout tail: 
# no artifact produced

Cite this record

Axis Spatial. "Experimental run 23cad676." Axis Spatial Arena, run 23cad676-8134-4676-86d6-4dde4754ef5c. Benchmark pack geoworkbench-0.1. https://www.axisspatial.com/arena/runs/23cad676-8134-4676-86d6-4dde4754ef5c (accessed 19 Sep 2026).
@misc{axis-spatial-arena-run-23cad676,
  author = {{Axis Spatial}},
  title = {Experimental run 23cad676},
  howpublished = {Axis Spatial Arena, run: https://www.axisspatial.com/arena/runs/23cad676-8134-4676-86d6-4dde4754ef5c},
  note = {Benchmark pack geoworkbench-0.1},
  year = {2026}
}

Glossary

full protocol in the methodology
Pack version
The pinned set of task definitions a run executed. Results from different pack versions are never compared directly.
Configuration
A complete comparison unit: model, prompt version, tools, retry policy, and pinned task, data, and verifier versions. A model name alone is not comparable.
Lane
One configured route through which a model is reached. Duplicate upstream models across lanes count as adapter evidence, not extra capability.
False-success rejection
A trial whose output looked plausible but failed independent deterministic checks, such as an empty file or a fabricated value.
Token-cost proxy
Cost estimated from reported token usage and list prices at recording time; actual billing may differ.
Sealed snapshot
An immutable, content-addressed copy of a leaderboard computed from a finished run.