Evaluation:

The first section of the evaluation is to evaluate your data, was your data clearly supporting the hypothesis? (for example: were all the points on a linear fit? were all the error bars crossing the linear fit? does the linear fit pass through just 2 points?) Obviously each example mentioned indicates a different level of 'support' for your hypothesis. Are the error bars big or small? Large error bars indicate a spread of results, so are these reliable, should they be done again, or more trials taken to eliminate outliers? Here you are trying to evaluate the quality of your data.

You need to address limitations in your research, being clear on which factors were not involved in your model (eg temperature change, friction, air resistance) and the effect these might have had on the data. Would more information have helped here? Was there any limitation in terms of understanding the mathematics found in the research, or using simplifications?

You need to address limitations in your method in order of decreasing impact, so first pick the largest uncertainty you have identified in your work. Then you need to explain how this could be reduced (This should be about better equipment/resources not available) and the expected impact of reducing the uncertainty. Would altering this uncertainty change the gradient of your results, or the error bars leading to the uncertainty in the gradient? How would you do this - explain the how (e.g., how would using a digital thermometer actually improve the data? If you simply say it will then that is a statement and a level 1-2 answer. Using a digital timer is a statement, explaining how to get a time from video analysis software and how to change the frame rate to improve precisions is a level 5-6 answer.) 

You should aim to go through this process a minimum of 2-3 times.

More often than not the reason why your results do not match your hypothesis is because the theory used is making some assumptions that are not actually valid, sometimes called experimental limitations (such as that there is no air resistance, no energy loss, no change in temperature, there is an ideal gas, the material is perfectly elastic.) How would taking account of these asssumptions/limitations help explain your results? How could you widen the experimental focus/research question to take account of these?

As a further improvement, could you extend the independent variable further? I always like to start imagining the limits of the variables. For instance, an investigation into static friction will return a horizontal line, but that cannot hold to zero mass as at zero mass there is nothing solid to exert a frictional force so an extension would be to investigate the smallest surfaces possible to try to understand where that change in the conclusion happens. With some thinking this can be done for most experiments. Then there are extensions due to scaling up small experiments (for example, a toy car being made to make predictions on motorway travel).

This section is the hardest to write but also the one examiners look at to separate out the best work.