This pass was about making Hugin feel less like a prototype and more like a public evidence tool people can actually trust. The scanner was already doing a lot of useful work; the product needed to explain that work more clearly and avoid offering features that looked automatic or expensive without the right controls behind them.
Cleaner public pages
The case-file and journal surfaces were tightened down. The case index now reads more like a compact dossier list, with live source counts and evidence mix signals visible without forcing the reader through a long page. The journal is now a simple publication feed instead of a half-dashboard with extra navigation cards competing with the entries.
Search also got a better empty state. If the archive has no match for a Reddit community or account, Hugin now gives the reader a useful next step: scan a post, open the browser companion, or request that target for review.
Watch became a request queue
The old watch framing made automatic monitoring sound too easy. That is risky: recurring scans cost money, and they can create noise if every public request immediately becomes a scheduled job.
The public watch path is now a request queue. A visitor can suggest a subreddit or account and describe the pattern they care about, but the target is saved inactive. It only joins internal recurring scans after manual approval, and the scanner still runs through the existing per-run caps.
More visible receipts
Reports now surface more of their reasoning earlier. Source coverage, "show your work" explanations, and limitation counts are easier to find, so an inconclusive or suspicious verdict does not feel like a black box.
That matters because Hugin should not just say "trust the label." It should show which public signals were available, which ones were missing, and why the scanner could or could not reach a stronger conclusion.
Cost controls are visible
The public limits endpoint now exposes the public action pause state, and the ops page shows the capped public routes, daily spend budget, and fail-closed policy in one place. That makes it easier to answer the practical question: "Can this feature quietly run up a bill?"
The answer should be no. Public requests stay inactive, public compute routes are budgeted, and the limiter fails closed when it cannot verify the cap.
The direction
The next version of Hugin should keep doing this: fewer fake-feeling buttons, more source-backed receipts, and clearer boundaries between public evidence, human review, and automated scanning. The strongest product surface is not the loudest verdict. It is the one where a reader can see exactly how the report got there.