Copy-on-Write and updating the owning DataFrame

Copy-on-Write is the default and only mode in pandas 3. A derived pandas object behaves independently when you modify it; writing to a temporary column selection does not update its parent frame. This makes ownership more predictable, but breaks old chained-assignment recipes. If you intend to update the parent, perform one assignment on that parent using loc. If you intend to return a changed result, make that ownership explicit.

The worked function constructs its own frame, creates a separate derived column for demonstration and changes only the owning frame. The input records are never mutated, and the demonstration column must retain its original values.

Select the score Series and keep it as preview. Build a missing-score mask and assign zero into the owning frame with frame.loc[mask, "score"]. The preview still represents the pre-update values under Copy-on-Write. An explicit assertion makes that behavior observable. Return the changed frame with the same nullable schema and row order.

Trace a missing score followed by a score of five. The result is zero and five, while preview still begins with a missing value. Reading a column is not an instruction to link later writes to its parent. State which object each assignment is supposed to change.

The controlled mistake replaces the owning loc assignment with fillna on a selected temporary Series and discards that new result. It runs without performing the promised parent update. Chained assignment through frame["score"][mask] is another broken pattern and can produce ChainedAssignmentError warnings. Do not suppress those warnings to make a migration appear complete.

Copy-on-Write is not a deep-copy guarantee for arbitrary Python objects stored inside object columns. The lesson uses numeric extension values; nested mutable objects require a separate ownership contract. Likewise, returning a NumPy view can introduce read-only or sharing rules that need their own documentation.

Change the replacement for missing scores from zero to negative one. Keep a single loc assignment to the owning frame, preserve the unchanged preview and leave observed values alone. This is a deliberately small policy edit after a larger ownership explanation.

Predict the result for missing, zero and five before running the practice cases. Only the missing row should change. Do not replace an observed zero, modify the preview to satisfy its assertion or add a global option that restores an older mode. The task is to express the same ownership guarantee with a different declared replacement value.

Step 1: Replace only missing scores with -1 using an owning loc assignment. Step 2: Retain the working imports, input validation and table schemas. Change the stated policy, run the supplied practice cases, then explain the newly accepted or rejected boundary.

The solution changes one assigned scalar. The lab compares the actual returned frame, verifies input records are unchanged and turns unexpected warnings into failures. The preview assertion independently checks that deriving a Series and changing the owning frame does not make the preview silently change too. This avoids proving ownership solely from a final displayed table.

The tests include all-observed input, all-missing input and an empty frame. No-op updates should preserve the schema and order. Explain why a correct result and correct ownership are separate acceptance properties: a function can return the desired numbers while unexpectedly modifying another object used elsewhere in a pipeline.

Explain the practical migration from chained assignment to a single loc assignment. Identify the owning object, the target rows and the target column in that statement. Describe the independent behavior of preview, and the limit of this claim for nested Python objects. Avoid describing Copy-on-Write as a switch that makes every operation free or every object recursively immutable.

On the PC, type the short owning update and reconstruct the missing-value mask from memory. In the app, reason about which object changes. Next, normalize text using vectorized string methods while preserving missing identifiers instead of converting them into literal text.

Optional stretch — after the core task works: Add one independently reasoned boundary case and explain its expected value before running it.

Computer practice: https://applydjinn.com/typorithm/pandas-lab-20261106.zip. Install pinned dependencies in an isolated Python environment. Run verify.py, then change the working exercise baseline and run --practice pd-cow. The real pandas tables, exact schemas, unchanged inputs, CSV parser options and owned in-memory SQL reconciliation use owned fixtures only. No production database, feed, account or paid service is used. This lab executes Python; the PC typing trainer and native reading progress do not execute it.