← All learning journeys

Run and reproduce a lab.

Keep a result another researcher can rerun and check.

0 of 4 checkpoints complete. Continue at step 1 ↓

  1. STEP 1 / 4

    Freeze the environment

    Select a lab. Download and extract the ZIP. Follow the README to create its environment.

    Foundations. An environment is a separate set of programs installed only for this lab.

    Read the other explanation levels

    Applied. Use the verified Linux dependency lock when available; record any deviation instead of silently mixing versions.

    Research. Pin transitive dependencies and retain executable inputs. Numerical libraries, parallel reductions and compiler choices can influence reproducibility.

    Explore this research question ↗

    Checkpoint

    Is recording only the solver name enough?

    Sign in to write your answer and keep it.

    Show explanation

    Keep its version, dependency lock, script hash, inputs and operating system. The package provides version pins and a run receipt.

  2. STEP 2 / 4

    Predict the reference answer

    Read the lab equation and assumptions. Calculate the expected answer before you run anything.

    Foundations. A reference answer lets you see when a result is wrong.

    Read the other explanation levels

    Applied. The absorber predicts exp(−ΣL). Constant diffusion with fixed endpoint concentrations produces a linear steady profile when there is no source.

    Research. Use a curved source-driven diffusion solution to measure mesh error. A linear finite-element method can already represent a linear solution exactly, so that alone cannot demonstrate convergence order.

    Explore this research question ↗

    Checkpoint

    What does agreement with an analytical solution establish?

    Sign in to write your answer and keep it.

    Show explanation

    It verifies this restricted numerical case. It does not validate a real fusion chamber model.

  3. STEP 3 / 4

    Run a controlled refinement

    Run the lab command. Compare the errors for both particle counts, or for all three meshes.

    Foundations. Change one setting at a time. Then you know what caused the difference.

    Read the other explanation levels

    Applied. Monte Carlo standard error typically scales as N⁻½. For a smooth problem, piecewise-linear concentration interpolation can show second-order error reduction under mesh refinement.

    Research. Separate statistical fluctuations, finite-mesh error, solver tolerances and model-form bias. Check conservation and convergence rather than judging a plot by appearance.

    Explore this research question ↗

    Checkpoint

    Does more computation remove every kind of uncertainty?

    Sign in to write your answer and keep it.

    Show explanation

    More histories reduce sampling noise; mesh refinement reduces discretization error. Neither fixes incorrect physics or material data.

  4. STEP 4 / 4

    Keep the receipt and limits

    Keep results/receipt.json, the lock file and your input changes together. Write down one physical process the lab leaves out.

    Foundations. Keep enough information that another person can repeat what you did.

    Read the other explanation levels

    Applied. The receipt includes solver version, script hash, reference comparison and measured runtime. Installation time is separate.

    Research. Treat the successful benchmark as a verified baseline. Add effects incrementally with independent reference checks and an uncertainty budget.

    Explore this research question ↗

    Checkpoint

    Can the plant-cycle button replace this solver run?

    Sign in to write your answer and keep it.

    Show explanation

    No. The cycle explains the sequence of physical processes. These downloaded labs run real, deliberately restricted calculations on your computer.

Explore the two code labs →

Sources and further reading

Teaching content reviewed against public primary sources on 9 September 2026. The simplified examples have their own assumptions and verification checks.