Claude File Upload to Check a Draft Report Against Its Tables

Create a file named s01-draft-report.txt (or a Word document) with the text below. All of it is invented. The screening value 50 is an invented screening value, not a real standard.

Site S-01 draft monitoring report (invented)

1. Summary
Five monitoring wells (MW-1 to MW-5) at Site S-01 were sampled on 14 and 15 September 2026 for Analyte A. Three of the five wells were above the screening value of 50 ug/L.

2. Results
Table 1 lists the results. The highest result was 140 ug/L at MW-5. The average detected result was 55 ug/L. Analyte A was not detected at MW-3, where the reporting limit was 5 ug/L.

3. Discussion
Two wells (MW-2 and MW-5) show Above SV in Table 1. Conditions are stable compared with June.

Table 1 (invented). Analyte A, results in ug/L.
Row | Location | Sample date | Result | Qualifier | Screening value | Comparison
2 | MW-1 | 2026-09-14 | 12 | | 50 | At or below SV
3 | MW-2 | 2026-09-14 | 68 | | 50 | Above SV
4 | MW-3 | 2026-09-15 | 5 | ND | 50 | Not detected
5 | MW-4 | 2026-09-15 | 50 | | 50 | At or below SV
6 | MW-5 | 2026-09-15 | 140 | | 50 | Above SV

Spreadsheet summary for Table 1 (computed in the spreadsheet): Above SV 2. At or below SV 2. Not detected 1. Average of detected results 67.5. Highest detected result 140 at MW-5.

What you should see: One short file with text and a table. Troubleshooting: Anthropic's upload page lists PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX as supported document types. For XLSX files, code execution and file creation must be enabled in your account.

1. Open Claude and start a new chat.
2. Click the "+" button in the lower left corner of the chat box.
3. Select "Add files or photos" and choose your file. Dragging the file into the chat window also works.
4. Wait for the file name to appear in the message box.

What you should see: The attached file name above or in the message box. Troubleshooting: If the table is a picture inside a PDF, Claude may not read it well. Use a text-based version.

Paste this prompt. It asks Claude to compare and list, never to calculate.

The attached draft report is invented practice material for Site S-01. Do not calculate anything and do not use outside knowledge. Compare only the text against Table 1 and the spreadsheet summary lines in the file.
1. List every number that appears more than once in the file, with every place it appears (section, sentence or table row).
2. List every mismatch: a sentence whose number or wording differs from Table 1 or from the spreadsheet summary lines. Quote the sentence, quote the table cell or summary line, and name both locations.
3. List every statement that no table in the file supports. Quote it. Do not say whether it is true.
4. Do not rewrite the report and do not suggest new text.
Label the output "Draft check for my review".

What you should see: Three lists with quotes and locations. Troubleshooting: If Claude suggests corrected sentences, tell it to remove them. Your job is to edit the draft.

The practice draft has these items. Compare the output against them.

Mismatches (2):

1. "Three of the five wells were above the screening value" in Summary. Table 1's comparison column shows 2 Above SV (MW-2 and MW-5), the summary lines say 2, and the Discussion says two wells.
2. "The average detected result was 55 ug/L" in Results. The summary line says 67.5. The value 55 is what you get when MW-3's reporting limit of 5 is counted with the four detected results, which is why it differs.

Correct statement (1): "The highest result was 140 ug/L at MW-5" matches Table 1 row 6 and the summary line.

Unsupported statement (1): "Conditions are stable compared with June." Table 1 has no June data in this file, and a statement about stability is a conclusion for the project manager.

Repeated numbers to expect: 50 in the Summary and in column E of rows 2 to 6. 140 in Results and in row 6. MW-5 in Results, Discussion and row 6. The reporting limit 5 for MW-3 in Results and row 4. Two for the Above SV count in Discussion and in the summary lines.

If Claude missed one of the three items above or added a mismatch that is not real, note it. The model output is a lead, never a result.

1. Open your draft in your word processor.
2. For each mismatch, open Table 1 and the spreadsheet and confirm the right value.
3. Change "Three of the five" to match the table's comparison column.
4. Change the average to the spreadsheet's value.
5. Either delete "Conditions are stable compared with June" or mark it for the project manager and the licensed professional, who decide any statement about conditions over time.
6. Run the prompt again on the revised file and confirm the lists are now short.

What you should see: The second run reports no mismatches and one fewer unsupported statement.

1. In your word processor, search for each number that appears in the text, such as 50, 140 and 55.
2. Confirm that each hit agrees with Table 1.
3. Search for "above", "below", "stable", "trend" and "no change". Read each hit against the table.