recallnest 2.2.0 → 2.3.0

package/eval/continuity/results-2026-03-19.md

@@ -1,453 +0,0 @@
-# RecallNest Continuity Eval
-
-- Generated: 2026-03-19T06:36:12.211Z
-- Cases: 21
-- Passed: 21/21
-- Average score: 100.0%
-
-| Case | Profile | Score | Pass | Stable | Patterns | Cases | Checkpoint |
-|------|---------|-------|------|--------|----------|-------|------------|
-| writing_style_continuity | writing | 100% | yes | 2 | 0 | 0 | no |
-| writing_style_sparse_continue | writing | 100% | yes | 2 | 2 | 0 | no |
-| visual_style_continuity | default | 100% | yes | 2 | 0 | 0 | no |
-| project_context_continuity | default | 100% | yes | 2 | 1 | 2 | no |
-| generic_named_recallnest_task_purity | default | 100% | yes | 2 | 1 | 2 | no |
-| project_scope_entity_continuity | default | 100% | yes | 2 | 2 | 2 | no |
-| workflow_pattern_gap_check | default | 100% | yes | 1 | 3 | 0 | no |
-| recallnest_checkpoint_continuity | default | 100% | yes | 3 | 3 | 3 | yes |
-| project_scope_checkpoint_continuity | default | 100% | yes | 3 | 2 | 2 | yes |
-| project_task_pivot_recall | default | 100% | yes | 3 | 2 | 2 | yes |
-| associative_recall_vague_query | default | 100% | yes | 2 | 1 | 1 | no |
-| non_recallnest_named_task_isolation | default | 100% | yes | 1 | 1 | 0 | no |
-| bridge_project_pin_filter | default | 100% | yes | 1 | 1 | 0 | no |
-| scoped_bridge_pin_isolation | default | 100% | yes | 2 | 2 | 2 | no |
-| scoped_foreign_project_pin_collision | default | 100% | yes | 3 | 1 | 2 | no |
-| scoped_foreign_project_entity_collision | default | 100% | yes | 2 | 2 | 3 | no |
-| scoped_foreign_project_task_result_collision | default | 100% | yes | 2 | 3 | 3 | no |
-| vague_nest_task_entity_recall | default | 100% | yes | 2 | 1 | 1 | no |
-| cross_language_recall | default | 100% | yes | 3 | 0 | 1 | no |
-| scope_isolation_check | default | 100% | yes | 2 | 2 | 2 | no |
-| legacy_null_scope_isolation | default | 100% | yes | 2 | 2 | 2 | no |
-
-## Case Notes
-
-### writing_style_continuity
-- Task: 继续写我的AI小木屋的文章
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 0
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Pinned: 用户写作语气偏好: 用户稳定偏好口语化、不端着、可以自嘲但不说教。表达应像朋友聊天，避免居高临下、用力过猛或碎嘴。 Snippet: 根据你的档案： - **医学出身，文化口工作，AI 野路子，不是程序员** - 公众... | Pinned: 用户视觉审美偏好: 用户常用视觉风格是手绘涂鸦风加高对比撞色；在内容包装和配图生成时，应优先沿用这一审美方向，除非任务目标明确要求其他风格。
-- Pattern preview: -
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: 口语化, 不端着
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: Fresh windows should carry over stable writing-style preferences.
-
-### writing_style_sparse_continue
-- Task: 不要让我重复前情，接着写
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Pinned: 用户写作语气偏好: 用户稳定偏好口语化、不端着、可以自嘲但不说教。表达应像朋友聊天，避免居高临下、用力过猛或碎嘴。 | Pinned: 用户视觉审美偏好: 用户常用视觉风格是手绘涂鸦风加高对比撞色；在内容包装和配图生成时，应优先沿用这一审美方向，除非任务目标明确要求其他风格。
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: 口语化, 不端着
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A sparse writing prompt should still recover the user's durable writing-style preferences.
-
-### visual_style_continuity
-- Task: 给文章做封面和配图
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 0
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Pinned: 用户视觉审美偏好: 用户常用视觉风格是手绘涂鸦风加高对比撞色；在内容包装和配图生成时，应优先沿用这一审美方向，除非任务目标明确要求其他风格。 Snippet: 1. 给刚才写的文章生成配图（封面1张+插图1张就够，测试... | Pinned: 用户写作语气偏好: 用户稳定偏好口语化、不端着、可以自嘲但不说教。表达应像朋友聊天，避免居高临下、用力过猛或碎嘴。 Snippet: 根据你的档案： - **医学出身，文化口工作，AI 野路子，不是程序员** - 公众...
-- Pattern preview: -
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: 手绘涂鸦, 高对比撞色
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: Visual direction should survive without the user restating it.
-
-### project_context_continuity
-- Task: 一起继续排查 RecallNest 的连续性问题
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 1
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: RecallNest scope fallback cleanup Problem: resume_context stayed too narrow on project scope and surfaced ongoi... | Case: RecallNest sparse startup context cleanup Problem: resume_context was returning noisy transcript fragments and ...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: Fresh windows should at least recover the active project context.
-
-### generic_named_recallnest_task_purity
-- Task: 继续整理 RecallNest 实施清单，不要让我重复前情
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 1
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: RecallNest scope fallback cleanup Problem: resume_context stayed too narrow on project scope and surfaced ongoi... | Case: RecallNest sparse startup context cleanup Problem: resume_context was returning noisy transcript fragments and ...
-- Checkpoint summary: -
-- Matched stable any: recallnest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: RecallNest scope fallback cleanup
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A generic named RecallNest continue prompt should keep project continuity context while suppressing same-project transport or smoke task results.
-
-### project_scope_entity_continuity
-- Task: 继续这个项目，不要让我重复前情
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped mixed-project pin collision leaked foreign project summaries Problem: When `resume_context` had an expli... | Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*...
-- Checkpoint summary: -
-- Matched stable any: RecallNest, checkpoint_session, resume_context
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: When the project scope is known, stable project entities should survive even if the task prompt is sparse.
-
-### workflow_pattern_gap_check
-- Task: 开个新窗口继续做同一个终端项目
-- Score: 100%
-- Pass: yes
-- Stable items: 1
-- Pattern items: 3
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Task focus: 开个新窗口继续做同一个终端项目
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Cross-window continuity handoff Tools: resume_context, latest_checkpoint Use when: When opening a f... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: -
-- Matched stable all: -
-- Matched patterns any: search_memory, resume_context, checkpoint
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: This is a stretch case. It should start passing as more workflow patterns are promoted into durable memory.
-
-### recallnest_checkpoint_continuity
-- Task: 继续 recallnest
-- Score: 100%
-- Pass: yes
-- Stable items: 3
-- Pattern items: 3
-- Case items: 3
-- Checkpoint present: yes
-- Stable preview: Entity: RecallNest is the shared memory layer for Claude Code, Codex, and Gemini CLI. | Checkpoint summary: RecallNest Phase 3/4 并行推进中。核心三件套 store_memory、checkpoint_session、resume_context 已上线，当前继续收 resume_... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Cross-window continuity handoff Tools: resume_context, latest_checkpoint Use when: When opening a f... | Workflow pattern: Checkpoint before switching windows Tools: checkpoint_session Use when: When leaving an active codi... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: RecallNest scope fallback cleanup Problem: resume_context stayed too narrow on project scope and surfaced ongoi... | Case: RecallNest sparse startup context cleanup Problem: resume_context was returning noisy transcript fragments and ... | Case: Continuity eval checkpoint isolation Problem: continuity eval results drifted because they were reading whichev...
-- Checkpoint summary: RecallNest Phase 3/4 并行推进中。核心三件套 store_memory、checkpoint_session、resume_context 已上线，当前继续收 resume_context compose。
-- Matched stable any: -
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: Phase 3, resume_context, store_memory
-- Forbidden matches: -
-- Notes: When a project checkpoint exists, resume_context should prefer checkpoint-led continuity over noisy transcript fragments.
-
-### project_scope_checkpoint_continuity
-- Task: 继续这个项目，不要让我重复前情
-- Score: 100%
-- Pass: yes
-- Stable items: 3
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: yes
-- Stable preview: Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Checkpoint summary: RecallNest 项目范围下已完成 continuity baseline setup、doctor baseline 和 headless Claude Code smoke，下一步继续扩... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped mixed-project pin collision leaked foreign project summaries Problem: When `resume_context` had an expli... | Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*...
-- Checkpoint summary: RecallNest 项目范围下已完成 continuity baseline setup、doctor baseline 和 headless Claude Code smoke，下一步继续扩 continuity eval case。 Entities: seed:continuity, smoke:clau...
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: doctor, smoke:claude-continuity, continuity eval case
-- Forbidden matches: -
-- Notes: Project-scoped checkpoints should compose cleanly when the scope uses the explicit `project:` key.
-
-### project_task_pivot_recall
-- Task: 回到 RecallNest 项目，处理主动 recall regression
-- Score: 100%
-- Pass: yes
-- Stable items: 3
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: yes
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Checkpoint summary: RecallNest 已补上 HTTP auto-recall，下一步把 task-pivot 的主动 recall 变成 smoke 和 eval 的稳定回归信号。 | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Recall before repo exploration Tools: resume_context, search_memory Use when: When a fresh window c... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*... | Case: Scoped recall leaked conversational durable pins Problem: Scoped resume_context could still surface raw `/` bri...
-- Checkpoint summary: RecallNest 已补上 HTTP auto-recall，下一步把 task-pivot 的主动 recall 变成 smoke 和 eval 的稳定回归信号。 Entities: search_memory, smoke:claude-continuity
-- Matched stable any: RecallNest, auto-recall
-- Matched stable all: -
-- Matched patterns any: search_memory
-- Matched cases any: -
-- Matched checkpoint any: smoke, eval
-- Forbidden matches: -
-- Notes: A named project task pivot should still surface continuity context and search_memory guidance even without an explicit continue prompt.
-
-### associative_recall_vague_query
-- Task: 之前弄过那个什么 Nest 的记忆系统
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 1
-- Case items: 1
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: 我把这次问题顺手沉淀成一个 RecallNest case，后面再遇到“同主题升级信息被 ingest 吞掉”的情况，可以直接回忆这条修复路径。
-- Checkpoint summary: -
-- Matched stable any: RecallNest, checkpoint, memory
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: Vague associative query should still retrieve relevant memories through fuzzy matching, not require exact keywords.
-
-### non_recallnest_named_task_isolation
-- Task: A2A code Claude SDK calling error
-- Score: 100%
-- Pass: yes
-- Stable items: 1
-- Pattern items: 1
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Task focus: a2a
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: a2a
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: An unscoped named external task should fall back to task focus instead of leaking RecallNest entities or unrelated style pins.
-
-### bridge_project_pin_filter
-- Task: 继续 telegram ai bridge 项目
-- Score: 100%
-- Pass: yes
-- Stable items: 1
-- Pattern items: 1
-- Case items: 0
-- Checkpoint present: no
-- Stable preview: Task focus: telegram
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: -
-- Checkpoint summary: -
-- Matched stable any: telegram
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: An external bridge task should not surface raw conversational pins or unrelated RecallNest/style context.
-
-### scoped_bridge_pin_isolation
-- Task: 继续 RecallNest continuity bridge 适配
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*... | Case: Scoped recall leaked conversational durable pins Problem: Scoped resume_context could still surface raw `/` bri...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A scoped RecallNest task should not surface raw bridge transcript pins even when durable pinned assets mention bridge terms.
-
-### scoped_foreign_project_pin_collision
-- Task: 继续 RecallNest bridge README 适配
-- Score: 100%
-- Pass: yes
-- Stable items: 3
-- Pattern items: 1
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Entity: RecallNest is the shared memory layer for Claude Code, Codex, and Gemini CLI.
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped mixed-project pin collision leaked foreign project summaries Problem: When `resume_context` had an expli... | Case: Scoped recall leaked conversational durable pins Problem: Scoped resume_context could still surface raw `/` bri...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A scoped RecallNest task with overlapping bridge and README terms should not surface pinned summaries from a different project.
-
-### scoped_foreign_project_entity_collision
-- Task: 继续 RecallNest MCP transport 适配
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 3
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: RecallNest MCP transport rollout Tools: resume_context, search_memory, eval:continuity Use when: Wh... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: RecallNest MCP transport regression Problem: Scoped RecallNest continuity work around MCP transport could still... | Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*... | Case: Borderline transcript dedup swallowed incremental A2A details Problem: Transcript ingest was dropping same-topi...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A scoped RecallNest task with shared tool nouns should not surface entity summaries from a different project.
-
-### scoped_foreign_project_task_result_collision
-- Task: 继续 RecallNest continuity MCP transport rollout
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 3
-- Case items: 3
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: RecallNest MCP transport rollout Tools: resume_context, search_memory, eval:continuity Use when: Wh... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Same-project generic task results needed specificity gating Problem: Even after foreign-project leakage was fix... | Case: RecallNest MCP transport regression Problem: Scoped RecallNest continuity work around MCP transport could still... | Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: RecallNest MCP transport rollout
-- Matched cases any: RecallNest MCP transport regression
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A scoped RecallNest continuity task with shared tool nouns should not surface foreign-project patterns, cases, or workflow fallback results.
-
-### vague_nest_task_entity_recall
-- Task: 之前弄过那个什么 Nest 的记忆系统
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 1
-- Case items: 1
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: 我把这次问题顺手沉淀成一个 RecallNest case，后面再遇到“同主题升级信息被 ingest 吞掉”的情况，可以直接回忆这条修复路径。
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: A vague associative Nest task should still recover RecallNest without dragging in unrelated writing or external-task context.
-
-### cross_language_recall
-- Task: continue my AI writing project
-- Score: 100%
-- Pass: yes
-- Stable items: 3
-- Pattern items: 0
-- Case items: 1
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Pinned: 用户写作语气偏好: 用户稳定偏好口语化、不端着、可以自嘲但不说教。表达应像朋友聊天，避免居高临下、用力过猛或碎嘴。 Snippet: 根据你的档案： - **医学出身，文化口工作，AI 野路子，不是程序员** - 公众... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: -
-- Case preview: Case: Borderline transcript dedup swallowed incremental A2A details Problem: Transcript ingest was dropping same-topi...
-- Checkpoint summary: -
-- Matched stable any: AI, 写作, 公众号
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: English prompt should still retrieve Chinese-language durable memories when the scope is known.
-
-### scope_isolation_check
-- Task: 继续我的项目
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo... | Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped mixed-project pin collision leaked foreign project summaries Problem: When `resume_context` had an expli... | Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: When querying with project:recallnest scope, should NOT leak memories from other projects like content-alchemy or digital-clone.
-
-### legacy_null_scope_isolation
-- Task: 之前我记得有个关于
-- Score: 100%
-- Pass: yes
-- Stable items: 2
-- Pattern items: 2
-- Case items: 2
-- Checkpoint present: no
-- Stable preview: Entity: RecallNest exposes continuity through both HTTP API and MCP, so different agents can share one memory layer a... | Entity: RecallNest continuity revolves around three primitives: store_memory for durable facts, checkpoint_session fo...
-- Pattern preview: Workflow pattern: Scoped project continuity recall Tools: resume_context, latest_checkpoint, search_memory Use when: ... | Workflow pattern: Promote recurring continuity workflow Tools: store_workflow_pattern, /v1/pattern Use when: When the...
-- Case preview: Case: Scoped entity recall leaked foreign project entities via shared tool nouns Problem: With an explicit `project:*... | Case: Borderline transcript dedup swallowed incremental A2A details Problem: Transcript ingest was dropping same-topi...
-- Checkpoint summary: -
-- Matched stable any: RecallNest
-- Matched stable all: -
-- Matched patterns any: -
-- Matched cases any: -
-- Matched checkpoint any: -
-- Forbidden matches: -
-- Notes: Legacy memories without proper scope should not leak into scoped queries. The scope policy should filter them out.
-

package/eval/continuity/runbook.md

@@ -1,110 +0,0 @@
-# Continuity Eval Runbook
-
-## Goal
-
-Measure whether a fresh window can recover stable background through `resume_context`, instead of depending on manual reminders.
-
-## Command
-
-```bash
-bun run eval:continuity
-```
-
-If you also want the continuity eval run to emit workflow observations into the dedicated observation store, use:
-
-```bash
-bun run eval:continuity --record-observations --observation-scope eval:continuity
-```
-
-Generate a dated markdown report:
-
-```bash
-bun run src/eval.ts --mode continuity \
-  --cases eval/continuity/cases.json \
-  --output eval/continuity/results-YYYY-MM-DD.md
-```
-
-Generate JSON for tooling:
-
-```bash
-bun run src/eval.ts --mode continuity \
-  --cases eval/continuity/cases.json \
-  --json \
-  --output /tmp/recallnest-continuity.json
-```
-
-## What To Check
-
-- Pass rate
-- Average score
-- Which cases only recover `stableContext`
-- Which cases still fail on `relevantPatterns` or `recentCases`
-- Whether a checkpoint is present when `scope` or `sessionId` is provided
-
-## Before Running
-
-1. Confirm your normal memory index is available.
-2. Seed the canonical workflow patterns if you want continuity eval to use durable `patterns` memory instead of only fallback logic:
-
-```bash
-bun run seed:patterns
-```
-
-3. Seed the continuity cases and stable project memories if you want project continuity cases to use durable `cases` / `entities` instead of only task fallback:
-
-```bash
-bun run seed:continuity
-```
-
-4. If you want to test checkpoint-aware continuity, write a checkpoint first with `checkpoint_session` or `POST /v1/checkpoint`.
-5. Keep the cases file small and high-signal. Four to eight cases is enough.
-
-## Claude Code Smoke Test
-
-Use this when you want a real Claude Code acceptance check instead of only local eval fixtures:
-
-```bash
-bun run smoke:claude-continuity
-```
-
-This headless smoke test:
-
-- runs `claude -p` with a fresh-window `继续 RecallNest` prompt
-- verifies `resume_context` appears before any visible repo exploration tools
-- runs a second prompt that requires `checkpoint_session`
-- stores raw `stream-json` artifacts under `/tmp/recallnest-claude-smoke-*`
-- writes success / missed / failure workflow observations into `data/workflow-observations` by default
-
-Notes:
-
-- the script pre-allows RecallNest MCP tools because non-interactive `dontAsk` mode otherwise denies them
-- the script prints `[smoke] Running continue case...` and `[smoke] Running checkpoint case...` so the slower checkpoint path is visible while it runs
-- use the saved JSONL artifacts as the source of truth for tool order
-- if Claude's prose mentions repo state without matching `Read` / `Bash` / `Grep` / `Glob` events, treat that as an inference, not a verified observation
-- if the warning points at `resume_context`, the inherited source is usually an older checkpoint summary; overwrite it with a clean checkpoint after you actually inspect the repo
-- the smoke script now fails if `checkpoint_session` itself writes repo-state claims without visible repo tools, because that contaminates later continuity handoffs
-- qualified copies still count: `git status ... but not verified in this window` is still a failure if no repo tool ran in that window
-- RecallNest now strips repo-state text from saved checkpoint content, so a raw `checkpoint_session` request with repo-state text is a warning, but the saved checkpoint output must stay clean
-- set `RECALLNEST_RECORD_WORKFLOW_OBSERVATIONS=0` if you only want JSONL artifacts without writing workflow observations
-- the checkpoint case can take materially longer than the continue case because it waits for both `resume_context` and `checkpoint_session`; that slower path is expected and is not the same thing as a lingering zombie process
-
-## After Running
-
-1. Save the markdown report into `eval/continuity/results-YYYY-MM-DD.md`.
-2. Compare it with the previous run instead of eyeballing one score in isolation.
-3. If a case fails, decide which layer is missing:
-   - capture gap
-   - checkpoint gap
-   - composition gap
-   - retrieval gap
-
-## Reading Failures
-
-- Stable context missing:
-  durable memories are not strong enough yet
-- Patterns missing:
-  reusable workflow knowledge has not been promoted enough
-- Cases missing:
-  solved problems are still trapped in raw transcripts
-- Checkpoint missing:
-  active work state was not written for the tested `scope` or `sessionId`

package/eval/continuity/seed-cases.md

@@ -1,54 +0,0 @@
-# Continuity Seed Cases
-
-这组 case 用来回答一个更具体的问题：
-
-`开新窗口以后，RecallNest 是否还能把稳定背景带回来？`
-
-## 设计原则
-
-- 优先测稳定上下文，不测“机械续上一个话题”
-- 问法尽量贴近真实工作，而不是关键词堆砌
-- 允许先暴露 gap，再逐步补齐 durable memory 和 checkpoint
-
-## 当前覆盖
-
-1. 写作风格连续性
-   检查口语化、不端着、可自嘲这类稳定偏好是否会被带回
-2. 视觉风格连续性
-   检查手绘涂鸦、高对比撞色等审美方向是否会被带回
-3. 项目背景连续性
-   检查 `RecallNest` 这类当前项目背景是否可见
-4. 工作流模式缺口检查
-   检查 `search_memory` / `resume_context` / `checkpoint` 是否已经足够稳定地进入 pattern recall
-
-## 如何新增 case
-
-新增 continuity case 时，优先填写这些字段：
-
-- `task`
-- `profile`
-- `expectStableAny`
-- `expectPatternsAny`
-- `expectCasesAny`
-- `scope` 或 `sessionId`（只有当你明确想测 scoped continuity 时才加）
-
-示例：
-
-```json
-{
-  "name": "recallnest_checkpoint_resume",
-  "task": "继续推进 RecallNest 的连续性层",
-  "profile": "default",
-  "scope": "agent:codex",
-  "sessionId": "codex-2026-03-16-001",
-  "expectStableAny": ["RecallNest", "Claude Code", "Codex", "Gemini CLI"],
-  "expectCheckpointAny": ["resume_context", "continuity", "checkpoint"]
-}
-```
-
-## 解释结果
-
-- `stableContext` 命中高，说明稳定背景已经能被新窗口带回
-- `relevantPatterns` 命中低，通常说明高价值 workflow 还没有被结构化存够
-- `recentCases` 命中低，通常说明历史解决方案还停留在 transcript，而没被提升成 case memory
-- `latestCheckpoint` 缺失，不一定是 bug，也可能只是当前没有为该 `scope` / `sessionId` 写入 checkpoint