Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.
Rust
# Curated 20-task iteration subset for [email protected].
#
# Used for fast inner-loop A/B tests of routing logic. Same 20 every run so
# deltas are apples-to-apples. See:
# Used by benchmark/run-baseline.sh via --task-list-file for deterministic iteration subsets.
#
# Composition: 5 easy + 8 medium + 6 hard + 1 expert across 13 categories.
# Wall-time projection at -n 8 (3-mode sequential): ~45-60 min total.
# Caveat: 20-task noise floor is +/-2-5 pp; do NOT quote as a headline number.
# --- easy (5) — weak-tier floor; verifies cheap tier still catches simple work
broken-python # software-engineering
cosign-keyless-signing # security
convolutional-layers # machine-learning
jq-data-processing # data-processing
mlflow-register # data-science
# --- medium (8) — discriminator zone; where the classifier matters most
acl-permissions-inheritance # system-administration
cpp-daemon-sighup-segfault # debugging
git-repo-forensics # forensics
iris-dataset-classification # data-science
monorepo-changelog-cli # software-engineering
parking-lot-pathfinding # general
rsa-jwt-token-api-redis-blacklist # security
todos-api # software-engineering
# --- hard (6) — strong-only / near-miss zone; tests the ceiling
application-debug # debugging
breast-cancer-mlflow # machine-learning
db-migration-local-storage # backend
pdf-table-parsing # parsing
sakila-sqlite-queries # sql
server-log-analysis # general
# --- expert (1) — top-of-difficulty probe
book-portfolio-analysis # data-engineering