The gradebook argument nobody wins
Everyone wants the numbers to mean something. They disagree about what.
The gradebook is where a learning platform stops being a document store and starts making claims. Once a number appears next to a person's name, somebody will use it to make a decision, and the moment that happens the exact rules by which the number was produced become important in a way they were not the day before. Most gradebook disputes are not about arithmetic. They are about whether the arithmetic represents what people thought it represented.
The classic version involves weighting. A trainer sets up a course where the final assessment is worth sixty per cent, then adds a formative quiz for practice, and the platform dutifully includes the quiz in the total because nothing told it not to. Nobody notices until a learner sits on a boundary. At that point three people look at the same screen and reach three different conclusions about what the course rules were, and none of them is being unreasonable, because the rules were never written anywhere except in the configuration.
In competency-based training the problem is sharper still, because the underlying judgement is not a number at all. A learner is either competent against the requirement or they are not yet. Forcing that into a percentage creates a figure that looks precise and means nothing, and once it exists somebody will average it, rank it, or put it in a dashboard. Where the platform can be made to display a plain outcome instead, that is worth the configuration effort it costs.
The durable fix is documentation outside the software. A short note attached to the course that says, in ordinary words, what contributes to the result and what does not, resolves in one paragraph the arguments that otherwise consume a week. It also survives the platform, which the configuration will not.
One further consideration applies wherever a result carries consequences outside the classroom. A gradebook is a calculation performed by software according to a configuration that can be changed by anybody with sufficient rights, and it retains no history of what the configuration was on the day a result was produced. If the weighting is altered in March, every result generated in February now displays differently, and nothing anywhere records that this happened. Where results matter, the defence is an exported snapshot taken at the point of finalisation and stored outside the system, which is a five-minute habit that answers a question no amount of later investigation can.