Final survey data in.
Insights out in 10 minutes.

Routing and masking compliance, checked against the master questionnaire — the pass you run the moment the soft-launch dataset lands.

• vc_skipAnswered when not asked, blank when asked — full skip-logic compliance per routed variable.
• vc_gateItem-by-item gating across matched batteries (each statement asked only when its base item qualifies).
• vc_funnelBrand-funnel integrity: no usage without awareness, no consideration without salience.
• vc_multiMulti-punch hygiene including exclusive-option breaches — "None of the above" ticked alongside a real answer.
• vc_single · vc_gridSingle-punch and grid punches validated against the code frame; out-of-range and missing flagged per cell.
• vc_codesUndefined codes — any punch with no matching value label anywhere in the file.
• vc_rank · vc_rotationTied or incomplete rankings, corrupted rotation orders, multi-punch slots that skip positions.

Every numeric entry held inside the bounds the questionnaire promised — including the allocation grids where respondents type a zero across the whole row.

• vc_rangeFloor / ceiling on any numeric write-in — age, spend, percentages, 0–100 ratings — single variables or whole blocks.
• vc_sumConstant-sum / auto-sum compliance: allocations must hit the target total, with a configurable tolerance. Blanks count as zero, so incomplete allocations surface too.
• vc_flatlineThe all-zero allocation grid — a respondent entering 0 in every numeric field of a matrix — caught as its own named flag.
• vc_outlierStatistical outliers by z-score or IQR — the ₹8,00,000 monthly grocery spend in a sample that tops out at ₹50,000.
• vc_datesInterview timestamps that end before they start; any start / end pair sanity-checked.
• vc_emptyQuestions that are blank for the entire base — the pipe that never fired, the punch that never landed.

The quality-triage layer: who is real, who is rushing, who is on autopilot. Each flag carries the evidence, so triage decisions are defensible.

• vc_dupeDuplicate respondent IDs within the file — plus identity collision against prior waves via a known-ID baseline you carry forward run to run.
• vc_speedSpeeders by median-fraction (adapts to actual survey length) or an absolute cut-off — with the option to freeze the baseline on soft-launch completes so the cut-off never drifts mid-field.
• vc_flatlineStraightliners: the same punch down an entire grid, with a minimum-items threshold for partly-filled batteries.
• vc_patternAlternating responders (1-2-1-2 down the grid) — the inattentive pattern plain straightlining checks miss.
• vc_textdupCopy-paste verbatims — one respondent reusing an answer across questions, and identical answers appearing across different respondents.

Verbatim hygiene before a single response reaches coding — and a compliance pass your DPO will actually like.

• vc_texteffortLow-effort filler ("na", "dk", "nothing", "-") counted across a respondent's open-ends, with configurable phrase lists and thresholds.
• vc_textformatType confusion: "many" typed into an income box, "1234" typed into a name field.
• vc_piiEmails and phone numbers left inside verbatims — the report shows only the PII type found, never the contact detail itself.
• vc_ruleFree-form cross-question logic: write any condition that should never be true for a clean interview ("under 18 yet marked complete") and label it for the QA report.
• csharpAnd when the mandate doesn't fit any template — a sandboxed scripting block. See the escape hatch below.

Every check accepts a filter expression, so bases are honoured: respondents never asked a question aren't flagged for skipping it.

Flags land per respondent, per variable, with the offending value and rule attached — a QA report you can hand straight to the panel provider.

The validation script is an artifact. Soft launch, main field, top-up sample, next wave — same rules, zero drift, one click.

Coded variables, nets, bands

Ordered IF / ELSE-IF logic builds segments and bands with value labels attached, ready to tab immediately. Any-punch nets, counts, row statistics and formula variables cover the rest.

Recode · reverse · convert

Scale collapses with explicit handling for unlisted codes, reverse-coded battery items fixed before averaging, text fields converted to clean numerics. All in place, all filter-aware, all logged.

RIM / rake weighting

Multi-factor rim weighting to gender, age, region or any quota set — per wave or whole sample. Output includes achieved-vs-target per cell and the weighting-efficiency figure your QA sign-off needs.

Banner-ready cross-tabulation

• →Z-test for column proportions with letter notation (A, B, C…), at 90% / 95% / 99% confidence.
• →Column %, row %, total % or raw counts — with SPSS value labels enriched automatically.
• →Three base rows on every weighted table: unweighted, weighted, and effective base (Kish design-effect), so significance is honest about what weighting costs.
• →Descriptive rows (mean, median, std-dev, min, max) alongside the distribution.

The DP-grade table engine

• →Rows and columns can be anything you'd write in a tab spec — any base, any cut, no pre-coding needed.
• →Nets and sub-nets with row basing: percentage a child row against its parent net instead of the column total.
• →Measures beyond counts: means (Welch's t-test), medians, percentiles — with significance letters on all of them.
• →Table-level weights with per-column overrides, low-base suppression, and whole banks of tables — up to 32 variants — run in one go.
• →Any finished table can be handed to the AI for a written insight summary plus suggested charts — grounded in that exact table, nothing else.

Gauge, pie, donut, bar, column, stacked bar, line, area, radar, heatmap, scatter, bubble, treemap, radial bar, polar area, box plot — each one checked and tidied automatically, so a broken chart never reaches a client.

Every analysis step behind the dashboard is captured as a recipe. When the next wave lands, replay reproduces the same dashboard on the new data — pure computation, no AI variance, no regeneration lottery. Need something new this wave? Regenerate-replay keeps every existing chart and adds new ones.

Read-only share links for clients, with owner-side view analytics — visits, unique visitors, referrers — so you know the deck actually got opened.

Brief in, sourced dashboard out

• →Write the brief — market context, competitor scan, category deep-dive — and the AI researches it into a structured dashboard: sources, summaries, key findings, charts.
• →Live progress while it works, regeneration history when the brief evolves, and manual override when an analyst wants to reshape the output.
• →Delivered on the same share-link infrastructure as survey dashboards — one link to the client, view analytics for you.

The master document and the data finally agree

• →Upload the questionnaire Word file and it becomes a structured, editable question list — sections, options, routing and programming notes all captured, with quality scores so you know exactly what to double-check.
• →An automatic matcher then links each questionnaire question to the SPSS file's variables — confirmed, suggested and unmatched, each with the evidence behind the match.
• →That mapping is what lets validation rules reference the questionnaire's own routing — the master document and the data finally agree on what "Q12" means.