Product death is quieter than you think — and signal scoring catches it early

Nobody writes a press release when a product dies. Founders quietly update their bios. The app keeps its store listing for years after the last release, and most of the time nobody even agrees on the exact date it happened. But across 79,573 tracked products in 18 categories, death still leaves a shape, and it starts showing up weeks before anyone, founder included, admits the thing is over.

What goes quiet first

Signal scoring doesn't wait for an announcement. It watches four things that move independently of any founder's instincts for good PR: review velocity, press mentions, update cadence, and the liveness probe that checks whether a product is still reachable, on its own site and its own API. Across the 353 confirmed ALIVE-to-dead transitions logged in the corpus, those four don't fail all at once. They fail in a fairly consistent order, consistent enough that it's worth charting instead of just describing it.

Here's that order, compressed into roughly the ten weeks that lead up to a recorded transition:

The shape of the decline

Put the median dying product's signal score next to a healthy baseline from the same category and the divergence isn't subtle once anyone bothers to look for it. The chart below is illustrative: every product's curve looks a little different, and some die faster than others. But the shape itself, a decline that starts well before the last public commit and steepens as review velocity and press mentions both flatten, holds across the 353 transitions logged so far.

It rhymes with what CB Insights found analyzing 431 VC-backed companies that shut down since 2023. "Ran out of capital" shows up in about 70% of the post-mortems, but it's the last domino, not the first one. Poor product-market fit, the thing that actually starves a company of cash, appears in 43% of the same post-mortems, usually months earlier. Our corpus doesn't track anyone's bank balance, but it tracks the public exhaust of that same erosion: a number worth watching instead of a funding round worth waiting on.

353 down, 812 back

Death isn't the only surprising transition in the corpus. In the same window, 812 products that had been flagged dead came back. Sometimes it's a new commit after a long silence, or a relaunched pricing page. Sometimes it's a founder who never actually closed the doors and just went quiet, off raising money or wrestling with something that had nothing to do with the product itself. That number is more than double the 353 that actually died, and it matters just as much. It's the reason Launch Sentinel doesn't call a product dead off one slow month.

A signal score has to sit under the dead threshold long enough to rule out a quiet quarter or a founder on leave. An app stuck between store review cycles can look the same way for a while without being anywhere close to dead. Call it dead too early and the corpus buries a product that's about to ship its best release. Call it too late and the entire warning window this piece is about quietly closes.

What the corpus doesn't catch

This pattern works best on products that leave a public trail: app-store reviews, press mentions, a changelog, a marketing site with something to ping. A lot of dying products fit that description; plenty don't. A single-SKU hardware product sold only through one retailer won't show up the same way. Neither will a tool built and used entirely inside one company, nor an API with no public status page. The corpus says so honestly instead of guessing. Sudden death still happens too. An acquisition can shut a product down overnight. So can a regulatory order, or a founder who simply walks away. None of that decays, it just stops, and no signal score sees it coming.

Reading the signal before it's a headline

Good judgment still matters here. What the signal removes is the wait for someone else's judgment to turn into a headline. An investor watching a portfolio company's score erode across two quarters has a very different conversation with the founder than one who only finds out from a shutdown post weeks later. A competitor gets something similar when a rival's press mentions go quiet: a window nobody else in the market has yet, useful whether that means moving on a customer list or just not panicking about a launch that was never going to land. And it holds for anyone deciding whether to keep building on an API that quietly stopped answering its own liveness probe three weeks before the vendor's status page admitted anything.

The 353 flips already happened; they're logged. The next one is already writing itself into somebody's signal score right now, and the only real edge is being the one watching before the obituary runs.