Skip to content

fix(memory): skip fully-wildcard L1 signatures from L2 bucketing - #323

Open
MatthewZhuang wants to merge 1 commit into
MemTensor:v1.1.2from
MatthewZhuang:fix/l1-signature-bucketing
Open

fix(memory): skip fully-wildcard L1 signatures from L2 bucketing#323
MatthewZhuang wants to merge 1 commit into
MemTensor:v1.1.2from
MatthewZhuang:fix/l1-signature-bucketing

Conversation

@MatthewZhuang

Copy link
Copy Markdown
Collaborator

Summary

  • 无工具且抽不到 tag/errCode 的 L1 会生成全通配签名 _|_|_|_,L2 按精确 candidateKey 分桶后会把无关回合并成一条政策
  • 新增 isBucketableSignature:四段里至少有一段不是 _ 才允许进 L2 candidate pool / 诱导
  • 全通配签名仍写在 L1 上,User Memory 写入和检索不动;error|python|_|_compact|summary|_|_ 这类半通配仍可分桶

Validation

  • rebase 到 MemTensor/v1.1.2 后:npm test -- tests/algorithm/plugin-algorithms.test.ts tests/service/evolution/policy-induction.test.ts tests/service/user-memory/user-memory.test.ts tests/service/evolution/reward.test.ts — 117 passed
  • 新测试覆盖:两条无工具通配 L1 不进池、不诱导 L2;「我喜欢吃苹果」仍写入 User Memory

Empty no-tool traces were collapsing into _|_|_|_ and inducing mixed L2 policies. Keep the L1 and user-memory paths unchanged.

Co-authored-by: Cursor <cursoragent@cursor.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants