This release has a mix of visible changes (mostly suggested by the community) and a few invisible ones that should make your day to day experience a bit more reliable.
The biggest visible update is on the Discovery page. There are also some improvements around Hive API node handling, AutoVote setup, and the mobile menu.
Discovery now has multiple ways to find posts
The Discovery feed is no longer just one algorithm. Based on some feedback from my previous post you can now switch between three tabs:
- Discovery: the existing social discovery feed, now tuned around fresher posts from the last few days.
- Most Reblogged: a new feed that looks at posts being reblogged recently by stronger accounts.
- Mixed: a first version of a blended feed that combines discovery, following, subscribed communities, ranking feeds, and snaps or waves.
Each algorithm also has its own explanation in the info modal, so you can see what kind of signal is being used instead of having to guess.
For the Most Reblogged feed, post previews also show the reblog count used by that feed. That should make the result easier to understand: if a post appears there, you can immediately see that the signal is coming from recent reblogs, not just votes or trending score.
This is still an area I expect to keep tuning, but the goal is simple: give you more than one path to find interesting Hive content without forcing everything through the same ranking logic. Also based on some direct feedback from users we may introduce a bit of randomness in the results.
Hive API calls should recover better when nodes misbehave
This one is less visible, but important.
PeakD now retries several Hive API calls and automatically rotates to another available node when a call fails or returns an invalid result. This is a step up from the "select the best available node and stick with it" we used to rely on. The new logic is only enabled if you have selected the option to automatically switch Hive API node in your settings (if you want to use a specific node for whatever reason the website will honor your preference).
The practical result should be fewer random broken loads when one RPC node has a bad moment. It does not make the network magically perfect, but it gives PeakD a better chance to recover without making you refresh manually.
AutoVote is easier to set up from a profile
If you use PeakD AutoVote, there is now a shortcut directly on user profiles.
When you are viewing another account, you can open the profile actions menu and add that account to AutoVote from there. If you already have a rule for that account, PeakD opens it for editing instead.
The AutoVote rule form was also pulled into a shared modal, so the same add/edit experience is used from both the AutoVote settings page and the profile shortcut. It still supports vote percentage, daily and weekly limits, and trail percentage behavior.
This is a small workflow change, but it saves the back-and-forth of copying a username, going to settings, opening the AutoVote panel, and pasting it there.
The writing menu works better on mobile
The user menu also got some cleanup around publishing shortcuts.
There is now a direct Write a Snap option, the menu labels were tightened up, and the icon sizing is more consistent. Drafts and Templates also have mobile navigation between them, so moving between those two dashboard pages is less awkward on smaller screens.
Nothing huge here, but these are the kind of small changes that matter when you use the writing tools often.
Other useful fixes
A few more changes included in this release:
- the wallet HIVE price display now shows one more decimal for USD values
- the witnesses table now shows the major chain version more cleanly in the header
- the light/dark mode switcher uses icon fonts instead of unicode characters that were not always rendered correctly
- snap previews no longer show the low-reputation indicator for the author inside the editor preview
As usual, let me know if the new Discovery tabs feel useful or if one of the feeds gives you results that look off. Discovery ranking is the kind of feature that gets better with real feedback, not just code changes.
