Description
EN. ContextSeek models contradictions via refutes / supersedes links and
has scaffolding in src/contextseek/domain/conflicts.py and
invalidation.py. Design and implement an engine that, on write or during
compact(), detects when a new item contradicts existing ones (semantic
similarity + LLM adjudication), creates the correct typed links, and resolves
the conflict per policy (supersede stale, lower confidence, or flag for review).
This is the core of "self-evolving" context and touches the EvolutionEngine.
Tasks: define detection (candidate selection → LLM judge with structured
verdict); define resolution policies (configurable); wire into the write/compact
path; propagate confidence through the evidence chain; add a thorough test
matrix including false-positive guards; document the policy model.
Acceptance: contradictory writes produce refutes/supersedes links and a
deterministic, policy-driven resolution; non-conflicting writes are untouched;
tests cover edge cases.
ZH. ContextSeek 用 refutes / supersedes 链接建模矛盾，并在
src/contextseek/domain/conflicts.py 与 invalidation.py 里有脚手架。请设计并实现
一个引擎：在写入或 compact() 时，检测新条目是否与既有条目矛盾（语义相似 + LLM 裁
决），建立正确的类型化链接，并按策略消解冲突（淘汰过时项、降置信度、或标记待人工复
核）。这是"上下文自进化"的核心，会触及 EvolutionEngine。
任务： 定义检测（候选选取 → LLM 裁判给出结构化判定）；定义可配置的消解策略；接入
写入/compact 路径；让置信度沿证据链传播；写完整测试矩阵含误报防护；文档化策略模型。
验收： 矛盾写入会产出 refutes/supersedes 链接和确定的、由策略驱动的消解；不矛
盾的写入不受影响；测试覆盖边界情况。
Description
EN. ContextSeek models contradictions via
refutes/supersedeslinks andhas scaffolding in
src/contextseek/domain/conflicts.pyandinvalidation.py. Design and implement an engine that, on write or duringcompact(), detects when a new item contradicts existing ones (semanticsimilarity + LLM adjudication), creates the correct typed links, and resolves
the conflict per policy (supersede stale, lower confidence, or flag for review).
This is the core of "self-evolving" context and touches the EvolutionEngine.
Tasks: define detection (candidate selection → LLM judge with structured
verdict); define resolution policies (configurable); wire into the write/compact
path; propagate confidence through the evidence chain; add a thorough test
matrix including false-positive guards; document the policy model.
Acceptance: contradictory writes produce
refutes/supersedeslinks and adeterministic, policy-driven resolution; non-conflicting writes are untouched;
tests cover edge cases.
ZH. ContextSeek 用
refutes/supersedes链接建模矛盾，并在src/contextseek/domain/conflicts.py与invalidation.py里有脚手架。请设计并实现一个引擎：在写入或
compact()时，检测新条目是否与既有条目矛盾（语义相似 + LLM 裁决），建立正确的类型化链接，并按策略消解冲突（淘汰过时项、降置信度、或标记待人工复
核）。这是"上下文自进化"的核心，会触及 EvolutionEngine。
任务： 定义检测（候选选取 → LLM 裁判给出结构化判定）；定义可配置的消解策略；接入
写入/compact 路径；让置信度沿证据链传播；写完整测试矩阵含误报防护；文档化策略模型。
验收： 矛盾写入会产出
refutes/supersedes链接和确定的、由策略驱动的消解；不矛盾的写入不受影响；测试覆盖边界情况。