The task and original case
Write a dataset README that lets someone reproduce a result from the exact input revision. The fictional CSV has id,value and unit columns and two versions with different bytes. A shared filename alone does not establish that two people used the same data.
| Case element | Value or requirement |
|---|---|
| Inputs | A fictional CSV has columns id,value,unit and two revisions with different bytes. |
| Build | A README, input manifest and reproduction instructions. |
This is an original invented exercise. It is independent practice, with no employer assessment or open-vacancy claim. Download the inputs, attempt your own result, then open the separate answer key.
Data and worked inputs
Build a result another person can follow
Document the schema, value units, row identity and original invented provenance. Retain revision-one and revision-two files separately. Compute a SHA256 hash for each actual file and record its byte size and revision name in a manifest. Do not type a guessed hash or infer byte equality because two tables look similar on screen.
GitHub's README guidance describes the role of repository documentation. Add the runtime version, transformation command, output location and expected result for each supplied revision. A minimal reproduction path should identify the exact input manifest entry, perform the declared transformation and compare the actual result with the expected output. Keep commands free of private credentials and machine-specific secret paths.
Sources[1]
Check the worked result
The correct artifact records both revision identities and shows whether their hashes differ. The exercise specifies different bytes, so identical hashes would require investigating your selected files or hash procedure before accepting the receipt. Your own computed hashes, not a generic example string, belong in the manifest.
Record both byte identity and semantic changes. A newline-only edit can change a hash without changing parsed rows; a changed value can affect the calculation. Explain what the two revisions actually differ in after inspecting them.
The supplied revision A values normalize to a combined 4.5 meters; revision B normalizes to 4.6 meters. Reproduce both totals from the retained value and unit columns, and calculate hashes from your actual downloaded files rather than using these totals as a substitute for byte identity.
Make the failure visible
Try reproducing the report with the input path accidentally pointing to the other revision. A manifest check should detect the mismatch before a misleading comparison. Also test a missing unit or duplicate ID against the stated schema. A README that says only “run the script” cannot explain a changed output or an invalid input.
Revise it and ask someone to reproduce it
Give a peer a clean copy of the files and README. Ask them to reproduce the declared output and record any missing step or environment assumption. Revise the instructions, then repeat the process with the second revision. Keep a dated reproduction receipt describing what actually ran and whether its checks passed.
- Hashes are computed from the exact retained input bytes.
- Commands identify runtime, input revision and expected output.
- Byte changes and semantic changes are explained separately.
Describe the work accurately
After completing your own version, you could adapt this project bullet: “Documented an independent dataset exercise with schema, units, hashed revision identities and peer-tested reproduction instructions.”
Use Projects or Independent practice as the heading. Keep the input files, result and revision together. Describe only your own completed work; the invented numbers are model outputs, not employer results or measured business impact.