A queue of live adjudications. A trace walk showing how each one was reached. A precedent panel surfacing every comparable claim across the book's history. A knowledge-gaps view showing where claims memory is thin. Four surfaces, embedded in the systems your teams already use.
Live and historical adjudications across the book, in one log. Filter by line of business, team, wording section, version, or examiner. Open a row to walk its full trace.
The queue is the substrate’s read surface, the same data the API returns, rendered for humans. Examiners use it to spot comparable claims; auditors use it to walk back through valid time; compliance uses it to verify a coverage call was consistent with how the book handled the same facts before.
Click any row in the queue, and the trace walk renders the full causal chain: the claim, the wording bound at the loss date, the guideline evaluated, comparable precedent cited, where the call diverges, approvals received, settlement.
Every step is a sealed event with a clickable identifier. The trace is not a summary or a regenerated narrative; it is the underlying log, rendered. What you read is what is in the substrate, and what a regulator or ombudsman would read too.
Guideline WORDING §quantum-bands places comparable injuries in band B3. The proposed settlement reads B4. Flagged at 09:14:02 GMT.
For every live claim, the precedent panel surfaces the most relevant prior adjudications, ranked by structural similarity across the typed graph, not by text similarity.
Voyage embeddings give the first pass; the typed-edge graph re-ranks on relationship match: same coverage trigger, same wording version, same injury class. The panel returns sealed claims with their outcomes attached, so the examiner sees not just “we've handled this before” but “here's how it settled, and whether it held up”.
The substrate is honest about where it has thin precedent. The gaps view surfaces claim classes where the book has fewer than three sealed comparables, or where the comparables it does have predate the current wording.
The point of this surface is to flag, not to hide. An examiner adjudicating in a thin-precedent class knows it. A new edge type might be needed. A wording amendment might be queued. It is also where emerging exposure shows up first.
The exclusion was amended in 2026-Q1. Claims adjudicated before that point are not comparables under the current wording. New claims in this class adjudicate alone.
Three prior claims exist but predate the rehabilitation clause added in a later revision. The examiner is shown the prior outcomes flagged as wording-mismatched.
Two sealed claims in this class. The substrate flags this to the examiner so the call isn’t silently grounded in too-thin precedent.
Five prior claims exist; their settlements diverge sharply (two low-banded, three high). The panel surfaces the conflict rather than averaging it, an early signal of severity drift.
Eight event types. Twenty-seven typed edges. A bitemporal log. An append-only ledger. Two endpoints. We will walk you through any of it.