Level 6 · In practice

6.5Field feedback

Goal: benefit from practitioners' aggregated experience, no folklore.

Estimated time: 9 min · If you only have 5 min: read “The essentials in 30s” + the diagram.

The essentials in 30 seconds

Isolate: venv from the second project avoids global debt paid in conflicts.
Prove: bug → red test → fix → green, and hints at boundaries: regressions warn.
Read: traceback bottom-up and baseline before GPU — engineering before prestige.

Nabil thread — his 12-month plan

After one year, Nabil sets his plan: isolate each project in a venv, write the reproducing test before fixing, and prefer a simple baseline before any heavy model. No folklore: consolidated principles, verified logs, and code reread each quarter.

Start from the real need of the moment: one principle to consolidate, not a prestigious catalogue.

What practitioners recommend

Anonymous synthesis of widely shared best practices from official documentation, community feedback and operations guides. Five lessons to apply as-is.

Misconceptions to dismiss

“Global is enough for a small script”, “fix fast without a test” and “deep learning first”: these three received ideas cost dearly — isolation, proof by test and a simple baseline always pay.

✓ What practitioners repeat (anonymous synthesis)

venv from the second project: isolating avoids “it worked on my machine”; global is debt.
The reproducing test first: bug → red test → fix → green, bugs fixed this way warn.
Baseline before GPU: a simple model and a good split settle most business needs.
Read the traceback bottom-up: the last line names, the top locates, search does the rest.
Docs that verify: hints at boundaries and mypy in CI, reviews talk business instead of guessing.

Common misinterpretation: “the field = anecdotes to collect”. No: it is a practised transferable method (venv, red test, baseline, traceback, hints), not folklore. Lasting skill is knowing how to isolate, prove and deliver.

Text version of the diagram (SEO / screen readers / translation): Field course: 1) Isolate (venv from the second project, pinned requirements). 2) Prove (red test before fix, hints + mypy in CI). 3) Stay lean (baseline before GPU, traceback read bottom-up). Lesson: what is neither tested, nor typed at boundaries, nor pinned-versioned, will break on a Friday evening.

Ready-to-say script — choosing your next field habit (30 s)

“My venv: [activated / to create]. My test: [red written / to write]. My baseline: [simple / to frame]. My traceback: [read bottom-up]. My hints: [at boundaries].”

Copy-paste template — reproduction test (pytest)

def total(montants: list[float]) -> float:
    return sum(montants)


def test_total_avant_fix() -> None:
    assert total([10.0, 20.5]) == 30.5
    assert total([]) == 0.0

VSE / small budget

Favour practice and peers: replayed venv, red test, feedback asked from the contact.

SME Algiers

Seek locally available resources and check practical terms before committing.

Remote / hybrid

Prioritise written traces at a distance (shared cards, reviewed code) and reviews after each month.

Intercultural

Prioritise local transposition as soon as rules differ: question peers, compare practices.

48h challenge — apply tomorrow

Apply 1 field lesson (venv, red test or bottom-up traceback) to your current bug and note the time saved.

Section exit checklist

I know which field practices to apply (venv, red test, baseline).
My bug → test → fix ritual is started.
I reread the 5 lessons and know where to return when needed.

Self-assessment 1-5: clarity of my ritual?regularity of my practice?

← Previous : Toolbox (V3)Next : Going further (V3) →

Check your understanding

Nabil fixes a total bug returning every month. What does the field say?

💡

Golden field rule: what is neither tested, nor typed at boundaries, nor pinned-versioned, will break on a Friday evening.