For SurveyLoopr, SurveyCTO and ODK Central

Every interview, checked
before it costs you.

Each interview is checked the moment it syncs, while the team is still in the field. A local AI helps fix what's wrong, and nothing leaves your computer.

7 days free. Or download it and explore the sample.

Made by ARCED International. Used daily at ARCED Foundation.

Every interview, the moment it syncs.

Connect SurveyLoopr, SurveyCTO or ODK once. New interviews arrive on a schedule; the raw data stays untouched.

Checks written for you.

DataFlow reads your form and sets up the checks. Describe a new one in plain words.

Every problem comes with a fix.

One ranked list. The local AI explains each flag and prepares the fix. You approve it.

Proof your client can open.

A Field Report with progress, team performance and every problem fixed. One file, no login.

Clean data is how you level up.

Stars reward care, not speed. Old problems block new levels.

Made by
ARCED International
In daily use at
ARCED Foundation
.do
Built on
arceddataflow
Their Stata workflow, since 2023

Field data goes wrong in the same four ways.

Errors found after the team has left

Checked on every sync

Hundreds of flags, no idea where to start

One ranked list, a fix ready for each

Quality locked in one person's do-file

Checks written for you, from your form

Fixes that overwrite the raw data

Every fix logged, reversible, reproducible

The checks nobody had time to write.

Added on import and on every sync. Rules the AI reads from your form are marked, and noisy checks arrive switched off.

Missing answers

Household ID is answered

Required and always asked, so it should never be blank.

2 flaggedReview

Don’t know / refused

Not too many “don’t know” answersToo noisy

Flags interviews where more than 20% of answers are don’t know or refused (codes -888, -999).

31 flaggedReview

Other, specify

Review “other” answers in Specify other main crop

Free-text answers often belong to an existing option.

37 flaggedFYI

Across questions

More children in school than people in the household

Added by AI

${children_school} > ${hhsize} · Ask who the children in school are

—Review

Land acres given but owns no land

Added by AI

${owns_land} = 'no' and ${land_acres} > 0 · Check whether the household owns land

—Review

Interviews that may be made up

Six statistical signs per enumerator, compared with the rest of their team.

Travel no one could make

If the form records GPS, this check adds itself: interviews an enumerator could only reach at over 60 km/h. A wrong clock, or not done in person.

“Other” answers that already fit

Matched to existing options by a small model on your computer, in any language.

Look closer with the personal-data check

A small model can also read comments, notes and unusual variables for names, numbers and sensitive topics the rules miss. It runs on this computer, so nothing is sent anywhere.

Download (0.8 GB)Not now

These look personalTicked ones will be protected

Respondent’s name

respondent_name

looks like a name600 filled

Respondent’s phone number

phone_number

looks like a phone number480 filled

deviceid

identifies the phone used600 filled

Personal data the rules miss

Found in comments and notes, then locked in a separate encrypted store.

A private AI for every research assistant.

It runs on the laptop, on the graphics card or the processor. No cloud, no API keys, no data sent anywhere.

It never makes up numbers. DataFlow counts; the AI explains.

Ask DataFlow

on this computer

Ask in your own words, or try one of these, from your data:

AI can make mistakes, so check carefully. Nothing changes until you approve it.

Built for research

Reproducible for the journal. Ready for your IRB.

Format

Stata (.dta)CSVExcel (.xlsx)R

Include

Codebook

Every variable with its label, type, value labels, how filled it is and where it came from.

Do-file and raw data

raw.csv as received, and dataflow.do, which replays every correction, step and label in Stata.

Correction log

corrections.csv: every correction with who, when and why, including ones later undone. Always included.

De-identified · 525 ready interviewsChoose folder and download

For the journal

Every download includes dataflow.do, which replays each correction from the raw data in Stata.

Stata do-file that rebuilds the clean data
Correction log with who, when and why
Tamper-evident activity log
Raw data never modified

Outputs after each step

Each step can write its data to a folder you choose, every time new interviews arrive.

After Import: the vault

Raw interviews with their personal data, encrypted in one file. It opens in VeraCrypt too, as an ordinary folder, for anyone without DataFlow.

Create the vault

After Protect: de-identified raw data

The interviews as they arrived, with protected variables left out.

After Clean: cleaned data

With every correction, in the formats you choose.

For the IRB

Raw interviews with personal data live in an encrypted vault that also opens in VeraCrypt.

VeraCrypt-compatible encrypted vault
Personal data found and locked away
De-identified downloads by default
Nothing leaves the computer, AI included

A day with the sample study.

Livelihoods Round 2 ships inside DataFlow: 600 household interviews, 8 enumerators, 4 villages. These are its real screens.

Today: one decision at a time, with a suggested fix

Overview: 525 of 600 interviews are ready to use

Ask DataFlow, the local assistant, proposing to fix five mistyped incomes, waiting for approval

Corrections kept on top of the raw data, with who, when, why and Undo

Datadom: the study as a kingdom at level 12, with its next goal

8:00 AMToday: one decision at a time, with a suggested fix

8:00 AM
One decision at a time
87% done, 10 decisions left. Each comes ranked and explained, with a fix ready.
10:30 AM
48 interviews arrive, already checked
31 new flags, grouped into 2 problems, while the team is still in the villages.
11:00 AM
The AI does the legwork
Five mistyped incomes found, every change shown first. You approve.
3:00 PM
Logged and reversible
Who, when and why for every correction, with Undo. The raw data is untouched.
6:00 PM
The client sees it too
A Field Report the PI opens without installing anything.
6:05 PM
And the study levels up
Old problems block new levels. Clean data comes first.

87%

of the checking done by DataFlow

525 of 600

interviews ready for analysis

decisions left for a person

bytes sent to the cloud

Built inside a field team.

DataFlow grew out of arceddataflow, the Stata workflow ARCED Foundation cleans its surveys with, and it's in daily use there. Its duplicate and outlier checks are tested to flag exactly what arceddataflow flags, so switching changes the tool, not the results.

arceddataflow on GitHub

ARCED Foundation

From the makers of SurveyLoopr

Simple pricing. Every feature in every plan.

Monthly

$19per seat a month

Choose monthly

Team

7-day free trial

$99a year

5 seats included

Save 91% against 5 monthly seats

Start free trial

Extra seats $4.99 a month each, billed yearly

Made by ARCED International. Used daily at ARCED Foundation.

Every check and connector. Local AI. Stata, R, Excel and CSV exports. Field Reports. Windows, macOS and Linux.

Paying for a SurveyLoopr server? One seat is included. Paying by invoice? Talk to us

Questions, answered.

Does my survey data go to your servers?

No. DataFlow runs on your computer and talks only to your own survey server. Your account just tells the app which plan you're on.

Does the AI send data anywhere?

No. The models run on your computer and work offline. No AI service, API key or account is involved.

Can I submit the cleaning code with my paper?

Yes. Every download includes a Stata do-file that rebuilds the clean data from the raw data, plus the full correction log.

What do I tell my IRB?

Data stays on study computers, encrypted at rest. Raw data with personal information lives in a VeraCrypt-compatible vault, and downloads are de-identified unless someone unlocks them, which is logged.

Which survey platforms work?

SurveyLoopr, SurveyCTO (encrypted forms too) and ODK Central directly, and CSV or Excel exports from any other tool, KoboToolbox included.

What happens when my trial or plan ends?

Nothing is locked away. DataFlow becomes read-only: you can open every project and export everything.

Start with the sample study.

600 interviews with real-looking problems, ready to explore. No account needed.

Windows macOS Linux

Made by ARCED International. Used daily at ARCED Foundation.

Every interview, checkedbefore it costs you.

Every interview, the moment it syncs.

Checks written for you.

Every problem comes with a fix.

Proof your client can open.

Clean data is how you level up.

Field data goes wrong in the same four ways.

The checks nobody had time to write.

Interviews that may be made up

Travel no one could make

“Other” answers that already fit

Personal data the rules miss

A private AI for every research assistant.

Reproducible for the journal. Ready for your IRB.

For the journal

For the IRB

A day with the sample study.

One decision at a time

48 interviews arrive, already checked

The AI does the legwork

Logged and reversible

The client sees it too

And the study levels up

Built inside a field team.

Simple pricing. Every feature in every plan.

Monthly

Team

Questions, answered.

Start with the sample study.

Every interview, checked
before it costs you.