DynamicVibePython learning
How to study · reference & help

One lesson at a time

Read the idea, then follow the named exercise steps. On a phone, use Idea, Code and Output to switch views. Review days begin with recall before the rebuild. A self-review mark is not a scored answer.

Run, inspect, repair

Run loads Python in your browser on the first use, so you need an internet connection. Read Output and the check result. Stop interrupts a run; it keeps your code. Reset restores the selected starter, replacing that exercise’s draft. Use an authored hint before asking for help.

Keep your work

Code drafts, progress and Notes save on this device when browser storage is available. They do not sync through your login. Use Backup & records above to download saved learning records, preview an import or undo the latest import. Copy unsaved text first; clearing browser data can remove local work. A save warning means you should copy your work before leaving.

Practice files and reports

Each browser Run starts in a fresh temporary workspace. A sample file exists only when that run’s code creates it; these are not your computer’s files. On report tasks, use Preview to inspect the actual text and Download to request a copy before another Run, Reset, exercise change or leaving. Reports do not save through tutor login. Read files and write reports.

Get tutoring when you need it

Lessons and checkpoints do not need tutor login. Unlock the tutor through Training’s shared tutor access, then use its return link to come back here. Your question is only sent when you press Send. Do not paste passwords or private production data.

Python words in plain language

Assignment and type
frames = 96 binds a name to an integer. "96" is text, not the same numeric value. Values and names.
List and index
A list keeps items in order. Positions start at zero; the final nonnegative index is length minus one. List lookups.
Loop and indentation
A for loop processes each item. Indentation groups its repeated statements; a statement outside the block runs separately. Loops.
Parameter, argument and Return
A parameter names an input in a definition; an argument supplies its value in a call. Return gives a result to the caller; print displays it. A function that only prints returns None. Functions.
Dictionary and schema
A dictionary labels values with keys. A schema states the required fields and types; a missing fact is not automatically zero. Dictionaries · Validation.
File, CSV and JSONL
A file stores bytes/text. CSV needs a quoting-aware parser; JSONL contains one JSON value per line. Browser practice files are temporary, not computer files. CSV · JSONL.
Denominator and evidence
A rate divides a count by its declared eligible total. Keep exclusions visible. A quiz score, practical check and self-review are different evidence. Quality rates.

Read the error, then make one repair

SyntaxError / IndentationError
Python could not parse the code. Inspect the named line and the preceding line for a missing colon, unmatched quote/bracket or inconsistent indentation. Review blocks.
NameError
A name has no value here. Compare case and spelling, and check that assignment happens before use. Do not replace an unknown parameter with a fixed sample.
TypeError / ValueError
A wrong kind of value or an unacceptable value can fail an operation. Inspect the supplied input and declared policy; do not broadly suppress the error. Narrow error handling.
IndexError / KeyError
The requested position or key is missing. Inspect the collection and interface instead of guessing a fallback. Indices · Keys.
Run stays loading / runtime or checker unavailable
This is not proof that your answer is wrong. Stop if available, copy your draft, check your connection and retry. If it repeats, keep the visible error and use the readable lesson/hints. Never reset or clear browser storage as the first repair.
Expected sample prints, but a check fails
Read the named contract and test changed/empty inputs. The helper may return None, use a fixed value or stop too early even when the sample looks right. A failed check is not a request to copy the expected output.

Press Escape while working inside this panel to close it and return focus to its summary. Without JavaScript, use the summary again. For local Studio commands, use terminal troubleshooting.

← Route Week 8 · Day 40 of 40

Human-data audit pipeline

Loading today's steps…
Lesson idea Tap to fold
Recall

Name the audit trail before calculating a metric.

Show the answer

Raw rows, invalid rows, duplicates removed, clean rows, and the rule for every exclusion.

Today

Build a defensible audit, not a polished-looking number

The four guided tasks trace, rebuild, fill and repair a small pandas audit. Their inputs are valid JSON object lines with typed nonempty fields when present; they demonstrate first-global-ID retention, acceptance and reference agreement, not a complete ingestion boundary. Rebuild returns an empty DataFrame with item_id/label columns for empty input. The independent final project extends the policy to malformed JSON, invalid types, blank lines and explicit issue evidence.

Use a validation → retention → aggregation → report pipeline. Validate before reserving a cleaned global ID, keep first valid occurrences, count each exclusion exactly once and retain source line numbers. Acceptance counts pass labels; agreement counts label/gold equality, including fail/fail. Overall rates use retained rows, not the average of model percentages. None records an absent denominator instead of an invented zero.

The final brief specifies build_audit, save_audit and test_audit. Return reusable data, save the actual data as JSON and write tests that catch empty, malformed and duplicate mistakes. Allow 95 minutes in two sessions: the guided tasks take about 50 minutes and the independent project about 45, plus the separate quiz. Explanations and portfolio readiness are self-reviewed; software checks alone do not certify learning. Use the Studio course map for the next project path.

# Keep the retained-row denominator, not model means.
groups = [
    {"rows": 1, "matches": 1},
    {"rows": 3, "matches": 1},
]
rows = sum(group["rows"] for group in groups)
matches = sum(group["matches"] for group in groups)
rate = matches / rows if rows else None
print("Retained agreement:", rate)
print("No observations:", None)
assert rows == 4
assert matches == 2

Two matches among four retained rows means 0.5 agreement; averaging the group percentages would answer a different question.

Explain exclusions, policy and sample limits before interpreting a rate. Preserve the downloaded report and your code; local browser storage is not a cloud backup.

Official reference for this lesson

Predict, explain, then test

These three comparisons are self-reviewed, not scored. Coding checks run new inputs; the section quiz records separate knowledge evidence.

Account for evidence

Which counts must add up to physical input lines?

Compare your answer · self-reviewed

Blank + rejected + duplicates + kept. A line belongs to one category, with one-based issue evidence for exclusions other than blanks.

Validate first

An invalid row appears before a valid row with the same ID. Which may reserve the ID?

Compare your answer · self-reviewed

Only the first valid cleaned global ID reserves identity. Invalid earlier rows must not suppress valid evidence.

Read a rate

One model has 1/1 matches and another 1/3. What is overall agreement?

Compare your answer · self-reviewed

2/4 = 0.5 over retained rows, not the unweighted mean of the model percentages.

Independent final audit: evidence and a real report

Follow this independent policy rather than copying the narrower guided demonstration:

  1. Write build_audit(raw) for a JSONL string; reject non-string callers with TypeError. Use raw.splitlines(): a trailing line terminator does not add a phantom line. Start fresh on each call.
  2. Classify every line in order: whitespace-only is blank; malformed JSON is rejected with reason json; JSON that is not a dictionary is rejected with reason object. A dictionary is rejected with reason fields unless item_id and model are nonempty strings after stripping outer whitespace, and label/gold are exact string pass or fail values. Allow extra fields; do not coerce types.
  3. After validation, retain the first valid cleaned global item_id, regardless of model. Later valid occurrences are duplicates with reason duplicate. An invalid earlier occurrence does not reserve an ID. Keep line numbers one-based and issue entries as {line: number, reason: text}, in source order. Blank lines have no issue entry.
  4. Return exactly lines, blank, rejected, duplicates, kept, acceptance, agreement, issues and models. Count exclusions separately so lines = blank + rejected + duplicates + kept. models maps each cleaned model to rows, accepted, matches, acceptance, agreement and disagreements: an input-order list of retained item IDs where label differs from gold.
  5. Count pass as accepted and equality of label/gold as a match, including fail/fail. Overall acceptance/agreement use all retained rows. Per-model rates use that model’s retained rows. Round defined fractions to four decimals only after calculating. Return None for overall rates when kept is zero; models is then empty. Counts are native integers, not booleans.
  6. Write save_audit(raw), which calls build_audit, writes its actual returned report as UTF-8 JSON plus a final newline to project-report.txt, and returns that same data. Do not write a fixed sample or a Python repr. The browser collects the sample file before its checks rerun new cases; use the explicit download button to keep it.
  7. Write test_audit() containing at least three real assertions: empty input has zero kept rows, malformed JSON is counted as rejected, and repeated valid global IDs are counted as duplicates. Call it before saving the supplied sample. The checker also tests whether these assertions detect those rules when deliberately broken.
  8. Explain one exclusion, the global-ID retention policy and the denominator behind a rate. These explanations are self-reviewed, not automatically scored. The report describes this retained sample and reference; it does not prove general model quality or certification. Allow 50 minutes for the four guided tasks and 45 for this independent task, in two sessions; quiz time is separate.

Select the final project in the controls. The quiz practical link selects it without running code or awarding evidence. Review JSON, validation, writing and verifying report files and meaningful tests.

Read the unfinished starter without running Python
# Build and test a new audit policy independently.
import json

# Return counts, issue evidence and retained group rates.
def build_audit(raw):
    return {}

# Write the returned report, not a fixed sample.
def save_audit(raw):
    return build_audit(raw)

# Replace the placeholder with three meaningful checks.
def test_audit():
    assert True

# Keep the supplied six physical input lines.
raw = (
    '{"item_id":"a1","model":"Canal",'
    '"label":"pass","gold":"pass"}\n'
    '{"item_id":"a2","model":"Canal",'
    '"label":"fail","gold":"pass"}\n'
    '{"item_id":"a1","model":"Museum",'
    '"label":"fail","gold":"fail"}\n'
    'broken\n{}\n '
)
test_audit()
report = save_audit(raw)
print("Kept:", report["kept"])
print("Rejected:", report["rejected"])
print("Duplicates:", report["duplicates"])
print("Report: project-report.txt")

The sample retains two Canal rows, with acceptance/agreement 0.5 and disagreement a2. It excludes one duplicate on line 3, malformed JSON on line 4 and invalid fields on line 5, plus a blank sixth line. Expected output is Kept: 2, Rejected: 2, Duplicates: 1 and Report: project-report.txt on separate lines. Download the actual UTF-8 report explicitly; a passing quiz does not pass this project.

Next: Four Studio readiness tasks and targeted refreshers · Python Studio course map · All training and shared tutor login. Do not treat a software pass or a self-review tick as certification.

Now

Choose an exercise to load its prompt.

day-40.py Python · task packages shown · runs here
Output

Loading exercises…


    
    
    

      
      

        
        
      

    

  













Section 8 checkpoint
Answer five questions, inspect explanations and use the targeted refreshers. The final audit and its real report are a separate practical task. Neither requires tutor login.




  

    
AI Tutor

    
The tutor mounts here when JavaScript is available. The lesson above stays readable without it.