Preregister and Archive a Domandata Study

Domandata can link a survey to a permanent preregistration record from AsPredicted, OSF Registries, or another provider with an HTTPS record URL. It also fingerprints the instrument, logs amendments, and can publish a verification certificate after collection. The preregistration provider and the repository that eventually receives response data do not have to be the same service.

Set up the registration before fielding if the study protocol requires preregistration. The clearest record starts before the public survey link is live.

AsPredicted: create and approve the preregistration on AsPredicted, then generate its shareable PDF. Paste the permanent URL ending in .pdf into Deploy → Preregistration. AsPredicted keeps private records accessible only to their authors, so Domandata cannot link or verify one until an author generates that permanent PDF URL.

OSF Registries: paste a public registration URL directly. Connecting an OSF token is optional and is only needed when you want Domandata to confirm a private OSF draft-registration link before saving it.

Another provider: paste its permanent HTTPS preregistration record URL. Domandata stores and displays the link without sending credentials to that provider.

1. Step 1: Create an OSF token if needed. A token lets Domandata confirm private draft registration links. Public registration links do not require one.

2. Step 2: Open Settings. In Domandata, go to Settings → Integrations → API Keys and find OSF under Research Repositories, then click Connect.

   Step 2: Open Settings

3. Step 3: Paste the token. Store the token only in the OSF connection in Settings → Integrations. Do not paste it into survey text, code blocks, or shared notes.

4. Step 4: Connect OSF. Confirm the connection before returning to the survey editor.

5. Step 5: Link the registration. In Deploy → Preregistration, paste the final OSF Registration URL or a draft registration link.

   Step 5: Link Registration From Deploy

Use the preregistration URL to connect the survey to the protocol or registration record. Before locking, confirm that the survey title, Flow, condition arms, quotas, and export variables match the registered design. Survey collaborators with Editor access can update the registration link, lock or unlock the survey, and submit amendments; ownership is not required. Viewers, commenters, and translators remain read-only.

• Preregistration URL: paste the permanent AsPredicted PDF, final OSF Registration, OSF draft registration, or another HTTPS record your team will cite.
• Lock and fingerprint: lock only after the survey version is ready to serve as the reference design.
• Verification: use verification details to document which survey state was associated with the open-science record.
• Amendments: record meaningful changes after preregistration, especially wording, routing, condition, quota, or measurement changes.

The replication package bundles the survey instrument and study documentation so collaborators, reviewers, or future analysts can understand what was collected. Build it from Export after the instrument, Flow, variables, and documentation are ready.

1. Open Export. Review the response export and file export areas before archiving documentation.
2. Review included files. Confirm the package includes the instrument, codebook, Flow diagram, security or review files, and any generated documentation you intend to share.
3. Review the privacy choices. Browser information, IP addresses, and device fingerprints are always stripped. The Privacy panel starts from The Public; choose Review to pick a different audience or change any column before depositing — see below.
4. Choose a destination. Harvard Dataverse is the default for a replication package because it contains response data and Dataverse supports reviewable drafts and restricted files. Zenodo remains available when immediate public publication is appropriate, or download a .zip for another repository.
5. Issue the certificate if needed. The Domandata verification record compares the collected instrument with the locked fingerprint and reports the final verified response count.

You don't need a repository account to get the package. In Export → Citation & Replication, choose Download → Replication Package (or Instrument Package for materials without responses). It builds the same files, with the same privacy choices described below, and saves them as a single .zip you can send to co-authors or archive yourself. A download doesn't create a citable repository record; use Harvard Dataverse by default for response-bearing packages, or Zenodo when immediate public publication is appropriate. Verification certificates are issued separately from Deploy.

Harvard Dataverse, Zenodo, and Download use the same Privacy panel as the Export tab — see Export Research Survey Responses for how presets and per-column choices work. Deposits are usually public, so it starts from The Public (who the file is for) unless the survey has saved its own deposit choices or you've picked a different replication-package default under Workspace Settings → Default Export & Display:

• Identifying and free-text columns are dropped. Questions asking for a name, email, or phone number, signatures, file uploads, voice recordings, open Text Response and Text Chat answers, panel or respondent IDs, and Silicon Sample prompts are left out of both response CSVs, Responses.sav/Responses.dta, the Data Dictionary, and the analysis templates. Whenever a package drops any column, Cleaning Log.csv also coarsens each response's timestamp to date-only and omits the country named in a VPN/proxy-detected quality flag, since either can otherwise help link a response back to other data.
• Rare demographic combinations are hidden. Columns that read as demographics (age, gender, region, occupation, income, education, and similar) have Rare Combos switched on. Every response is grouped by its combination of values across those columns, and that combination is replaced with suppressed (rare combination) for any respondent whose exact combination is shared by fewer than the minimum group size (5 by default). Everything else about that response, and every column for a respondent whose combination is common enough, is untouched.
• Choose something else column by column. Choose Review, pick a different audience, or set a column to Keep, Broaden (number ranges, month or year dates, 3-digit ZIP codes, or rare answers grouped as "Other"), Redact PII (free-text questions: replaces email-, phone-, and SSN-shaped text, optionally also names, places, and organizations), Coded ID (panel and respondent IDs become stable codes that still link across your team's exports), Mask (every value becomes anonymized), or Drop. Detection works from question type, validation, and wording, so check each group before depositing. The name-and-place scan runs inside Domandata and works from grammar patterns, so it can miss an unusual name or catch an ordinary word.
• The README says what was done. The package README and metadata.json list the preset; every dropped, redacted, masked, coded, and broadened column; any rare-combination suppression; and a disclosure risk check — how many respondents have a combination of demographic answers no one else shares, and the size of the rarest combination.
• Include GitHub Sync repo snapshot. If this survey has a GitHub Sync repo connected, this toggle (on by default) bundles a .tar.gz of the repo's current contents — including any author-drafted analysis pipeline committed there as GitHub Sync starter files — into the package as an extra file, so the analysis code travels with the deposit rather than only living on GitHub. See GitHub Sync.

The certificate (public at /verify/[surveyId]) checks two things against the preregistered record: that the instrument used for data collection still matches the locked fingerprint (or lists each amendment if it doesn't), and how many responses were collected against the promised sample size.

The sample-size count is a verified human respondent count, not a raw submission tally — it excludes Silicon Sample (AI-simulated) responses, your own test submissions from Preview or the live-draft link, and any response you marked excluded during quality review. Test submissions are identified by the server when they arrive, after it checks that the submitter is the survey's author, so they can't be relabeled as respondents later. A separate Respondent Verification section reports the full breakdown (total collected, simulated, excluded, flagged for review) and lists which fraud/quality signals — reCAPTCHA, IP intelligence, honeypot questions, attention checks, dwell-time and impossible-combination rules — were actually configured on the survey during collection. It only certifies against signals you turned on: a survey with none of those configured will say so plainly rather than imply protection it didn't have.

Two more lines speak directly to whether the data could have been manufactured or trimmed:

• Distinct network origins. Each response records a one-way hash of the respondent's IP address (the address itself is never stored), and the certificate reports how many distinct origins the verified responses came from. People on a shared network, like a campus or office, count as one origin, so treat this as a floor; many responses from very few origins is worth a closer look.
• Responses deleted after the survey was locked. Every deleted response is written to an append-only log that no one on the survey team can edit, whether it was deleted in the app, through the API or MCP, or by an automated retention rule. The certificate reports how many respondent responses were deleted after locking, and how many of those were deleted by a person. Clearing test responses before you lock doesn't count against you.

Responses can only arrive through the survey itself, and no one on the survey team can change what a respondent submitted. The only later additions are your quality-review decisions and optional AI transcripts attached to audio answers, and neither alters the original answer. Together with the instrument fingerprint, this gives reviewers a chain of custody from preregistration to the final dataset. It can't make fabrication impossible, but it makes it visible.

• Pre-Analysis Plans and Survey Codebooks
• How to Cite Domandata in Your Methods
• Prepare a Research Survey for IRB Review
• Export Instruments, Codebooks, and Flow Diagrams
• Publish and Share a Research Survey
• GitHub Sync