The Culminating Project, Society and Careers

A culminating performance task is a program you design and build yourself, from an idea to a working demonstration. The single most common reason one goes badly is not difficulty — it is scope. Pick a problem small enough that the core of it could be working within the first quarter of the time you have, then write the features down in priority groups and build strictly in that order. When time runs short, and it will, you drop from the bottom of the list rather than abandoning a half-built feature in the middle.

Break the project into milestones that are each provable by running something. "Work on the data structure" is not a milestone, because it can be 90 percent done forever. "The program loads a file of questions and prints them" is a milestone: it either runs or it does not. Then plan backwards from the deadline and leave the final fifth of your time for testing, fixing and documentation, because that work always exists and is always underestimated.

Losing work is a self-inflicted wound and it is entirely preventable. Keep the project in one folder, with each source file named exactly after the class it contains. Keep a dated copy every time something significant starts working, and never edit that copy — when a change breaks everything and you cannot see why, being able to return to the last version that ran is the difference between an hour lost and a weekend lost. Back up to a second location, because one copy is no copies. Version control such as Git does all of this properly and records why each change was made. And never rename a file to finalFINAL2: dates sort, adjectives do not.

Computing is not weightless. Every device consumes electricity while it runs, took energy and mined materials to manufacture, and becomes waste at the end of its life. Manufacturing dominates: for a typical laptop, making it accounts for far more energy than the years of electricity it will ever draw, which reorders the whole list of what actually helps — keeping a device in service longer beats almost any efficiency measure. Data centres deserve a separate mention, because they are where cloud services, streaming and machine learning actually run, and their two big costs are the computing itself and the cooling needed to remove the heat it produces. Efficiency there is measured by comparing total facility power with the power delivered to the computers alone; the closer that ratio is to 1, the less is wasted on everything other than computation.

These fields are worth knowing about now because they change what a programmer's job involves. Machine learning shifts effort from writing rules to gathering and checking data, which brings new responsibilities around bias and privacy. Quantum computing threatens some current encryption, which is why new standards are already being deployed. Cybersecurity has moved from a specialism to something every developer is expected to understand, and human-computer interaction keeps the field honest about the fact that software exists for people to use.