Implementation Defects
These findings were recorded during documentation validation. The documentation does not change runtime behavior or release artifacts to hide them.
Core README release pointer is stale
At Core a71c0eb2, the README still presents an older release-candidate path while GitHub publishes v2.0.2. This documentation resolves downloads from release metadata and exact source SHA rather than copying that statement.
Published v2.0.2 embeds a non-permitting EPoSE gate result
The Linux v2.0.2 artifact from workflow run 35385285493 identifies source 54308d84, passes its three binary help/version checks, offline-mainnet smoke, 239 EPoSe C++ tests and 43 EPoSe Python tests. However, the same embedded BUILD-INFO.json records:
epose_release_gate.overall_status = no-go
epose_release_gate.stable_permitted = false
epose_release_gate.candidate_bound = falseThe unresolved entries describe missing candidate-bound security/rehearsal evidence. This conflicts with treating the artifact as having passed that formal release-readiness ledger. It does not by itself show that the compiled EPoSe runtime is disabled; it means the stronger release-gate approval must not be claimed from this artifact. The operator pages distinguish executable/test evidence from formal readiness evidence.
Diagnostics release gap
Development source a71c0eb2 registers get_epose_diagnostics; release v2.0.2 does not. Documentation labels the method development-only rather than presenting it as a release feature.
Platform signing gaps
Current Core artifacts are unsigned. Current GUI Windows artifacts are unsigned and the macOS package is ad-hoc signed rather than Developer ID signed/notarized. Checksums provide integrity against the published digest, not publisher identity.
Last updated
Was this helpful?