Add model-in-loop long-horizon benchmark

Makefile

@@ -1,4 +1,4 @@
-.PHONY: quick lint test verify research-check research-check-unit research-check-deterministic research-check-smoke paper-corpus-check paper-corpus-internal-review-check paper-corpus-claim-check paper-adapter-conformance paper-baseline-run paper-graph-ablation paper-producer-coverage paper-output-promotion paper-candidate-admission paper-contract-projection paper-model-loop paper-llm-catalog-baseline paper-model-loop-analysis paper-toolinkos-parity paper-openapi-closure paper-harness-check goal-completion-benchmark arazzo-long-horizon-benchmark xgen-benchmark xgen-llm-benchmark xgen-scale-snapshot xgen-scale-snapshot-check xgen-scale-acceptance xgen-scale-sweep xgen-scale-gate-check xgen-scale-028-gate-check xgen-scale-contract-ablation bfcl-benchmark bfcl-llm-benchmark bfcl-sweep bfcl-027-gate bfcl-027-gate-check bfcl-028-gate bfcl-028-gate-check bfcl-failure-subset bfcl-inspect-failures bfcl-hard-cases release-check pypi-smoke public-smoke launch-evidence launch-evidence-check observability-evidence observability-evidence-check
+.PHONY: quick lint test verify research-check research-check-unit research-check-deterministic research-check-smoke paper-corpus-check paper-corpus-internal-review-check paper-corpus-claim-check paper-adapter-conformance paper-baseline-run paper-graph-ablation paper-producer-coverage paper-output-promotion paper-candidate-admission paper-contract-projection paper-model-loop paper-llm-catalog-baseline paper-model-loop-analysis paper-toolinkos-parity paper-openapi-closure paper-harness-check goal-completion-benchmark arazzo-long-horizon-benchmark arazzo-long-horizon-model-benchmark xgen-benchmark xgen-llm-benchmark xgen-scale-snapshot xgen-scale-snapshot-check xgen-scale-acceptance xgen-scale-sweep xgen-scale-gate-check xgen-scale-028-gate-check xgen-scale-contract-ablation bfcl-benchmark bfcl-llm-benchmark bfcl-sweep bfcl-027-gate bfcl-027-gate-check bfcl-028-gate bfcl-028-gate-check bfcl-failure-subset bfcl-inspect-failures bfcl-hard-cases release-check pypi-smoke public-smoke launch-evidence launch-evidence-check observability-evidence observability-evidence-check
 
 quick:
 	scripts/quick-check.sh
@@ -110,6 +110,7 @@ paper-model-loop:
 		--model "$$MODEL" \
 		--model-revision "$$MODEL_REVISION" \
 		--provider "$${PROVIDER:-openai-compatible}" \
+		--provider-profile "$${PROVIDER_PROFILE:-default}" \
 		--llm-url "$${LLM_URL:-http://localhost:8000/v1}" \
 		--repeats "$${REPEATS:-1}" \
 		--seed "$${SEED:-17}" \
@@ -186,6 +187,25 @@ arazzo-long-horizon-benchmark:
 		--token-budget "$${TOKEN_BUDGET:-2048}" \
 		--out "$${OUT:-/tmp/graph-tool-call-arazzo-long-horizon.json}"
 
+arazzo-long-horizon-model-benchmark:
+	@test -n "$${MODEL:-}" || (echo "MODEL is required"; exit 2)
+	@test -n "$${MODEL_REVISION:-}" || (echo "MODEL_REVISION is required"; exit 2)
+	@test -n "$${LLM_URL:-}" || (echo "LLM_URL is required"; exit 2)
+	poetry run python -m benchmarks.arazzo_long_horizon.model_run \
+		--model "$$MODEL" \
+		--model-revision "$$MODEL_REVISION" \
+		--provider "$${PROVIDER:-openai-compatible}" \
+		--llm-url "$$LLM_URL" \
+		--api-key-env "$${API_KEY_ENV:-OPENAI_API_KEY}" \
+		--catalog-size "$${CATALOG_SIZE:-1000}" \
+		--lengths "$${LENGTHS:-3,10,30}" \
+		--top-k "$${TOP_K:-8}" \
+		--token-budget "$${TOKEN_BUDGET:-2048}" \
+		--repeats "$${REPEATS:-1}" \
+		--seed "$${SEED:-17}" \
+		--timeout "$${TIMEOUT:-180}" \
+		--out "$${OUT:-/tmp/graph-tool-call-arazzo-model-long-horizon.json}"
+
 xgen-benchmark:
 	poetry run python -m benchmarks.xgen_tool_graph.run --suite all