Implementing level by level fallback

Question

I have a class ScoreStrategy that describes how to calculate points for a quiz:

public class ScoreStrategy
{
    public int Id { get; set; }

    public int QuizId { get; set; }

    [Required]
    public Quiz Quiz { get; set; }

    public decimal Correct { get; set; }

    public decimal Incorrect { get; set; }

    public decimal Unattempted { get; set; }
}

Three properties Correct, Incorrect and Unattempted describe how many points to be assigned for a response. These points can also be negative. The score strategy applies to all questions in the quiz, thus there can only be one ScoreStrategy per quiz. I have two subclasses:

public class DifficultyScoreStrategy : ScoreStrategy
{  
    public QuestionDifficulty Difficulty { get; set; }
}

public class QuestionScoreStrategy : ScoreStrategy
{ 
     [Required]
     public Question Question { get; set; }
}

My questions have three difficulty levels(Easy, Medium, Hard; QuestionDifficulty is an enum). The DifficultyScoreStrategy specifies if points for questions of a specific difficulty need to be assigned differently. This overrides the base ScoreStrategy that applies to the entire quiz. There can be one instance per difficulty level.

Thirdly, I have a QuestionScoreStrategy class that specifies if points for a specific question have to be awarded differently. This overrides both the quiz-wide ScoreStrategy and the difficulty-wide DifficultyStrategy. There can be one instance per question.

While evaluating the responses of the quiz, I want to implement a level-by-level fallback mechanism:

For each question:

• Check if there is a QuestionScoreStrategy for the question and return the strategy if one is found.
• If not, fallback to DifficultyScoreStrategy and check if there is a strategy for the difficulty level of the question being evaluated and return it if a strategy is found.
• If not, fallback to the quiz-wide ScoreStrategy and check if one exists and return it if it does,
• If there is no ScoreStrategy either, use default as { Correct = 1, Incorrect = 0, Unattempted = 0 }(It would be great if I can make this configurable as well, something much like the .NET's elegant way:

options => {
    options.UseFallbackStrategy(
        correct: 1, 
        incorrect: 0, 
        unattempted: 0
    );
} 

).

Summary

^{I've summarized the above info in a table:}

Strategy Type	Priority	Maximum instances per quiz
QuestionScoreStrategy	1st (highest)	As many as there are questions in the quiz
DifficultyScoreStrategy	2nd	4, one for each difficulty level
ScoreStrategy	3rd	Only one
Fallback strategy (Default { Correct = 1, Incorrect = 0, Unattempted = 0})	4th (lowest)	Configured for the entire app. Shared by all quizzes

I have a container class called EvaluationStrategy that holds these score strategies among other evaluation info:

partial class EvaluationStrategy
{
    public int Id { get; set; }

    public int QuizId { get; set; }

    public decimal MaxScore { get; set; }

    public decimal PassingScore { get; get; }

    public IEnumerable<ScoreStrategy> ScoreStrategies { get; set; }
}

What I have tried:

I have added a method called GetStrategyByQuestion() to the same EvaluationStrategy class above(note it is declared as partial) that implements this fallback behavior and also a companion indexer that in turn calls this method. I have declared two HashSets of types DifficultyScoreStrategy and QuestionScoreStrategy and an Initialize() method instantiates them. All the score strategies are then switched by type and added to the appropriate HashSet, there can only be one ScoreStrategy per quiz, which will be stored in defaultStrategy:

partial class EvaluationStrategy
{
    private ScoreStrategy FallbackStrategy = new() { Correct = 1, Incorrect = 0, Unattempted = 0 }; 
    private ScoreStrategy defaultStrategy;
    HashSet<DifficultyScoreStrategy> dStrategies;
    HashSet<QuestionScoreStrategy> qStrategies;


    public void Initialize()
    {
        qStrategies = new();
        dStrategies = new();
        // Group strategies by type
        foreach (var strategy in strategies)
        {
            switch (strategy)
            {
                case QuestionScoreStrategy qs: qStrategies.Add(qs); break;
                case DifficultyScoreStrategy ds: dStrategies.Add(ds); break;
                case ScoreStrategy s: defaultStrategy = s; break;
            }
        }
    }

    public ScoreStrategy this[Question question] => GetStrategyByQuestion(question);
    

    public ScoreStrategy GetStrategyByQuestion(Question question)
    {
        if (qStrategies is null || dStrategies is null)
        {
            Initialize();
        }
        // Check if question strategy exists
        if (qStrategies.FirstOrDefault(str => str.Question.Id == question.Id) is not null and var qs)
        {
            return qs;
        }
        // Check if difficulty strategy exists
        if (dStrategies.FirstOrDefault(str => str.Question.Difficulty == question.Difficulty) is not null and var ds)
        {
            return ds;
        }
        // Check if default strategy exists
        if (defaultStrategy is not null)
        {
            return defaultStrategy;
        }
        // Fallback
        return FallbackStrategy;
    }
}

---

This method seems a bit clumsy and doesn't quite feel right to me. Using a partial class and adding to EvalutationStrategy doesn't seem right either. How do I implement this level-by-level fallback behavior? Is there a design pattern/principle I can use here? I know many things in the .NET framework fallback to default conventions if not configured. I need something similar. Or can someone simply recommend a cleaner and elegant solution with better maintainability?

---

NOTE/ADDITIONAL INFO: The ScoreStrategys and EvaluationStrategy for all quizzes are stored in a database managed by EF Core(.NET 5) with TPH mapping:

modelBuilder.Entity<ScoreStrategy>()
                .ToTable("ScoreStrategy")
                .HasDiscriminator<int>("StrategyType")
                .HasValue<ScoreStrategy>(0)
                .HasValue<DifficultyScoreStrategy>(1)
                .HasValue<QuestionScoreStrategy>(2)
                ;
modelBuilder.Entity<EvaluationStrategy>().ToTable("EvaluationStrategy");

I have a single base DbSet<ScoreStrategy> ScoreStrategies and another DbSet<EvaluationStrategy> EvaluationStrategies. Since EvaluationStrategy is an EF Core class, I'm a bit skeptical about adding logic(GetStrategyByQuestion()) to it as well.

Answer 1

With Polly

There is a 3rd party library called Polly which defines a policy called Fallback.

With this approach you can easily define a fallback chain like this:

public ScoreStrategy GetStrategyByQuestionWithPolly(Question question)
{
    Func<ScoreStrategy, bool> notFound = strategy => strategy is null;

    var lastFallback = Policy<ScoreStrategy>
        .HandleResult(notFound)
        .Fallback(FallbackStrategy);

    var defaultFallback = Policy<ScoreStrategy>
        .HandleResult(notFound)
        .Fallback(defaultStrategy);

    var difficultyFallback = Policy<ScoreStrategy>
        .HandleResult(notFound)
        .Fallback(() => GetApplicableDifficultyScoreStrategy(question));

    var fallbackChain = Policy.Wrap(lastFallback, defaultFallback, difficultyFallback);
    fallbackChain.Execute(() => GetApplicableQuestionScoreStrategy(question));
}

I've extracted the strategy selection logic for QuestionScoreStrategy and DifficultyScoreStrategy like this:

private ScoreStrategy GetApplicableQuestionScoreStrategy(Question question)
    => qStrategies.FirstOrDefault(str => str.Question.Id == question.Id);

private ScoreStrategy GetApplicableDifficultyScoreStrategy(Question question)
    => dStrategies.FirstOrDefault(str => str.Difficulty == question.Difficulty);

Pros

• There is a single return statement
• The policy declarations are separated from chaining
• Each and every fallback can be triggered by different conditions
• Primary selection logic is separated from the fallbacks

Cons

• The code is really repetitive
• You can't create a fallback chain by utilizing a fluent API
• You need to use a 3rd party library

Without Polly

If you don't want to use a 3rd party library just to define and use a fallback chain you do something like this:

public ScoreStrategy GetStrategyBasedOnQuestion(Question question)
{
    var fallbackChain = new List<Func<ScoreStrategy>>
    {
        () => GetApplicableQuestionScoreStrategy(question),
        () => GetApplicableDifficultyScoreStrategy(question),
        () => defaultStrategy,
        () => FallbackStrategy
    };

    ScoreStrategy selectedStrategy = null;
    foreach (var strategySelector in fallbackChain)
    {
        selectedStrategy = strategySelector();
        if (selectedStrategy is not null)
            break;
    }

    return selectedStrategy;
}

Pros

• There is a single return statement
• The fallback chain declaration and evaluation are separated
• It is simple and concise

Cons

• It is less flexible: each fallback selection is triggered by the same condition
• Primary selection is not separated from fallbacks

Answer 2

You can sort the sequence of ScoringMethods by your priority.

First you sort by whether str is QuestionScoreStrategy and str.Question.Id == question.Id.

Then you sort by whether str is DifficultyScoreStrategy and str.Question.Difficulty == question.Difficulty.

(Note that since false comes before true, you'll have to invert the conditions)

Then you can just do FirstOrDefault() ?? defaultStrategy.

Example:

var defaultStrategy = new() { Correct = 1, Incorrect = 0, Unattempted = 0 };

var selectedStrategy = Strategies.OrderBy(str => 
    !(str is QuestionScoreStrategy questionStrat && questionStrat.Question.Id == question.Id)
).ThenBy(str =>
    !(str is DifficultyScoreStrategy difficultySrat && difficultySrat.Difficulty == question.Difficulty)
).FirstOrDefault() ?? defaultStrategy;

You can easily add more "levels" to this by adding more ThenBy clauses.