# Wansiri Research Note # WRN-P01 --- Does Better Research Produce Better Live Decisions? ### Live validation of the first stage of the Wansiri research pipeline **Publication type:** Production-validation research\ **Status:** Ongoing experiment\ **Observation requirement:** Minimum 30 days before substantive conclusions are drawn ------------------------------------------------------------------------ ## Abstract A strategy can survive historical validation and still fail when exposed to the live market. But there is another question that is often overlooked: **Does the research process that selected the strategy actually improve live decisions?** WRN-001 established the first stage of the Wansiri research framework: a profitable backtest is not sufficient evidence of a genuine trading edge. WRN-P01 takes the next step. The output of Stage 1 has been deployed into production for live validation. The objective is not simply to determine whether the resulting strategies make money. The experiment tests whether **Stage 1 of the research process improves the quality of live strategy selection, decision-making and execution relative to appropriate controls**. This creates a parallel production-validation branch alongside the continuing historical research pipeline. The experiment is ongoing. No conclusion about the incremental value of Stage 1 should be drawn until sufficient live evidence has been collected and analysed. ## 1. From Backtesting to Live Validation WRN-001 established a simple principle: > **A profitable backtest is the beginning of research, not the end.** The Wansiri research process therefore does not move directly from a profitable backtest to deployment. Strategies first pass through statistical and robustness validation. But even a strategy that survives those tests has only been evaluated using historical information. Production introduces a new information set. The question becomes: > **Does the additional research performed at each stage actually > improve the quality of decisions when exposed to unseen market > conditions?** This is why production validation runs in parallel with the research pipeline. Each major research stage can be evaluated for its incremental contribution. The framework does not assume that a research improvement works. It is designed to measure whether it does. ## 2. What WRN-001 Tested The Stage 1 research described in WRN-001 did not evaluate each strategy configuration under one generic market condition. The research campaign evaluated configurations across multiple historical market regimes. Each configuration was therefore tested under different regime conditions. This produced regime-conditioned evidence about how each configuration performed across different market environments. Only configurations satisfying the Stage 1 validation requirements were allowed to progress into the downstream research process. WRN-001 therefore did not simply produce a list of profitable strategies. It produced a validated strategy universe containing configurations with historical evidence across different regime conditions. WRN-P01 takes that regime-conditioned research output into production. ## 3. The Stage 1 Production Experiment The first production experiment begins with the validated strategy universe generated by WRN-001. The production system observes the current market regime and uses the historical research results to determine which validated configurations are currently eligible. It then ranks eligible candidates and selects a preferred configuration for production. Production is therefore not testing the strategies against regimes for the first time. That historical regime testing has already occurred in WRN-001. The new question is whether the regime-conditioned research results can be converted into effective live decisions when the system is exposed to a continuously changing market. This creates a more complex problem than simply putting one backtested strategy into production. The live system has to manage the universe of validated strategies simultaneously. ## 4. The 48-Regime Problem The production system evaluates **48 different regime states**. These states can overlap, so several regime states can be active simultaneously. Because the WRN-001 strategy universe contains configurations that were historically validated under different regime conditions, multiple configurations can become eligible at the same time. The production system must resolve this contention. ``` text Historical WRN-001 Research | v Configurations tested across multiple regimes | v Validated Strategy Universe | v Current Live Market | v 48 Regime States | v Multiple Active States | v Multiple Valid Candidates | v Ranking / Conflict Resolution | v Production Decision | v Live Execution ``` This introduces an important distinction. A strategy can be historically valid. The regime-specific evidence can be valid. The ranking can be valid. Yet the combination of all three can still produce a poor live decision. Production therefore provides information that individual historical strategy tests cannot. ## 5. What Is Being Tested? The experiment evaluates several distinct questions. ### 5.1 Strategy validity Do strategies that survived WRN-001 continue to exhibit evidence of an edge in unseen live data? ### 5.2 Selection validity Does the Wansiri ranking process select better candidates than appropriate controls from the same strategy universe? ### 5.3 Decision-system validity Can the production infrastructure correctly resolve overlapping regime states and competing strategies that have already passed historical validation? ### 5.4 Execution validity Can the selected strategy actually be executed as intended on the target venue? ### 5.5 Research value Most importantly: > **Did WRN-001 improve the quality of the live decision process?** This final question is the reason for the experiment. ## 6. Controlled Production Testing The experiment includes control groups constructed from the same underlying strategy universe. This is important. The comparison is not simply between one Wansiri strategy and an unrelated benchmark. The objective is to determine whether the selection and validation process itself adds information. The live comparison is designed to examine: - Selected strategies. - Control strategies. - Performance by regime. - Selection events. - Strategy contention. - Realised execution. The resulting evidence is intended to help distinguish between the performance of the underlying strategy universe and the incremental value created by the Wansiri research process. ## 7. Production Is Also an Execution Experiment The live test is not only about strategy performance. It is also the first time the complete decision can be observed moving through the production infrastructure. The system must: - Identify the current regime. - Determine eligible strategies. - Resolve competing candidates. - Select the intended configuration. - Generate the appropriate orders. - Send them to the target venue. - Correctly capture the resulting executions. The experiment therefore observes execution characteristics such as: - Implementation Shortfall. - Slippage. - Tick-to-quote timing. - Tick-to-trade timing. - Fill behaviour. - Execution failures. This allows strategy performance to be considered separately from implementation performance. A strategy can have a genuine edge but lose economic value through execution costs. Conversely, poor live performance does not necessarily invalidate the underlying strategy if the difference can be attributed to implementation. The experiment is intended to collect evidence that helps distinguish these possibilities. ## 8. The 30-Day Rule Each stage-level production experiment has a minimum observation period of **30 days before substantive conclusions are drawn**. This is deliberately a minimum observation period, not a claim that 30 days automatically establishes statistical significance. During the observation period, the system collects: - Strategy performance. - Control performance. - Regime observations. - Selection events. - Contention events. - Execution data. - Operational information. Technical or infrastructure problems can be identified and corrected during the test. However, a few days of profit and loss are not sufficient grounds to declare that a research stage has succeeded or failed. The objective is to accumulate sufficient live evidence before making substantive conclusions. The appropriate sample size and statistical analysis will depend on the observations collected; the 30-day minimum is not, by itself, a guarantee of adequate statistical power. ## 9. Production as a Research Feedback Loop The production branch does not terminate the research process. It feeds information back into it. ``` text Research | v Historical Validation | v Production | v Live Observations | v Execution Analysis | v Research Feedback | v New Hypotheses | v Further Research ``` Production observations may reveal: - Weaknesses in regime definitions. - Unexpected strategy interactions. - Problems with ranking. - Execution costs not captured by the model. - Differences between historical and live behaviour. These observations become new research inputs. They are not simply used to retrospectively optimise the existing backtest. Instead, they generate new hypotheses that can enter the research process. ## 10. What the Experiment Is Designed to Learn The purpose of WRN-P01 is not to prove that the Stage 1 strategies will be profitable. That would be the wrong standard. The experiment is designed to determine whether the research process produces better decisions. Possible outcomes include: - Stage 1 improves live strategy selection. - Stage 1 produces no measurable improvement. - Stage 1 improves strategy selection but introduces problems elsewhere in the decision process. - Execution costs reduce an otherwise valid edge. - Live observations reveal weaknesses in historical assumptions. Each outcome can provide useful information. The value of production testing is not simply whether profit and loss is positive. It is the information gained about which parts of the research-to-execution chain are working. **No outcome is claimed in this note. The experiment remains ongoing.** ## 11. Conclusion WRN-001 asked: > **Is a profitable backtest enough evidence of a genuine trading > edge?** The answer was no. WRN-P01 asks the next question: > **Does the research process that follows the backtest actually improve > live decisions?** That is what this experiment is designed to test. The strategies entering production have been tested across multiple historical market regimes as part of WRN-001. They have then passed the relevant Stage 1 validation requirements and entered a validated strategy universe. Production now uses that historical, regime-conditioned evidence to make live selections across **48 potentially overlapping regime states**. The system must resolve competing validated candidates, execute the selected strategy and measure the resulting trading and implementation outcomes. At the same time, appropriate control groups allow the experiment to assess whether the research and selection process itself adds measurable value. The experiment uses a minimum **30-day observation period** before substantive conclusions are drawn. That period is a minimum observation requirement, not proof of statistical significance. This creates a different philosophy of quantitative research: **Don't assume that better modelling produces better trading.** **Test whether it does.** Every major improvement to the research pipeline should eventually face the same question: > **Did this actually make the live decision better?** That is the purpose of the Wansiri production-validation branch. **WRN-P01 is the first experiment, and it remains ongoing.**