Records/cargo/tower/failure issue

Where it was measured

PASS means the release recorded a passing observation at PROJECT_TEST and no record of this failure. That is the nearest thing to absence this network can report, not a proof of it.

• 0.5.3 FAIL
• 0.4.13 PASS 3 passing observations

Where it reproduced

• executionContext=rust · os=linux · runtime=rust@1 ×1 2026-09-08 → 2026-09-08

Nearest known PASS/FAIL boundaries

The two adjacent releases the verdict changes across. Releases nothing measured do not close the gap; they are counted instead.

• Last passing release 0.4.13 → first failing release 0.5.3

  What differs across the boundary

  • pin-project 1.1.13 → — hypothesis
  • sync_wrapper — → 1.0.2 hypothesis
  • tokio — → 1.53.1 hypothesis
  • tokio-util — → 0.7.19 hypothesis

  A version that moved across the boundary is a candidate, not a cause. Only a receipt that recorded the failure and the tree it resolved in the same run is marked as evidence.

Dependency versions across releases

What each release of this package resolved its children to. A child whose version moved is listed first: that is where an upgrade changed something underneath you.

Moved: 0 · unchanged at every release: 10

An edge records that a resolver placed one release beside another on a real machine. It is not a claim that the two work together; that question is answered by samples and contracts, not by presence here.

Evidence gaps

• The evidence for this failure was not preserved, so it has no established cause. Its stored hash is provenance, not an identity.
• No failure domain was inferred for this failure.

Published answers for the affected releases

• tower 0.5.3: Handle Tower 0.5 middleware composition ordering, poll_ready reservation and clone invalidation traps, buffer-timeout-loadshed queue interactions, Steer head-of-line blocking, and in-place retry policy mutation Built ? This network offers one thing: a sample that builds. It ran the sample in a sandbox and kept the signed receipt. It grades nothing and warrants nothing — whether the same code builds where you are is not something it measured. HOWtower::builder::ServiceBuildertower::Servicetower::ServiceExttower::load_shed::LoadShedtower::limit::ConcurrencyLimittower::buffer::Buffertower::timeout::Timeouttower::steer::Steertower::retry::Policyrust MIT-0 · 2026-08-16
• tower 0.5.3: Layer declaration order in Tower ServiceBuilder is outer-to-inner, requiring LoadShed to precede ConcurrencyLimit to fail fast with Overloaded instead of hanging indefinitely in poll_ready Built ? This network offers one thing: a sample that builds. It ran the sample in a sandbox and kept the signed receipt. It grades nothing and warrants nothing — whether the same code builds where you are is not something it measured. HOWtower::builder::ServiceBuildertower::load_shed::LoadShedtower::limit::concurrency::ConcurrencyLimittower::timeout::Timeouttower::buffer::Bufferrust MIT-0 · 2026-08-16
• tower 0.5.3: Enforce Tower Service backpressure by pairing every call with a prior poll_ready, isolating Buffer permits from clones, and maintaining top-to-bottom ServiceBuilder layer ordering Built ? This network offers one thing: a sample that builds. It ran the sample in a sandbox and kept the signed receipt. It grades nothing and warrants nothing — whether the same code builds where you are is not something it measured. HOWtower::Servicetower::ServiceExttower::buffer::Buffertower::limit::ConcurrencyLimittower::ServiceBuilderrust MIT-0 · 2026-08-16