Chronological
backtesting.
Model long/flat ETF strategies with next-eligible-open execution. Inspect targets, orders, fills, fees, cashflows, holdings, and equity through a shared accounting ledger.
Private quantitative researchAccounting. Causality. Reproducibility.
Research beyond
the equity curve.
Combine reproducible backtesting, rigorous validation, and AI-assisted research in one connected workflow. Designed for integrated Claude API tools and Claude Code access through MCP.
Integrated AI toolsClaude APIClaude Code compatibilityMCP
Every result has a history.Keep the evidence connected.
A more deliberate
way to research.
Bring rigor to your research.
Register your interest in blackaxiom-research.
Follow every result
back to its evidence.
An equity curve is the beginning of the question.
Inspect the decisions, executions, and accounting behind it.
Research trace / demo_run_004
Synthetic research example- Frozen inputs
- Configuration
- Targets
- Orders & fills
- Cash & holdings
- Report
Account equity
Net of costsBefore costs
Inspect execution
DrawdownMaximum: −1.97%
Accounting ledger
Shared account · USD| Event / date | Execution | Quantity | Fill | Fee | Cash after | Holdings after |
|---|---|---|---|---|---|---|
| Selected 001 / Jan 07 | BUY · ETF-A | 500 | $100.00 | $25.00 | $49,975.00 | 500 |
| Selected 002 / Jan 10 | SELL · ETF-A | 500 | $101.00 | $25.25 | $100,449.75 | 0 |
| Selected 003 / Jan 13 | BUY · ETF-A | 480 | $102.00 | $24.48 | $51,465.27 | 480 |
Illustrative mechanics. No performance claims.
View underlying equity data
| Date | Net equity | Before costs | Drawdown |
|---|
From research question
to inspectable evidence.
Explore a Claude-assisted workflow for strategy logic, experiments, results, and execution. Keep each proposed action connected to its evidence.
AI research workspace
Synthetic AI research exampleConversation & tools
Awaiting reviewResearcher
Compare these two ETF configurations, explain the difference in drawdown, and propose a cost-sensitivity experiment using development data.
Illustrative assistant response
Configuration B has a smaller drawdown in the selected development sample, but higher execution costs. That observation alone does not establish a better strategy.
I propose replaying both configurations under the same fee assumptions before drawing a conclusion.
Selected authorized evidence4 artifacts
Compared development runsA / B
Attached findings to evidencereport:02
Drafted cost-sensitivity planReview
Proposed experiment
Approval required- Change
- Execution fees: 0, 5, 10, 20 bps
- Fixed
- Inputs, signals, execution delay, seed
- Scope
- Development data only
- Budget
- 2 configurations × 4 fee settings = 8 replays
- Output
- Cost, drawdown, and ledger comparison
This control advances a local simulation. It does not submit a research job.
Inspect the proposal before approving the simulated execution.
Explore evidence boundaries
See how unsupported requests are represented.
Select evidence → Compare → Explain → Draft → Review & approveLocal demonstration · no live jobs
One research system.
Connected evidence.
From frozen inputs to inspectable conclusions.
A single thread through the research process.
Chronological
Input & configuration
discipline.
Work with supplied frozen snapshots, strict input preflight, explicit availability, layered configuration, and defined execution assumptions.
snapshot.lockconfig.resolvedpreflight.report
Research planning
& recovery.
Describe experiments before execution. Preserve task identities, resume verified work, compare replays, and protect records through backup and restore.
plan → task → replayResume verified work ↳
Reproducibility
& lineage.
Connect inputs, configurations, runs, artifacts, fitted states, code, environments, and seeds through persistent identities and provenance.
Selection
& validation.
Explore bounded parameter searches, causal walk-forward selection, training and scoring boundaries, retained alternatives, frozen procedures, and controlled final evaluation.
Stress
& uncertainty.
Examine costs, execution delays, parameter neighborhoods, matched baselines, dependent-return inference, and uncertainty with explicit assumptions and evidence limitations.
Portfolio
research.
Compare manually selected candidates and construct equal-weight or inverse-volatility portfolios. Distinguish independent sleeves from shared-account execution, with opt-in bounded combination and allocation search.
CFD-program
evaluation.
Study contract rules through causal account replay, fees and withdrawals, coherent scenarios, matched-risk baselines, and horizon-specific funding and cashflow outcomes.
Evidence registry
& reporting.
Organize immutable research records, annotations, selections, economic assessments, and deployment history. Inspect versioned metrics, decision traces, comparisons, and linked evidence.
Research visualization
& paper replay.
Explore artifact-linked charts, execution markers, temporal masks, lineage diagrams, and supplied-input paper replay through a connected research interface.
Research
copilot.
Turn a research question into a structured hypothesis, rationale, and experiment plan. Keep the proposed next step connected to selected evidence.
Strategy
assistant.
Draft or revise strategy logic, configuration, and tests within approved authoring boundaries. Changes remain proposals until reviewed.
Experiment
planner.
Define parameters, temporal masks, objectives, dependencies, and work budgets before execution.
Results
analyst.
Explain metrics, costs, uncertainty, and differences between authorized runs. Distinguish an observed result from an inferred explanation.
Execution-trace
assistant.
Connect targets, orders, fills, cashflows, and holdings to explanations of a result. Ask why a target did not produce a fill.
Validation
assistant.
Identify assumptions, missing checks, and evidence limitations. Generated explanations do not replace computed validation outcomes.
Portfolio
assistant.
Compare selected candidates and explain allocation choices, independent sleeves, and shared-account trade-offs.
Research report
assistant.
Prepare evidence-linked summaries that distinguish observations, assumptions, and proposed next steps.
Two ways in.
One research system.
Model reasoning and research tool access have distinct roles. Both paths are designed around shared evidence and authorized operations.
Model inference
Claude, integrated into
the research workflow.
The integrated workspace is designed to call the Claude API directly through the application’s AI service. Plan experiments, draft strategy changes, and interpret selected evidence inside blackaxiom-research.
- Provider
- Claude API
- Model
- Not configured in this preview
- Connection
- Illustrative · not connected
- Tool scope
- Selected, authorized research operations
- Context
- Development comparison A / B
Selected research context would be sent to Anthropic for model processing. API credentials belong in the trusted application service.
Tool connectivity
Your research workflow,
connected to Claude Code.
The planned MCP integration will bring blackaxiom-research’s authorized research tools to Claude Code and compatible clients. Inspect evidence, prepare plans, and follow approved jobs through the same research services.
- Inspect configurations and metadata
- Prepare and review experiment plans
- Submit approved jobs and follow progress
- Retrieve reports and execution traces
Claude Code compatibility is described here as an external MCP client path. No live MCP server or verified client connection is provided by this landing page.
Shared research servicesIllustrative architecture
Inside the platformIntegrated AI workspace
Trusted serviceApplication AI service
Model inference ↔Claude Messages API
Authorized operationsResearch services
Accounting · experiments
metrics · evidence
External clientClaude Code
Tool connection layerMCP gateway
Shared evidence ↔Supported research tools
The integrated workspace connects to the application AI service, which calls Anthropic’s POST https://api.anthropic.com/v1/messages endpoint and requests authorized research operations. Claude Code connects through the MCP gateway to those same services.
Integrated API calls use the application’s configured API credentials and billing. Claude Code authentication and billing follow the researcher’s Claude Code setup.
Generated changes remain proposals until approval and service checks are satisfied. Computed metrics and validation outcomes come from research services.
Research boundaries
Supplied inputs. Authorized evidence. Declared budgets. Preserved selection history. Final evaluation evidence stays outside tuning context. No broker orders.
A result is only
as useful as the
evidence
behind it.
Accounting first.
Derive results from actual execution and holdings. Keep fees, cashflows, and account state inside the explanation.
Causal discipline.
Preserve information boundaries throughout fitting, selection, and evaluation. Make what was known, and when, explicit.
Inspectable conclusions.
Keep assumptions, uncertainty, limitations, and provenance attached to the result. Follow a conclusion back to its evidence.
Research with intelligence.
Decide with evidence.
Bring backtesting, validation, portfolio research, and AI-assisted reasoning into one connected process. Explore the integrated Claude workflow or the planned Claude Code path through MCP.
Request Access