Download and install
The .dmg on macOS or the .exe on Windows. It is an ordinary program, with no dependency to install first and no server to configure.
2 minutesYour spreadsheet becomes system data
without retyping
Partiu Importador is the desktop app that brings the data living in spreadsheets today into Partiu Formatura: graduates, installments, contracts, products, employees, and suppliers. You point to the file, review what will be saved, and only then run the import.
Version 0.3.0 · free for Partiu Formatura customers · updates automatically
94import destinations
30modules covered
2,098mappable fields
58field transformations
10file formats
Data import
Weeks of work in one afternoon
Migrating from another system, receiving a list from the institution, uploading your supplier base: everything that is manual data entry today becomes a file and a click.
Reads the file you have
CSV, delimited TXT, fixed-width TXT, XLSX, XLS, ODS, JSON, and JSONL. Encoding detected automatically, choice of tab, header row, and range, plus a sample of the raw file before any interpretation.
Mapping with suggestions
The app suggests the destination field for each column from its name and content, showing the confidence of each guess. Where the suggestion does not fit, you can use a constant, column join, split, conditional, or expression.
Value matching with the system
The spreadsheet says “Gold Plan” and the system has an ID. The matching screen arrives pre-linked by similarity, and anything you fix once comes back ready in the next import.
Two-way deduplication
Within the file itself and against the database, using each destination's natural key: taxpayer ID, code, document number. A repeated row becomes a warning, not a duplicate record.
Templates that last
Mapping is saved by column name, not position. Next month's spreadsheet can arrive with columns in a different order and the template still works, and it runs the same on another machine.
Undo the import
Each load keeps the IDs it created. If something went wrong, rollback removes exactly what that job saved, without touching anything else.
Bulk documents
A folder with thousands of files named by taxpayer ID becomes attachments on the right records. The app checks the matches and shows orphan files before uploading the first byte.
Database diagnostics
Before importing, the app asks the system what your database really has and warns you if any catalog field does not exist in that environment. Nothing is silently half-saved.
Getting started
From download to your first import
There is no server to install, no API key to generate, and no template spreadsheet to fill in first. Five steps separate the installer from your first rows saved in the system.
Sign in with your account
The same login you use in the Partiu Formatura dashboard: password, a code by email, SMS or WhatsApp, or a QR code scanned with the mobile app. Company and permissions come from the login itself.
1 minuteConfirm the environment
Production, staging, or local. The environment stays visible in the app bar at all times and is highlighted whenever it is not production checking it here avoids the most expensive mistake in the process.
10 secondsRun the Diagnostics
The app asks your database what it actually has and compares it to the catalog, destination by destination. A catalog field missing in that environment shows up now, not after the load.
1 minuteImport a slice first
Pick the destination, point to the spreadsheet, map it, dry-run it, and execute with a few rows. Once checked in the dashboard, the same template runs the whole file without redoing anything.
10 to 20 minutes the first timeA real first load usually starts with a small slice of the spreadsheet: ten rows, dry run, execution, and a check in the dashboard. Once the slice is approved, the same template runs the whole file without redoing anything.
How it works
Seven steps between the spreadsheet and the system
Each step answers one question and only unlocks the next one when it is resolved. You can always go back; you cannot skip ahead with a required field still open. Below, what is up to you and what the app handles on its own in each of them.
1
Destination
Where does the data go?
You do
Search by destination name, by module, or by field search for “CPF” and see which destinations have it. When the destination requires a class, the scope is chosen here.
The app does
Loads from your database the catalog of allowed destinations, with fields, required flags, the size of each column, and the natural key used for deduplication.
Moves on when
A destination is selected and, if it requires a scope, the class is too.
94 destinations30 modulesSearch by fieldClass scope
2
File
How should this file be read?
You do
Point to the file and fix whatever came out wrong: delimiter, spreadsheet tab, header row, range, rows to skip, and column ranges in fixed-width TXT.
The app does
Detects the encoding, including legacy Latin-1, proposes the delimiter, and shows the raw sample before interpreting any field that is where you see the header starting on row 5.
Moves on when
The app can read the columns and return at least one data row.
10 formatsAutomatic encodingRaw sampleFixed-width layout
3
Mapping
Which column feeds which field?
You do
Review the suggestion field by field and resolve what is left: join first and last name, split the address, create a constant for the class, write a conditional or an expression.
The app does
Proposes the match from the column name and its content, shows the confidence of each guess, pushes pending required fields to the top, and applies transformations with a before-and-after preview.
Moves on when
Every required field of the destination has a defined source.
Auto-map with confidenceJoin and splitExpressions58 transformationsLearned aliases
4
Value matching
What is this value inside the system?
You do
Confirm the suggested matches and fix the crooked ones. “Gold Plan”, “GOLD”, and “gold 12x” become the same id.
The app does
Pulls the dictionary from the system itself, filtered by your company, pre-matches by similarity, and flags what is an exact match and what is a suggestion.
Moves on when
No required value is left unmatched whatever is left becomes an error in the dry run, not a crooked record.
System dictionarySimilarityExact vs. suggestedSaved in the template
5
Rules
What should happen when a row is not simple?
You do
Choose create, update, or upsert; define what happens with duplicates and with errors; mark which fields may be overwritten; set batch size, parallelism, and row filters.
The app does
Stores everything in the template and applies it row by row in the dry run and in the execution, with identical behavior in both.
Moves on when
Always: the step ships with safe defaults already selected.
Create · update · upsertDuplicate policyAllowed fieldsBatch and parallelismRow filters
6
Dry run
What would happen if I ran it right now?
You do
Read the report, fix the mapping or the spreadsheet, and dry-run again as many times as you want. The report can be exported and sent to whoever prepared the file.
The app does
Processes the entire spreadsheet without touching the database: groups errors by type with the row number, points out duplicates inside the file and against the database, shows the diff of records that already exist, and the exact JSON that would be sent.
Moves on when
A new template only executes after at least one dry run.
Nothing is savedErrors by typeDiff before updateJSON previewExportable report
7
Execution
Is the load holding up?
You do
Follow the progress and, if needed, pause, resume, reprocess only the rows with errors, or undo the entire job.
The app does
Sends in parallel batches with automatic retry when the network wavers, saves the state of every row on your computer, and uses an idempotency key so a lost response never becomes a repeated record.
Moves on when
It finishes on its own and a crash halfway continues from the exact point the next time you open it.
Parallel batchesPause and resumeReprocess errorsUndo by jobIdempotency
Input
The file you have already works
There is no official format or required template. The app opens whatever came from the other system, your accountant, or the institution's office, including the fixed-width TXT nobody reads anymore. Before interpreting anything, it shows the raw file: that is where you instantly notice the delimiter is wrong or the header starts on row 5.
CSVDelimited TXTFixed-width TXTXLSXXLSXLSMXLSBODSJSONJSONL
- Encoding detected automatically, including legacy Latin-1 files with broken accents
- Delimiter, tab, header row, range, and rows to skip
- Positional layout by column ranges, for fixed-width TXT files
- Sample of the raw file before the app interprets any field
Assisted mapping
Every column finds its field
The grid is organized by destination field, not by file column: the right question is “is the required taxpayer ID filled in?”, not “where did column F go?”. The app suggests the match from the column name and content, shows the confidence of each guess, and keeps pending required fields at the top.
- Suggestions from column name and content, with a confidence score
- Source by column, constant, column join, split, conditional, or expression
- Pending required fields highlighted at the top of the grid
- Learned aliases: the fix you make today comes back ready in the next spreadsheet
Required dry run
See the result before saving
Finding an error after saving four thousand records is the worst possible outcome of an import. That is why the app processes the entire spreadsheet without touching the database and shows a report: how many rows will be added, how many are duplicates, which have errors, and why. You fix the mapping and run the dry run again, as many times as you like.
- Errors grouped by type, with the row number of each occurrence
- Preview of the exact JSON that will be sent to the system
- Quality alerts: text that will be truncated, empty required fields, out-of-range values
- Exportable report to send to whoever prepared the spreadsheet
58 chainable transformations
The spreadsheet does not need to arrive ready
Dates in four different formats, taxpayer IDs with and without masks, amounts with thousands separators, names in all caps, phone numbers missing the ninth digit. The app cleans each field on input, with a before-and-after preview using the real data from your file, and the transformation is saved in the template for next time.
- Text: case, accents, punctuation, prefix, suffix, trimming, regex replacement
- Numbers: Brazilian or US decimals, currency, percentages, rounding
- Dates: dd/MM/yyyy, ISO, Excel serial, written-out dates, and configurable order
- Documents: CPF, CNPJ, postal code, phone numbers with the ninth digit, and email domain correction
The 58 transformations
Six families, chainable on the same field
A transformation takes a value and returns the corrected value. Because the output of one is the input of the next, the “Name” field can go through space cleanup, casing, and capitalization in an order you define and the preview shows the whole chain running on your own data.
Text
Cleanup and shape for anything typed by hand: extra spaces, all-caps names, punctuation and emoji that came along for the ride.
TrimCollapse spacesCasingProper nameRemove accentsRegexPrefix and suffixTruncateSlugMask
Examplemaria DA silvaMaria da Silva
Number
Brazilian and American decimals detected from the content, currency and units stripped, sign and scale corrected.
BR/US decimalStrip currencyMultiply and divideRoundParentheses as negativePercentageUnits
Example$4,800.504800.50
Date
Recognizes the shapes that show up in a real spreadsheet, including the serial number Excel keeps underneath the displayed format.
dd/MM/yyyyISOddMMyyyyExcel serialWritten outdmy/mdy orderAdd daysAccounting periodAge
Example452312023-11-21
Documents
The fields that carry their own rules in Brazil: documents with and without masks, phone numbers missing the ninth digit, emails with a mistyped domain.
CPFCNPJPostal codePhone and ninth digitE.164Normalize emailFix domainLicense plateExtract from free text
Examplejoao@gmial.conjoao@gmail.com
Semantic
For when the spreadsheet value has to become another vocabulary: yes/no, state abbreviation, internal code.
Configurable booleanState by nameLookup table
ExampleRio Grande do NorteRN
Structure
For files that are not a clean table: a value that depends on another column, a field buried inside JSON.
ConditionalExtract value from JSON
Example{"contact":{"phone":"84999..."}}84999…
Execution built to take the load
Eighty thousand rows, no fear
The load runs in parallel batches, with automatic retries when the network fluctuates. The state of each row is saved on your computer: if the power goes out at row 61,400, the app reopens and continues from 61,401. If a response is lost along the way, the idempotency key ensures the record is not added twice.
- Configurable batches and parallelism, with pause and resume
- Row state saved on your computer: closing the app does not lose work
- Reprocess only rows with errors, without repeating good ones
- History of every run, with undo per job
The rules step
You decide what happens in every situation
Step 5 is where the import stops being generic: it is where you settle what to do with a duplicate row, a broken row, an empty cell, and a record that already exists in the system. All of it is saved in the template.
Write mode
Defines what the load is allowed to do with the database. It is the first choice in the step, and it changes what the other options show.
- Create: only inserts new records
- Update: only touches what already exists
- Upsert: creates what is missing and updates the rest
Repeated row
Repetition is detected on two fronts inside the file itself and against the database by the natural key of the destination: document number, code, identifier.
- Skip the repeated row
- Update the existing record
- Create it anyway, when that makes sense
Row with an error
A validation error in one row does not have to condemn the other 79,999. You choose how much the load tolerates.
- Skip the row and carry on
- Stop the import right away
- Stop after N errors
Fields allowed to overwrite
In upsert mode the app shows a field-by-field diff of the records that already exist before the UPDATE. Only the fields you mark go through.
- Field-by-field selection
- Diff against the current system value
- Update the phone without touching the address
Empty cell
The default is conservative: a blank column preserves what is already saved. An incomplete spreadsheet does not wipe a record.
- Keep the current value (default)
- Write the empty value, if you explicitly ask for it
Pace of the load
How much the API receives at a time and how many batches run in parallel. An unstable network calls for smaller batches; a local database takes more.
- Batch size
- Simultaneous batches
- Automatic retry on network failure
Row filters
Not every row in the file should go in. The filter cuts before the dry run, so the report already comes out with only what matters.
- Import only rows matching the condition
- Drop repeated headers and totals
- Slice a range for the first test
Value matching
Spreadsheets speak in names, the system in IDs
“Gold Plan”, “GOLD”, and “gold 12x” are the same thing to whoever typed them and three different things to the database. The matching screen arrives pre-linked by similarity: you review it, fix anything that looks off, and move on. A spreadsheet with forty distinct plan values stops being a two-hour problem.
- Pre-matching by similarity, marked as exact or suggested
- Dictionary pulled from the system itself, filtered by the user company
- Values without a match become dry-run errors, not bad records
- The confirmed match is saved in the template
Reconciliation
Spreadsheets that depend on each other
Migrating systems is rarely a single spreadsheet: there are classes, graduates, plans, and installments, each pointing to the previous one through a code that does not exist in Partiu Formatura yet. Reconciliation combines several loads into one import: reference loads go first and, when the system returns the real IDs, they replace the temporary codes in the loads that depended on them.
- Reference loads go before dependent loads, in the order you define
- The ID returned by the system replaces the temporary code in later loads
- A missing or ambiguous match stops the import before the affected load is sent
- Works for migrating an entire system, not just one table
Files, not just data
A folder with thousands of documents
Scanned contracts, IDs, receipts, product photos. You point to the folder and say where the key is in the file name; the app figures out who each file belongs to and checks everything before uploading the first byte.
Point to the folder
Thousands of files at once, straight from your computer or the company drive. Nothing is sent yet.
Say where the key is
The app extracts the taxpayer ID, enrollment number, or code that identifies the owner from the file name: the whole name, before the separator, digits only, or a pattern you define.
Review the matches
The review happens before the first byte: each file appears with the name of whoever will receive it, and orphan files are kept separate for you to fix.
Upload with a queue
Progress per file, retries for failures, and resume from where it stopped. Closing the app does not restart the upload.
Key extracted from the file name:Full nameBefore the separatorSpecific partDigits onlyCustom patternFolder name
After the load
The work does not end at “completed”
Importing is routine: the same spreadsheet comes back next month, someone asks what was added last week, and every now and then you need to roll something back.
Reusable templates
The entire configuration (destination, mapping, transformations, value matching, and rules) saved under a name. Matching is by column name, so the order can change. You can favorite, export, and take it to another machine.
Detailed history
Each run records how many rows were added, how many failed, in which environment, and with which file. Details open row by row, with the error returned by the server.
Resume days later
Each row payload is kept in the local database. An interrupted load continues from the exact point in another session, without the original file, and the duplicate check is redone on resume.
Reprocess only what failed
Good rows stay as they are. The error report includes the values of each row, so you can fix the exported spreadsheet and reimport only those rows.
Undo the job
Rollback uses the IDs that load verifiably created. It is not “delete what was created today”: it deletes exactly those records.
Catalog diagnostics
Compares, destination by destination, what the app can write with what your database actually has. A field that does not exist in the table would be silently discarded; here it shows up before the load.
Features
Every feature exists because of an accident
Importing is an operation that fails quietly: the record that never arrived, the field dropped on save, the retry that duplicated everything. The list below is what the app does so none of that goes unnoticed.
FeatureWhy it exists
Per-row state on your computerAn 80,000-row load that dies at row 61,400 continues from 61,401. Closing the app loses nothing.
Mandatory dry runFinding the error after saving four thousand records is the worst possible outcome of an import.
Check before uploadingUploading 3,400 files only to discover 200 orphans wastes bandwidth, time, and patience.
Idempotency keyA response lost on the way back must never become a second identical record.
Deduplication on two frontsThe repetition may be inside the file and also between the file and the database.
Undo by jobThe rollback deletes the ids that load provably created, not “everything that came in today”.
Resume in another sessionThe payload of every row is stored: days later the load continues without the original file.
Compare before updatingIn upsert mode the field-by-field diff appears before the UPDATE, and only what is marked overwrites.
Empty does not eraseA blank cell preserves the saved value, unless you explicitly choose otherwise.
Template by column nameNext month the spreadsheet may arrive in a different order and the setup still holds.
Learned aliasesThe mapping correction you make today comes back ready on the next similar spreadsheet.
Catalog diagnosticsA declared field that does not exist in the table would be dropped silently on save.
Visible environment selectorImporting into the wrong database is the most expensive mistake, and it happens by distraction.
Quick command (⌘K)Someone who imports every week should not hunt for the same screen in the menu every time.
Coverage
94 destinations across 30 modules
Each destination knows its table fields, what is required, the size of each column, and how to identify a repeated record. The app checks this against your database before it starts.
Finance 8 destinations
Installments, unified installments, payment plans, accounts payable and receivable, categories, cost centers, and budgets
Graduation 7 destinations
Graduates, classes, subclasses, institutions, courses, events, and custom enrollment fields
People Management 6 destinations
Employees, positions, departments, time records, benefits, and job openings
Inventory and WMS 6 destinations
Products, categories, warehouses, locations, balances, and movements
Procurement 6 destinations
Suppliers, requisitions, quotes, orders, items, and receipts
Production Planning 5 destinations
Product structures, routings, work centers, orders, and time entries
CRM 5 destinations
Accounts, contacts, opportunities, activities, and territories
Fleet 4 destinations
Vehicles, fueling, maintenance, and traffic fines
Accounting 3 destinations
Chart of accounts, journal entries, and reconciliations
Occupational Health and Safety 3 destinations
Occupational exams, risks, and mandatory training
Routes and Trips 3 destinations
Routes, stops, and trips
Projects 3 destinations
Projects, tasks, and time entries
Fixed Assets 3 destinations
Assets, depreciation, and movements
Store and POS 3 destinations
Store products, categories, and terminals
Tickets 3 destinations
Ticket types, issued tickets, and coupons
Kitchen 3 destinations
Menus, recipe cards, and batches
Communication 3 destinations
Announcements, polls, and contact lists
Event Sales 2 destinations
Event consumables and sales
Galleries 2 destinations
Galleries and photo-to-graduate links
Contracts 2 destinations
Contracts and signers
Kids 2 destinations
Children and guardians
Pricing 2 destinations
Inputs and cost compositions
Field Team 2 destinations
Freelancers and shifts
Rewards Club 2 destinations
Partners and offers
Albums 1 destination
Album orders
Training 1 destination
Learning paths and enrollments
Access Control 1 destination
Access credentials
Legal 1 destination
Cases and deadlines
Collections 1 destination
Collection workflows and events
Fundraising 1 destination
Raffles and prize draws
No surprises
Importing means writing to production
The app handles every load with the care of someone who knows the data on the other side is real.
The environment is always in plain sight
Importing into the wrong database is the most expensive mistake there is. The selected environment is visible at all times and flashes when it is not production.
Dry run before the first load
A new template only runs after a dry run. No import starts blind.
Nothing lands in the wrong place
The server only writes to the fields the destination declares, always within the user's company. The file does not choose where to write.
Audit trail for every load
Who imported, when, in which environment, how many rows were added, and how many failed. Everything recorded per company.
Sensitive data protected
Taxpayer IDs, salaries, and bank details are masked in technical logs, and imported content is neutralized so it cannot become an executable formula when re-exported.
Resending does not duplicate
Each row carries an idempotency key. A response lost on the way back does not become a repeated record.
Empty cells do not erase
On updates, a blank column preserves what is already saved, unless you explicitly say otherwise. An incomplete spreadsheet does not wipe records.
You choose what gets overwritten
On upsert, a field-by-field diff appears before the UPDATE and only the fields you select are applied. Updating a phone number does not have to touch the address.
Files checked before they are accepted
When uploading documents: extension on the allowed list, executables rejected, binary signature matching the extension, and file names free of directory traversal.
The app
A native program for the heavy lifting
A 300 MB file is not uploaded anywhere before it needs to be: reading, mapping, and dry runs happen on your computer. Only the finished record goes to the system.
macOS and Windows
Native app, with a .dmg installer for Apple Silicon and .exe or .msi for Windows 10 and 11. Same engine and same features on both.
Automatic updates
New versions are announced inside the app and installed with one click, with the signature verified before applying. Nobody gets stuck on an old version.
Environment selector
Production, staging, or local, visible at all times and highlighted when it is not production. Importing into the wrong database is the most expensive mistake there is.
Quick command (⌘K)
Reach any destination, section, or action from the keyboard, without hunting through menus.
Sign in your way
Sign in with the same dashboard account: password, access code by email, SMS, or WhatsApp, or QR code scan from the mobile app.
Work saved locally
Templates, history, row state, and the upload queue live in a database on your computer. A power outage in the middle of a load does not lose what was already processed.
Also in the browser
Partiu Importador Web
The same importer, opened in the browser with nothing to install. You sign in with the same dashboard account and get the same 7-step wizard, with the mandatory dry run before anything is saved. The spreadsheet is still read on your computer: only the finished record goes to the system.
What comes with it
The same 94 destinations, the same file formats, and the same wizard: destination, file, mapping, transformations, rules, dry run, and execution. History with report, resume, and undo, templates, reconciliation of several loads, bulk file uploads, and graduate documents.
Browsable catalog
On the web you can browse every destination and field, see examples and the duplicate rule for each one, and download the XLSX template ready to fill in.
Your work kept in the browser
Templates, history, and the state of each row are kept in the browser itself, separated by user and by company. Whoever shares the computer does not see the work of a colleague, and you can back up and restore.
When to use the web version
On a computer where you cannot install programs, on a borrowed machine, or for a quick load away from your desk. Open the browser, sign in, and import.
When to prefer the app
For files above 80 MB, which is the web version limit, and for anyone who imports every week: the app reads much larger files, switches environments, and updates itself.
One tab at a time
So the history does not get mixed up, the web version works in one tab at a time. If you open another one, it offers to continue there and the first one saves everything before handing over.
Frequently asked
What people ask before downloading
Do I need to prepare the spreadsheet in a specific format?
No. The app adapts to the file you have: it picks the tab, header row, delimiter, and encoding, and cleans fields on input. If you prefer to start from a template, each destination also offers a ready-made spreadsheet to download.
What if I get the mapping wrong and import incorrectly?
The dry run shows the result before anything is saved, and every executed import can be undone: the app keeps the identifiers it created and removes exactly those records, without touching anything else.
How many rows can it handle?
Loads of tens of thousands of rows are the normal case. Uploads run in parallel batches, with automatic retries on network failures, and the state of each row is saved on your computer so it can resume.
Does the app work on Mac and Windows?
Yes, it is a native app for macOS (Apple Silicon) and Windows 10/11. Both downloads are on this page, and the app updates itself when a new version is released.
Who can use the importer?
Company administrative users, with a login to the system itself. Every import is restricted to the signed-in user's company and recorded in the audit trail.
Can I import documents and photos, not just spreadsheets?
Yes. In addition to data, the app uploads files in bulk: you point to the folder, it extracts the key from each file name (taxpayer ID, enrollment number, code), checks who each file belongs to, and only then uploads.
Can I migrate from another system with several linked spreadsheets?
That is what reconciliation is for. You build the import with several loads (classes, graduates, plans, installments) and say which depends on which. Reference loads go first, and the IDs returned by the system replace the temporary codes in later loads. If any match is missing or ambiguous, the import stops before sending the affected load.
What happens if the spreadsheet has a row that already exists in the system?
It depends on what you choose in the rules step: skip, update, or create anyway. The app identifies repeats using the destination's natural key (taxpayer ID, code, document number), both within the file and against the database, and shows the total number of duplicates during the dry run.
Could updating an existing record erase data that is already filled in?
A blank cell preserves the saved value by default. And before the UPDATE, the app shows a field-by-field diff of existing records: only the fields you select are overwritten.
Is the entire spreadsheet uploaded to the server?
No. Reading, mapping, transformations, and the dry run happen on your computer. Only each row's finished record goes to the system, through the Partiu Formatura API, with the user's login and company.
Is there a browser version?
Yes. Partiu Importador Web does the same import in the browser, with nothing to install, the same login, the same 7-step wizard, history with undo, and templates. It accepts files up to 80 MB; for larger files and for anyone who imports often, the desktop app is the best fit.
How is the app updated?
Automatically. When a new version is released, the app notifies you and installs it with one click, verifying the package signature before applying it. There is no need to come back to the website to download it again.
Do I need special permission to use it?
You need to be a company administrative user, with a login to the system itself. Every import is restricted to the signed-in user's company and recorded in the audit trail, with who imported, when, in which environment, and what was added.
Download Partiu Importador
Free for Partiu Formatura customers. Choose the version for your system, install it, and sign in with the same login you already use in the dashboard. Updates arrive automatically.
Download for macOSApple Silicon · .dmgDownload for WindowsInstaller · .exeDownload MSI for WindowsEnterprise · .msi
Requirements: macOS (Apple Silicon) or Windows 10/11 · version 0.3.0. The MSI package is for enterprise mass deployment; for normal use on Windows, use the .exe installer.
Bring your data into the system
Download the app, sign in with the same account you already use in the dashboard, and import at your own pace. The dry run shows the result before anything is saved, and undo stays available afterwards.