Benchmark prominence

README.md

@@ -10,6 +10,8 @@
 
 OME-Arrow uses [Open Microscopy Environment (OME)](https://github.com/ome) specifications through [Apache Arrow](https://arrow.apache.org/) for fast, queryable, and language agnostic bioimage data.
 
+> 📐 Benchmark results that inform OME Arrow's design decisions are available in the [ome-arrow-benchmarks](https://github.com/WayScience/ome-arrow-benchmarks) repository.
+
 <img height="200" src="https://raw.githubusercontent.com/wayscience/ome-arrow/main/docs/src/_static/references_to_files.png">
 
 __Images are often left behind from the data model, referenced but excluded from databases.__
@@ -266,7 +268,8 @@ Notes:
 
 ## Benchmarking lazy reads
 
-Use the lightweight benchmark utility in `benchmarks/` to compare lazy tensor read paths (TIFF source-backed, Parquet planes, Parquet chunks):
+Use the lightweight benchmark utility in `benchmarks/` to compare lazy tensor read paths (TIFF source-backed, Parquet planes, Parquet chunks).
+For more detailed benchmark results and analysis, see the [ome-arrow-benchmarks](https://github.com/WayScience/ome-arrow-benchmarks) repository.
 
 ```bash
 uv run python benchmarks/benchmark_lazy_tensor.py --repeats 5 --warmup 1
@@ -342,4 +345,5 @@ OME Arrow is used or inspired by the following projects, check them out!
 - [`coSMicQC`](https://github.com/cytomining/coSMicQC): performs quality control on microscopy feature datasets, visualized using CytoDataFrames.
 - [`pycytominer`](https://github.com/cytomining/pycytominer): supports feature profiling, normalization, and downstream analysis workflows for image-based profiling datasets.
 - [`iceberg-bioimage`](https://github.com/WayScience/iceberg-bioimage): defines warehouse-oriented patterns for connecting bioimage formats and analytical tables at scale.
+- [`ome-arrow-benchmarks`](https://github.com/WayScience/ome-arrow-benchmarks): contains benchmark results and analysis that inform the design choices in OME Arrow.
 - [`CytoTable`](https://github.com/cytomining/CytoTable): converts image-based profiling outputs into analysis-ready tabular formats such as Parquet.

docs/src/presentations.md

@@ -3,3 +3,5 @@
 ## OME Community Meeting 2026
 
 - Poster PDF: [OME-Arrow 2026 poster (WayScience/ome-arrow)](https://github.com/WayScience/ome-arrow/blob/main/docs/presentations/2026-OME-community-meeting/ome-arrow-2026-poster.pdf)
+
+To see benchmark results that help inform OME Arrow's design decisions, check out the [ome-arrow-benchmarks](https://github.com/WayScience/ome-arrow-benchmarks) repository.

docs/src/why-ome-arrow.md

@@ -40,6 +40,7 @@ OME Arrow is not a universal replacement for OME-Zarr.
 ## Preliminary benchmark signal
 
 Preliminary results in [ome-arrow-benchmarks](https://github.com/WayScience/ome-arrow-benchmarks) show that outcomes are highly workload- and layout-dependent.
+These benchmarks provide important insights into the performance tradeoffs of different OME Arrow layouts and access patterns.
 In the repository's synthetic wide-table plus image-column runs, Arrow-table-native backends can reduce full-table read time and storage size relative to some alternatives, while write performance varies by backend.
 In the OME-Arrow-only benchmark that compares against directory-per-image OME-Zarr and TIFF layouts, full write/read timings and random-read timings diverge in different directions depending on operation type.
 In that same OME-Arrow-only setup, Lance showed random-read timing similar to OME-Zarr (about 0.020 seconds vs about 0.019 seconds average), suggesting Lance can be a practical OME-Arrow-based option for large image repositories when table-native workflows are desired.