Flat grids, near-flat grids, zig-zag, staircase and repeating-cycle patterning, extreme and midpoint response styling, non-differentiation against the sample and reverse-keyed contradictions.
Automated data cleaning · built in-house
Every respondent scored, every exclusion explained.
ResClean is the respondent quality engine behind every MIU dataset. It reads your raw export, auto-detects the questionnaire, runs 24 fraud and quality checks and returns one auditable Data Quality Score per interview — with the evidence that produced it. It runs inside your browser, so the file never reaches a server.
24fraud & quality checks per respondent
6+1sub-scores rolled into one Data Quality Score
0files uploaded — processing is local
The score system
Six dimensions. One number you can defend.
Each sub-score runs 0–100 where 100 is clean. You set the weights and the Keep / Review / Remove cut-offs, and both are written into every export so the run can be reproduced on the next wave.
Interview length against the sample median, hyper-speeding, seconds per question, seconds per open end, Qualtrics-style per-page timers and implausibly long idle sessions.
Keyboard mash and non-language text, AI-written prose, low-effort filler, off-topic answers, verbatims duplicated across respondents and paste detection from typing speed.
Batch entry with machine-regular arrival gaps, duplicate identifiers, near-identical answer vectors, repeat IPs, /24 subnet clusters, click-farm device signatures and geo mismatches.
How much of the questionnaire the respondent actually answered, measured across the questions you selected rather than the whole file.
Pass rate on the trap and instructed-response items you designate. Codes and answer labels both work, and several acceptable answers can be allowed.
The weighted roll-up. Under your removal cut-off the record is dropped; between the cut-offs it is retained but flagged for a human look; a single critical finding always earns review, whatever the mean says.
What it catches
Twenty-four checks across six families.
No single rule proves fraud. ResClean combines behavioural, statistical, textual and metadata signals, then reports which ones fired and which could not run because a column was not available.
Pattern
- Straightlining
- Near-straightlining
- Zig-zag / staircase / cycles
- Extreme response style
- Midpoint response style
- Reverse-item contradictions
Timing
- Speeders
- Hyper-speeders
- Per-question pace floor
- Per-page speeding
- Pasted verbatims
- Implausibly long sessions
Open text
- Gibberish / keyboard mash
- Likely AI-written prose
- Low-effort filler
- Blank verbatims
- Off-topic answers
- Duplicated verbatims
Fraud & duplicates
- Duplicate identifiers
- Near-identical answer vectors
- Batch entrants
- Repeat IP addresses
- Subnet clusters
- Click-farm signatures
- Geo mismatch
- Failed attention checks
- High item non-response
Why ResClean
Where it goes further than a generic quality tool.
Most platforms score the answers they can parse and ask you to trust the verdict. ResClean is built for research operations: it reads production export formats, understands survey structure, and hands you the audit trail.
| Capability | Manual review in Excel/SPSS | Generic quality platform | ResClean by Miures |
|---|---|---|---|
| Native SPSS .sav input | Needs SPSS licence | Usually CSV / XLSX only | Read directly — labels, measure levels, long strings, all compression modes |
| Question-type detection | Manual mapping | Column-level guesses | Grids, multi-punch, ranking, numeric and open ends grouped by stem, with confidence and reason |
| Multi-header exports | Manual clean-up | Often breaks | Qualtrics three-row and Decipher two-row headers detected automatically |
| Codes and labels together | Two files, manual lookup | Codes only | Pair a labels export with a codes export and read real answer text everywhere |
| AI-written verbatim detection | Analyst judgement | Score with no reasons | Score plus the specific triggers — phrasing, structure, register, sentence uniformity, typing speed |
| Fraud clustering | Pivot tables, if attempted | Duplicates and IPs | Batch-entry gap regularity, /24 subnets, device + duration signatures, near-identical answer vectors |
| Audit trail | Whatever was written down | Score export | Flags, sub-scores, reason text, thresholds used and a printable methodology report |
| Tracker reuse | Re-specified each wave | Project-level settings | Saved cleaning profile — thresholds, question map, traps and reverse-keyed items — reapplied in one click |
| Data residency | Your machine | Vendor cloud | Your browser. Nothing is uploaded, stored or logged |
| Who decides | You | Automated threshold | You. A critical flag triggers review, never a silent delete, and any disposition can be overridden |
Comparison reflects MIU's own capability and the general shape of manual and platform-based alternatives. Individual vendors differ — ask us what a specific tool does and does not cover before switching.
Interactive visual demo
From raw export to an auditable clean dataset.
Four stages, each producing a file you can inspect. Nothing happens off-screen.
Ready for a project-specific estimate?
Open the enquiry form with this service selected, review the prefilled details and click Send enquiry for review.
FileRowsFormattracker_w3.sav1,240SPSSqualtrics_raw.csv8603 header rows
Stage 1 of 4
Read the export as it actually comes out
Value and label structure, multi-row headers and SPSS dictionaries are parsed locally — no reformatting first.
- CSV, TSV, XLSX, .sav, .zsav
- Qualtrics and Decipher header rows
- Codes paired with a labels export
QuestionTypeConfidenceQ5r1–r8Grid 1–590%Q4r1–r5Multi90%Q9oeOpen end85%
Stage 2 of 4
Auto-detect the questionnaire, then approve it
Columns are grouped by stem and classified with a confidence and a reason. Tick, untick or re-type anything before scoring.
- Grids, multi-punch, ranking, numeric, open ends
- Metadata roles: ID, IP, LOI, timestamps, device, geo
- Trap items and reverse-keyed items marked by hand
CHSTSOESBAS
DQS 613 flags · disposition: remove
Stage 3 of 4
Score every interview across six dimensions
Twenty-four checks run in seconds, even on files of 100,000+ records, and each flag carries the evidence that triggered it.
- Straightlining, speeding, patterning
- Gibberish and AI-written verbatims
- Batch entry, duplicates, click-farm clusters
Raw data→Flag file→Clean data
1,164 retained76 removed
Stage 4 of 4
Decide, then hand over something auditable
Analysts review flagged clusters and you can override any disposition. The export carries the original columns plus flags, scores and reasons.
- Retained and removed data as separate sheets
- Printable quality report with the thresholds used
- Cleaning profile saved for the next wave
1 / 4
Service tiers
Buy the engine, or the engine plus our analysts.
Quoted per completed interview with a project minimum. Add open-ended screening per record, and take a discount from wave 2 of a tracker when the saved profile is reused.
Automated screening
The ResClean run, the scored dataset and the quality report. You make the exclusion calls.
- Six sub-scores and a DQS per interview
- Flag file with reason codes
- Printable methodology report
- 1 business-day planning TAT
Managed cleaning
Everything in screening, plus an analyst reviewing every flagged record and cluster before delivery.
- Human review of all consequential exclusions
- Client-agreed thresholds and decision log
- Retained and removed datasets delivered separately
- 2 business-day planning TAT
Forensic cleaning
For disputed data, suspect suppliers or trackers where the removal rate has to be defended.
- Supplier / source quality scorecards
- Wave-over-wave incidence comparison
- Cluster investigation and recontact review where evidence allows
- 3 business-day planning TAT
What you receive
Files a client or committee can audit.
- Cleaned workbook: original columns plus flags, sub-scores, DQS and reason text.
- Retained and removed records as separate sheets, never silently dropped.
- Flag matrix — one row per respondent, one column per check.
- Question map showing what was scored and what was excluded from scoring.
- Quality report with disposition split, score distribution, flag incidence and cluster table.
- The exact thresholds and weights used, so the wave can be reproduced.
- Reusable cleaning profile as JSON.
Honest limits
What we will not claim.
- No tool can guarantee zero fraud. Fraud evolves and legitimate respondents sometimes look unusual.
- AI-verbatim detection is probabilistic. It reports its reasons and is meant to prompt a read, not to prove authorship.
- Checks that need an unmapped column are skipped — and the report says so rather than implying a clean bill of health.
- Geo mismatch needs both a stated and an IP-derived country.
- Final confidence still depends on recruitment source, respondent verification and questionnaire design.
Questions
The things buyers ask first.
Does my file get uploaded?
No. ResClean parses and scores it inside your own browser tab. Nothing is uploaded, stored or logged, so respondent PII never leaves your machine — and there is no server-imposed file-size limit.
Which formats can it read?
CSV, TSV, Excel .xlsx and SPSS .sav / .zsav, including value labels, measure levels, long variable names and very long strings. Qualtrics three-row and Decipher two-row header exports are detected automatically.
Can it really spot AI answers?
It scores AI-likeness from assistant phrasing, markdown structure, stacked essay connectives, metronomic sentence lengths, impersonal register, length against the sample norm and typing speed. The triggers are reported with the score. Read the verbatim before acting.
Will it delete data automatically?
Never silently. Each record gets Keep, Review or Remove from thresholds you control, every decision carries its reason, and you can override any of them before export. Removed records are delivered, not discarded.
Can you clean a study you did not run?
Yes — cleaning is a standalone line item. Send the raw export, the questionnaire and any known concerns; it works regardless of who programmed or fielded the study.
What about tracker waves?
Save the thresholds, question map, trap items and reverse-keyed items as a profile and reapply it to the next wave in one click — so wave-on-wave removal rates are comparable rather than re-specified.
Start with your own data
Run ResClean on a real file before you talk to anyone.
Open the engine, drop in an export and look at the flags. If you want our analysts on it, send the questionnaire, the raw file and the quality concerns you already suspect.