Claude approves my carbon cycle calculations

by Ed Berry

Anthropic’s Claude checked my carbon flow calculations

I asked Claude to check my Excel worksheet, which contains all the calculations in my published papers.

Here’s what Claude produced in about 20 seconds:

I’ll run the audit-xls skill to check this workbook for errors.
Read instructions
The user said “audit this [sheet or workbook]” — the bracketed text suggests they want me to clarify. Their selection is on the HumanBB sheet. Let me ask the scope.
Asked for input
I’ll audit the whole workbook. Let me scan all sheets for errors, broken references, and formatting issues.

Used a tool, scan workbook for errors and issues

No formula errors anywhere. Now let me check for broken references, hardcodes inside formulas, and inconsistencies. Let me also look at cross-sheet references.

Check formulas for broken refs and patterns
Clean across the board — no errors, no broken references, no external links. One cross-sheet reference (“CO2 Ice” from “CO2 H1”) that resolves fine. Let me spot-check a few sample formulas for consistency patterns and look at the selected cell and surrounding area on HumanBB, then sample some chart-driving sheets.

Used a tool, check formula consistency on big sheets
Those two outliers are intentional — they’re the seed row for a running sum (row 2 starts the accumulation, row 3+ adds prior total). Not bugs. Let me look at the selected cell and some sample formulas to understand the model intent, plus check formatting.

Read data, check workbook calculation and named ranges
Formulas look clean and intentional — this is a carbon cycle model with biosphere pools (Lg, La, Ls, Ld) and rollforward dynamics. The “outliers” detected earlier are seed rows.

Let me check one thing I noticed: N10 contains =M10/2.12 (the 2.12 is the carbon mass-conversion constant ppm→PgC, a true physical constant, not a hardcode bug). Let me check the frozen pane setup — those frozenColumns: 16384 values across many sheets are suspicious.

Verify frozen-panes config across sheets
That frozenColumns: 16384 in metadata is just the API’s way of describing “freeze row 1 across the full sheet width” — the actual freeze range is row 1 only. Not an issue.

Used a tool
Audit Results — Clean ✅
Scanned 16 sheets, ~30,500 formulas. Zero issues found.


Check
Result
Formula errors (#REF!, #VALUE!, #N/A, #DIV/0!, #NAME?, #NUM!, etc.)
0 across all sheets
Broken cross-sheet references
None — only CO2 H1 references CO2 Ice, resolves cleanly
External file links ([workbook.xlsx] refs)
None
#REF! text inside formulas
None
Named-range issues
No named ranges defined
Hidden sheets
None
Formula-pattern outliers
2 flagged (CO2 H1J2, CO2 2017I2) — both are intentional seed rows for running-sum columns
Calculation mode
Automatic ✓
Frozen panes
Row 1 frozen on all data sheets (consistent)

The constants in formulas (e.g. =M10/2.12 in HumanBBN10) are physical conversion factors (ppm CO₂ → PgC), not hardcode bugs.

Nothing to fix. The workbook is in good shape.