Execution software. Risk controls. Your codebase.
Connect quantitative tools to your agents, or commission the execution and risk infrastructure your desk needs. We define the scope, build the system and hand over the bespoke code.
Software engineering, not investment management.
Start with the tools.
quant-swarm 0.4.0 · BetaRuns locally in your MCP client. Some tools require a configured service and credentials.
uvx --from quant-swarm==0.4.0 swarm-warden-mcp
Requires uv and Python 3.11 or later. The MCP client starts the process. Tool schemas can be listed without a token; tool calls require access configuration.
Setup instructions. Not connected to your IDE.
The npm launcher is not available through the public registry. Use the uvx setup.
Setup instructions. Not connected to your IDE.
If Cursor did not open, use the configuration below. Merge this server entry with your existing configuration. Do not replace other servers. Restart or reload the client, inspect the tool list, then configure access locally.
{
"mcpServers": {
"swarm-warden": {
"command": "uvx",
"args": ["--from", "quant-swarm==0.4.0", "swarm-warden-mcp"]
}
}
}Setup instructions. Not connected to your IDE.
MCP servers you can add to your own agent and inspect before connecting.
Pre-trade controls gate what any automated process may do, and you can inspect them.
Commissioned deliverables are defined in your statement of work and handed over.
Inspect the control layer.
Inspect historical tests and explore the control layer. Every panel identifies its source, period and limitations. Historical results are not live account performance.
Pre-trade controls
Demonstration / Control flowSource evaluator: execution-service risk gates (C#, .NET 8) · code-reviewed scenario diagram
| Check | Status | Why |
|---|---|---|
| Choose a scenario above to inspect the demonstration trace. | ||
Evidence and method
Historical backtest
Research backtest / Verification incompletePeriod: 2021-08-18 → 2026-08-20 · Originating project: momentum factor research (daily ETF universe) · Cost model: 5 bps/side + 2 bps slippage + 1 bps half-spread
Real market data (IEX daily bars), net of the configured cost model. Not a live account result.
Evidence and method
Regime context
Demonstration / Synthetic exampleLabelled timeline (synthetic) · Classifier: market.regime · Nothing here is inferred from a live feed or from later data.
Illustrative labels only — not a reviewed artifact.
| Regime label | Recorded window | Classifier |
|---|---|---|
| Trend-up (equity) | 2026-07-06 → 2026-08-14 | market.regime · illustrative |
| Choppy / mean-reverting | 2026-06-02 → 2026-07-03 | market.regime · illustrative |
| Risk-off / drawdown | 2026-05-05 → 2026-06-01 | market.regime · illustrative |
| Low-volatility drift | 2026-04-01 → 2026-05-04 | market.regime · illustrative |
Evidence and method
Tournament / research evidence
Demonstration / Synthetic exampleCohort: demonstration (synthetic) · Ruleset: 0.4.0 · Evaluation period: 2026-08-01 → 2026-08-31 · Status: illustrative
Elo measures relative results within a declared pool; it is not a return forecast.
| Genome | Elo | Outcome | Δ bps | Status |
|---|---|---|---|---|
| 8f3a1c02d9e4 | 1524 | beats champion | +12.4 | rated |
| 7b0e55af21c8 | 1511 | beats champion | +8.9 | rated |
| 3cd9f1e07b26 | 1504 | inconclusive | +1.1 | rated |
| a5d2c84b90f3 | 1496 | loses | −4.7 | rated |
| 9e71b3d05a42 | 1488 | loses | −9.2 | rated |
| f0a8c6d21b7e | — | pending | — | not rated |
Evidence and method
Choose a capability. Inspect the contract.
See the inputs, outputs and connection requirements before you add a tool to your agent.
Warden
Inspect order constraints, feature provenance and execution-cost assumptions.
Inspect Warden tools →Gym
Probe historical fragility and compare candidates on matched paths.
Inspect Gym tools →Data
Inspect derived market context and point-in-time feature coverage.
Inspect Data tools →Tournament
Submit a candidate for hosted comparison and inspect the recorded outcome.
Inspect Tournament tools →Try one safe example: warden.cost_check
Demonstration / Synthetic exampleDemonstration. This request is not sent to a server.
Request (tool arguments)
Response (recorded fixture)
Evidence and method
Separate research, controls and execution.
Strategy evaluation informs candidates. Risk controls govern order eligibility. Execution services handle the approved path. The public MCP interface does not transfer the proprietary engine.
Research & evaluation
Candidate strategies are evaluated on historical data under declared train/test boundaries. Only reviewed results inform later stages.
evaluation · candidatesPre-trade controls
Every order passes conditional risk checks before it is eligible. A blocked or reduced order stops or narrows here — never silently downstream.
policy · eligibilityExecution services
Approved orders are routed by dedicated execution services with idempotency and recovery behaviour defined up front.
execution · approved pathPublic panels on this site show demonstrations and reviewed historical artifacts only. No browser request here reaches private execution, training or data-mutation services.
Three ways to engage.
Fixed-scope engineering engagements. Deliverables, dependencies, acceptance tests and IP terms are agreed in the statement of work.
Execution infrastructure
Turn an execution requirement into an operable system.
Agreed broker/venue adapter, idempotency and recovery tests, deployment definition and a runbook. Scope includes only agreed instruments and venues.
Risk-control engineering
Make pre-trade constraints explicit, testable and observable.
Deterministic checks, failure/recovery scenarios, an audit-event schema and operator controls. Not a regulatory-compliance certification.
Research-to-production review
Identify the gap between a research result and a deployable service.
Reproducibility review, data/feature-boundary assessment and a prioritised remediation and acceptance plan. No guaranteed profitable strategy.
Deliverables, dependencies, acceptance tests and IP terms are agreed in the statement of work. Ongoing support is optional and separately scoped.
Built for your team. Handed over to your team.
You own the commissioned deliverables defined in your statement of work. Pre-existing IP, hosted services and third-party licences are identified separately before work begins.
Handover acceptance matrix
| Deliverable | Handover position |
|---|---|
| Bespoke source and tests | Client-controlled repository |
| Infrastructure definitions | Agreed client environment |
| Dependency / third-party licence inventory | Delivered and acknowledged |
| Build / deploy instructions | Clean-environment reproduction |
| Operational runbook | Start / stop / rollback / recovery walkthrough |
| Secrets | Provisioned in client control, never source |
| Data / model artefacts | Transferred or licensed only where explicitly scoped |
| Background engine | Excluded unless expressly licensed |
| Hosted MCP access | Independent service entitlement, not engine ownership |
Compare delivery models, not sales promises.
| Full-time hire | Generalist consultancy | The 1.21 Initiative engagement | |
|---|---|---|---|
| Commercial model | Employment and ongoing team costs | Time-based or fixed-scope by agreement | Fixed scope, deliverables and acceptance terms agreed in writing |
| Technical coverage | Depends on the individual and supporting team | Depends on the assigned delivery team | Quantitative software engineering within the agreed scope |
| Commissioned IP | Normally governed by employment terms and applicable law | Governed by supplier contract | Client ownership of commissioned deliverables as defined in the statement of work |
| Existing IP | Employer/employee terms and third-party rights apply | Supplier background-IP terms may apply | Pre-existing engine/methods remain separate; any required licence is explicit |
| Operational continuity | Requires internal documentation and staffing | Depends on contracted handover/support | Runbooks, deployment instructions and agreed knowledge transfer included |
| Exit and support | Team-owned operations after employment changes | Contract-dependent | Client operation against documented dependencies; optional, separately scoped support |
Bring the system constraint.
Tell us what must be built, what it must connect to and how you will judge acceptance. The next step is a technical review with the person responsible for the build.
Three stages, transparently
Three steps describe the information we need — not mandatory sales stages. A short, non-confidential brief is enough to start.
Engineering due diligence
For engineering due diligence, a walkthrough of selected evidence can be arranged under a mutual NDA after qualification and disclosure approval.
Subject to qualification, disclosure approval and a signed mutual NDA. No confidential material before both are in place.
Software engineering, not investment management.