Pareto Cost Frontier
Find the cheapest solution that holds quality.
Replace recurring model calls with tested deterministic modules and schema adapters. Every candidate is measured against the no-extra-material baseline on quality, cost, tokens, calls, and latency.
Baseline vs. cheapest verified compositions
A cost-cutting submission enters the Pareto frontier only when it preserves the challenge quality floor and strictly reduces tokens, model calls, cost, or latency.
Compose an editable 3D scene and motion pack
Baseline without extra material (baseline:no_extra_material): quality 0.72, USD 0.065, 16,000 tokens, 4 model calls, 3400 ms.
Cheapest Pareto plan: parse_scene_det -> synthesize_motion_det at USD 0.0000, 0 tokens, 0 model calls, 53 ms (saves USD 0.0650 and 16,000 tokens per run).
| Typed edge | Model-backed step | Deterministic replacement | USD saved | Tokens saved | Calls saved |
|---|---|---|---|---|---|
parsed_scene_graph/v1 -> verified_motion_pack/v1 | synthesize_motion_hybrid | synthesize_motion_det | USD 0.0120 | 2,800 | 1 |
raw_scene_brief/v1 -> parsed_scene_graph/v1 | parse_scene_llm | parse_scene_det | USD 0.0180 | 4,200 | 1 |
Deduplicate and normalize multilingual records
Baseline without extra material (baseline:no_extra_material): quality 0.74, USD 0.092, 24,000 tokens, 5 model calls, 4800 ms.
Cheapest Pareto plan: csv_unicode_norm_det -> minhash_cluster_det at USD 0.0000, 0 tokens, 0 model calls, 43 ms (saves USD 0.0920 and 24,000 tokens per run).
| Typed edge | Model-backed step | Deterministic replacement | USD saved | Tokens saved | Calls saved |
|---|---|---|---|---|---|
normalized_rows/v1 -> verified_entity_table/v1 | minhash_plus_tiebreak_hybrid | minhash_cluster_det | USD 0.0060 | 1,400 | 1 |
raw_customer_csv/v1 -> normalized_rows/v1 | csv_prompt_cleaner_llm | csv_unicode_norm_det | USD 0.0280 | 6,400 | 2 |
Verify MCP tool schemas and corner-case refusals
Baseline without extra material (baseline:no_extra_material): quality 0.68, USD 0.048, 11,500 tokens, 3 model calls, 2600 ms.
Cheapest Pareto plan: extract_json_schema_det -> corner_case_fuzzer_det at USD 0.0000, 0 tokens, 0 model calls, 53 ms (saves USD 0.0480 and 11,500 tokens per run).
Every step in this challenge is already zero-model deterministic.
Synthesize the lowest-cost plan
loop-engine worker-pipeline arena --origin https://arena.teleon.dev --action plan_composition_menu --challenge-id ch_creative_motion_01 --objective-mode cheapest_solution
Candidate graphs, synthesized plans, and volunteer retrieval probes are tracked separately from Stage-1 orthogonal-family acceptance and Stage-2 holdout measured benefit. The no-extra-material baseline stays eligible on every challenge board.