AI average audit: separate blank, text and numeric zero

Which cell types belong in this calculation?

An AI average audit should begin with an inclusion ledger, not a polished answer. The arithmetic mean divides the sum of included numeric values by their count. This exercise asks you to decide which source cells belong in that denominator. It is not a second lesson on choosing between a mean and a median.

Use a stipulated Excel range-reference contract: numeric cells count, including numeric zero; genuinely empty cells and text cells do not count. A text numeral that looks like zero is still text in this supplied card. Visual similarity is not evidence of matching data types. The source supplies the types explicitly; we are not claiming to diagnose an actual workbook by looking at its displayed values.

Write this rule first: “Practice V1; B1–B7; fictional score units; numeric cells only; include observed numeric zero; preserve empty and text cells without conversion.” These are teaching inputs, not customer records, an actual model response or measured company performance. No account or spreadsheet application is needed for the manual core.

The contract concerns a reference to a cell range, not every possible value typed directly into a function’s argument list. Dates, logical values, errors and formulas returning empty text are outside this small card. For a broader starting point, use the artificial intelligence guide. Here the finished artifact will show a decision beside every source identifier before it shows an average.

Preserve seven source records and their stipulated types

Copy the following source card without converting or filling anything. The word empty is explanatory notation for a genuinely empty source cell, not text entered into that cell. The text numeral is explicitly declared to be text. Quotation marks help describe the type; they do not create an additional observation.

Give each identifier one ledger row with four fields: source value, stipulated type, inclusion decision and reason. Keep excluded records visible. Excluding a cell from a numeric calculation does not authorize deleting its source row or treating the underlying case as nonexistent.

The included set is B1, B2, B3 and B4. B5 and B7 have no supplied numeric score. B6 is a text cell under this exercise contract. Its type is not automatically evidence of an import mistake, nor permission to repair it. If a real source owner later authorizes a correction, that belongs in a new source version with a new calculation.

The leading-zero CSV check protects identifier text during transfer. This task neither imports a file nor changes an identifier. It reads a pretyped numeric-measurement card and records which supplied cells a reference-based aggregation would include.

How do the sum and numeric count separate?

Add the four included values on a calculator: 0+8+16+24=48. Count their identifiers separately: B1 through B4 make four numeric cells. The reference mean is 48 divided by 4, which is 12 fictional score units. Save all three labels: sum 48, numeric count 4 and mean 12. The answer alone cannot reveal a denominator mistake.

Now calculate a deliberately wrong alternative. If numeric zero is excluded as if it were empty, the sum remains 48 but the count falls to 3. The result becomes 16. Matching sums therefore do not establish matching calculations. B1 contributes nothing to the sum but still contributes one to the numeric count.

A different incorrect claim counts every listed record, giving 48 divided by 7. Keep that result as the exact fraction 48/7 when diagnosing the mistake. A rounded decimal is unnecessary here. It does not describe the supplied numeric-cell mean, because the count includes three records that the written contract excludes.

These are not competing business recommendations or proposals to choose whichever answer looks better. Two rows deliberately violate the same fixed contract. The mean and median audit instead uses complete numeric observations to distinguish summary measures. Do not add a median or sorted list to this exercise: the typed inclusion ledger is the new work product.

What decisions do the three proposed claims receive?

Claim one says: “Seven source identifiers require a denominator of seven.” Request a correction. The source contains seven records, but only four are numeric cells under this range contract. Cite the per-ID decisions rather than relying on the result displayed by a calculator. The denominator must be explainable from the original card.

Claim two says: “Zero adds nothing, so excluding it is equivalent to ignoring an empty cell.” Reject that for this contract. B1 is numeric and belongs in the count. The difference between 48/4=12 and 48/3=16 is the evidence. An observed zero and an unknown value may contribute differently even when a quick glance at a sum hides that distinction.

Claim three says: “The correct mean of 12 proves the actual mean score for all seven underlying cases.” Mark it unsupported. No numeric score is supplied for two empty cells, and the intended meaning of the text entry has not been established. The calculation describes the supplied numeric cells, not a fully observed population of cases.

Save an unresolved source-owner question: “Why are B5 and B7 empty, what does the text in B6 mean, and who may authorize any correction?” Do not answer it yourself. The Excel prompt templates contain a separately specified default for a missing discount in a row formula. That domain-specific instruction is not permission to fill unknown scores with zero in this card.

What do the two official reference pages establish?

Microsoft Support documents that AVERAGE ignores text, logical values and empty cells within a range or reference, while retaining numeric zero. Its COUNT page explains counting numeric entries in a referenced range rather than counting all visible source records. These are two pages from one organization, not independent institutions.

Keep their scope narrow. We have not run Excel or a model to produce an observed workbook result. This is a manual comparison between a typed fictional card and documented reference rules. If you independently test a workbook, record its actual setup and result separately. Do not describe that optional future test as something this article performed.

A card with no numeric cells has a numeric count of zero. An arithmetic mean is then undefined; do not invent a zero mean. Write “no numeric observations; mean undefined.” Errors require separate handling in real ranges: a visible COUNT does not by itself prove that another aggregate has a valid result. Error cells are deliberately absent from this source.

The Notebook table export audit preserves unspecified programme durations when transferring data. This card involves no export or generated Studio table. It checks an inclusion count on already stipulated cell types. The shared rule against inventing missing values is acknowledged, but the operative artifact here is the per-ID numeric denominator, not transfer verification.

Which evidence must survive saving and reopening?

On Windows, open Start, type Notepad and launch the text editor. Head the document “Typed-cell denominator audit V1.” Add the rule card and every source row with its value, type, inclusion decision and reason. Include sum 48, numeric count 4, mean 12, and the two deliberately wrong denominator results.

Before saving, add all three claim decisions with their supporting source or calculation evidence. Include the unresolved source-owner question without an invented answer. Press Ctrl+S, select Documents and save as typed-cell-audit-V1.txt. Close the editor and reopen the file from that folder. Confirm that all seven identifiers remain, B1 is included, B6 stays text and B5/B7 remain unknown rather than numeric zeros.

For an optional second opinion, supply only the fictional card to an assistant you already use. Ask: “Preserve each source ID and stipulated cell type. Apply only the numeric cell-range reference contract. Show inclusion decisions, sum, count and mean; include numeric zero, exclude text without converting it and leave empty cells unfilled. Evaluate the three claims against the original card. Do not generalize the result to every underlying case.”

Compare any response with your saved manual key. If no AI comparison was performed, say so explicitly. No subscription purchase or live workbook edit is necessary. A suggested fifteen-minute plan allows three minutes for boundaries, four for ledger decisions, four for arithmetic and claims, and four for saving and reopening. This is a cognitive planning allowance, not a measured speed guarantee.

In earlier public writing about perfection and effectiveness, I described delaying projects while pursuing every detail. Applied here, that principle means finishing one reproducible record before expanding the task. It is not a claim that I ran a spreadsheet experiment or achieved a customer result. The final handoff should let another reader reconstruct the denominator without trusting fluent wording alone.