Why 18 categories mask the real intelligence — and how to read across them
Launch Sentinel ships 18 category tags. The real signal shows up when a product's behavior stops matching the label it launched under.
Eighteen is a schema decision made two years ago. Somebody drew that many category boxes into Launch Sentinel's taxonomy, and every one of 79,573 tracked products got sorted into exactly one of them the week it launched. The box doesn't move again after that. The product usually does.
Filter the corpus by category and you get a clean, defensible slice: developer tools, consumer apps, fintech, whatever you pick. What you don't get is where that slice has actually drifted since launch day. A tool that shipped as a code-formatting utility and now sells an agent platform to enterprise IT still shows up under "Developer tools" on the category page. Its signal-score curve doesn't look like developer tools anymore. Neither does its press mix, or its buyer. The label froze. Everything underneath kept moving.
The clearest drift right now runs through developer tools
Pick any starting point in the corpus and this pattern shows up somewhere. Developer tools is just the loudest version of it at the moment. GitHub's 2025 Octoverse report found that six of the ten fastest-growing open source projects by contributors that year were AI infrastructure or tooling, not general-purpose developer utilities (GitHub, "What the fastest-growing tools reveal about how software is being built," 2025). The fastest-growing corner of what used to be a stable, boring category is now behaving like the category next to it.
Launch Sentinel doesn't relabel a product because its behavior started matching a different category better, and there's no reason it should. The category field records intent at launch. It was never built to track what a company does eighteen months in. The signal score, the press-velocity curve, and the liveness probes are the parts that actually track what's happening today, and none of them check the category field before they move.
Same-category correlation is falling
Run the correlation the other way and the drift turns into a number, even while the tag stays put. Products inside the same category are supposed to move together: same buyers, same competitive pressure. Their seasonal swings should line up too. That correlation has been loosening for the pair we watch closest, developer tools and AI/ML infrastructure. Two years ago the two barely tracked each other. Lately, a meaningful share of the movement in one shows up in the other a quarter later.
Illustrative pattern: built to show the shape of the crossover between developer-tools self-correlation and its correlation to AI/ML infra, rather than a literal per-product ledger. The two lines nearly touched in Q1 '26. By Q2 '26, cross-category correlation had overtaken the within-category baseline.
A category tag records where a product was born. It stops recording anything about where it's going.
How three category pairs actually behave
Zoom out from developer tools and AI/ML infra, and the same lean shows up across a handful of other boundaries at once. It's rarely a full category swap; hardware doesn't relabel itself as software. It's closer to a lean: a category starts producing products that score and sell like the category next door, long before anyone updates the tag.
Read the bottom row first. That's the actual finding: the tag is static by design. It's the fastest way to narrow 79,573 products down to something you can look at. It was never built to keep up with a product that's changed what it does since the week it launched.
Reading across categories without breaking the taxonomy
None of this argues for scrapping the 18 categories. A stable taxonomy is what makes the corpus comparable at all, and it's still the fastest filter available. The fix is reading past the label instead of stopping at it.
Weigh correlation to the neighbor alongside growth inside the box
A product's own category growth tells you if the box is hot. Its correlation to the category next door tells you whether it still belongs in that box, or has quietly become a member of a different one.
Watch the press mix as closely as the press volume
Coverage volume inside a category can hold steady while the outlets covering it change entirely. When trade press for a "developer tools" launch starts running in AI-beat coverage instead of dev-focused outlets, the tag has fallen behind the buyer even though the product hasn't.
Treat category as the opening filter, then dig past it
Use it to narrow 79,573 down to a workable set. From there, signal score and liveness carry most of the weight, with cross-category correlation filling in what they miss. The category name is where analysis starts.
The same quarter that produced 12 breakouts across the corpus also produced 353 deaths and 812 back-from-dead transitions. None of that movement reshuffled a single category tag. Death and revival happen at the product level, inside whatever box a product happened to land in at launch. The category tells you where to start looking. It has never told you what you'd find.
Read your market across every category at once
Launch Sentinel scores 79,573 products across 18 categories every week, and tracks the cross-category correlation a static tag can't show you.