Release notes · 12 August 2026
Findings you can defend.
DataDigger 4.0 is a trust release. Reports now surface three to four times more findings per run, and every one of them is verifiable: each number is checked against the data that produced it, each finding carries inspectable evidence, and the wording is calibrated to how big a difference actually is. The engine now catches the classic data traps — mixed currencies, cancelled orders inflating revenue, misleading conversion rates — and flags them instead of reporting them as fact. And your data is more private than ever: personal identifiers never reach the AI at all.
New features
3 items
Evidence behind every finding
Every finding now comes with its evidence: the exact result rows it was computed from, a plain-language explanation of how it was calculated, and its verification status. Evidence is available on historic runs, can be deleted per run at any time, and is covered by account deletion.
Every number verified against your data
Before a report ships, every figure in every finding is checked against the analysis rows that produced it. A finding whose numbers can't be reproduced from its own data is corrected or removed — it never reaches your report.
Personal data never reaches the AI
Emails, names, and phone numbers are now detected automatically and masked before any AI model is involved — the AI reasons about anonymous placeholders, and the real values are restored only in your report. Combined with EU-based processing, your customers' identities stay entirely within the data pipeline.
Improvements
5 items
Three to four times more findings per run
The reporting engine was rebuilt to give every analysis a fair chance at becoming a finding, instead of summarizing a handful and discarding the rest. Restated duplicates are collapsed automatically, so more findings never means more filler.
Language that matches the math
DataDigger now measures how big every difference actually is before describing it. A 1-point gap between segments is reported as parity — useful information in itself — while a genuine 3× difference leads the report. No more drama over noise.
Guards against the classic data traps
The engine now detects when money columns mix currencies (and refuses to sum them blindly), states which order statuses a revenue figure includes, questions implausible conversion rates computed across data sources, and adds context when a 'top segment' is simply the biggest group rather than the best performer.
Aggregate findings are checked for hidden reversals
Key comparisons are cross-checked against their segment-level breakdowns. When an overall ranking flips inside individual groups (Simpson's paradox), the finding says so explicitly.
Accurate time-period handling
Date ranges are now profiled exactly across the full dataset, so trends no longer show false 'collapses' from partially covered months or years — incomplete periods are labelled as such.
Bug fixes
2 items
Refreshing the page no longer loses a running discovery
A discovery run started from the product page now survives a page refresh and resumes automatically, instead of disappearing while the analysis kept running in the background.
Scheduled runs always have a destination
A schedule could previously be created without a save folder, leaving its results inaccessible. A destination folder is now required, and the creation flow was cleaned up along the way.
The privacy-first data discovery platform.
© 2026 Imput. All rights reserved.