Library, not the tab

N-way is pip. The lab stays two.

A vs B in a browser is the right default: you can see the cut. Three, four, ten strategies on the same file is a table, not ten columns of cards. That table already ships.

Why the tab refuses a third column

Each extra column is another full pass, another list, another color. Three or more names also want avg size, quality, and elapsed, which is a table. The library already prints that table.

compare: one file, many names, one table

chunking-strategy compare doc.txt -s fixed_size,sentence_based,recursive_character

# JSON if you are piping into something else
chunking-strategy compare doc.txt -s fixed_size,sentence_based,markdown_chunker --format json

Each row is strategy, n_chunks, avg_size, quality_score (0.4 size consistency + 0.4 sentence-boundary + 0.2 span coverage; CDC / code / JSON skip the .!? term), and elapsed.

Python equivalent:

from chunking_strategy import create_chunker
for name in ["fixed_size", "sentence_based", "recursive_character"]:
    r = create_chunker(name).chunk(text)
    print(name, len(r.chunks), r.quality_score)

benchmark: timings, repeats, custom files

When you care about speed across files, or you want a custom algorithm in the same run:

chunking-strategy benchmark doc.txt --strategies "fixed_size,sentence_based"
chunking-strategy benchmark --quick doc.txt
chunking-strategy benchmark doc.txt --custom-algorithms ./my_chunker.py
chunking-strategy custom benchmark my_chunker.py --compare-with fixed_size --compare-with sentence_based

Reports land in chunking_benchmarks/ (or --output-dir / CHUNKING_BENCHMARK_OUTPUT_DIR) as console + JSON + CSV. See examples/17_benchmark_demo.py and BENCHMARKING.md in the repo if you need every flag.

What to do in the lab instead

Pick the two names you are actually arguing about. Look at the underline. Copy the snippet. Then run compare on the full corpus with those names plus the ones you did not put on screen. Hierarchical recursive still compares as two columns; the tree is the parent/child list, not a third strategy slot.

Bring-your-own strategies plug into the same compare / benchmark commands. That path is Bring your own.

MIT. Lab matches chunking-strategy==0.5.0.