#data infrastructure optimization

[millenniallust4death]

If a reblog doesn't add anything, the love flows up to the last person in the chain who did.

This line sent me. It's not a community feature.

From a data science perspective, “love flows up” is a heuristic for node collapsing to manage graph sparsity. Instead of tracking large numbers of “empty” reblogs that add zero feature variance, Tumblr attributes engagement to the last reblog with actual content (text or media). This effectively prunes low-signal nodes and focuses on the parts of the reblog chain that drive interaction. This creates cleaner engagement signals for feed ranking, targeted recommendations, and ads.

[millenniallust4death]

Here's a summary of my shit posts about the Tumblr update from last night.

From Tumblr’s perspective:

a silent reblog = one step, no added value

a high-branch reblog = a new hub of activity

the OP big number note total acts as a stopping cue (why explore when one number signals a post's popularity?)

Tumblr is moving from aggregate cascade measurement to node-level engagement measurement. My guess? This change is likely driven by the need for clean, directly queryable activity per node. Hiding the single note total on the OP also (potentially) shifts what people pay attention to and changes user behaviour to make high-branch nodes more visible and valuable. “Love flows up” fits this model too. It avoids wasting engagement on empty reblogs and concentrates it on the last post that actually added "content". This would likely lead to more reblogs, longer sessions as people click through the chain, higher visibility for contributions that add new layers of content, and clearer signals about what people interact with.

I don’t actually know what Tumblr’s internal user data looks like, so this is just my cynical take. But reconstructing the reblog network data for Martin and Bosco’s post from the outside via the API took a significant amount of effort when I ran the cascade metrics. Blaze is also one signal of the original system’s limitation. Unlike other platforms, you currently can’t target a direct audience for promotion. Node-level engagement data could support this by enabling finer-grained user group clustering.

As a Tumblr ancient, I’d be THRILLED to be proven wrong. I don’t want my community fractured in the pursuit of monetization.

[millenniallust4death]

The wording in the "Where's that comment?" post is very interesting. Tumblr presents the new feature as a way to find comments from yourself and the people you follow, as if this is a user benefit I have secretly longed for.

A reblog is an expensive form of engagement, but commenting is relatively cheap. From Tumblr's perspective, comments likely represent a large pool of engagement signals that currently do little to help content move through the network. By incorporating comments into content discovery, Tumblr can increase circulation while extracting more value from that data for feed ranking, recommendations, and targeted ads.