v0.6.2 Announcement

English

Agent Memory Bridge v0.6.2 is out.

This release broadens the canonical benchmark fixture instead of pretending the current small set is enough.

What changed:

1. Retrieval coverage is wider

• more overlap-heavy memory cases
• more overlap-heavy signal cases
• context-compaction ambiguity is now part of the canonical fixture

1. The benchmark stays honest

• memory_expected_top1_accuracy = 1.0
• file_scan_expected_top1_accuracy = 0.636
• the fixture is harder than before, so those numbers mean more

1. The learning-quality rollout still stays controlled

• classifier calibration from v0.6.1 remains in place
• signal lifecycle still holds claim -> extend -> ack / expire / reclaim
• no MCP surface break

This is the point of the release:

• broaden retrieval proof before tuning more weights
• expose overlap and ambiguity before claiming the ranking is stable

GitHub:
https://github.com/zzhang82/Agent-Memory-Bridge

Release:
https://github.com/zzhang82/Agent-Memory-Bridge/releases/tag/v0.6.2

简体中文

Agent Memory Bridge v0.6.2 发布了。

这次发布的重点不是继续调一堆 ranking 权重，而是把 canonical benchmark fixture 扩大，让它更接近真实的重叠检索场景。

主要变化：

1. retrieval 覆盖面更广了

• 增加了 overlap-heavy 的 memory case
• 增加了 overlap-heavy 的 signal case
• 现在把 context compaction 的歧义 case 也放进了 canonical fixture

1. benchmark 继续保持诚实

• memory_expected_top1_accuracy = 1.0
• file_scan_expected_top1_accuracy = 0.636
• 题库比之前更难，所以这些数字也更有意义

1. learning-quality rollout 仍然是受控的

• v0.6.1 的 classifier calibration 继续保留
• signal lifecycle 仍然是 claim -> extend -> ack / expire / reclaim
• 没有 MCP surface break

这版真正想做的是：

• 在继续调权重之前，先把 retrieval proof 扩大
• 在说“排序稳定了”之前，先让 overlap 和 ambiguity 暴露出来

GitHub:
https://github.com/zzhang82/Agent-Memory-Bridge

Release:
https://github.com/zzhang82/Agent-Memory-Bridge/releases/tag/v0.6.2