Challenge · Glass Box
This Glass Box view pulls in the challenge logs and explains, in plain language, what the system did, why it stepped in, and which cases may need a human to look closer.
What this page helps you see
Each case shows what the system did and when it happened, so nobody has to read raw logs first.
We surface the rule, reason, and suggested correction that led to the system response.
This helps judges, reviewers, and teammates quickly spot risky cases and decide whether a human should step in.
Important note
We are using challenge-supplied sample logs to demonstrate how this review layer would work in a real production system. In Holmes, a surface like this could help teams catch AI hallucinations, spot when data refresh jobs fail or run with stale inputs, trace why a model response was blocked or corrected, and flag cases that need a human to step in before decisions reach people in the real world.
That matters because this platform touches sensitive civic information that could inform housing decisions, connectivity planning, and public-facing explanations. If a system like Holmes is going to be trusted by industry teams, city partners, or operations reviewers, it needs a clear audit trail that explains what happened, why it happened, and when someone should investigate further.
Event Playback
Click any recent event to have Holmes explain what happened in simple terms.
Loading the latest challenge log traces.
Loading the latest challenge log traces.
Loading the latest challenge log traces.
Loading the latest challenge log traces.
Loading the latest challenge log traces.
Audit Breakdown
Click an action or trigger below and Holmes will explain what it usually means.
Holmes Explainer
Select a recent timeline event to inspect it.
Click a recent timeline event and Holmes will explain what happened, why the system stepped in, and what a reviewer should take away.
Holmes Breakdown
Select an action or policy trigger above to inspect it.
Click any action bar or policy trigger card above and Holmes will explain what it means in simple terms.
Read-only CSV and JSONL inhibitor logs. We do not modify the provided sample_logs baselines; we ingest and render them.
The dashboard parses event timestamps, actions, severities, and policy triggers into summaries, timelines, and reviewer-friendly inspection views.
A compliance or demo reviewer can quickly answer what happened, why the intervention occurred, and what should be checked next without reading raw logs.