Confidence You Can See: Newrali's New Scoring Engine Grades Every Piece of Evidence in Real Time

Every investigation eventually runs into the same question: how sure are we, actually? Analysts have been answering that with letter grades and gut feel for decades — useful, but static. Grade a source once, and that grade just sits there, even after three more pieces of corroborating (or contradicting) evidence show up.

This release changes that. Newrali’s evidence weights are now real probabilistic math: live confidence scores that update the moment new evidence lands, visible everywhere from the investigation graph to the tables you already work in. It’s the biggest change in this cycle, and it touches almost everything else we shipped.

A scoring engine that shows its work

Every node in your investigation — every source, every piece of evidence — now carries a confidence gauge computed on the fly instead of baked in once and left to go stale. Retune a weight, and every badge across the workspace updates instantly, rather than quietly drifting out of date.

You can see the math behind it, too. A new Scoring card opens two panels: Completion, showing per-field relative weights and contribution bars with one-click even balancing, and Confidence, plotting every evidential field on a 0–1 scale. If your team grades sources using the Admiralty system — the reliability-and-credibility scale widely used across intelligence and OSINT work — it ships with the standard’s own weights out of the box, and any workspace can override per grade and reset back to default whenever they want.

Even better, the model tunes itself against how your team actually works. Each option and grade shows how often it’s genuinely chosen, with a footer accounting for values still pointing at choices you’ve since retired — so adjusting the model is informed, not a guess.

And the scores aren’t siloed to the graph anymore: confidence and completion now show up in your tables as sortable hairline meters, averaged into group headers, with ring gauges on section heads — the same numbers as the badge on the canvas, by construction. One score, everywhere you look.

The investigation reads itself back to you

A live score is useful. Knowing why it moved is more useful. Newrali can now replay your event log through the scoring model to chart how completion and confidence have shifted over time, alongside four built-in checks that flag exactly what’s dragging the numbers down. The underlying material gets organized like an analyst would actually argue it — events as the headline, sources folded in beneath — instead of a flat dump of everything in front of you.

That same context now heads the sources panel and scrolls with it, and every check in the legend doubles as a filter, so getting from “what’s dragging this down” to the actual rows responsible is one click. Grade evolution also got its own dedicated page, leading with the answer instead of asking you to parse a dense matrix crammed into a grid cell.

Faster to move through, easier to trust

A lot of this cycle went into making the day-to-day faster without making anything less reliable.

Tables got real filtering (an add-filter cascade with live counts and pinned selections), grouping by up to two dimensions at once, shift-click multi-select with a docked selection bar, and rows that can refuse selection with a visible reason instead of failing silently. The whole app now shares one keyboard vocabulary — arrow keys or J/K to move, X to check, Esc to clear and close, number keys to switch tabs — documented on its own help page. And every sources table now pulls from a single shared definition, so no two lists in the product can quietly disagree about what “linked” means.

The sources panel itself is now a full docked sheet that opens instantly — it loads the source first and streams everything else in behind it, with rows preloading on hover and the route caching so paging back through your history reopens from memory. Events are first-class now too: a source can report on an event, and grouping by event pulls in its entire chain. And citations are a real, always-current table linking every source to every report that cites it — add sources into a report straight from that table, or type / in the editor to insert them at your cursor.

Also worth knowing

• Properties, tags, and templates are finally one library. Two systems that never talked to each other are now unified, field visibility can be scoped to a person, team, or role, and templates read like real documents instead of a wall of unlabeled checkboxes.
• Safer deletion, built for the audit trail. Sources, investigations, and reports now soft-delete instead of disappearing outright — the record survives, so the audit trail and event log still resolve; deleting an investigation cleanly cascades to everything under it; and the delete itself is gated behind an explicit permission that no default role carries. Nothing is erased without a record.
• Billing is a real page now — live addresses, cards, and invoices, per-seat pricing on add-on modules, and trials that finally behave the way trials are supposed to.
• A long list of smaller fixes, including a real-time editing bug that could silently drop content, tables that finally render correctly on public share pages, and a smoother four-step invite flow.

Try the Scoring card on your next investigation

The theme this cycle was simple: confidence shouldn’t be something you write down once and hope stays true. It should move with your evidence, show its reasoning, and look the same no matter where in the product you’re looking at it. That’s what the new scoring engine does — and everything else we shipped this cycle, from faster tables to safer deletion, was built to support it.

If you’re new here, the previous release notes cover the rebuild this cycle stands on, and Special Operations is where the same tables and canvas run at event scale.