Add competitive bilateral contracting scenarios in role-specific task-data format
Add all ten bilateral-contracting variants under the competitive interest category, in the same folder layout as the existing benchmark data:
contracting////{input,reference}
The two tasks are two_agent (buyer and seller) and four_agent (two commercial leads and two terms specialists). Both use the same ten contract scenarios. This adds 160 small UTF-8 Markdown and JSON files; no binary assets or software helpers are required.
Classification: contracting is competitive in both this dataset contribution and the corresponding a2-failures run-log contribution. contracting/README.md records the classification. The category update changes documentation only; all 160 task-data files remain byte-for-byte unchanged.
Each role receives its private contract brief and a blank filing template in input/. Commercial leads and bilateral agents additionally have a scoring_reference.json in reference/, staged only after the conversation by the benchmark runner. Specialists receive only their side's term economics; leads receive only their side's reservation price. The seller specialist holds the actual operational metric.
The economics, integrity facts, output schemas, and evaluator criteria are preserved. A deterministic exporter in the companion code branch reproduces this dataset subtree. The runner configuration changes requires_task_data to true and directs agents to read the staged role files. These files should be used together with that companion code change; existing historical runs retain their original scorer.
Validation: all 160 exports verified against the generator; 19 tests passed, including the existing contracting suites, preservation of role briefs, blank templates, reference round trips for all ten variants, and rejection of stale or mismatched references.
lgtm. 🙏