An archive of software postmortems, preserved exactly as published.
Postmortems are the most honest writing our industry produces — engineers explaining, in detail, how their systems actually failed. And they rot. Blogs get redesigned, companies get acquired, links die. The institutional memory of how software breaks is scattered across dead URLs.
Preserve them whole. Not summarized, not bulleted — the original narrative as the author wrote it, because the reasoning is the value. Vector search over the full text so you can find incidents by failure mode rather than by remembering which company had the outage.
Keeping full original text instead of summaries makes the corpus large and makes search harder to tune — a summary index would have been smaller, faster, and cheaper to run. It would also have thrown away the thing worth archiving. Cost and complexity were the right thing to spend here.
A searchable archive that grows without manual curation — GitHub Actions handle ingestion and indexing on a schedule, so adding sources is a config change rather than an afternoon.
05 Built with
06 Screens
.png)
.png)
.png)
.png)
.png)