performance - How can avoid people using my code for evil? -

I'm not sure that this is the right place, but it seems like a decent place to ask.

My current job includes manual analysis of large data sets (at many levels, each is done by more sophisticated and faster experienced analysts). Almost a year ago, I started developing some utilities to track the analyst's performance by comparing results from the first to the last level. First of all, it worked quite well - we used the in-shop as a simple indicator to help make efforts to focus on shopping and work better overall.

In recent times, however, the results have been taken out of context, in a way I never intended that management (especially one person) has the EPR (Listed Performance Report - \ It's an Air Force But I think that exists in other similar areas) and have started using the results of these tools to directly affect the same paperwork. The problem is not that who uses these results, but how. I have made it clear to everyone that the results are simply error-prone.

There are many inevitable obstacles to generate this data, which I have worked to minimize with some nifty heuristics and thus took it in proper context, though they are a useful tool From the context, now they are being used, they do more harm than good.

The manager (s) in question is taking the result whether an analyst is performing well or is poor. The results are getting moderate and individual scores are being ranked above (good) or below (poor) average. This error and there is no relation to the inherent margin of sample bias, for any kind of proper interpretation There is no connection. I am aware of at least one person whose performance rating was marked below for a 'Accuracy Percentage', which was below one percentage point below the average (when the normal difference of error from the calculation method is approximately two to three Percent).

I am in the process of writing a formal report in the system errors ("Beginners Guide to Financial Strategic Analysis"), but not all signals have any effect.

I want to reduce intentional tool tools (a path that I would like to save but I firmly think in the circumstances), I am thinking that anyone here is effective before similar circumstances Dealt with by the way?

Update: Thanks for the responses - Around many good ideas.

If someone is curious, then I am moving forward the direction of 'control of interpretation, education, and interpretation' I have started rebuilding my tools so that I try and refuse the error and track And automatically create any number and graph, which includes the documentation (whereas raw data while hiding ambiguous references they are currently 'magic' Excel Sheets)

Specifically, I hope that the status of the situation in the visual representation of the error and the ranking system (Account Error, Standard Deviation, etc) will be helpful.

Modify the output to include either error information (so if there is an error +/- 5%, Do not output 22%, output 17% - 27%), or to educate those against whom this error is being used against so that they can use themselves in defense, when used against them.