Mine Alpha
Find research opportunities across evidence sources and market behaviour.
In development · Predvix Alpha OS
An autonomous research system for the full quantitative workflow.
Mine opportunities, turn evidence into testable strategies, run controlled experiments, challenge promising results, and preserve what the research learns.
1 Research question
Which post-earnings signals remain predictive after realistic costs and out-of-sample validation?
UniverseHK large-cap equities
Horizon1–20 trading days
2 Hypotheses
H1 · Revision breadthEstimate revisions after surprise
H2 · Volume persistenceAbnormal participation after event
H3 · Sentiment divergencePrice response vs. text signal
3 Experiments
EXP-01 · Revision × driftBaseline · costs included
EXP-02 · Volume filterWalk-forward validation
EXP-03 · Sentiment spreadChallenge in progress
4 Candidates
STR-12Revision continuation
STR-18Sentiment divergence
STR-21Liquidity reversal
5 Validation
0.68illustrative score
Out-of-sample Passed
Replication Passed
Cost stress Passed
Stability Review
The complete research loop
One system from question to evidence.
Alpha OS is designed as a connected research process rather than a collection of disconnected AI features.
Form Hypothesis
Express the economic idea as an explicit, testable claim.
Build Strategy
Translate the hypothesis into a transparent strategy specification.
Backtest
Evaluate the strategy with controlled assumptions and realistic costs.
Challenge
Stress the result for leakage, fragility, overfitting and bad assumptions.
Validate
Use holdouts, replication and robustness tests before trust.
Learn
Store evidence and failures as reusable institutional memory.
Alpha Mining
Search systematically for promising questions.
Alpha OS is intended to combine financial research, market and event data, price and volume behaviour, prior experiments, and portfolio needs into a structured opportunity pipeline. The objective is not to produce endless random strategies, but to identify hypotheses worth spending research budget on.
Strategy Intelligence
Turn an idea into an explicit experiment.
Research ideas are translated into auditable definitions covering universe, signals, entry and exit logic, position sizing, risk constraints, data requirements, execution assumptions, and the validation protocol. This creates a common language between researchers, agents, and the backtesting engine.
Alpha Command
Backtest as an experiment, not a screenshot.
Alpha Command is the experimentation layer: repeatable backtests, parameter studies, cross-security analysis, regime breakdowns, realistic cost assumptions, and versioned results. Every meaningful run can be tied back to the research object and assumptions that produced it.
Challenge & validation
Promising results should receive more scrutiny, not less.
The research-integrity layer is designed to check data timing, parameter sensitivity, execution assumptions, holdout discipline, statistical fragility, and independent replication. Its job is to make autonomous research harder to fool—even when an initial backtest looks attractive.
Core research principle
Invalidate First.
Finding an attractive backtest is not the end of research. Alpha OS is designed to challenge promising results, test their assumptions, and reject strategies that do not survive scrutiny.
Alpha OS doesn't just search for reasons a strategy works. It actively searches for reasons it doesn't.
2.31Sharpe
Look-ahead issueDetected
Parameter instabilityHigh
Execution assumptionsCorrected
Data-snooping riskReview
0.68Sharpe
Decision
Rejected
Does not meet robustness threshold.
Illustrative example only. Values shown above are not historical Predvix investment performance.
Research principles
The operating rules behind Alpha OS.
These principles describe how we want autonomous quantitative research to behave. They are methodology goals for a product still in development, not claims that every capability is already production-ready.
Write the hypothesis before optimizing.
Separate the economic idea from the code. Define what should happen, why it should happen, the required data, and what would count as evidence against it.
Make experiments rerunnable.
Record strategy definitions, data assumptions, parameters, costs, execution assumptions and the environment needed to understand a result later.
Attack the strongest-looking results.
Promising backtests deserve more scrutiny. Test nearby parameters, timing shifts, cost shocks, regimes, leakage risks and alternative explanations.
Protect data that has not influenced development.
Development, validation and holdout periods should be treated differently. Once a holdout influences strategy choices, the research record should say so.
Rebuild important findings independently.
Where practical, a separate implementation should receive the hypothesis—not the original code—and attempt to reproduce the result.
Store failed research as knowledge.
Rejected hypotheses, broken assumptions and failed replications are useful information. Research memory should compound from negative results as well as successes.
Who Alpha OS is for
Built for serious quantitative research.
The same research loop, whether you run a desk or run your own book. Discipline shouldn't depend on headcount.
Institutions
Systematise research without lowering the bar.
Give a team one auditable path from question to evidence, with every experiment versioned and every rejection retained.
- Proprietary desks Scale idea throughput, keep verification strict
- Family offices Auditable internal research around investment hypotheses
- Asset managers Connect idea generation with disciplined evidence
- Research teams Institutional memory across experiments and researchers
Individuals
Institutional method, without an institution.
The same challenge-first workflow a desk would run, so a strategy you build alone still has to survive the tests that matter.
- Independent quants Turn research questions into reproducible experiments
- Systematic retail traders Test an idea properly before risking capital on it
- Aspiring quant researchers Learn the loop by working inside it
- Finance students Practise research that would survive real scrutiny
Research memory
The research should get smarter even when a strategy fails.
Alpha Forge is the knowledge layer behind Alpha OS: sources, claims, hypotheses, experiments, outcomes and contradictions can remain connected rather than disappearing into old notebooks and chat threads. Each serious investigation is kept as a versioned research object: question, mechanism, universe, datasets, strategy definition, experiment protocol, results, robustness checks, contradictory evidence, replication outcome and status.
Source-linked conclusions
Connect findings back to research, datasets, assumptions and the experiment that produced them.
Versioned research history
Retain strategy definitions, parameters, environment information, results and subsequent challenges.
Institutional failure memory
Record why an idea failed so future research can avoid repeating the same mistakes.
Compounding context
Use accumulated evidence to improve future hypothesis prioritization and research design.
Early access
Help us build Alpha OS around real research workflows.
We are speaking with serious quantitative researchers and design partners while the product is in development.