diff --git a/README.md b/README.md
index 7b89679..f94aafe 100644
--- a/README.md
+++ b/README.md
@@ -10,9 +10,9 @@
 [![Papers](https://img.shields.io/badge/papers-989-brightgreen.svg)](papers/index.md)
 [![PDFs](https://img.shields.io/badge/local_PDFs-534-orange.svg)](papers/pdfs/)
 [![Memory products](https://img.shields.io/badge/memory%20products-38-purple.svg)](products/)
-[![Benchmarks](https://img.shields.io/badge/benchmarks-21-blueviolet.svg)](benchmarks/)
+[![Benchmarks](https://img.shields.io/badge/benchmarks-23-blueviolet.svg)](benchmarks/)
 [![Surveys](https://img.shields.io/badge/meta_surveys-6-yellow.svg)](docs/meta-surveys.md)
-[![Updated](https://img.shields.io/badge/updated-2026--07-lightgrey.svg)](docs/signals.md)
+[![Updated](https://img.shields.io/badge/updated-2026--08-lightgrey.svg)](docs/signals.md)
 
 </div>
 
@@ -55,13 +55,13 @@ separate buckets.
 
 | Area | Current coverage | Entry point | Use it when you need to |
 |---|---:|---|---|
-| Paper index | 989 scraped papers + manual radar additions through 2026-07 | [`papers/index.md`](papers/index.md) | Searchable entry point for agent-memory papers; the 989 count is the 2026-05 scrape baseline. |
+| Paper index | 989 scraped papers + manual radar additions through 2026-08 | [`papers/index.md`](papers/index.md) | Searchable entry point for agent-memory papers; the 989 count is the 2026-05 scrape baseline. |
 | Paper stubs | 988 stubs | [`papers/stubs/`](papers/stubs/) | Track papers that are covered but not yet fully read. |
 | Local PDFs | 534 files | [`papers/pdfs/`](papers/pdfs/) | Re-read sources and audit paper notes. |
-| Full / seed paper notes | 7 full + 15 seed | [`papers/`](papers/) | Use human-read notes for architectural decisions. |
+| Full / seed paper notes | 7 full + 19 seed | [`papers/`](papers/) | Use human-read notes for architectural decisions. |
 | Memory product notes | 38 notes | [`products/`](products/) | Compare memory layers, memory SDKs, managed memory, and memory-enabled agents. |
 | Product page archives | 37 snapshots | [`products/archives/`](products/archives/) | Audit product claims after source pages change. |
-| Benchmark catalog | 21 catalog rows | [`benchmarks/index.md`](benchmarks/index.md) | First-class benchmark records plus stub-backed candidate rows and a usage-claim ledger. |
+| Benchmark catalog | 23 catalog rows | [`benchmarks/index.md`](benchmarks/index.md) | First-class benchmark records plus stub-backed candidate rows and a usage-claim ledger. |
 | Claims ledger | Structured YAML ledger | [`benchmarks/claims/claims.yaml`](benchmarks/claims/claims.yaml) | Separate vendor claims, paper evaluations, critiques, and reproductions. |
 | Cost-savings lane | Seed landscape | [`docs/cost-savings-landscape.md`](docs/cost-savings-landscape.md) | Find agent-memory papers, methods, code paths, and products that reduce token, latency, or runtime cost. |
 | Survey and taxonomy | 1 living survey + 6 meta-survey records | [`docs/agent-memory-survey.md`](docs/agent-memory-survey.md) · [`docs/meta-surveys.md`](docs/meta-surveys.md) | Build a field-level view before choosing an implementation. |
@@ -74,7 +74,7 @@ Start with the path that matches your question:
 | Goal | Read these first |
 |---|---|
 | Get the field overview | [`docs/agent-memory-survey.md`](docs/agent-memory-survey.md), then [`docs/taxonomy.md`](docs/taxonomy.md) |
-| Read the latest source refresh | [`docs/memory-radar-2026-07.md`](docs/memory-radar-2026-07.md), then [`docs/signals.md`](docs/signals.md) |
+| Read the latest source refresh | [`docs/memory-radar-2026-08.md`](docs/memory-radar-2026-08.md), then [`docs/signals.md`](docs/signals.md) |
 | Find relevant papers | [`papers/index.md`](papers/index.md), then full notes under [`papers/`](papers/) |
 | Compare memory products | [`docs/products-landscape.md`](docs/products-landscape.md), [`docs/product-memory-architectures.md`](docs/product-memory-architectures.md), [`docs/product-architecture-diagrams.md`](docs/product-architecture-diagrams.md) |
 | Check why a product was included or rejected | [`docs/product-discovery-log.md`](docs/product-discovery-log.md) |
@@ -159,7 +159,7 @@ flowchart LR
 | [`docs/taxonomy.md`](docs/taxonomy.md) | Shared vocabulary for classifying agent-memory systems and memory-kernel responsibilities. |
 | [`docs/meta-surveys.md`](docs/meta-surveys.md) | External meta-survey index from late 2025 through 2026 H1. |
 | [`docs/research-radar.md`](docs/research-radar.md) | Workflow for turning papers, products, and benchmark evidence into ImpactReports and ADR inputs. |
-| [`docs/memory-radar-2026-07.md`](docs/memory-radar-2026-07.md) | 2026-07 weekly refresh across papers, products, GitHub projects, and reviewer decisions. |
+| [`docs/memory-radar-2026-08.md`](docs/memory-radar-2026-08.md) | 2026-08 weekly refresh across papers, products, GitHub projects, and reviewer decisions. |
 | [`docs/cost-savings-landscape.md`](docs/cost-savings-landscape.md) | Focused lane for agent-memory token reduction, budgeted retrieval, runtime-cost methods, code paths, and product practice signals. |
 | [`docs/information-sources.md`](docs/information-sources.md) | Source catalog for papers, products, communities, and zh-CN information channels. |
 | [`docs/related-work.md`](docs/related-work.md) | Discovery-input attribution and scrape provenance. |
@@ -180,11 +180,11 @@ flowchart LR
 
 | Path | Purpose |
 |---|---|
-| [`papers/`](papers/) | 7 full paper notes, 15 seed notes, and the master [`index.md`](papers/index.md). |
+| [`papers/`](papers/) | 7 full paper notes, 19 seed notes, and the master [`index.md`](papers/index.md). |
 | [`papers/stubs/`](papers/stubs/) | 988 generated stubs for papers not yet fully read. |
 | [`papers/pdfs/`](papers/pdfs/) | 534 archived PDFs, about 1.8 GB. See the archival policy below. |
 | [`papers/_scrape/`](papers/_scrape/) | Reproducibility artifacts: scrape script and dedup JSON. |
-| [`benchmarks/`](benchmarks/) | 21 benchmark catalog rows, including protocol notes, stub-backed candidate rows, and the note template. |
+| [`benchmarks/`](benchmarks/) | 23 benchmark catalog rows, including protocol notes, stub-backed candidate rows, and the note template. |
 | [`benchmarks/claims/`](benchmarks/claims/) | Usage-event ledger for benchmark mentions, vendor claims, critiques, and reproductions. |
 | [`benchmarks/archives/`](benchmarks/archives/) | Optional source-page snapshots for benchmark pages, repositories, or dataset cards. |
 | [`docs/benchmarks-landscape.md`](docs/benchmarks-landscape.md) | Benchmark landscape by capability, usage type, and evidence independence. |
diff --git a/README_cn.md b/README_cn.md
index 0766987..8374ee9 100644
--- a/README_cn.md
+++ b/README_cn.md
@@ -10,9 +10,9 @@
 [![Papers](https://img.shields.io/badge/papers-989-brightgreen.svg)](papers/index.md)
 [![PDFs](https://img.shields.io/badge/local_PDFs-534-orange.svg)](papers/pdfs/)
 [![记忆产品](https://img.shields.io/badge/memory%20products-38-purple.svg)](products/)
-[![Benchmarks](https://img.shields.io/badge/benchmarks-21-blueviolet.svg)](benchmarks/)
+[![Benchmarks](https://img.shields.io/badge/benchmarks-23-blueviolet.svg)](benchmarks/)
 [![Surveys](https://img.shields.io/badge/meta_surveys-6-yellow.svg)](docs/meta-surveys.md)
-[![Updated](https://img.shields.io/badge/updated-2026--07-lightgrey.svg)](docs/signals.md)
+[![Updated](https://img.shields.io/badge/updated-2026--08-lightgrey.svg)](docs/signals.md)
 
 </div>
 
@@ -53,13 +53,13 @@
 
 | 板块 | 当前覆盖 | 入口 | 适合用来 |
 |---|---:|---|---|
-| 论文索引 | 989 篇抓取论文 + 截至 2026-07 的手工 radar 新增 | [`papers/index.md`](papers/index.md) | 搜索 agent-memory 论文和发现线索；989 是 2026-05 抓取基线。 |
+| 论文索引 | 989 篇抓取论文 + 截至 2026-08 的手工 radar 新增 | [`papers/index.md`](papers/index.md) | 搜索 agent-memory 论文和发现线索；989 是 2026-05 抓取基线。 |
 | 论文 stub | 988 个 stub | [`papers/stubs/`](papers/stubs/) | 跟踪已覆盖但尚未 full 阅读的论文。 |
 | 本地 PDF | 534 个文件 | [`papers/pdfs/`](papers/pdfs/) | 复读来源和审计论文笔记。 |
-| full / seed 论文笔记 | 7 个 full + 15 个 seed | [`papers/`](papers/) | 为架构决策引用人工阅读笔记。 |
+| full / seed 论文笔记 | 7 个 full + 19 个 seed | [`papers/`](papers/) | 为架构决策引用人工阅读笔记。 |
 | 记忆产品笔记 | 38 个笔记 | [`products/`](products/) | 对比 memory layer、memory SDK、managed memory 和带记忆的 agent 产品。 |
 | 产品页面快照 | 37 个快照 | [`products/archives/`](products/archives/) | 在源页面变化后审计产品 claims。 |
-| Benchmark 目录 | 21 个 catalog 行 | [`benchmarks/index.md`](benchmarks/index.md) | 理解 memory benchmark、stub-backed 候选行及其 claims 来源。 |
+| Benchmark 目录 | 23 个 catalog 行 | [`benchmarks/index.md`](benchmarks/index.md) | 理解 memory benchmark、stub-backed 候选行及其 claims 来源。 |
 | Claims ledger | 结构化 YAML ledger | [`benchmarks/claims/claims.yaml`](benchmarks/claims/claims.yaml) | 区分厂商自报、论文评测、方法批评和独立复现。 |
 | 成本节省专题 | seed landscape | [`docs/cost-savings-landscape.md`](docs/cost-savings-landscape.md) | 查找能减少 token、延迟或运行成本的 agent-memory 论文、方法、代码和产品。 |
 | 综述与 taxonomy | 1 份活综述 + 6 条 meta-survey 记录 | [`docs/agent-memory-survey.md`](docs/agent-memory-survey.md) · [`docs/meta-surveys.md`](docs/meta-surveys.md) | 在选型或设计前建立领域视角。 |
@@ -72,7 +72,7 @@
 | 目标 | 先读这些 |
 |---|---|
 | 快速理解领域 | [`docs/agent-memory-survey.md`](docs/agent-memory-survey.md)，再读 [`docs/taxonomy.md`](docs/taxonomy.md) |
-| 阅读最新来源刷新 | [`docs/memory-radar-2026-07.md`](docs/memory-radar-2026-07.md)，再看 [`docs/signals.md`](docs/signals.md) |
+| 阅读最新来源刷新 | [`docs/memory-radar-2026-08.md`](docs/memory-radar-2026-08.md)，再看 [`docs/signals.md`](docs/signals.md) |
 | 找相关论文 | [`papers/index.md`](papers/index.md)，再看 [`papers/`](papers/) 下的 full note |
 | 比较记忆产品 | [`docs/products-landscape.md`](docs/products-landscape.md)、[`docs/product-memory-architectures.md`](docs/product-memory-architectures.md)、[`docs/product-architecture-diagrams.md`](docs/product-architecture-diagrams.md) |
 | 查看产品为什么入库或被拒绝 | [`docs/product-discovery-log.md`](docs/product-discovery-log.md) |
@@ -154,7 +154,7 @@ flowchart LR
 | [`docs/taxonomy.md`](docs/taxonomy.md) | 分类 agent-memory 系统和 memory-kernel 职责的共享词表。 |
 | [`docs/meta-surveys.md`](docs/meta-surveys.md) | 2025 年末到 2026 H1 的外部 meta-survey 索引。 |
 | [`docs/research-radar.md`](docs/research-radar.md) | 把论文、产品、benchmark 证据转成 ImpactReport 和 ADR 输入的工作流。 |
-| [`docs/memory-radar-2026-07.md`](docs/memory-radar-2026-07.md) | 2026-07 周更来源刷新，覆盖论文、产品、GitHub 项目和 reviewer 分流结论。 |
+| [`docs/memory-radar-2026-08.md`](docs/memory-radar-2026-08.md) | 2026-08 周更来源刷新，覆盖论文、产品、GitHub 项目和 reviewer 分流结论。 |
 | [`docs/cost-savings-landscape.md`](docs/cost-savings-landscape.md) | agent-memory token reduction、预算检索、运行成本方法、代码路径和产品实践信号专题。 |
 | [`docs/information-sources.md`](docs/information-sources.md) | 论文、产品、社区和中文信息源 catalog。 |
 | [`docs/related-work.md`](docs/related-work.md) | 发现线索归因和抓取来源记录。 |
@@ -175,11 +175,11 @@ flowchart LR
 
 | 路径 | 用途 |
 |---|---|
-| [`papers/`](papers/) | 7 个 full 论文笔记、15 个 seed 笔记和主索引 [`index.md`](papers/index.md)。 |
+| [`papers/`](papers/) | 7 个 full 论文笔记、19 个 seed 笔记和主索引 [`index.md`](papers/index.md)。 |
 | [`papers/stubs/`](papers/stubs/) | 988 个尚未 full 阅读论文的生成 stub。 |
 | [`papers/pdfs/`](papers/pdfs/) | 534 个本地 PDF，约 1.8 GB。详见下方存档策略。 |
 | [`papers/_scrape/`](papers/_scrape/) | 可复现产物：抓取脚本和 dedup JSON。 |
-| [`benchmarks/`](benchmarks/) | 21 个 benchmark catalog 行，包含协议笔记、stub-backed 候选行和笔记模板。 |
+| [`benchmarks/`](benchmarks/) | 23 个 benchmark catalog 行，包含协议笔记、stub-backed 候选行和笔记模板。 |
 | [`benchmarks/claims/`](benchmarks/claims/) | benchmark 提及、厂商自报、方法批评和复现的 usage-event ledger。 |
 | [`benchmarks/archives/`](benchmarks/archives/) | benchmark 页面、repo 或 dataset card 的可选审计快照。 |
 | [`docs/benchmarks-landscape.md`](docs/benchmarks-landscape.md) | 按能力、使用方式和证据独立性整理的 benchmark 全景。 |
diff --git a/benchmarks/claims/claims.yaml b/benchmarks/claims/claims.yaml
index d08a015..5cefc41 100644
--- a/benchmarks/claims/claims.yaml
+++ b/benchmarks/claims/claims.yaml
@@ -703,3 +703,43 @@ events:
       - benchmarks/memprobe.md
       - https://arxiv.org/abs/2606.24595
     extract_confidence: medium
+  - event_id: memsecbench-origin-2026
+    benchmark_id: memsecbench
+    source_kind: paper
+    source_id: benchmarks/memsecbench.md
+    source_date: 2026-07
+    actor: MemSecBench authors
+    actor_type: paper_authors
+    usage_type: originates_benchmark
+    claim_direction: methodological_only
+    compared_systems: []
+    reported_metrics: []
+    experimental_setup: "Memory-poisoning lifecycle benchmark from persistence to downstream consequence and selective repair; local note is seed quality."
+    reproduction_status: not_reproduced
+    evidence_level: primary_pdf
+    independence_class: survey_only
+    comparability_notes: "Origin protocol only; attack set, backend comparisons, and repair scores are not normalized in this ledger."
+    evidence_refs:
+      - benchmarks/memsecbench.md
+      - https://arxiv.org/abs/2607.27080
+    extract_confidence: medium
+  - event_id: setoka-origin-2026
+    benchmark_id: setoka
+    source_kind: paper
+    source_id: benchmarks/setoka.md
+    source_date: 2026-07
+    actor: Setoka authors
+    actor_type: paper_authors
+    usage_type: originates_benchmark
+    claim_direction: methodological_only
+    compared_systems: []
+    reported_metrics: []
+    experimental_setup: "Hierarchical user-understanding benchmark for personalized agents over heterogeneous data; local note is seed quality."
+    reproduction_status: not_reproduced
+    evidence_level: primary_pdf
+    independence_class: survey_only
+    comparability_notes: "Origin protocol only; data provenance, privacy boundary, and persona-inference labels are not normalized."
+    evidence_refs:
+      - benchmarks/setoka.md
+      - https://arxiv.org/abs/2607.27056
+    extract_confidence: medium
diff --git a/benchmarks/index.md b/benchmarks/index.md
index f55ed6f..a4a0c4d 100644
--- a/benchmarks/index.md
+++ b/benchmarks/index.md
@@ -49,6 +49,8 @@ official repository, dataset card, or independent reproduction.
 | MemSyco-Bench | seed | paper-origin | memory-induced sycophancy | scope / conflict resolution / update / valid personalization | [`memsyco-bench.md`](memsyco-bench.md) | yes |
 | MemLeak | seed | paper-origin | multimodal deletion leakage | deletion compliance / provenance / residual image leakage | [`memleak.md`](memleak.md) | yes |
 | MemDelta | seed | paper-origin | memory-evaluation baseline control | component delta / model-family sensitivity / write-path cost | [`memdelta.md`](memdelta.md) | yes |
+| MemSecBench | seed | paper-origin | memory poisoning lifecycle | persistence / consequence / selective repair | [`memsecbench.md`](memsecbench.md) | yes |
+| Setoka | seed | paper-origin | hierarchical user understanding | explicit retrieval / abstract persona inference / heterogeneous data grounding | [`setoka.md`](setoka.md) | yes |
 
 ## Evidence Ledgers
 
diff --git a/benchmarks/memsecbench.md b/benchmarks/memsecbench.md
new file mode 100644
index 0000000..919a345
--- /dev/null
+++ b/benchmarks/memsecbench.md
@@ -0,0 +1,72 @@
+---
+title: MemSecBench — memory poisoning persistence-to-repair benchmark
+benchmark_id: memsecbench
+name: MemSecBench
+aliases:
+  - MemSecBench
+status: seed
+origin_type: paper_origin
+origin_source: https://arxiv.org/abs/2607.27080
+first_public_date: 2026-07
+domain: memory_security
+modality: text
+task_grain: poisoning_persistence_consequence_repair
+capability_axes:
+  - persistence
+  - downstream_consequence
+  - selective_repair
+  - memory_backend_comparison
+data_nature: attack_and_repair_tasks
+metrics:
+  - persistence_rate
+  - consequence_rate
+  - repair_success
+judge_type: source_protocol_check
+code_available: check
+data_available: check
+license: check
+known_limitations:
+  - seed note; attack set, backend setup, and repair metrics need full read
+canonical_sources:
+  - https://arxiv.org/abs/2607.27080
+confidence: medium
+memory_modules:
+  - evaluator-benchmark
+  - policy-privacy
+  - memorydiff-generator
+last_revised: 2026-08-03
+---
+
+# MemSecBench
+
+## What It Measures
+
+MemSecBench evaluates memory poisoning across the full lifecycle from
+persistence, to downstream action consequence, to selective repair.
+
+## Protocol
+
+The arXiv abstract frames the benchmark around malicious instructions that enter
+long-term memory, are recalled later, and influence real actions. It emphasizes
+tracking the same malicious semantics across persistence, consequence, and
+repair, with comparisons across memory backends.
+
+## Baselines and Reported Results
+
+No normalized scores are logged here. Any reported backend comparison remains a
+paper-origin claim until the setup is fully mapped.
+
+## Validity / Contamination / License Caveats
+
+Seed note only. The attack corpus, memory backends, repair mechanism, and judge
+protocol must be reviewed before use as a production security gate.
+
+## Related Papers
+
+- Paper:https://arxiv.org/abs/2607.27080
+
+## Impact Use
+
+- `policy-privacy`:candidate benchmark for persistent memory poisoning.
+- `memorydiff-generator`:repair semantics may inform selective rollback.
+- Ready for ImpactReport:no, upgrade after full protocol read.
diff --git a/benchmarks/setoka.md b/benchmarks/setoka.md
new file mode 100644
index 0000000..e9b4be8
--- /dev/null
+++ b/benchmarks/setoka.md
@@ -0,0 +1,73 @@
+---
+title: Setoka — hierarchical user-understanding benchmark
+benchmark_id: setoka
+name: Setoka
+aliases:
+  - Setoka
+status: seed
+origin_type: paper_origin
+origin_source: https://arxiv.org/abs/2607.27056
+first_public_date: 2026-07
+domain: personalized_agent_memory
+modality: heterogeneous_user_data
+task_grain: hierarchical_user_understanding
+capability_axes:
+  - explicit_fact_retrieval
+  - abstract_persona_inference
+  - heterogeneous_data_grounding
+  - personalization
+data_nature: heterogeneous_personal_data_tasks
+metrics:
+  - retrieval_accuracy
+  - persona_understanding
+  - inference_quality
+judge_type: source_protocol_check
+code_available: check
+data_available: check
+license: check
+known_limitations:
+  - seed note; data provenance, privacy boundary, and task construction need full read
+canonical_sources:
+  - https://arxiv.org/abs/2607.27056
+confidence: medium
+memory_modules:
+  - evaluator-benchmark
+  - retriever-reranker
+  - policy-privacy
+last_revised: 2026-08-03
+---
+
+# Setoka
+
+## What It Measures
+
+Setoka evaluates whether personalized agents can infer hierarchical user
+understanding from heterogeneous data, rather than only retrieve facts explicitly
+stated in conversations.
+
+## Protocol
+
+The arXiv abstract positions Setoka as a benchmark for deeper user
+understanding: explicit memory retrieval plus abstract personal characteristics
+needed for personalized assistance.
+
+## Baselines and Reported Results
+
+No normalized results are recorded. Treat any paper scores as origin-protocol
+claims until the benchmark setup and data construction are fully reviewed.
+
+## Validity / Contamination / License Caveats
+
+Personalized memory benchmarks can encode sensitive user data assumptions. Full
+review should check data provenance, consent model, privacy risks, and whether
+abstract-persona labels are stable enough for kernel decisions.
+
+## Related Papers
+
+- Paper:https://arxiv.org/abs/2607.27056
+
+## Impact Use
+
+- `evaluator-benchmark`:candidate protocol for user-modeling depth.
+- `policy-privacy`:privacy and consent caveats are first-order.
+- Ready for ImpactReport:no, upgrade after full protocol read.
diff --git a/docs/README.md b/docs/README.md
index 2a6837d..8784647 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -15,7 +15,7 @@ decide what to read before opening the paper index or the product notes.
 4. [`weekly-memory-refresh-runbook.md`](weekly-memory-refresh-runbook.md) —
    Codex weekly refresh runbook for papers, products, GitHub discovery, and
    benchmarks.
-5. [`memory-radar-2026-07.md`](memory-radar-2026-07.md) — latest current-source
+5. [`memory-radar-2026-08.md`](memory-radar-2026-08.md) — latest current-source
    refresh across papers, products, GitHub projects, and reviewer decisions.
 6. [`cost-savings-landscape.md`](cost-savings-landscape.md) — focused map of
    agent-memory papers, algorithms, code paths, and products that reduce token,
@@ -49,7 +49,7 @@ decide what to read before opening the paper index or the product notes.
 |---|---|
 | [`research-radar.md`](research-radar.md) | Generic Radar loop: paper/product/benchmark -> ResearchItem or BenchmarkItem -> ImpactReport -> sandbox -> ADR. |
 | [`weekly-memory-refresh-runbook.md`](weekly-memory-refresh-runbook.md) | Weekly Codex automation contract for source search, subagent review, verification, and PR output. |
-| [`memory-radar-2026-07.md`](memory-radar-2026-07.md) | 2026-07 weekly refresh with must-add, update-existing, watchlist, adjacent, and reject decisions. |
+| [`memory-radar-2026-08.md`](memory-radar-2026-08.md) | 2026-08 weekly refresh with must-add, update-existing, watchlist, adjacent, and reject decisions. |
 | [`information-sources.md`](information-sources.md) | Source catalog for papers, products, communities, and zh-CN information channels. |
 | [`related-work.md`](related-work.md) | Positioning against sibling agent-memory awesome-lists. |
 
diff --git a/docs/benchmarks-landscape.md b/docs/benchmarks-landscape.md
index 46d7676..1d1925a 100644
--- a/docs/benchmarks-landscape.md
+++ b/docs/benchmarks-landscape.md
@@ -35,6 +35,8 @@ language: zh-CN
 | memory-induced sycophancy | [`MemSyco-Bench`](../benchmarks/memsyco-bench.md) | whether retrieved memory should influence factual reasoning, conflicts, updates, and personalization | seed,origin paper logged; results/resources not normalized |
 | multimodal deletion leakage | [`MemLeak`](../benchmarks/memleak.md) | residual recovery after deletion via correlated text and retained images | seed,origin paper logged; image/data/setup not normalized |
 | baseline-control methodology | [`MemDelta`](../benchmarks/memdelta.md) | component-controlled memory-vs-RAG/full-context evaluation and write-path cost discipline | seed,methodology event logged; not an end-agent leaderboard |
+| memory poisoning lifecycle | [`MemSecBench`](../benchmarks/memsecbench.md) | persistence, downstream consequence, and selective repair of malicious memory semantics | seed,origin paper logged; attack/backend/repair setup not normalized |
+| hierarchical user understanding | [`Setoka`](../benchmarks/setoka.md) | explicit facts plus abstract personal characteristics over heterogeneous user data | seed,origin paper logged; data/privacy/task setup not normalized |
 
 ## B. 初始交叉统计
 
@@ -61,6 +63,8 @@ language: zh-CN
 | MemSyco-Bench | 1 | origin paper logged; memory-induced sycophancy protocol not yet normalized |
 | MemLeak | 1 | origin paper logged; multimodal deletion-leakage protocol not yet normalized |
 | MemDelta | 1 | methodology event logged; use for claims discipline, not direct benchmark ranking |
+| MemSecBench | 1 | origin paper logged; poisoning persistence/consequence/repair protocol not yet normalized |
+| Setoka | 1 | origin paper logged; hierarchical personalization protocol not yet normalized |
 
 ### B2. Evaluation uses / baseline comparisons
 
@@ -106,6 +110,8 @@ language: zh-CN
 | MemoryAgentBench | 1 | Survey mention only |
 | DynamicMem | 0 | Origin protocol only |
 | MEMPROBE | 0 | Origin protocol only |
+| MemSecBench | 0 | Origin protocol only |
+| Setoka | 0 | Origin protocol only |
 
 ### B7. Independent or methodological pressure
 
@@ -160,7 +166,11 @@ enters the catalog.
    becomes a kernel evaluation priority.
 9. Upgrade GateMem if shared-memory governance or enterprise scoped recall becomes
    a kernel priority.
-10. Add independent reproduction rows only when the source gives enough setup
+10. Upgrade MemSecBench if persistent memory poisoning repair becomes a kernel
+    security priority.
+11. Upgrade Setoka if hierarchical user understanding becomes a personalization
+    benchmark priority.
+12. Add independent reproduction rows only when the source gives enough setup
    detail to distinguish reruns from marketing summaries.
 
 ## F. Maintenance Contract
diff --git a/docs/memory-radar-2026-08.md b/docs/memory-radar-2026-08.md
new file mode 100644
index 0000000..d7be68d
--- /dev/null
+++ b/docs/memory-radar-2026-08.md
@@ -0,0 +1,95 @@
+---
+title: 2026-08 Memory Radar refresh
+date: 2026-08-03
+status: current-source-refresh
+language: zh-CN
+---
+
+# 2026-08 Memory Radar refresh
+
+本页记录 2026-08-03 的 weekly radar refresh。主 agent 从最新 `origin/main`
+创建 `codex/weekly-memory-radar-2026-08-03`,并用 paper/product/GitHub discovery
+子 agent 做候选检索。所有 GitHub/list/catalog 信号只作为 discovery;最终收录只依赖
+primary paper/product sources。
+
+注意:2026-07-27 周更 PR `#18` 仍未合并到 `origin/main`,其中 Databricks Managed
+Agent Memory、Memora 和多条 7 月论文 seed 已在开 PR 中。本轮不重复这些 PR `#18`
+条目,只记录相对当前 `origin/main` 的新候选和官方产品行为更新。
+
+## 执行模型
+
+| Lane | 角色 | 输出 |
+|---|---|---|
+| papers/conferences | `researcher` + main agent | arXiv primary paper 候选与 duplicate 风险 |
+| products/platforms | `researcher` | 官方产品文档、release notes、changelog |
+| GitHub/benchmarks | `researcher` | repo/list/dataset discovery signals only |
+| repo-map | `explore` | counts、landing files、现有条目和别名风险 |
+| source/relevance review | main agent + later reviewer | must-add / update-existing / watchlist / adjacent / reject |
+
+## Must-add / update-existing
+
+### Papers and benchmarks
+
+| Action | Item | Why it matters | Local anchor |
+|---|---|---|---|
+| must-add | Zero-Mem | 把 memory write/read path 的 LLM 调用成本单独提出,要求评估 memory operations 的 token/latency 开销 | [`../papers/zero-mem-zero-token-memory-operations.md`](../papers/zero-mem-zero-token-memory-operations.md) |
+| must-add | Memory Provenance Laundering | 说明 consolidation 可能把低可信 observation 洗成高可信 user/workflow history,补 source-authority 生命周期维度 | [`../papers/memory-provenance-laundering.md`](../papers/memory-provenance-laundering.md) |
+| must-add | MemHarness | 从 replay 走向 reconstructed memory,要求检索结果匹配当前 decision state 而不是只看语义相似度 | [`../papers/memharness-memory-reconstructed-not-replayed.md`](../papers/memharness-memory-reconstructed-not-replayed.md) |
+| must-add | Filesystem-Based Memory | 把 Markdown/file-tree memory 作为真实 agent substrate,强调组织漂移、冲突、过期和可持续性 | [`../papers/filesystem-based-memory-llm-agents.md`](../papers/filesystem-based-memory-llm-agents.md) |
+| must-add | MemSecBench | 追踪 memory poisoning 从持久化到后续 action consequence 再到 selective repair | [`../benchmarks/memsecbench.md`](../benchmarks/memsecbench.md) |
+| must-add | Setoka | 个性化 agent 评测从显式事实召回扩展到异构数据上的层级用户理解 | [`../benchmarks/setoka.md`](../benchmarks/setoka.md) |
+
+### Products
+
+| Action | Product | Why it matters | Local anchor |
+|---|---|---|---|
+| update-existing | Google Agent Platform Memory Bank | 官方 setup docs 继续把 topics、TTL、自定义配置和 Agent Runtime read/write 集成作为 Memory Bank 使用面;仍只支持 product behavior claim | [`../products/google-memory-bank.md`](../products/google-memory-bank.md) |
+| update-existing | Mem0 | 官方 changelog 出现 n8n / Zapier integration,说明 workflow automation memory 接入面扩大;不是架构或 benchmark 结论 | [`../products/mem0.md`](../products/mem0.md) |
+| update-existing | Zep | 官方 changelog 给 graph node/edge valid-time 排序、opaque page token 和 `is_null` date filter,加强 temporal graph query control | [`../products/zep.md`](../products/zep.md) |
+| update-existing | AWS AgentCore Memory | Bedrock Agents Classic maintenance-mode docs 与 AgentCore release notes 继续把 AgentCore memory 作为迁移/新建 agent runtime 能力面 | [`../products/aws-agentcore-memory.md`](../products/aws-agentcore-memory.md) |
+| update-existing / watchlist | TencentDB Agent Memory | v2 beta/team memory 信号相关,但主 release 在 2026-07-21/22;本轮只记录为 watch/update signal,不升级性能 claims | [`../products/tencentdb-agent-memory.md`](../products/tencentdb-agent-memory.md) |
+| update-existing | Hindsight | official `v0.8.6` release 增加 `list_memory_units` ingest-age filtering,作为 API surface / lifecycle 查询信号 | [`../products/hindsight.md`](../products/hindsight.md) |
+
+## Watchlist / adjacent / reject
+
+| Decision | Item | Reason |
+|---|---|---|
+| watchlist | MemTX / MemTxn / ChronoMem | transactional commit、source-supported updates、semantic rollback 与 memory governance 强相关,但本轮避免一次性扩大 seed 数;下轮优先精读 |
+| watchlist | MemLens / MemChain / Know It, Act on It / Beyond Retrieval Analytic Memory | 都是强候选,分别覆盖 interactive analytics、interpretable traces、preference utilization、multimodal analytic memory;先记录,等待 full source read |
+| watchlist | Databricks Managed Agent Memory / Memora | 已在未合并 PR `#18` 中覆盖,本轮不重复写入当前分支 |
+| watchlist | OWASP Agent Memory Guard / Mnemoverse MCP Memory / AMBIENT | GitHub/API/catalog discovery 相关,且 OWASP/Mnemoverse 有 primary follow-up;本轮未完成 product note 或 benchmark note source read |
+| watchlist | Letta Memory Filesystem | 官方 changelog 页缺少可核验日期;功能相关但不按本周窗口升级 |
+| watchlist | inspeximus / ai-memory-mcp / memgres / sqlite-graph-memory | GitHub-only 或 vendor/self-claimed benchmark 信号;先保留 discovery,不升级为产品/benchmark 证据 |
+| adjacent | Microsoft Foundry Local Agentic Retrieval | 2026-07 release note 说 agentic memory management for long conversations,但与既有 Foundry Agent Service Memory surface 边界不清 |
+| adjacent | HAM-VLN / TransMem / hidden-state memory items | 与 memory 有关,但偏 embodied navigation、hidden-state/context compression 或模型内部记忆,不是本轮 core agent-memory evidence |
+| reject as performance evidence | GitHub stars, README benchmark numbers, vendor benchmark rows, catalog placement | 只能用于 discovery 或 vendor/product behavior,不能支持性能/质量结论 |
+| source mismatch | no new mismatched canonical URL promoted | 本轮收录条目均使用直接 arXiv 或官方产品 URL;未把 search snippet 日期当作 source fact |
+
+## Evidence gaps / next verification
+
+| Item | Gap | Why not blocking |
+|---|---|---|
+| Zero-Mem / provenance laundering / MemHarness / filesystem memory | 尚未 full read PDF、code/data/license 和 exact setup | 本轮只登记 seed note,不登记 score claim |
+| MemSecBench / Setoka | benchmark task construction、metrics、data provenance 和 reported result 未完整归一化 | claims ledger 只记录 origin protocol event |
+| Product updates | official docs/changelog support product behavior only | 未写任何 independent benchmark、quality 或 superiority claim |
+| PR `#18` overlap | open July 27 PR 未合并 | 本轮显式排除重复条目,避免 current-main 分支冲突扩大 |
+
+## Trend synthesis
+
+- **Memory governance is moving into transaction/provenance territory**:provenance
+  laundering、MemSecBench、MemTX/MemTxn/ChronoMem watchlist 都把 memory 从"召回"
+  推向 authority、commit、rollback、repair。
+- **Memory cost needs write/read-path accounting**:Zero-Mem 和 MemDelta 类工作都说明
+  final-answer tokens 不是 memory 系统总成本。
+- **Human-editable substrates are becoming first-class**:filesystem memory 与
+  Letta Memory Filesystem 信号共同强化 Markdown/file-tree memory 的工程现实性。
+- **Personalization benchmarks are getting deeper**:Setoka 和 Know It, Act on It
+  把个性化从事实召回推进到用户理解与行动使用。
+
+## Review notes
+
+- 收录项均要求 primary source:论文用 arXiv,产品用官方 docs / changelog / release notes。
+- GitHub stars、README 性能数字、MCP catalog 和 awesome-list placement 只作 discovery。
+- Author-reported benchmark results are paper-origin claims;vendor numbers are vendor claims。
+- 本轮 benchmark catalog 从 21 行增至 23 行;paper seed notes 从 15 增至 19;product
+  note count 不变,Hindsight 等只做 update-existing。
diff --git a/docs/product-discovery-log.md b/docs/product-discovery-log.md
index 6ce0064..dedb869 100644
--- a/docs/product-discovery-log.md
+++ b/docs/product-discovery-log.md
@@ -156,3 +156,19 @@ language: zh-CN
 | update-existing | Mem0 | changelog highlights 2026-06-27 | 更新 `products/mem0.md`;expiration controls 作为 lifecycle signal |
 | update-existing | Redis Agent Memory Server | Redis blog 2026-07-01 | 更新 `products/redis-agent-memory-server.md`;作为产品定位/实现建议,非 benchmark |
 | watchlist | Cloudflare Think harness | Cloudflare docs | 暂不拆产品;与 Cloudflare Agent Memory 有重叠,先观察是否形成独立 memory product |
+
+## 11. 2026-08-03 weekly refresh delta
+
+| Decision | Item | Source | Action |
+|---|---|---|---|
+| update-existing | Google Agent Platform Memory Bank | setup docs / Memory Bank docs | 更新 `products/google-memory-bank.md`;topics、TTL、自定义配置和 Agent Runtime read/write 集成是产品行为证据 |
+| update-existing | Mem0 | changelog highlights | 更新 `products/mem0.md`;n8n / Zapier workflow integration 是接入面扩展,不是 benchmark |
+| update-existing | Zep | official changelog | 更新 `products/zep.md`;valid-time ordering、opaque page token、`is_null` date filter 强化 temporal graph query control |
+| update-existing | AWS AgentCore Memory | Bedrock Agents Classic maintenance docs / AgentCore release notes | 更新 `products/aws-agentcore-memory.md`;迁移与 runtime memory surface 是 vendor product behavior |
+| update-existing / watchlist | TencentDB Agent Memory | GitHub release / README_CN / Tencent Cloud article | 更新 `products/tencentdb-agent-memory.md`;v2 beta/team memory 是相关信号,性能/benchmark claims 仍不升级 |
+| update-existing | Hindsight | GitHub release `v0.8.6` | 更新 `products/hindsight.md`;ingest-age memory-unit filtering 是产品/API 行为,benchmark 数字仍是 vendor/affiliated claim |
+| watchlist | OWASP Agent Memory Guard | OWASP project page + GitHub | memory poisoning guard 相关,但 roadmap/评测证据不足,不新增产品计数 |
+| watchlist | Mnemoverse MCP Memory | official docs + GitHub | MCP persistent memory 相关,但本轮未完成 product note/source read |
+| watchlist | AMBIENT / inspeximus / ai-memory-mcp / memgres / sqlite-graph-memory | GitHub/site discovery | benchmark 或 MCP memory 信号,但多为 GitHub-only 或 vendor self-claim,不升级 |
+| watchlist | Letta Memory Filesystem | official changelog | 功能相关,但本轮未核验日期;不按 weekly window 升级 |
+| adjacent | Redis / Cloudflare / Microsoft Foundry Local signals | official docs/blogs | 没有清晰新 core memory product delta,保留为 discovery/watchlist |
diff --git a/docs/products-landscape.md b/docs/products-landscape.md
index b70dbc4..1c74bc7 100644
--- a/docs/products-landscape.md
+++ b/docs/products-landscape.md
@@ -68,6 +68,14 @@ language: zh-CN
 
 ## A. 按领域分类
 
+### 2026-08 source refresh note
+
+2026-08-03 周更只更新 existing product behavior,不新增核心产品计数。Google
+Memory Bank、Mem0、Zep、AWS AgentCore Memory、TencentDB Agent Memory 和 Hindsight 的变化分别
+记录在产品笔记与 [`product-discovery-log.md`](product-discovery-log.md)。这些变化
+来自官方 docs / changelog / release sources,只能支持 product behavior 或 vendor
+self-report,不能支持独立 benchmark 结论。
+
 ### A1. Agent memory 专门层(memory-as-a-product)
 
 | 名称 | 笔记 | Mode | Audience | 一句话 |
diff --git a/docs/signals.md b/docs/signals.md
index e044e5b..1e5dbda 100644
--- a/docs/signals.md
+++ b/docs/signals.md
@@ -20,6 +20,9 @@ Radar 动作 enum:`stub` `seed-note` `deep-note` `impact-report` `archive-only`
 
 | 日期 | 来源 | 类型 | 一句话 | Radar 动作 |
 |---|---|---|---|---|
+| 2026-08-03 | [Zero-Mem](https://arxiv.org/abs/2607.29377) / [Memory Provenance Laundering](https://arxiv.org/abs/2607.29167) / [MemHarness](https://arxiv.org/abs/2607.28272) / [Filesystem-Based Memory](https://arxiv.org/abs/2607.26637) | paper | 周更 radar 追加 zero-token memory operations、provenance authority、reconstructed memory retrieval、filesystem memory substrate 四条 seed note | `seed-note` ✅(见 `papers/`) |
+| 2026-08-03 | [MemSecBench](https://arxiv.org/abs/2607.27080) / [Setoka](https://arxiv.org/abs/2607.27056) | paper | benchmark catalog 新增 persistent memory poisoning lifecycle 和 hierarchical user understanding 两个 origin-protocol seed | `seed-note` ✅(见 `benchmarks/`) |
+| 2026-08-03 | [Google Memory Bank setup](https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/memory-bank/setup) / [Mem0 changelog](https://docs.mem0.ai/changelog/highlights) / [Zep changelog](https://help.getzep.com/changelog) / [AWS AgentCore release notes](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html) / [Hindsight v0.8.6](https://github.com/vectorize-io/hindsight/releases/tag/v0.8.6) | product | 官方产品源补充 Memory Bank setup/TTL/topics、Mem0 workflow integrations、Zep valid-time query controls、AWS AgentCore migration/runtime signal、Hindsight ingest-age filtering | `deep-note` ✅(更新 `products/`) |
 | 2026-07-06 | [A-TMA](https://arxiv.org/abs/2607.01935) / [Mandol](https://arxiv.org/abs/2606.29778) / [Forensic Trajectory Signatures](https://arxiv.org/abs/2606.30566) | paper | 周更 radar 追加 state-aware ghost memory、agglomerative memory-native storage、memory-poisoning trajectory forensics 三条 seed note | `seed-note` ✅(见 `papers/`) |
 | 2026-07-06 | [MemSyco-Bench](https://arxiv.org/abs/2607.01071) / [MemLeak](https://arxiv.org/abs/2606.29788) / [MemDelta](https://arxiv.org/abs/2606.29914) | paper | benchmark catalog 新增 memory-induced sycophancy、多模态删除泄漏、controlled baseline methodology | `seed-note` ✅(见 `benchmarks/`) |
 | 2026-07-06 | [AWS AgentCore release notes](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html) / [Google release notes](https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes) / [OpenAI Business release notes](https://help.openai.com/en/articles/11391654-chatgpt-business-release-notes) / [Mem0 changelog](https://docs.mem0.ai/changelog/highlights) / [Redis guide](https://redis.io/blog/build-smarter-ai-agents-manage-short-term-and-long-term-memory-with-redis/) | product | 官方产品源补充 AgentCore streaming、Memory Bank model default、ChatGPT org memory controls、Mem0 expiration 和 Redis two-tier guidance | `deep-note` ✅(更新 `products/`) |
diff --git a/papers/filesystem-based-memory-llm-agents.md b/papers/filesystem-based-memory-llm-agents.md
new file mode 100644
index 0000000..938cb90
--- /dev/null
+++ b/papers/filesystem-based-memory-llm-agents.md
@@ -0,0 +1,61 @@
+---
+title: Filesystem-Based Memory for LLM Agents
+arxiv_id: 2607.26637
+source: arXiv:2607.26637
+date: 2026-07
+domain: systems
+core_claim: |
+  Deployed agents increasingly use filesystem trees, often Markdown files, as
+  durable memory. The paper frames organization, evolution, stale/conflicting
+  memory, and sustainability as first-class evaluation questions for this
+  substrate.
+evidence_level: medium
+code_available: check
+license: check
+memory_modules:
+  - parser-chunker
+  - ingest-adapter
+  - semantic-dedup
+  - audit-ui
+status: seed
+last_revised: 2026-08-03
+urls:
+  - https://arxiv.org/abs/2607.26637
+---
+
+# Filesystem-Based Memory for LLM Agents
+
+## Problem statement
+
+Many practical agents maintain memory as a directory tree of human-readable
+files. The arXiv abstract argues that research has under-tested the default
+assumption that agents can keep such stores organized as memories accumulate,
+conflict, and go stale.
+
+## Core claim
+
+Filesystem memory is a memory substrate, not just an implementation detail. It
+needs its own organization, evolution, and sustainability evaluation.
+
+## Decision relevance
+
+- `parser-chunker`:Markdown and file-tree boundaries become memory-unit
+  boundaries.
+- `semantic-dedup`:conflict, staleness, and reorganization are central quality
+  dimensions for file memory.
+- `audit-ui`:human-editable memory surfaces create audit advantages that graph
+  or vector-only stores may not have.
+
+## Caveats
+
+Seed quality only. This note does not yet validate the benchmark setup, agent
+workloads, code, or any reported improvement.
+
+## Sources
+
+- arXiv:https://arxiv.org/abs/2607.26637
+
+---
+
+> *Ymem project binding: see
+> [`../docs/ymem-binding/relevance-index.md`](../docs/ymem-binding/relevance-index.md).*
diff --git a/papers/index.md b/papers/index.md
index 3a0a8c8..f298dcb 100644
--- a/papers/index.md
+++ b/papers/index.md
@@ -19,6 +19,18 @@ removed.
 - By year: 2026=388, 2025=309, 2024=129, 2023=72, 2022=6, 2021=4, 2020=3, 2018=2, 2017=1, undated=75
 - By source-count: 6 sources=6, 5 sources=18, 4 sources=32, 3 sources=66, 2 sources=127, 1 sources=740
 
+## 2026-08 manual radar additions
+
+These entries were added by the 2026-08-03 weekly radar refresh. They are not
+part of the 2026-05-19 nine-list scrape statistics above.
+
+- [Zero-Mem: Zero-Token Memory Operations for LLM Agents](zero-mem-zero-token-memory-operations.md) — 2026-07 — seed — [arxiv](https://arxiv.org/abs/2607.29377)
+- [Memory Provenance Laundering in LLM Agents](memory-provenance-laundering.md) — 2026-07 — seed — [arxiv](https://arxiv.org/abs/2607.29167)
+- [MemHarness: Memory Is Reconstructed, Not Replayed](memharness-memory-reconstructed-not-replayed.md) — 2026-07 — seed — [arxiv](https://arxiv.org/abs/2607.28272)
+- [Filesystem-Based Memory for LLM Agents](filesystem-based-memory-llm-agents.md) — 2026-07 — seed — [arxiv](https://arxiv.org/abs/2607.26637)
+- [MemSecBench: Tracking Agent Memory Poisoning](../benchmarks/memsecbench.md) — 2026-07 — benchmark seed — [arxiv](https://arxiv.org/abs/2607.27080)
+- [Setoka: Hierarchical User Understanding in Personalized Agents](../benchmarks/setoka.md) — 2026-07 — benchmark seed — [arxiv](https://arxiv.org/abs/2607.27056)
+
 ## 2026-07 manual radar additions
 
 These entries were added by the 2026-07-06 weekly radar refresh. They are not
diff --git a/papers/memharness-memory-reconstructed-not-replayed.md b/papers/memharness-memory-reconstructed-not-replayed.md
new file mode 100644
index 0000000..f03c49b
--- /dev/null
+++ b/papers/memharness-memory-reconstructed-not-replayed.md
@@ -0,0 +1,61 @@
+---
+title: MemHarness: Memory Is Reconstructed, Not Replayed
+arxiv_id: 2607.28272
+source: arXiv:2607.28272
+date: 2026-07
+domain: retrieval
+core_claim: |
+  Retrieved experiences should be adapted to the current decision state instead
+  of replayed verbatim. The paper frames memory retrieval as reconstruction
+  across abstraction gaps between stored experience and current task state.
+evidence_level: medium
+code_available: check
+license: check
+memory_modules:
+  - retriever-reranker
+  - dream-consolidator
+  - evaluator-benchmark
+status: seed
+last_revised: 2026-08-03
+urls:
+  - https://arxiv.org/abs/2607.28272
+---
+
+# MemHarness: Memory Is Reconstructed, Not Replayed
+
+## Problem statement
+
+Experience retrieval is often implemented as replay: fetch a prior trace and
+inject it into context. The arXiv abstract argues that this can cause negative
+transfer when the stored experience is abstract or mismatched to the current
+state.
+
+## Core claim
+
+MemHarness treats useful memory as reconstructed for the current decision
+situation. This makes it relevant to retrieval/reranking and consolidation
+design, because memory usefulness depends on state alignment rather than only
+semantic similarity.
+
+## Decision relevance
+
+- `retriever-reranker`:retrieval should test applicability to the current state,
+  not only nearest-neighbor similarity.
+- `dream-consolidator`:stored experiences may need abstracted, reusable forms
+  that can be reconstructed rather than replayed.
+- `evaluator-benchmark`:future tests should separate helpful transfer from
+  harmful memory replay.
+
+## Caveats
+
+Seed quality only. Full task suite, baselines, and reported results are not
+normalized in this repo.
+
+## Sources
+
+- arXiv:https://arxiv.org/abs/2607.28272
+
+---
+
+> *Ymem project binding: see
+> [`../docs/ymem-binding/relevance-index.md`](../docs/ymem-binding/relevance-index.md).*
diff --git a/papers/memory-provenance-laundering.md b/papers/memory-provenance-laundering.md
new file mode 100644
index 0000000..4435f4e
--- /dev/null
+++ b/papers/memory-provenance-laundering.md
@@ -0,0 +1,62 @@
+---
+title: Memory Provenance Laundering in LLM Agents
+arxiv_id: 2607.29167
+source: arXiv:2607.29167
+date: 2026-07
+domain: security_privacy
+core_claim: |
+  Long-term memory consolidation can rewrite low-trust observations into
+  apparently trusted user history or workflow support. The paper frames this as
+  provenance laundering and proposes a non-amplification firewall for persistent
+  memory.
+evidence_level: medium
+code_available: check
+license: check
+memory_modules:
+  - policy-privacy
+  - ingest-adapter
+  - memorydiff-generator
+  - audit-ui
+status: seed
+last_revised: 2026-08-03
+urls:
+  - https://arxiv.org/abs/2607.29167
+---
+
+# Memory Provenance Laundering in LLM Agents
+
+## Problem statement
+
+Persistent memory lets untrusted observations survive across sessions. The
+arXiv abstract identifies a failure mode where consolidation rewrites external
+observations as trusted history while preserving the downstream action trigger.
+
+## Core claim
+
+The paper argues that filters, sanitizers, and tool guards do not by themselves
+enforce source authority after a fact has been consolidated into memory. The
+proposed non-amplification firewall is decision-relevant because it treats
+source authority as a memory-lifecycle invariant.
+
+## Decision relevance
+
+- `policy-privacy`:memory records need source-authority metadata that survives
+  extraction and consolidation.
+- `memorydiff-generator`:diff review should expose authority upgrades, not only
+  textual changes.
+- `audit-ui`:review surfaces should make source laundering visible before a
+  persistent record affects action selection.
+
+## Caveats
+
+Seed quality only. Do not use any attack success rate, defense result, or
+benchmark claim until the PDF and setup are fully mapped.
+
+## Sources
+
+- arXiv:https://arxiv.org/abs/2607.29167
+
+---
+
+> *Ymem project binding: see
+> [`../docs/ymem-binding/relevance-index.md`](../docs/ymem-binding/relevance-index.md).*
diff --git a/papers/zero-mem-zero-token-memory-operations.md b/papers/zero-mem-zero-token-memory-operations.md
new file mode 100644
index 0000000..e90588b
--- /dev/null
+++ b/papers/zero-mem-zero-token-memory-operations.md
@@ -0,0 +1,60 @@
+---
+title: Zero-Mem: Zero-Token Memory Operations for LLM Agents
+arxiv_id: 2607.29377
+source: arXiv:2607.29377
+date: 2026-07
+domain: systems
+core_claim: |
+  Agent memory operations can avoid extra LLM calls outside final answering by
+  using deterministic structured access paths. The paper frames memory cost as a
+  write/read-path systems problem, not only a retrieval-quality problem.
+evidence_level: medium
+code_available: check
+license: check
+memory_modules:
+  - ingest-adapter
+  - retriever-reranker
+  - evaluator-benchmark
+status: seed
+last_revised: 2026-08-03
+urls:
+  - https://arxiv.org/abs/2607.29377
+---
+
+# Zero-Mem: Zero-Token Memory Operations for LLM Agents
+
+## Problem statement
+
+Many agent-memory systems use LLM calls to extract, summarize, merge, or mediate
+memory before the final answer. The arXiv abstract argues that these operations
+create recurring token and latency cost, and may obscure the original evidence
+behind intermediate memory records.
+
+## Core claim
+
+Zero-Mem proposes zero-token memory operations: no LLM invocation outside final
+question answering for memory access. This seed note records the idea as a
+systems-design pressure on memory kernels, not as a reproduced efficiency claim.
+
+## Decision relevance
+
+- `ingest-adapter`:raises the bar for when a memory write actually needs an LLM
+  extraction step.
+- `retriever-reranker`:useful baseline for deterministic retrieval or structured
+  access paths before adding LLM-mediated memory operations.
+- `evaluator-benchmark`:future cost evaluations should count write/read-path LLM
+  calls, not only final-answer tokens.
+
+## Caveats
+
+Seed quality only. Full protocol, compared systems, code, and any reported
+cost/quality results need source read before use in ImpactReports.
+
+## Sources
+
+- arXiv:https://arxiv.org/abs/2607.29377
+
+---
+
+> *Ymem project binding: see
+> [`../docs/ymem-binding/relevance-index.md`](../docs/ymem-binding/relevance-index.md).*
diff --git a/products/aws-agentcore-memory.md b/products/aws-agentcore-memory.md
index e615fd9..ae53b01 100644
--- a/products/aws-agentcore-memory.md
+++ b/products/aws-agentcore-memory.md
@@ -12,7 +12,7 @@ memory_modules:
   - retriever-reranker
   - policy-privacy
 status: seed
-last_revised: 2026-07-06
+last_revised: 2026-08-03
 archive: archives/aws-agentcore-memory-overview.md
 ---
 
@@ -58,6 +58,13 @@ record streaming 作为当前能力面:memory record create / update / delete 
 Kinesis,用于下游审计、同步或增量处理。该能力是产品行为证据,可支持"AgentCore Memory
 暴露 evented memory lifecycle"这一判断,但不支持任何独立性能结论。
 
+## 3.3 2026-08 refresh
+
+2026-08-03 周更复核时,AWS Bedrock Agents Classic maintenance-mode docs 建议新建或
+迁移工作负载使用 AgentCore,而 AgentCore release notes / developer guide 继续把
+memory 作为 agent runtime 的内建或 BYO 能力面。该信号支持"AgentCore Memory 是 AWS
+agent runtime 迁移路径的一部分"这一产品行为判断,不支持性能或质量结论。
+
 ## 4. 决策相关性 / Decision relevance
 
 - **对照点**:AWS 把 memory 当成 agent runtime 的可配置云资源,说明 hyperscaler
@@ -87,6 +94,7 @@ Kinesis,用于下游审计、同步或增量处理。该能力是产品行为证
 - Metadata:https://aws.amazon.com/about-aws/whats-new/2026/05/agentcore-longterm-memory-metadata/
 - Harness GA:https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-harness-is-now-generally-available-go-from-idea-to-production-grade-agent-in-minutes/
 - Release notes:https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html
+- Bedrock Agents Classic maintenance mode:https://docs.aws.amazon.com/bedrock/latest/userguide/agents-classic-maintenance-mode.html
 
 ---
 
diff --git a/products/google-memory-bank.md b/products/google-memory-bank.md
index a8eb519..7b379e5 100644
--- a/products/google-memory-bank.md
+++ b/products/google-memory-bank.md
@@ -12,7 +12,7 @@ memory_modules:
   - retriever-reranker
   - policy-privacy
 status: seed
-last_revised: 2026-07-06
+last_revised: 2026-08-03
 archive: archives/google-memory-bank-overview.md
 ---
 
@@ -55,6 +55,14 @@ generation 的默认模型从 Gemini 2.5 Flash 改为 Gemini 3.5 Flash。该更
 memory extraction / generation 仍会随平台模型配置变化;它是产品行为证据,不代表
 memory ranking 机制或质量有独立复现。
 
+## 3.3 2026-08 refresh
+
+2026-08-03 周更复核官方 setup / integration docs 时,Memory Bank 继续作为
+Gemini Enterprise Agent Platform 的 managed memory instance 暴露。文档强调可用
+topics、TTL 等 custom configurations,并可与 Agent Runtime 集成 read / write
+memory。该更新只支持"Memory Bank 提供可配置托管 memory 与 runtime 集成"这一
+产品行为判断,不支持独立质量或 benchmark 结论。
+
 ## 4. 决策相关性 / Decision relevance
 
 - **对照点**:Memory Bank 是 hyperscaler 级 scoped memory store 的典型样本。
diff --git a/products/hindsight.md b/products/hindsight.md
index 3927fd5..70a1df9 100644
--- a/products/hindsight.md
+++ b/products/hindsight.md
@@ -13,7 +13,7 @@ memory_modules:
   - dream-consolidator
   - evaluator-benchmark
 status: seed
-last_revised: 2026-05-31
+last_revised: 2026-08-03
 archive: archives/hindsight-overview.md
 ---
 
@@ -41,6 +41,11 @@ retain / recall / reflect 三类操作,也可以作为 wrapper 接到现有 LLM
 截至 2026-05-31 查询,GitHub metadata 约 15.2k stars / 857 forks,MIT
 license,最新 release 为 `v0.7.1`(2026-05-28)。
 
+2026-08-03 周更复核到官方 GitHub release `v0.8.6`(2026-07-29)。该 release
+包括 `list_memory_units` 按 ingest age 过滤(`created_before`)以及多篇 memory
+positioning blog/docs 更新。这里仅记录为产品行为与 API surface 更新;Hindsight
+LongMemEval 相关数字仍按 vendor/affiliated claim 处理,不升级为独立复现。
+
 ## 3. 关键技术选择
 
 - **接口**:HTTP API、SDK、LLM wrapper、embedded Python server
@@ -86,6 +91,7 @@ license,最新 release 为 `v0.7.1`(2026-05-28)。
 - GitHub:https://github.com/vectorize-io/hindsight
 - Docs:https://hindsight.vectorize.io/
 - Release:https://github.com/vectorize-io/hindsight/releases/tag/v0.7.1
+- Release v0.8.6:https://github.com/vectorize-io/hindsight/releases/tag/v0.8.6
 
 ---
 
diff --git a/products/mem0.md b/products/mem0.md
index 1c561bb..76309b2 100644
--- a/products/mem0.md
+++ b/products/mem0.md
@@ -11,7 +11,7 @@ evidence_level: medium (open-source library, blog claims need independent benchm
 code_available: yes
 license: Apache 2.0
 status: seed
-last_revised: 2026-07-06
+last_revised: 2026-08-03
 ---
 
 # Mem0
@@ -123,6 +123,16 @@ search toward retention policy. It should not be mixed with benchmark claims.
 
 > 来源:https://docs.mem0.ai/changelog/highlights
 
+## 2026-08 workflow integrations
+
+Mem0 changelog highlights surfaced n8n and Zapier integrations around the
+2026-08-03 refresh window. This is product-behavior evidence that Mem0 is
+expanding durable memory into automation workflows where facts can be stored and
+recalled across tool runs. It is not evidence for memory quality, benchmark
+rankings, or architectural superiority.
+
+> 来源:https://docs.mem0.ai/changelog/highlights
+
 ## Notes
 
 (随版本更新追踪)
diff --git a/products/tencentdb-agent-memory.md b/products/tencentdb-agent-memory.md
index e6b06fa..77a0db4 100644
--- a/products/tencentdb-agent-memory.md
+++ b/products/tencentdb-agent-memory.md
@@ -14,7 +14,7 @@ memory_modules:
   - dream-consolidator
   - audit-ui
 status: seed
-last_revised: 2026-06-29
+last_revised: 2026-08-03
 archive: archives/tencentdb-agent-memory-overview.md
 ---
 
@@ -55,6 +55,15 @@ Tencent Cloud VectorDB。
 - **接入**:OpenClaw plugin、Hermes Gateway adapter、agent tools
   `tdai_memory_search` / `tdai_conversation_search`
 
+## 3.1 2026-08 watch/update signal
+
+2026-08-03 周更复核到 TencentDB Agent Memory v2 beta / team memory 方向的官方和
+仓库信号:team memory 将 Chat Memory、Skill、Wiki/Link Graph、CodeGraph 等作为可
+复用资产,并暴露 owner/version/status、sharing/equipping 以及 private/team/
+restricted/agent ACL 可见性。由于主 release 时间在 2026-07-21/22,本轮只作为
+update/watchlist 信号记录;README 或云文章里的准确率/token claims 仍按 vendor
+self-report,不升级为独立 benchmark evidence。
+
 ## 4. 决策相关性 / Decision relevance
 
 - **对照点**:它把"可调试记忆"作为核心卖点,不像纯 vector DB 只返回相似度列表。
@@ -92,6 +101,7 @@ Tencent Cloud VectorDB。
 - archive: [`archives/tencentdb-agent-memory-overview.md`](archives/tencentdb-agent-memory-overview.md)
 - 腾讯云产品页:https://cloud.tencent.com/product/agm
 - GitHub:https://github.com/TencentCloud/TencentDB-Agent-Memory
+- Releases:https://github.com/TencentCloud/TencentDB-Agent-Memory/releases
 - Historical GitHub alias:https://github.com/Tencent/TencentDB-Agent-Memory
 
 ---
diff --git a/products/zep.md b/products/zep.md
index 9bffa71..f1febef 100644
--- a/products/zep.md
+++ b/products/zep.md
@@ -12,7 +12,7 @@ memory_modules:
   - dream-consolidator
   - memorydiff-generator
 status: full
-last_revised: 2026-05-19
+last_revised: 2026-08-03
 archive: archives/zep-overview.md
 ---
 
@@ -56,6 +56,15 @@ healthcare 等行业的预置 entity schema 模板。
 > event ledger row: `zep-locomo-2026` in
 > [`../benchmarks/claims/claims.yaml`](../benchmarks/claims/claims.yaml)。
 
+## 3.1 2026-08 changelog refresh
+
+Zep official changelog entries around the 2026-08-03 refresh window added graph
+query controls such as valid-time ordering for node/edge lists, opaque page
+tokens, and `is_null` date filters. These support the product-behavior claim that
+Zep is exposing more temporal graph control to applications. They do not change
+the evidence class of Zep benchmark numbers, which remain vendor/self-claim
+unless independently reproduced.
+
 ## 4. 决策相关性 / Decision relevance
 
 - **对照点**:Zep 占据和 memory kernel 几乎相同的生态位 — host-agnostic 的记忆层。它
@@ -99,6 +108,7 @@ domain × audience 表:
 - archive: [`archives/zep-overview.md`](archives/zep-overview.md)
 - 配套笔记:[`graphiti.md`](graphiti.md)(OSS 内核)
 - 官方:https://www.getzep.com、https://help.getzep.com/(docs 已迁移)
+- Changelog:https://help.getzep.com/changelog
 
 ---