Blog
Notes from building it
Mostly about the parts that turned out to be harder than they looked: scheduling that cannot double-send, history that cannot be rewritten, and one schema that has to serve ten industries without forking.
LatestAlerting3 min read
Scheduling notifications exactly once, and proving it
A reminder that fires twice trains people to mute the channel. Here is the ledger that makes duplicate alerts structurally impossible rather than unlikely.
Read the post
Everything else
Architecture
Why the event log has no update or delete grant, for any roleAn audit trail that the application can rewrite is not an audit trail. The guarantee has to come from privileges the running service does not hold.Security
Testing tenant isolation on every endpoint, on every pull requestMulti-tenant leaks are not caused by hard problems. They are caused by one handler that forgot a where clause — so the test has to be mechanical.Product
Building one schema for ten regulated industriesA dental lab and an aerospace machine shop track the same four nouns. Everything that differs between them belongs in configuration, not in branching code.Operations
Why the spreadsheet works right up until it does notThe spreadsheet is not failing because it is unsophisticated. It is failing because it smears four different things into one tab and has no idea what it does not know.Compliance
What an auditor actually asks forThe gap between showing that a reminder was sent and showing that someone acted on it is where most compliance systems fall down.
No schedule and no newsletter. We publish when something is worth writing down, which in practice means when we got something wrong and had to work out why.
Most of these posts started as a conversation.
If you are wrestling with the same problems, the questions you are about to ask are probably the ones we have been writing about. Bring them.