Cortext v1.2.1
Cortext v1.2.1
This patch release hardens runtime and storage correctness and improves durable-ingestion latency without changing the public API.
Highlights
- fixes Go native-handle lifetime and zero-value configuration behavior
- validates image buffers across bindings and corrects retention compatibility for legacy/media paths
- fixes working-memory decay and flush rollback, out-of-order timestamp arithmetic, emotion-cascade timing, SQLite bind error handling, graph-cache invalidation, and eviction-owned embedding cleanup
- includes the passive working-memory decay repair
- reduces durable-ingestion search work through coherent exact-result caches, indexed candidate selection, and SQL fallbacks
- audits every paper section, preserves shipped-versus-experimental distinctions, and regenerates the manuscript
Behavior and performance evidence
The merged change passed all 12 exact-head GitHub checks. An exact 500-event comparison preserved the event-behavior digest, canonical logical-database digest, and association/memory/signal counts while reducing mean engine processing latency from 47.846 ms to 9.492 ms and mean sequential end-to-end latency from 55.919 ms to 17.297 ms.
These measurements prove a material improvement on that workload. They do not prove flat storage cost, actual eviction behavior, or post-eviction throughput; the documented eviction frontier remains a planning extrapolation.
Packaging
PyPI and npm artifacts contain six native targets each and retain the existing on-demand model bootstrap rather than embedding model payloads in the registry packages.
Full change and review history: PR #2.