BundleSessionObservabilityTests' two gauge facts asserted absolute values
(Assert.Equal(0, baseline), Assert.Equal(2, ReadGauge())) against what is a
single process-global provider slot — the same latent shape as the row-32
QueueDepthGaugeTests flake, correct only while this file is the slot's sole
registrant. Converted to the delta-over-scoped-action pattern the row-66
rejection-counter facts in the same file already use: a new ReadGaugeDelta
helper measures how far the instrument MOVED across an action driven by this
test's own store, so an inherited non-zero baseline cancels out.
Two assertions got stronger in the process. The post-clear check was
Assert.Equal(0, ReadGauge()) — "the instrument reports zero", which any
unregistered slot satisfies; it is now "opening more sessions on the detached
store moves the gauge by nothing", which is the actual claim. And the
compare-and-clear fact gained a negative control: the superseded store's opens
must NOT reach the instrument, so a slot still pointing at oldStore can no
longer satisfy the assertion by coincidence when both stores hold equal counts.
Fail-before (temporary probe, not committed): with a foreign provider returning
7 registered before the service starts, the OLD body fails
`Assert.Equal(0, baseline)` with `Expected: 0 / Actual: 7`; the NEW delta body
passes the identical scenario.
Honest residual documented on the helper: a gauge has one slot and
last-writer-wins, so a registrant that clobbers it BETWEEN the two reads is
unrecoverable read-side — that needs row-32's mutual exclusion. It cannot arise
today (this is the only file in the assembly constructing a
BundleSessionEvictionService, and test assemblies are separate processes); a
second such file should bring the collection treatment with it.
Transport.Tests: 180/180 green.
Joseph Doherty
98c9458830