In my checklist, correctness stays first. So pushing it will also narrow down the scope of the cover range ~> reliability or not, depending on how we handle the 'uncover' case. Based on this, we need to 'talk'. And find out what's the most important criteria we need right now. For me, I stay in the role of service provider and just want my client to be happy by getting what they really need.

When clients push for algorithmic innovation, developers should first clarify the requirements and identify the quantifiable metrics the client values most. These requirements should drive the development process, ensuring the solution aligns with the client’s objectives. Developers must also measure key metrics—such as performance and reliability—to provide a full picture of the solution’s capabilities. In my view, one should focus on getting their requirements from clients rather than getting proposal algorithmic innovations. Often, existing algorithms, with minor adjustments, can meet these requirements effectively and with high reliability. Of course any suggestions should be welcomed, but has to come with requirements.

When clients advocate for cutting-edge algorithm innovations, it is essential to strike a balance between achieving technological progress and maintaining system reliability. While innovation can provide competitive advantages, it must not come at the cost of system stability or user experience. Careful planning, rigorous testing, and iterative implementation can ensure that advancements align with the system's overall robustness and reliability.

It is essential to conduct comparative tests for false negatives and false positives, highlighting to the company the advantages and disadvantages of each approach. It is important to explain that the entire process is empirical and that continuous adjustments will be necessary during production. Ongoing monitoring until a satisfactory level of confidence is achieved is crucial for the success of a cutting-edge algorithm. Additionally, it should be noted that new algorithms may fail in cases where previous versions succeeded, with the main advantage of new algorithms being the theoretical reduction in the overall error rate.

Balancing innovation with reliability is always a fine line, especially when dealing with a client eager to push the boundaries. 1. Always Have a Rollback Plan Things can go south, even with the best preparation. That’s why I always have a rollback strategy in place. 2. Balance Short-Term Gains with Long-Term Scalability I always advocate for scalable solutions that won’t bottleneck future progress. Building with the next 2-3 years in mind often pays off in reliability and performance. 3. Experiment in Isolation Always test new algorithms in isolated, sandboxed environments. Shadow testing and A/B testing help us catch potential issues without impacting production systems