Point the usage-reporting link at https://danielstephenson.dev/usage-reporting (the old link pointed at a private repository and returned 404).

Every event can now carry a random installation ID as the tag install, so trace can count installations instead of raw events. The program passes an explicit ID, or a file path whose first line holds one; the client creates a random UUID there on first run. Nothing is stored in a default location. The ID identifies no person, account or address; delete the file to reset it. Every opt-out (TRACE_USAGE_REPORTING=off, DO_NOT_TRACK=1, disabled) stops it, and a disabled client never creates one. See the README section "Every event carries a random installation ID".

This release also includes the unreleased changes since 0.1.1, which required the program's version: every event is tagged version (breaking constructor change, see the README).

drafted by Claude on behalf of Daniel Stephenson

close() now sends reports that are still queued before stopping, bounded by the client timeout (5 s). Previously they were discarded, so a program that reported and exited within milliseconds — a CLI — lost its event about a quarter of the time. Consumers that call close() on exit get the fix by re-vendoring trace_client.py; nothing else changes.

drafted by Claude on behalf of Daniel Stephenson

First release. One file, standard library only, Python 3.8+: TraceClient(base_url, application, key=..., enabled=...) then report(name, value=None, tags=None). Returns immediately, never raises, queue bounded at 256; no key or enabled=False is a no-op. Same wire format and promises as trace-client-java 0.1.0.

drafted by Claude on behalf of Daniel Stephenson