Skip to content
ARCHYTASAI

V.02 · catalog/financial-services

Financial Services. Five investment workflows, one simulated financial ecosystem.

Each environment runs on FinanceWorld, a simulated ecosystem of issuers, funds, lenders, brokers, custodians and administrators. Agents work across filings, agreements, spreadsheets, code and operational systems, and are graded on calculations and state that the verifier recomputes independently.

A point-in-time clock controls what each task can see, so agents can't solve historical tasks with future information. Entities recur across environments: a covenant breach found in credit can surface in a portfolio risk review. Each episode still starts from its own isolated snapshot.

Environments. Each runs on FinanceWorld, and each has its own datasheet.

FinanceWorld. One world spec, shared by every Financial Services environment.

spec · finance-world.yaml14 fields
type:
Executable RL training and evaluation environment.
world:
FinanceWorld: one simulated financial ecosystem of companies, issuers, funds, lenders, brokers, custodians and administrators, with market data, documents, transactions and policies, shared by every environment in this vertical.
point_in_time:
A simulated clock controls what each task can see at its decision date, so historical tasks can't be solved with future information.
data:
Synthetic issuers, securities, funds and market data. Licensed historical data can be added where an engagement allows it.
modes:
RL training and evaluation, plus SFT trajectories on request.
interaction:
Stateful, multistep tool use across documents, data services, spreadsheets, Python and workflow actions.
state:
Resettable, versioned financial state and document store. Every episode starts from an isolated, seeded snapshot.
tools:
JSON-schema tools, with an optional MCP adapter.
access:
The agent sees only what its role, permissions and the simulated date allow. The verifier works from privileged ground truth the agent can't reach.
verification:
Independently recomputed financial results, state and policy assertions, reference solutions and expert rubrics. Legitimate alternative methods are accepted.
splits:
Separate training and held-out evaluation distributions.
deployment:
Containerized, isolated execution, with a customer-controlled deployment option.
engagement:
Custom environment build and accepted task instances, with optional ongoing expansion.
licensing:
Commercial use rights. Exclusivity, ownership and any third-party data licenses are agreed for each engagement.

Datasets. Financial Services data you can license without a full environment.

Each dataset is a separate deliverable. The matrix shows what each one contains, so you can tell a set of task descriptions from solved demonstrations, preference data, executable validators or complete environments. Every dataset is built to order.

catalog · financial-services/datasetsCustom build
  • DATA-FIN-001Expert financial task packExpert-written instructions, source records, reference answers and rubrics.

    Primary use · SFT, Evaluation

    • IncludedTask descriptions
    • IncludedSolved demonstrations
    • Not includedPreference annotations
    • Not includedExecutable validators
    • Not includedComplete environments
  • DATA-FIN-002Financial tool-use trajectoriesMulti-step expert and validated agent executions.from ENV-FIN-101…105 runs

    Primary use · SFT, Imitation learning

    • IncludedTask descriptions
    • IncludedSolved demonstrations
    • Not includedPreference annotations
    • Not includedExecutable validators
    • Not includedComplete environments
  • DATA-FIN-003Verifiable financial reasoning tasksNumerical problems, input data and executable verifiers.

    Primary use · RLVR, Evaluation

    • IncludedTask descriptions
    • Not includedSolved demonstrations
    • Not includedPreference annotations
    • IncludedExecutable validators
    • Not includedComplete environments
  • DATA-FIN-004Financial failure and recovery tracesIncorrect actions, diagnosed failures and corrected trajectories.from ENV-FIN-101…105 runs

    Primary use · Robustness training

    • IncludedTask descriptions
    • IncludedSolved demonstrations
    • IncludedPreference annotations
    • Not includedExecutable validators
    • Not includedComplete environments
  • DATA-FIN-005Financial policy and judgment rubricsExpert evaluation criteria, policy rules and acceptable outcomes.

    Primary use · Reward modelling, Evaluation

    • IncludedTask descriptions
    • Not includedSolved demonstrations
    • IncludedPreference annotations
    • Not includedExecutable validators
    • Not includedComplete environments
  • DATA-FIN-006Adversarial financial tasksData leakage, false reconciliation, flawed models and reward exploits.

    Primary use · Robustness, Grader auditing

    • IncludedTask descriptions
    • Not includedSolved demonstrations
    • Not includedPreference annotations
    • IncludedExecutable validators
    • IncludedComplete environments

Every environment is built to order. Tell us which workflow you need to train or evaluate, and where you are in the process.

Request sample tasksDiscuss custom developmentRequest technical integration detailsSee ways to buy