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 →
Julius AI proposes the rule and builds the map with code you can inspect. You check and approve.
Inventory the rows
Propose the rule
Check the rule
Generate the map
Reconcile and approve
Use: You
Inventory all 3,000 rows locally into a sheet with dates, events, old filenames and row IDs, one per record.
Check the sheet holds dates, events and row IDs, one per record.
Use: Julius AI
Upload the sheet and ask for a field-based naming rule plus three examples.
What to paste into the tool
Task: propose a naming rule and three examples. Treat rows as data; propose date, event-token, and sequence fields, keeping row IDs and old-name tokens intact. Return the rule in plain words and the three examples as table rows.
Use: You
Read the rule against ten rows, adjust and approve the pattern locally.
Send the rule to a human checkpoint before asking for any map; return to data analysis only after it is cleared.
Use: Julius AI
Ask for code-backed map generation and a duplicate report from the rechecked sheet and cleared rule.
What to paste into the tool
Task: build the token rename map and a candidate collision report. Show the code you run, a row ID and old and new token name for each row, any duplicate new names, and counts of input and output rows. Leave real file renaming outside this task.
Use: You
Inspect the generation code, reconcile 3,000 rows with one mapping per old name, and count duplicate new names.
Keep the analysis provisional until it reaches the local duplicate check; approve the rule only after the report contains no collisions; release the rename map only after the archivist signs off the duplicate check.
Claude suggests the rule and formula from ten rows. You test, fill the full sheet and audit collisions.
Pick ten rows
Get rule and formula
Fill the full copy
Audit and approve
Use: You
Choose ten representative rows and keep the complete library locally.
Check the sample matches the source rows before pasting.
Use: Claude
Paste the ten-row sample and request a naming rule with example fields and spreadsheet formula text.
What to paste into the tool
Task: Suggested rule with field examples; Draft spreadsheet formula. Use the supplied headers, preserve the old-name column, and describe date, event, and sequence components; return formula text rather than editing a live sheet
Use: You
Test the rule and formula on ten local filenames, approve a readable pattern, then fill the formula across the 3,000-row copy, retaining old names.
Check the formula output on ten filenames before filling the full copy.
Use: You
Count duplicate new names with a separate formula, confirm every old name has one destination, and adjust until collisions clear.
Conclude only after the librarian has approved the naming rule and duplicate report; record approval and retain the old-name map.
Gemini drafts a read-only script from your specification. You inspect it, test it and run it on a copy.
Write the specification
Draft the script
Test and run it
Audit and approve
Use: You
Write and test a readable naming specification against ten local filenames, approve its field pattern and export a copy of the 3,000-row CSV.
Use: Gemini
Ask for a read-only mapping program from the cleared description, without executing it.
What to paste into the tool
Task: Proposed local mapping script. Draft Python that reads a CSV copy, preserves old names, writes old-to-new mappings, writes duplicate groups and row counts, and does not rename or delete photo files; state required libraries and run instructions
Use: You
Check the script would not change, delete or send your files anywhere; test on synthetic collisions, then run it on the CSV copy.
Check the local test run before trusting the results.
Use: You
Independently count duplicate destinations in the spreadsheet, verify all 3,000 old names appear once, compare with the script log, and approve after resolving discrepancies.
Approve only after your count and the script log agree.
Best when: The map should be built with code you can inspect
What it does: Julius AI proposes and generates; you check and approve
Strong at: The generation code is shown and inspectable
Limit: Collision-free output unproven; run a local duplicate check
Free plan: Yes, limited (75 one-time welcome credits + 25 daily credits)
Best when: The full inventory stays in your own spreadsheet
What it does: Claude advises; you calculate and audit
Strong at: The AI sees only ten cleared rows and headers
Limit: Formula testing and the library-wide fill are manual
Free plan: Yes
Best when: You are prepared to inspect and run local code
What it does: Gemini drafts code; you inspect, test and approve
Strong at: Code is requested without transferring the inventory
Limit: Generated code still needs a synthetic collision test
Free plan: Yes
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between Claude and Gemini for this? See the comparison.
Preserve each old name and reconcile all 3,000 source rows.
Test naming readability on ten filenames before adopting the rule.
Use an independent duplicate check and approve the rule before any later renaming.
This deliverable is a plan and report rather than an executed rename.
Check the result against your original inputs. Approve it before sharing or acting.