Choose one of three approaches. The steps for that option appear below.
Choose an option above. On a phone, swipe the tabs left or right to see all options →
Cursor drafts a cell-indexed account; you trace all transformations, parameters and versions back to the notebook.
Collect labelled code and declared
Draft a cell-by-cell transformation account
Check every cell and finish
Use: You
Have ready: notebook code with stable cell labels, environment declaration and author-supplied limitations.
Use: Cursor
Open the code export in Cursor Ask mode; request a transformation register and methods paragraphs, then copy to a local check file.
What to paste into the tool
Task: Screened notebook text with cell labels, environment declarations and author limitation notes; Cell-indexed transformation register and proposed methods paragraphs. Read the cell labels in order and describe filters, joins, transformations and parameters using code references; list versions only from the supplied declaration and mark missing versions as not stated; draft methods paragraphs and use only the author limitation notes; do not run cells or infer findings
Use: You
Trace register and prose to cells; restore filters, joins and parameters, remove unsupported behaviour, check declared versions and retain author-supplied limitations.
Accept methods text only after all code references are reconciled. The notebook author releases methods text only after its code references have been reconciled.
You extract the operation ledger and references; GitHub Copilot drafts prose, then you check coverage and code agreement.
Write an operation ledger
Draft the methods narrative
Check ledger coverage and reject
Use: You
Have ready: labelled notebook cells, environment file and author limitations. Record every filter, join, transformation, parameter, literal setting and declared version.
Use: GitHub Copilot
Open the ledger as text in VS Code; ask GitHub Copilot Ask to draft prose without consulting other workspace files.
What to paste into the tool
Task: Freshly screened methods ledger with code and environment extracts; Methods narrative organized by ledger cell references and supplied limitations. Convert the ledger into methods prose; include every recorded filter, join, parameter and declared package version with its cell label; retain missing values as not stated; include limitations only from the author column; do not infer analytical results or execute the notebook
Use: You
Compare prose with ledger and retained code; check every row and label, correct omissions, reject invented behaviour and attach the author limitation statement.
Accept prose only when ledger coverage and code agreement are complete. The notebook author accepts the prose only after ledger coverage and code agreement are complete.
You write the methods; Gemini Code Assist flags discrepancies, then you check the code and settle every item.
Write methods with code citations
List methods-to-code discrepancies
Check discrepancies and finish cited
Use: You
Have ready: notebook code, recorded versions and author limitations. Write methods, cite transformations and parameters by cell, and append code and environment extracts.
Use: Gemini Code Assist
Place the paired prose and code extracts in a scratch file; ask Gemini Code Assist to compare descriptions with cited code blocks.
What to paste into the tool
Task: methods draft paired with matching code extracts; Methods-to-code discrepancy log with cell references. Compare every methods statement with the corresponding code extract; flag missing or extra filters, joins and parameter values with cell references; note version claims absent from the declaration; do not add limitations or run cells; produce a discrepancy log rather than rewritten methods
Use: You
Reopen every cited cell, settle each discrepancy, revise methods and check every filter, join and parameter is cited; retain only author-stated limitations.
The notebook author holds each discrepancy for manual adjudication. Close the discrepancy log before releasing the methods section.
Inference: A explains code before prose review, B renders a human operation ledger, and C compares a human methods draft with code excerpts
Best when: You want a cell-indexed explanation before checking methods prose.
What it does: Cursor drafts; you check and decide
Strong at: Connects supplied code cells to transformation descriptions and proposed methods paragraphs.
Limit: You must trace the entire account and own the limitations.
Free plan: Yes, limited agent features
Best when: You want to extract the facts yourself before narrative drafting.
What it does: GitHub Copilot drafts; you check and decide
Strong at: Turns your cell-referenced operation ledger into readable methods prose.
Limit: It relies on ledger completeness; you must reject invented behaviour.
Free plan: Yes
Best when: You already have methods prose and want a code comparison.
What it does: Gemini Code Assist drafts; you check and decide
Strong at: Flags description-to-code mismatches with cell references.
Limit: You must adjudicate discrepancies and retain only author-stated limitations.
Free plan: Yes, daily request caps
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between Cursor and GitHub Copilot for this? See the comparison.
Never paste credentials or full payment details.
Trace filters, joins and parameters to code cells and restrict limitations to the author statement.
Check the result against your original inputs. Approve it before sharing or acting.