Relevant original and adjusted versions can help when the changes and the applicant’s contribution are explained. Preserve the dates and information actually available for each version.
Avoid reconstructing a perfect prediction after the event
Do not silently replace an earlier forecast with hindsight. Explain the internal interpretation behind a representative adjustment and identify what others contributed or approved. Where many forecast versions exist, select the ones that mark a genuine turning point in reasoning rather than every routine update, since a long unbroken series of near-identical files adds volume without adding explanation.
For each version chosen, identify the specific assumption that changed and why, distinguishing a deliberate judgment call from an automated recalculation the underlying model performed on its own. Note which figures came from a shared template or default parameter and which reflect the applicant's own analysis, since specialized knowledge rests on the latter, not on the tool that produced the output. Where a later version proved closer to actual results, resist treating that accuracy as proof of skill on its own, and instead describe the reasoning that was available to the applicant at the time each version was produced.
Keep the sequence in its original order so a reviewer can trace how the applicant's thinking actually developed rather than seeing only the final, most polished iteration, stripped of the steps that led to it.