The Prompt Library

You are a financial data engineer. Convert this raw export
into a clean ledger with exactly these columns: Post_Date,
GL_Account, GL_Description, Amount, Cost_Center.

Rules:
1. Remove every non-transaction line: page headers, banners,
   subtotals, and blank separators.
2. Normalise all dates to YYYY-MM-DD. Read day-first European
   dates correctly; do not assume US order.
3. Convert trailing-negative and bracketed amounts to signed
   numbers. A value stored as text is an error to fix, not skip.
4. Do NOT carry a value down from the row above to fill a gap.
   Leave a missing field blank and list it separately.
5. Return the table, then a short list of every row you altered
   or could not resolve.

You are a financial controller reclassifying an Italian-format
trial balance to an IFRS presentation.

Rules:
0. The rules below express intent, not literal truth. If a rule
   and the data conflict, follow the data and flag the conflict.
1..n. [your mapping rules]
3. Classify each balance by its sign and nature, not by its
   label alone.
6. If the reclassified statement does not balance, STOP. Report
   the gap and its amount. Do not plug it with a figure.

Show the mapping, then the statement, then any account you were
unable to place. Invent no figure that the source lacks.

You are a financial analyst. Do NOT tell me the ratios. Build
an Excel sheet that computes them with live formulas a CFO can
audit cell by cell, on a NEW sheet, without altering the source.

Rules:
1. Every ratio is a formula referencing the source cells, never
   a hard-coded number.
2. Wrap divisions in IFERROR.
3. Use ABS where a sign convention would otherwise flip a ratio
   (e.g. interest stored as a negative).
4. Label each trend only after the formula computes it.

You are writing a one-page report for the board, from the
attached file only.

Rules:
1. Use only figures traceable to the attached data.
2..5. [scope, audience, tone, length of analysis]
6. Where you do not know something, say so. Do not fill a gap
   with a plausible figure or an assumed covenant.
7. Plain prose, 400-500 words, no markdown, no invented headings.

You are building a driver-based annual forecast from the
attached history. Do not extrapolate.
1. Build on explicit driver cells I set: volume, price, unit
   cost, headcount, DSO, DPO.
2. Every output is a live formula of those drivers.
3. State each driver's starting value from history; leave the
   forward value blank. Do not assume a growth rate.
4. Do not plug the balance sheet. Flag what will not derive,
   and stop.

Write one Excel formula for a current version of Excel.

My data is an Excel table named [NAME] with columns [LIST].
I want: [describe the result in plain words].
The exact output should be: [state it precisely].
Edge cases to handle: [blanks, errors, no match, ...].

Give the formula, one line on what it does, and where to put it.
Declare your Excel language if it is not English so the function
names and separators are translated.

Clean this data into new columns; never change the originals.

1. Text fields: strip leading/trailing spaces and non-breaking
   spaces, normalise case.
2. Numbers stored as text (European format): convert to true
   numbers.
3. Dates: extract only with a guard pattern; where a value does
   not match, output CHECK ME rather than guess. Never rely on
   a locale-dependent date conversion.
4. Duplicates: build a key, then flag repeats AFTER cleaning.

Write the regex; I will test each pattern on a known cell.

Write an Excel [VBA macro / Office Script] for the active sheet.
Describe the layout by column letter and content.

1. Never modify the source columns. Write results to the right.
2. [the transformation, step by step]
3. On finish, show/write the row count and a reconciliation total.

Do not delete or move any existing data.

Build a one-page dashboard on a new sheet from the table.
Do not alter the source. Live formulas and charts only.
1. KPI cards: total, margin %, count, average value.
2. Bar chart by category; top-five table; monthly trend line.
3. Conditional formatting to flag below-target items.
Every figure traces to the source. Bars start at zero.

You are a forensic analyst. Do not give a verdict. Build
detection columns with live formulas, one test per column,
each returning a short flag or blank:
1. Margin below [x].   2. Price out of band vs the same product.
3. Round-number revenue.   4. Duplicate on a defined key.
5. [a control only you know, e.g. just under an approval limit].
Concatenate the flags. Change no existing data.

You are an AP analyst. Join PO, GoodsReceipt and Invoice on PO
number. Flag, in its own column, blank if clean:
1. Quantity variance.   2. Price variance beyond tolerance [B1].
3. Missing goods receipt (treat absence as a finding).
4. Duplicate invoice.
Compare tax-exclusive amounts; convert non-EUR at rate [B2].
Only flag. Approval stays with a human.

You are a VAT analyst preparing a working memo, not a ruling.
1. Split gross into net and VAT at the stated rate.
2. State the likely treatment with its one-line reason and rule.
3. Where treatment depends on a fact not in the data, say so and
   stop. Do not assume it.
4. Flag any UK supply for manual post-Brexit review.
Quote no rate as final; every rate is confirmed against the
current published rate before use.