The existing finance environments cover stock trading (finrl_env) and
financial document QA (finqa_env), but neither covers transaction-level
risk decisioning — the core loop of a payments fraud/risk team.
I'd like to propose contributing a Fraud Triage Environment: an agent is
streamed synthetic transactions one at a time and must choose to
approve/flag/escalate/block each one, with asymmetric rewards reflecting
the real cost trade-off (missing fraud costs more than over-flagging a
legitimate transaction).
I have a working implementation with passing tests following the existing
env conventions (models.py/client.py/server pattern, same as connect4_env).
Happy to open a PR if this fits the project's direction.
The existing finance environments cover stock trading (finrl_env) and
financial document QA (finqa_env), but neither covers transaction-level
risk decisioning — the core loop of a payments fraud/risk team.
I'd like to propose contributing a Fraud Triage Environment: an agent is
streamed synthetic transactions one at a time and must choose to
approve/flag/escalate/block each one, with asymmetric rewards reflecting
the real cost trade-off (missing fraud costs more than over-flagging a
legitimate transaction).
I have a working implementation with passing tests following the existing
env conventions (models.py/client.py/server pattern, same as connect4_env).
Happy to open a PR if this fits the project's direction.