Immersive Portals - CE 6.0.13

Minecraft 1.21.1 / NeoForge 21.1.228+ / Java 21

Fixed

• Recover failed portal chunk promotions while their tickets remain active. Failed neighbor-range results are requeued with a one-second game-time cooldown; vanilla accessible/block/entity-ticking futures are recreated only when failed and their current ticket permits that status. Pending and successful futures retain their identity; exceptional I/O remains an error. The player delivery queue retains unloaded chunks until actual block/light packets can be sent. Retry queue cleanup removes retained tickets too.
• Make Forgified Fabric networking 4.2.2+a92978fd19 receiver registration safe across both flows of one payload and distinct local listeners. Concurrent maps publish native handlers and receivers to network reads. A per-receiver-map registration lock preserves upstream duplicate return values and callback construction; dispatch and removal stay lock-free. Unsupported networking versions fail with a precise supported-version message.

Verification and release speed

• Start real shaderpack jobs immediately after check selection, overlapping core, UDP/TCP and fixture checks. Core, UDP/TCP, all 26 fixture/control lanes and all four real-pack lanes remain mandatory in the same aggregate gate.
• Reuse the exact main-tested release jar only after all 35 CI checks succeeded for the tagged commit, with matching repository/run attempt, artifact origin, version and checksum. Ineligible evidence verifies afresh; corrupt selected artifacts block publication.
• Show live scene/readiness changes and minute heartbeats without extending deadlines or counting these messages as game progress.
• Real-pack checks unload a client End chunk and inject failed server block/entity-ticking futures during nested reload. Recovery, real resend, full chunk data and unchanged portal/depth pixel assertions must pass.

The renderer versions and scoped Create/Flywheel fallback described in 6.0.12 remain supported. See the issue acceptance map for the retained regressions and required native lanes.

Measured CI bottleneck follow-up

• Capture diagnostic inner/native depth only on assertion frames, preserving the 120-frame scene warm-up, ten-frame spacing, three consecutive passing observations, every phase and every existing threshold. Reject stale scene/phase/frame depth explicitly in both the live client and artifact verifier.
• Record scene readiness, first-check and pass timings, frame counts and diagnostic readback cost to make further optimization evidence-driven. Actual before/after workflow and per-lane durations are reported in the PR after the exact-head runs complete.