AI-Written Python: The Text “False” Can Still Switch a Flag On

A catalogue export contains a reassuring instruction: false. The item should stay out of the next brochure. An AI-written Python script reads the file, converts that field with bool() and includes the item anyway. The script runs successfully. Its interpretation is the problem.

For a creator using AI to prepare small automations, this is an unusually useful review exercise. The word on the screen, the type of value in memory and the rule used to make a decision are three separate things. Checking them together can reveal an error that a successful run would leave hidden.

A word is not automatically a Boolean

A Boolean represents a true-or-false value. A string represents text. Python’s truth-value documentation explains that empty strings are false in a condition; non-empty strings are true. Applying bool() to text therefore tests whether that text is empty. It does not interpret an English instruction.

In Python, bool("false") returns True. So do bool("true") and bool("0"). The empty string, bool(""), returns False. Replacing an explicit conversion with if value: still uses truth-value testing, so it does not solve this particular problem.

The distinction matters wherever a text field controls inclusion, visibility or an optional processing step. The business question might be “should this record appear?” The code may instead be answering “does this field contain any characters?” Those questions need different tests.

Follow one fictional record through the import

Imagine a designer, Ada, preparing a seasonal product catalogue. Her CSV has two columns: reference and include_in_catalogue. Record P101 contains the text false; record P102 contains true. Her agreed rule is to include only P102. These records are illustrative, rather than an account of a real business.

Python’s CSV documentation says the ordinary reader returns strings without automatic type conversion, apart from a specified numeric-conversion option. Its DictReader maps fields to named keys. Under the default reading behaviour used here, Ada’s two flags arrive as text.

Selecting records with bool(row["include_in_catalogue"]) includes both P101 and P102. The output can look tidy: two valid references, no error message and a perfectly readable catalogue. A review that checks only whether a file was produced will miss the wrong selection.

Before asking for a fix, Ada can write the expected outcome beside the input. P101 must be excluded; P102 must be included. That gives the reviewer an independent basis for judging the script, rather than relying on the AI’s explanation of its own code.

Agree the accepted vocabulary

For this example, Ada chooses a narrow contract: accept the words true and false, ignore surrounding spaces and letter case, and reject everything else. That is a workflow decision. Another importer might deliberately require exact lowercase text without any spaces. The owner should choose the rule before the code implements it.

A parser for Ada’s contract first requires a string. It then obtains a comparison token using value.strip().casefold(). If the token equals "true", it returns the Boolean True. If it equals "false", it returns False. Otherwise, it raises an error for review.

The important improvement is the explicit mapping. It should be possible to point to the two accepted words and explain what each produces. An unfamiliar token does not silently acquire a meaning merely because the programme needs a yes-or-no answer.

Avoid the tempting shortcut of treating “anything other than true” as false. Under that approach, the misspelling flase disappears into the exclusion group alongside a deliberate false. Ada would lose the opportunity to correct an ambiguous instruction before building the catalogue.

Test the boundaries as well as the happy path

• "true": accept and include the record.
• "false": accept and exclude the record.
• "FALSE" and " false ": accept and exclude under Ada’s chosen cleaning rule.
• "0", an empty string and "flase": reject for review.
• None: reject because this parser requires text.

The conversions and the two-record selection were checked in Python 3.12.14. The generic Boolean conversion selected both references; the explicit parser selected only P102. These checks establish the stated example behaviour in that environment. They do not validate an uninspected production import.

MaryChuks.com’s guide to testing rejected values in AI-written regex applies the same useful discipline to reference formats. An acceptance rule needs examples of what must stay out, including inputs that look almost acceptable.

JSON makes the type distinction visible

A JSON source needs its own check. Python’s JSON decoder documentation maps JSON false to Python False, while a JSON string becomes a Python string. Accordingly, json.loads('false') produces the Boolean false; json.loads('"false"') produces text. Those two results were also checked in Python 3.12.14.

Quotation marks here change the value’s type. Ada should not assume that a field has the same representation simply because a CSV export and a JSON response display a similar word. Equally, a parser deliberately designed for CSV text should not be applied indiscriminately to values already decoded as Booleans.

Review the selection before creating the output

For a trial, ask the AI to produce a review report containing each reference, the original flag, the interpreted result and any validation error. Keep that report separate from the catalogue file. A successful validation step should account for every input record, including records held back for clarification.

Ada can inspect the complete two-record example directly. For a larger import, she could begin with a small labelled set, then reconcile the full report’s included, excluded and rejected groups against the input. If an unknown value appears, decide how to resolve it before letting the script produce the final deliverable.

The reusable lesson is to specify meaning at the boundary where data enters the workflow. Ask the AI what type arrives, which values are accepted and what happens to everything else. Then test the actual selection. A small conversion deserves attention when it decides which work the reader eventually sees.

Sources

• Python 3.12: Truth Value Testing.
• Python 3.12: CSV File Reading and Writing.
• Python 3.12: JSON encoder and decoder.

Documentation checked on 4 November 2026. The catalogue and records are hypothetical; the Python example results were tested as described.