brightlab.Made for curious mindsBrightlab
Year 7 · Systems
Can correct code still pass the wrong numbers to a model?
Go to the investigation ↓A unit-labelled pipeline trace and a regression-test contract.
Does the next stage understand what this number means?
A model’s inputs have a contract: units, order, scale and meaning. This synthetic environmental model expects Celsius but can receive Fahrenheit if a conversion is missing. Each step may run without an error while the complete pipeline gives a wrong result. Inspecting intermediate values and checking a known reference case reveals the mismatch.
Before this lesson: Identify a value, its units and its source. Useful earlier investigations: A claim needs somewhere to stand.
A model’s inputs have a contract: units, order, scale and meaning. This synthetic environmental model expects Celsius but can receive Fahrenheit if a conversion is missing. Each step may run without an error while the complete pipeline gives a wrong result. Inspecting intermediate values and checking a known reference case reveals the mismatch.
Identify a value, its units and its source. Useful earlier investigations: A claim needs somewhere to stand.
Review Fahrenheit = Celsius × 9/5 + 32. Prepare equivalent values 0°C/32°F, 20°C/68°F and 35°C/95°F. Explain that the model is an illustrative linear score, not a real environmental forecast.
ACARA V9 AI curriculum connection ↗ · Technologies ↗
A complete teaching sequence · 75 minutes
Pass a paper value through sensor, conversion, normalisation and model stages.
Ask: “Does the number 68 tell us enough without a unit?”
Listen for: “No, 68°C and 68°F mean different temperatures.”
Predict the model result when Fahrenheit bypasses conversion.
Ask: “Which stage first receives an unexpected meaning?”
Listen for: “The normaliser expecting Celsius.”
Change the input unit, follow every intermediate value and compare equivalent temperatures. Toggle conversion only while keeping the physical temperature fixed.
Ask: “Why did no syntax error appear?”
Listen for: “The values were still numbers, just wrong for the contract.”
Load the edge case that converts an already-Celsius value as if it were Fahrenheit.
Ask: “Can adding more conversion make things worse?”
Listen for: “Yes, the conversion must match the actual input unit.”
Specify explicit unit metadata and one conversion boundary. Write checks for three equivalent pairs and an invalid unit.
Ask: “How will a future edit reveal this bug again?”
Listen for: “The reference-pair test will fail.”
Submit a stage trace and regression-test statement.
Ask: “Should we retrain the model to hide a unit bug?”
Listen for: “First repair the contract; otherwise the meaning remains wrong.”
Correct steps always make a correct system.
A double conversion produces valid numbers but incorrect meaning, so simply adding a conversion step is insufficient.
Define a unit-aware boundary and three equivalent-temperature regression tests plus an invalid-unit rejection.
Ask for the first invalid intermediate value rather than only the final error. Check that physically equivalent inputs yield equal model inputs after repair.
| Criterion | Beginning | Secure | Extending |
|---|---|---|---|
| Pipeline diagnosis | Blames the final score | Finds the unit mismatch stage | Distinguishes missing from double conversion |
| Contract repair | Patches one numeric case | Defines units and tests equivalent inputs | Adds invalid-unit rejection and a reproducible regression suite |
Use only 0°C/32°F and a precomputed conversion table before introducing variable temperatures.
Add feature-order and normalisation-version contracts; create a test that catches a swapped-column bug.
Only fictional sensor readings are processed. This educational model must not control real equipment or environmental decisions.
Shared device? Turn remembering off. A project file lets you continue on another device.
Your browser is the laboratory
Send 68°F through a Celsius-trained pipeline without conversion. Predict the direction of the score error.
The physical temperature is 20°C, reported by a sensor as 68°F. The model expects Celsius, divides by 40, then computes 0.8x+0.1. Conversion is initially bypassed.
The experiment opens after your prediction.
There is no penalty for being surprised.
Apple Silicon · PyTorch MPS
Run thousands of Celsius/Fahrenheit equivalence cases through a linear model. Compare missing, correct and double conversion; verify correct-conversion parity within 0.00001 on the selected device.
Students extend the experiment in teams, documenting parameters, outputs and limitations.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download Mac Studio investigation ↓unzip y7-pipeline-mac-pathway.zip -d y7-pipeline-mac cd y7-pipeline-mac bash setup-mac.sh source .venv/bin/activate python experiment.py --device mps --output results marimo edit notebook.py
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
NVIDIA DGX Spark · PyTorch CUDA
Run thousands of Celsius/Fahrenheit equivalence cases through a linear model. Compare missing, correct and double conversion; verify correct-conversion parity within 0.00001 on the selected device.
Students extend the experiment in teams, documenting parameters, outputs and limitations.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download DGX Spark investigation ↓unzip y7-pipeline-dgx-pathway.zip -d y7-pipeline-dgx cd y7-pipeline-dgx bash run-dgx.sh # Open the localhost notebook URL printed by the container.
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
A unit-labelled pipeline trace and a regression-test contract.
Download the editable handout →