Milestone 6 of 8
Inspect the result numerically and visually
Inspect observed points, generation evidence, the selected line, and signed differences without hiding them behind one score.
The selected score is one view of the result. Inspect the values and plots together so that a plausible picture does not hide a numerical mistake.
Goal
Inspect the selected candidate numerically and visually. Plot the observed points, the source line used to generate them, the selected candidate line, and a separate view of signed differences.
Inputs
Use generated_data.csv, selected_result.json, and the full search evidence.
The source line is generation evidence and must be labelled separately from the
selected candidate. Signed differences use observed - predicted and retain
the input identity.
Keep units, axis labels, limits, legends, and the source and selected line identities clear. Do not connect observations as if they were a continuous measurement curve unless the figure explicitly says what the line means.
Deliverables
Implement the numerical and plotting part of src/report.py and save
output/fit_and_differences.png. The figure should contain:
- observed points with their input values;
- the configured source line labelled as generation evidence;
- the selected candidate line labelled as the grid result; and
- a separate signed-difference view with a zero reference.
Record selected score, minimum grid score, one reconstructed point, and a
short observation about the visible differences in report.md or the later
report section. Keep the plot data traceable to saved arrays.
Checks
Check that the figure uses the saved input, observed, selected prediction, and signed-difference arrays rather than a newly invented calculation. Read those arrays back and compare their lengths and identities. Check that the selected score and minimum grid score agree, that the zero line is present in the difference view, and that labels distinguish source from selected candidate.
Use a deliberate mutation of one plot input to confirm that the audit can detect a mismatch. A successful file existence check is not enough to prove that the figure shows the selected result.
Workspace
Keep report and plotting code in src/report.py; use src/main.py only to
orchestrate saved artifacts. Keep the figure under output/ and leave the
source data, full grid, and selected result available for read-back.
Hints
HintA plot is another artifact
HintSeparate the two lines
Review
Look at the numerical values and figure together. Can you identify the source line, selected line, each difference's sign, and the input behind one plotted point? Remove any claim that the plot alone establishes.
How to check your work
Checks compare figure data, labels, and the difference convention with the supplied fixture. The supplied fixture demonstrates one clear layout; it does not replace numerical inspection.