Reconova: A Complete Data-Quality Workspace for Messy Business Data

Most data problems do not announce themselves. An export opens cleanly, the columns look right, and the file goes into a CRM, a finance system, or a migration script. The damage shows up later: duplicate customers after an import, invoice references that lost their leading zeros somewhere between two tools, a status column that quietly contains four different spellings of the same value.

Reconova is the Motifuse product built for that category of problem. It is a complete data-quality and cleanup workspace: a place to profile a messy dataset, see exactly what is wrong with it, build a cleaning recipe one step at a time, validate the result against explicit rules, find near-duplicate records with explainable matching, and export a corrected file along with a verifiable record of what was done.

This article explains what Reconova does today, how the free browser workspace and the subscription cloud tier differ, how files and data are handled, and where the product's current limits are. The goal is that you can judge whether it fits your workflow before you ever open it — and use it well if you do.

What Reconova is

Reconova lives at motifuse.com/reconova and has two processing modes that share one interface.

The browser workspace is free and needs no account. You open the page, add a CSV, TSV, or JSON file, and everything — parsing, profiling, transformation, validation, deduplication on the loaded data, export — happens locally on your device. Files up to 25 MB and around 100,000 rows are supported, subject to your browser's memory. If you refresh mid-session, Reconova restores your pipeline and view; you re-open the file and continue where you left off.

The cloud workspace is a subscription tier for work that outgrows a browser tab. Signed-in subscribers upload larger files — CSV, TSV, JSON, NDJSON, XLSX, and Parquet — to private, encrypted object storage and run profiling, cleaning, and full-scale deduplication as tracked server-side jobs. Cloud processing is explicit: nothing is uploaded unless you choose Secure upload while signed in. Current file-size and job allowances per plan are listed on the pricing page.

Within Motifuse, Reconova sits in the premium products line: the free tier is a real, complete tool rather than a demo, and the subscription adds scale, cloud execution, automation, and workflow continuity rather than unlocking basics.

Why a data-quality studio, and not another converter

There is no shortage of one-shot utilities that convert a CSV to JSON or split a file in half. Motifuse publishes several itself, including a CSV to JSON converter, and for a quick format change they are exactly the right tool.

The gap appears when the job is not conversion but judgment. A file with 40,000 customer rows does not need reformatting; it needs someone to discover that 6 percent of email addresses are blank, that two exports were pasted together with different date conventions, and that "M. Rao", "Maya Rao", and "MAYA RAO " are probably the same person. Spreadsheet software can find some of this, but only through manual filtering that leaves no repeatable record — and it famously mangles identifier-like values along the way.

At the other end of the spectrum, enterprise data-preparation platforms handle all of this and far more — at a price, a learning curve, and a deployment footprint that make no sense for a weekly CRM import or a one-time migration.

Reconova is aimed at the middle: serious enough to profile, validate, and deduplicate data properly, light enough that the free tier runs entirely in a browser tab.

Core capabilities

Profiling and the explainable quality score

When a dataset loads, Reconova profiles every column: detected type, missing values, distinct values, common values, pattern signatures, and length distribution. The headline result is a data-quality score built from five separate factors — completeness, validity, consistency, uniqueness, and conformity.

The important design decision is that the score is explainable. Each factor is shown on its own, so a low score points at a cause: completeness drops when fields are empty, uniqueness drops when duplicate rows appear, conformity drops when values stray from the column's dominant pattern. You are never left with an unexplained "73" and no idea what to fix first. Cloud jobs produce the same profile for large files, rendered in the workspace with accessible charts.

A real data grid, not a read-only preview

The workspace grid behaves like a data tool: sort any column, resize and reorder columns, freeze the first columns while scrolling wide files, and edit a cell inline with a double-click. Edits are recorded as pipeline steps rather than silent mutations, so they preview, undo, and export exactly like any other change. A right-click menu offers quick actions — trim a column, filter to a value, convert a type — and the grid stays virtualized and responsive at a hundred thousand rows. A before-and-after toggle shows the source data against the latest dry run.

There is also a command palette (Ctrl-K or ⌘K) that searches datasets, columns, cleaning operations, and saved recipes from anywhere in the studio, and a read-only mobile review mode that presents each column's findings as cards on a phone.

Twenty-seven cleaning operations

Cleaning happens through an ordered pipeline of steps rather than in-place edits. The engine currently ships 27 operations across four families:

• Text and structure — trim whitespace, collapse repeated spaces, Unicode normalization, case conversion (lower, upper, title), strip invisible characters, strip HTML, strip emoji, rename and remove columns, filter rows, and remove exact duplicate rows.
• Find and replace — exact value replacement, dataset-wide find-and-replace, and regular-expression replacement with guarded patterns.
• Reshaping — split a column by delimiter or pattern, merge columns with a separator, extract the first pattern match into a new column, and derive a new column from a safe expression (no arbitrary code execution).
• Semantic normalizers — parse numbers out of currency-formatted text ("₹ 12,400" becomes 12400), normalize dates to ISO with ambiguous values flagged rather than guessed, normalize phone numbers to E.164 with a configurable default region, canonicalize emails, and clean up URLs.

Every step reports how many values it affected and how many it could not process, and the whole pipeline supports a non-destructive dry run with undo and redo. The original file is never modified.

Identifier safety by default

Reconova treats values such as "001234", long account numbers, and reference codes as identifiers by default, which means they pass through untouched. Converting a column's type is a deliberate act: you add an explicit conversion step, preview it, and choose what should happen to values that fail to convert. That single default prevents the most common and least visible spreadsheet injury — numeric coercion destroying leading zeros and long references.

Validation rules and reports

Beyond implicit profiling warnings, you can attach explicit validation rules to a dataset: required values, email, phone, URL, IP address, UUID, JSON, custom regex, numeric ranges, allowed-value lists, cross-column comparisons (an end date must not precede its start date), and uniqueness. Failing cells are highlighted in the grid, a downloadable validation report summarizes the failures, and rules are saved inside recipes — so the same checks run every time the recipe runs, in the browser or in the cloud.

Facets and a filter builder

Clicking a column opens its value distribution as facets with include and exclude controls, and a compound filter builder combines conditions with AND/OR logic. Filters drive the grid live, and when a filtered view is what you actually want to keep, one action commits the filter into the pipeline as reproducible steps.

Saved recipes

A recipe is a named, reusable cleaning pipeline — steps plus validation rules — stored in your account and synced across devices. Apply a recipe to a new dataset and Reconova checks column compatibility first, warning about missing columns instead of failing midway. Recipes export and import as portable JSON, and the same recipe that cleans a small file in your browser can run server-side against a large upload. Saved-recipe allowances vary by plan; every tier includes a working number of slots, listed on the pricing page.

Fuzzy deduplication and entity resolution

Exact duplicate removal is one step. The harder, more valuable problem is near-duplicates: the same customer entered three ways, the same vendor with and without punctuation. Reconova ships a full matching engine for this.

You choose which columns matter and how each should be compared — exact, edit-distance similarity (Levenshtein), Jaro-Winkler for name-like fields, token overlap for reordered words, Soundex for phonetic matches, or numeric tolerance for amounts. Blocking keeps the comparison fast by only scoring plausible candidate pairs, and matched records are grouped into duplicate clusters.

Every cluster is reviewable: each pair shows why it matched, score by score, rather than a bare confidence badge. You decide what to keep, and survivorship rules — keep the most common value, the longest, or the first — assemble a single golden record per cluster. Decisions are recorded as pipeline steps, so the result is reproducible.

In the browser this runs on the loaded dataset. Subscribers can run the same engine as a cloud job across a full upload, with a downloadable cluster report and the deduplicated dataset as outputs.

Dataset comparison and schema drift

Cloud datasets are snapshotted immutably, and any two snapshots can be diffed: rows added, removed, and changed (matched by a key column), columns added or removed, type changes, and the quality-score delta. It is the fastest way to answer "what actually changed between last month's file and this one" — and to catch schema drift before an import script does.

Data-quality certificates

When a cleaned dataset needs to be trusted by someone who was not in the room, a subscriber can issue a data-quality certificate against a cloud snapshot. The certificate records the dataset's quality score, row and column counts, and content hash, cryptographically signed. It comes with a public verification page — anyone with the link can confirm the certificate is genuine and that the dataset still matches its recorded hash, without any access to your workspace — and a PDF for attaching to a report. Editing the underlying data invalidates the certificate, which is precisely what makes it worth showing to an auditor.

Cloud cleaning jobs also emit a reproducibility manifest — input hash, recipe, engine version, output hash — so a result can be independently re-derived.

AI assistance, with a strict privacy boundary

Subscribers on eligible plans get optional AI assistance: ranked cleanup suggestions from the profile, plain-language explanations of the quality score, semantic column detection, and natural-language transforms ("split full_name into first and last") that come back as ordinary, validated pipeline steps you can inspect before applying.

The privacy boundary is structural, not a promise: only masked samples and aggregate profile statistics ever leave the server, never raw cell values, and a model's output cannot execute anything — it must parse into the same step registry every manual edit uses, or it is rejected.

Automation: API, webhooks, and schedules

Recurring work should not require a browser. Subscribers can create scoped API keys and drive Reconova from their own systems through a REST API — upload a file, run a recipe against it, poll the job, download the result. Webhooks post a signed payload to your endpoint when a job completes or fails, delivered from a durable outbox with retries. And schedules re-run a saved recipe against a project's latest upload daily, weekly, or monthly, enqueuing exactly one job per period.

Workspace collaboration and accountability

Workspaces support team members with role-based access — a view-only member can review data and comment without being able to change it. Comment threads attach to datasets, recipes, jobs, and certificates. An append-only activity log shows everything that happened in the workspace — uploads, jobs, recipe changes, key creation, deletions — and cannot be edited by anyone, including the workspace owner. Workspace settings hold shared defaults: export format, date format, locale, and the default phone region.

For teams with compliance obligations, a downloadable compliance pack assembles the evidence that already exists in the workspace: which columns held personal data per dataset, confirmation that samples were masked, each asset's retention deadline, and deletion receipts — generated fresh from current records, as JSON or CSV.

Export formats

Browser exports are generated on your device: CSV, JSON, Excel (.xlsx), SQL (CREATE TABLE plus typed INSERTs), XML, and TSV, plus the recipe itself as portable JSON and a reproducible bundle that pairs the cleaned data with the recipe that produced it. CSV and Excel exports escape formula-injection attempts by default.

Cloud jobs produce downloadable outputs on top of that: the cleaned dataset as CSV and canonical Parquet, run reports with per-step counts, validation reports, rejected-row reports, and duplicate-cluster reports. Downloads use short-lived signed links.

How files and data are handled

The privacy model differs by mode, and the difference is the point.

In browser mode, the file is read by your own browser and never transmitted. Local processing is not a magic privacy shield — the data is still on your device and visible to your browser, extensions, and operating system — but no copy exists on any Motifuse server. Session restore keeps only the pipeline and dataset descriptors, never your rows.

In cloud mode, uploads go directly to private object storage encrypted at rest with AES-256, under a randomly generated object key. Access is controlled per request, download links expire in minutes, and stored files carry fixed retention windows enforced by scheduled cleanup:

You can delete any asset yourself before its window ends; deletions are recorded and produce receipts that appear in the compliance pack. Detected personal data is masked in samples and reports, and server logs are content-free — they record that work happened, not what your data said. The Privacy Policy describes platform-wide practices, and the security page covers reporting and infrastructure posture.

Free and paid, honestly

The browser workspace is the free tier, and it is not a teaser: profiling, all 27 cleaning operations, validation rules, facets, fuzzy deduplication on the loaded dataset, recipes, and every local export format work without an account. For a great many files, it is all you need.

A subscription buys scale and workflow: larger files in more formats, server-side cleaning and full-dataset deduplication, dataset comparison, certificates, AI assistance, the API, webhooks, schedules, and team features. Plans differ mainly in capacity — cloud jobs and data volume per billing period, active projects, and recipe slots. Those numbers are managed centrally and shown live on the pricing page, so this article deliberately does not restate them.

There is currently no free trial of the cloud tier — the browser workspace itself is the honest trial, since it runs the same pipeline engine on your own machine. Subscriptions are billed through Razorpay, renew until cancelled, and cancellation takes effect at the end of the paid period.

Three practical scenarios

Cleaning a customer list before a CRM import

An operations coordinator receives a 30,000-row customer export assembled from two regional systems. Profiling shows the problem immediately: 7 percent of emails missing, a status column with "Active", "active", and "ACTIVE" as separate values, and hundreds of near-duplicate contacts. The pipeline is five steps — trim whitespace, title-case names, canonicalize the status values, normalize phone numbers to E.164, then run fuzzy dedupe on name plus email and review the clusters. Survivorship keeps the most complete record from each cluster. The CRM import goes through without creating duplicate accounts, and the saved recipe handles next quarter's export — or runs on a schedule without anyone opening a browser.

Normalizing vendor data for a finance close

A finance analyst preparing accounts-payable data needs vendor names consistent and invoice references intact. Case conversion and whitespace cleanup standardize the vendor column; a validation rule set — required references, amounts within range, dates in order — turns tribal knowledge into checks that run every time. Invoice references with leading zeros pass through untouched because Reconova treats them as identifiers unless told otherwise. The exported Excel file loads into the reporting tool with references intact, and the validation report documents what was checked.

Validating a migration export

A developer migrating a legacy system profiles the export before writing a single import script. The profile reveals a date column with two formats and a supposedly unique key column with 41 duplicate values — both cheaper to discover now than after the import. The full 80 MB export runs in the cloud rather than a sample, and when the source system produces a second export a week later, a snapshot diff shows exactly which rows changed and that a column quietly switched type. The fixes land before the import script ever runs.

Where quick free tools are still the right choice

Reconova does not replace single-purpose utilities, including Motifuse's own free tools. If you need to convert one small file, reformat a JSON snippet, or check a CSV's structure once, a focused tool with zero setup is faster — that is what pages like the CSV to JSON converter are for.

Reconova earns its place when the questions get bigger than the format: how healthy is this data, what exactly is wrong with it, are these two records the same customer, can I prove what I changed, and can I do the same thing again next month without doing it by hand.

Limitations to know before you rely on it

• The browser workspace tops out at 25 MB and roughly 100,000 rows, and wide files can hit browser memory before those limits.
• Cloud uploads accept CSV, TSV, JSON, NDJSON, XLSX, and Parquet; password-protected or corrupted files are rejected. Per-file size and per-period volume depend on your plan.
• Matching two different files against each other row-by-row (classic two-source reconciliation) is not a feature today. What exists is fuzzy deduplication within a dataset and snapshot-to-snapshot comparison of cloud datasets.
• Cloud outputs expire on the schedule above; Reconova is a processing workspace, not a storage product.
• Download links are short-lived by design; a link that expires is refreshed from the project page, not a data loss.
• An internet connection is required for the cloud tier; browser mode works offline once the page is loaded.
• Like any automated tool, it applies the rules you configure. Fuzzy matching in particular deserves review — that is why clusters show their reasons and nothing merges without your decision.

Getting started

The sensible path is to start free. Open Reconova, drop in a file you know is imperfect, and read the profile — the quality factors usually surface something within seconds. Build a small pipeline, dry-run it, save it as a recipe, and export.

If your files are larger than the browser handles, arrive as XLSX or Parquet, or belong to recurring work you want scheduled and auditable, compare the plans on the pricing page and sign in to the customer workspace. And if your team's need is a workflow specific to your business rules rather than a general tool, Motifuse also builds private custom tools delivered in the same workspace.

Frequently asked questions

The bottom line

Reconova has grown from a careful browser cleanup tool into a full data-quality workspace: explainable profiling, a 27-operation cleaning engine, explicit validation, fuzzy deduplication with golden records, reproducible recipes, cloud execution with certificates and an API — while keeping the two properties it started with. The free tier is genuinely useful on its own, and the privacy model is concrete rather than promotional: local by default, explicit, masked, and time-limited when the cloud is involved.

If messy exports are a recurring part of your work, the fastest way to evaluate it is the product itself: open the browser workspace with a real file and see what the profile finds.