Too many metrics
A return alone leaves drawdown behavior, trading costs and sample limitations unanswered.
Quantitative research · AI planned
Backtest with Data.
Understand Your
Strategy with AI.
AI Backtest Lab is a private quantitative research platform for historical crypto backtesting. Inspect strategy performance and risk before deciding what to test next. Claude-powered explanations and research assistance are planned.
Private research workspace · public early access planned.
Build the hypothesis. Backtest with data. Understand the risk. Validate again.
AI Backtest Lab / Research loop01 — 06
Historical price context
Historical windowClosed candlesResearch only
Define the hypothesisNot simulated performance
✳Claude insightsConcept · Planned
Not just “how did it perform?”
What should you test next?
Illustrative research view · No measured performance shown
Build → Backtest → Quantify → Explain → ValidateScroll to explore ↓
Hidden risks
Historical results can conceal overfitting, market sensitivity and assumptions that deserve scrutiny.
Slow research iterations
Turning a result into the next testable question still takes deliberate, manual work.
One research loop.
Six deliberate steps.
Every decision begins with quantitative evidence. Claude is planned to help translate that evidence into a better research question.
Step 01 / Research loop
Define strategy
Pick a strategy, parameters, symbol and historical window.
Private workspace · not publicly available Historical price context
Historical windowClosed candlesResearch only
Define the hypothesisNot simulated performance
Step 02 / Research loop
Run backtest
Execute a historical simulation with explicit data and cost assumptions.
Private workspace · not publicly available Historical price context
Historical windowClosed candlesResearch only
Inspect the evidenceNot simulated performance
Step 03 / Research loop
Measure results
Inspect recorded outcomes, drawdowns, costs and individual trades.
Private workspace · not publicly available Historical price context
Historical windowClosed candlesResearch only
Inspect the evidenceNot simulated performance
Step 04 / Research loop
Analyze with Claude
Planned: send structured metrics, costs and assumptions to Claude for risk interpretation and explanations.
Planned Claude integrationReport → Interpretation
What does the evidence actually say?
Claude will read the supplied report, identify limitations and explain the metrics used.
Planned workflow · Concept previewStep 05 / Research loop
Form new hypotheses
Turn an observation into a question you can test quantitatively.
Planned research workflowObservation → Hypothesis
What deserves another test?
A suggested test remains a hypothesis until a quantitative experiment validates it.
Planned workflow · Concept previewStep 06 / Research loop
Validate again
Use the quantitative engine to test follow-up hypotheses. Broader out-of-sample validation is planned.
Planned research workflowHypothesis → New test
One answer. A better question.
The researcher chooses the next experiment. AI has no authority to run or trade automatically.
Planned workflow · Concept previewBuilt on evidence.
Not AI guesswork.
Calculations stay in the engine. Strategy assumptions and recorded outcomes remain inspectable in the research workspace.
Historical backtesting
Trace a strategy through historical data and explicit execution assumptions.
Private workspace · not publicly availableStrategy parameters & versions
Keep strategy settings and source snapshots alongside research evidence.
Private workspace · not publicly availableTrade-level inspection
Inspect the chart, recorded costs and individual simulated trades.
Private workspace · not publicly availableComparative research
Explore version and cohort evidence within the supported research workflow.
Private workspace · scope-dependentMarket scope
Historical backtesting focuses on crypto. Vietnam equity research is an internal preview. Source-backed backtesting and execution are not available. Other market views provide informational context.
Planned: Claude explains
what the evidence says.
The planned Anthropic Claude API integration will receive structured backtest metrics, trade summaries, strategy parameters, data coverage and cost assumptions. It will help researchers interpret drawdowns, identify weaknesses and propose testable hypotheses. The quantitative engine calculates results and validates each follow-up experiment; Claude does not predict returns or place orders.
Claude API integration is planned and is not yet available in the current product.
01 / Quantitative inputThe engine supplies
The engine supplies
the evidence.
strategy_versionengine record
historical_windowengine record
recorded_costsengine record
risk_metricsengine record
data_limitationsexplicit
No exchange keys. No trading authority.
Only the research report and its context.
Illustrative report · Not a live AI response
Start with what the evidence supports.
01
What did the report measure?Observe
Interpret recorded outcomes and reference the metrics supplied by the engine.
02
What evidence is missing?Question
Highlight cost assumptions, sample limitations and the uncertainty of historical results.
03
What should we validate next?Test
Suggest an out-of-sample or sensitivity experiment. The researcher decides whether to run it.
Performance explanationPlannedRisk & weakness reviewPlannedRun-specific research Q&APlannedNext experiment suggestionsPlanned
Researchyoucanverify.
Our engine is responsible for computation. The planned Claude layer is responsible for interpretation. A compelling explanation remains a hypothesis until it survives another test.
Versioned research evidence
Explicit data and cost assumptions
Out-of-sample validation · planned
Transparent AI limitations
DatasetStrategyEngine resultAI report PlannedNext test
Backtesting is hypothetical. Past performance does not guarantee future results. AI-generated explanations may be incomplete and must be independently validated.
A platform with
clear responsibilities.
Research, interpretation and execution environments remain separate. The planned AI analysis service has no authority to place orders.
Backtest service
Historical simulations, inspectable research results and transparent assumptions.
Private foundationClaude AI research
Grounded report explanations, run-specific questions and follow-up hypotheses.
PlannedDemo sandbox runtime
Private sandbox infrastructure exists. Broader forward-validation remains planned.
Private infrastructureMarket indicator insights
Existing source-specific context views; broader public research access is a roadmap goal.
Private context viewsLive trading is outside this public product. Any future execution pilot remains conditional on separate risk review and explicit authorization.
Reliable research first.
Then smarter iteration.
A narrow, measurable path from quantitative backtesting to grounded AI research. Roadmap goals may change with evidence.
01
Q4 2026 — Q1 2027
Private backtest core exists · Claude POC plannedBacktesting core & Claude POC
Continue hardening auditable research evidence and define the first grounded Claude proof of concept.
02
Q2 — Q3 2027
PlannedValidation & demo forward-testing
Extend out-of-sample research and validate differences against isolated sandbox execution.
03
Q4 2027 — Q1 2028
PlannedAI research workspace
Run-specific questions, experiment comparisons and testable research hypotheses.
04
Q2 — Q3 2028
PlannedResearch platform beta
Carefully evaluated early access, shareable research reports and feedback from real users.
For people who prefer
evidence to hype.
AI Backtest Lab is an independently developed quantitative research startup for independent researchers, crypto strategy developers and systematic traders. We focus on repeatable strategy evaluation and planned AI-assisted explanations.
Our direction is an evidence-led research workflow: inspect the data and assumptions behind a result, use planned Claude analysis to understand its limits, and validate the next hypothesis through quantitative backtesting.
09 / What comes nextResearch better.
Try the interactive demo
Research better.
Understand more.
Test again.
Try the interactive demo Private research workspace · public early access planned. Contact the founder for partnerships or product questions, or request support using the links below.
A few clear answers