Prompt Chain in Claude: From a Results Table to a Reviewed Memo Draft

What This Builds

A memo can carry a sentence the table never supported, such as a line about what business confidence will do next. This chain puts a checking pass inside the drafting process. Claude first states what a results table says and does not say, then drafts a short memo from those statements only, then lists every claim in the draft next to the table cell behind it, and finally revises the draft to remove or mark whatever has no support. You check the output after each step. The project lead reviews the finished draft, and you sign off.

Prerequisites

• A Claude account. The paid plan is Pro ($20/month), and that is the whole ongoing cost of this build. Check claude.ai for what the free tier covers.
• A results table that your own spreadsheet or code computed, with the differences from Baseline computed there too
• A synthetic or public table, or approval to use the real one (see the data caution below)

The Concept

Think of a newsroom with a fact-checker who is not the writer. The writer works only from the notes. The checker walks through the finished piece, line by line, and asks where each claim came from. Here, one Claude chat plays both roles in turn, and you stand over it as the editor. The model never does arithmetic you rely on, because your sheet already did it, and it never adds a figure. Where the pasted table gives nothing, the model writes "not stated".

Data caution. Client data, documents produced in litigation, anything under a protective order or NDA, and non-public company data stay in the firm's approved systems. Work with synthetic tables, public data, column names and coded placeholders such as "Scenario A" or "[Client]" by default. Before pasting anything non-public, check Claude's privacy and data settings for whether chats are used to train models and how long they are kept, and confirm that the firm, the engagement and, on litigation matters, counsel approve the tool for that data.

Build It Step by Step

Part 1: Prepare the table in your own sheet

This table is invented, for practice. Build it in a spreadsheet with a header row in row 1 and the scenarios in rows 2 to 4.

Add three difference columns to the right, computed by the sheet so the model never subtracts:

• E2 =ROUND(B2-B$2,1) filled down to E4, for the output growth difference in points.
• F2 =C2-C$2 filled down to F4, for the employment difference in jobs.
• G2 =ROUND(D2-D$2,1) filled down to G4, for the price change difference in points.

The sheet should show these differences from Baseline:

Check them by hand. High output growth is 3.0 against 2.0, a gap of 1.0. High employment is 1,800 against 1,200, a gap of 600. High price change is 2.1 against 1.5, a gap of 0.6. Low is the mirror image in each column.

Part 2: Open a fresh chat and run Prompt 1

Start a new chat in Claude. Copy the whole table, with the difference columns, from your sheet. Paste the prompt below. Then replace the bracketed line with your table.

Below is a results table from a scenario exercise. Everything in it was computed outside this chat. Do not calculate anything and do not add any number.

Task: write a numbered list of statements about what the table says, each using only values shown in it. Then write a second numbered list of things the table does not say. Where something is missing from the table, write "not stated". Do not explain why any value is what it is, and do not describe any real-world event.

Table:
[PASTE THE TABLE WITH THE DIFFERENCE COLUMNS]

Checkpoint 1. Read each statement against the table. Every number must match a cell, in the same units. The second list should include points such as the table giving no reason for any scenario's values, and nothing about the margin of error. Delete or correct any statement with a number you cannot find. If Claude reworked a difference, compare it with the sheet's column.

Part 3: Prompt 2, the memo draft

In the same chat, send:

Write a draft client memo of about 400 words using only the statements from your first list in this chat. Do not add any figure, any cause or any statement about what will happen. Begin with the label "Draft for project lead review". Use the scenario names exactly as in the table. Where the statements give nothing, write "not stated". Use plain language for a reader who is not an economist.

Checkpoint 2. Read the draft against the Prompt 1 list. Look for forward-looking sentences, causes, rankings that the table does not show and any number that was rounded or restated. Do not fix them yet. Part 4 is designed to find them, and fixing them now hides what the checker can catch. Do confirm the label is at the top.

Part 4: Prompt 3, the claim check

Make a table of every claim in the memo draft above, one row per claim. Columns: claim number, the claim in the draft's own words, the table cell or cells that support it (give the scenario name and the column header), and the status "supported" or "no support in the table". After the table, list every claim with the status "no support in the table" again, on its own. Do not rewrite the memo in this step.

Checkpoint 3. Open the memo and the table beside this output. Check three things. First, each "supported" row points to the right scenario and column. Second, each cited value really appears in that cell. Third, no claim is missing from the claim table. The model is checking its own draft, so a claim can be missed. Go through the memo sentence by sentence and tick each one off against the claim table.

A typical unsupported claim in a draft like this one reads "Business confidence will recover under the High scenario". The table holds nothing on confidence and nothing on the future. Claude should list it under no support. If it does not, you have just caught a miss.

Part 5: Prompt 4, the revision

Revise the memo. Remove every claim you marked "no support in the table", or if one has to stay for the memo to read well, mark it in square brackets as [unsupported, for project lead decision]. Do not add any figure or any new claim. Keep the label "Draft for project lead review" at the top. After the memo, list what you removed or marked.

Checkpoint 4. Compare the revised memo with the claim table. Every unsupported claim should be gone or marked. Then run the final number check: for each figure in the memo, find it in your sheet. Check units, the plus and minus signs on the differences and the scenario names. Then send the memo to your project lead as a draft. You sign off on your own work, and the project lead reviews it.

Real Example: Three Invented Scenarios

Setup: The invented table from Part 1, with the sheet's difference columns pasted in. Input: Prompts 1 to 4 in one chat, with a checkpoint after each. Output, step by step (an illustration of what a reply might look like, not a guaranteed result):

• Prompt 1 lists, for example, that High shows output growth of 3.0 percent against 2.0 percent for Baseline, a difference of +1.0 point as the sheet states. Its "does not say" list includes the reasons for the scenario values and any uncertainty around them.
• Prompt 2 returns a short memo under the label "Draft for project lead review". Suppose it contains the sentence "Employment is 600 jobs higher in the High scenario than in Baseline." That sentence is fine, since +600 is in the table. Suppose it also contains "Confidence will recover under High." That one has no support.
• Prompt 3 returns a table. One row cites the High row and the Employment change difference column for the 600 jobs. Another row, for the confidence sentence, shows "no support in the table".
• Prompt 4 removes the confidence sentence and lists it among the removals.

Time saved: You get a first-pass review before the project lead sees the draft. That does not replace your own check or theirs.

What to Do When It Breaks

• The memo contains a number that is not in the table → Go back to Prompt 2 and repeat it, adding the sentence "Every number must appear in the table I pasted." Check the number against your sheet. The model must not supply it.
• Claude restates a difference that disagrees with the sheet → Your sheet is correct. Paste the difference columns again and ask Claude to quote them, not calculate.
• The claim table misses a sentence → That is expected sometimes. Read the memo line by line against the table yourself and add the row.
• Claude adds a cause or a prediction → Remove it. The table cannot support it. Repeat Prompt 4 with a reminder that no new claims are allowed.
• A long chat starts to drift → Start a new chat and paste the table and the revised memo into Prompt 3 again.
• The reply refers to real events or data you did not paste → Treat it as invented and delete it. No real data value belongs in this exercise.

Variations

• Simpler version: Run only Prompts 1 and 3. Draft the memo yourself and have Claude check it.
• Extended version: Use a Claude Project to hold the four prompts as saved instructions. Keep the rule about public, synthetic or approved material.

What to Do Next

• This week: Run the chain once on the invented table. Compare it with your own memo habits.
• This month: Try a public-data table from your own work, and keep your source log beside it.
• Advanced: Add a fifth step in which you paste the project lead's comments and ask for a list of the changes still to make.

Advanced guide for economist professionals. These techniques may need a paid Claude plan. Menus and settings change, so check Claude's current screens and help pages.