900 topic ideas – page 26 of 36.

Compare direct line of sight, one building corner, and several intervening walls at similar distances. Develop an additive attenuation model.

Measure both frequency bands at the same positions and behind the same materials. Compare range, attenuation, and data rate.

At many positions, record RSSI, throughput, latency, and packet loss. Search for thresholds or piecewise relationships.

Generate 2×2×2 scrambles of different lengths and determine the optimal solution using exhaustive search. Investigate the saturation effect.

Analyse all or a large sample of 2×2×2 states. Determine the distribution, mean, and variance of minimum solution length.

Measure one person’s solving time over many standardized practice sessions. Compare exponential, logarithmic, and power learning curves.

Compare beginner, CFOP, and computer-assisted methods in terms of move count, time, and variation. Create a Pareto analysis of speed and efficiency.

Vary heuristic strength in A\* or IDA\* search. Measure explored states, memory use, runtime, and solution quality.

Solve the same states using both move definitions. Investigate the distribution and ratio of the resulting minimum solution distances.

Present the same task with different levels of precise detail. Measure correctness, completeness, and number of unmet requirements.

Give the model zero, one, three, or five solved examples. Investigate how the number of examples changes accuracy.

Translate a fixed set of tasks as equivalently as possible. Compare correctness, response length, and consistency between languages.

Gradually add irrelevant or distracting information to tasks. Measure at what amount performance begins to decline.

Introduce controlled letter swaps, missing spaces, or punctuation errors. Model accuracy as a function of noise level.

Ask the same question in many independent runs. Determine agreement rate, response entropy, and variation in numerical results.

Request free text in one condition and a fixed table or JSON-like schema in another. Compare formatting and content errors separately.

Compare responses without a source requirement with responses requiring verifiable evidence. Independently check every cited source.

Systematically change the order of answer options for identical questions. Test whether particular positions are selected disproportionately.

Slightly change numerical values in a mathematics problem without changing its structure. Investigate sudden errors and local answer stability.

Create graded versions of the same images with different brightness and contrast. Determine accuracy and confidence.

Rotate images through fixed angles or cover a controlled proportion of each image. Model recognition rate as a function of angle or occlusion.

Train the same small classification model using increasing amounts of data. Create learning curves for training and test error.

Change the class proportions in the training dataset. Compare accuracy, precision, recall, F1 score, and confusion matrices.

Compare models or compression levels on the same task. Create a Pareto frontier using accuracy, runtime, memory use, and energy consumption.

Use your own water, CO₂, battery, or sound data. Compare linear or polynomial regression with a machine-learning model using unseen test data.