16 September 2026

From past chats

Give the Temporal queue an explicit rollover contract

The rate-limited Signal queue becomes more convincing once Continue-As-New is part of its design rather than an emergency escape hatch. Define the state carried into the next run as the queued items, any deduplication identifiers, and the earliest permitted processing time. Let the main Workflow—not a Signal handler—decide to roll over after draining handler work and checking Temporal’s suggested-history threshold. Then write one deliberately tiny-threshold test that sends Signals before, during, and after the rollover and proves that order and throttling survive the new Run ID. The useful design question is whether the backlog is small enough to remain Workflow state, or whether its possible size calls for an external queue with Temporal coordinating consumption.

Turn the timing discussion into an executable message timeline

A compact Workflow test can settle more than prose about concurrent handlers. Schedule a burst of three Signals at virtual time zero, make the consumer wait five virtual seconds between items, and query its state after each boundary. Assert FIFO order, at most one processing Activity in flight, and a stable backlog after an injected Activity failure. Add a replay test over a captured Event History before each Workflow-code deployment; that catches determinism regressions that a mocked happy path cannot. Which guarantee actually matters to the caller—durable acceptance, bounded queue delay, or idempotent external effects? Naming it now will prevent “exactly once” from becoming an accidental and untestable promise.

From the homepage

Preserve the ABC theme recreation as a reconstruction notebook

The Logic Pro recreation could become a small piece of broadcast archaeology rather than only a finished bounce. Save a cue sheet with section timestamps, tempo and key estimates, the evidence behind each sound choice, and the places where the source is too compressed or masked to be certain. Export the stereo mix, individual stems, MIDI, and a plain-text plug-in inventory alongside the Logic project; that leaves enough material to revisit the arrangement after today’s instruments disappear. A worthwhile next step is to search the NFSA catalogue for the original education segment and record any title, date, composer, or production clues that either confirm or challenge the reconstruction.

Map the identity layer cake before choosing another identity product

The awkwardness around acquisitions, divisions, and reporting lines suggests starting with an identity topology rather than an authentication redesign. For one troublesome application, draw the authoritative people sources, identity providers, relying parties, group ownership, provisioning paths, and offboarding triggers; annotate every manual bridge and every place the same person can acquire a second account. This separates login, lifecycle management, and authorisation—three problems that are often bundled under “auth”—and should reveal one narrow seam to simplify first, perhaps replacing a manual joiner/leaver feed with SCIM. The follow-up question is especially revealing: after someone changes division or leaves, which system is trusted to make access disappear, and how would you prove that it did?