chore: daily update 2026-07-27

2026-07-27 02:21:48 +08:00
parent 70c3a74a85
commit 5e20003f93
7 changed files with 138 additions and 38 deletions
@@ -1,5 +1,5 @@
 {
-  "updatedAt": "2026-07-26T17:31:08.313Z",
+  "updatedAt": "2026-07-26T18:21:48.752Z",
  "command": "update",
  "model": "MiniMax-M3"
 }
@@ -4,7 +4,12 @@ Concrete work tasks, follow-ups, approvals, and scheduled items extracted from G
 ## Active
-_No active commitments yet._
+### china-cde-cell-gene-therapy-rules-2026: verify CDE classification scope for Chinese pharma / CDMO conversations
 - Owner: me
 - Source: web-search — cisema.com 2026-07-24 (China CDE Defines Classification Rules for Cell & Gene Therapy Drugs)
 - Due: unscheduled
 - Status: open
 - Notes: Single credible regulatory-trade source. Pull the NMPA/CDE-direct release before the next vendor sync; check whether any active CellShot-TriArm or ClinoReactor-class deals depend on the new classification scope.
 <!--
 Structure for new entries:
@@ -23,7 +23,7 @@
 - **Wiki state:** initialized — Hacker News, the local Obsidian vault, and Tavily web search are active and feeding `/sources/hackernews.md`, `/sources/git-repo.md`, `/sources/web-search.md`, and `/themes.md`.
 - **Connectors:** Hacker News, git-repo, and web search active. Gmail, Slack, Notion, and X are not yet configured — see "Adding sources" below.
- **Last update:** see [/.last-update.json](/.last-update.json) — this run was a web-search (`web-search-1`) source-specific update.
+- **Last update:** see [/.last-update.json](/.last-update.json) — this run was a 3-connector refresh (`hackernews-1` + `web-search-1` + `obsidian-robin-1`) after the 2026-07-27 `~/.openwiki/INSTRUCTIONS.md` scope file was loaded. No new vendor-anchored evidence; wiki evidence indexes updated to document the new pull coverage.
 ## How to use this wiki
@@ -15,8 +15,8 @@ the vault, not summarize their content.
 - **Ahead of origin/main:** 3 commits (the `CellShot-TriArm-Demo` plan v1 + v2
  pair, un-pushed as of this run).
 - **Auth:** local path only — no remote, no credentials
- **Latest fetch:** `2026-07-26T17:25:17Z`
+- **Latest fetch:** `2026-07-26T18:16:12Z` (no change vs. 18:12:31 prior pull)
- **Raw manifest:** `~/.openwiki/connectors/git-repo/raw/2026-07-26T17-25-10-637Z/manifest.json`
+- **Raw manifest:** `~/.openwiki/connectors/git-repo/raw/2026-07-26T18-16-06-806Z/manifest.json`
 - **Source config:** `~/.openwiki/connectors/git-repo/config.json`
 ## Working-tree status
@@ -33,7 +33,10 @@ Since the last fetch (2026-07-16T08:07Z), 6 commits have landed on
 pair (111_Planimetry + 260528_CS_vs_OS app note) on 2026-07-17, and a new
 `CellShot-TriArm` active deal cluster: kickoff meeting record on 2026-07-17
 followed by three demo-plan revisions (v1, v2 老师版, v2 纯文本版) drafted for
-a 2026-07-20 demo. Dirty working tree is unchanged from the previous run.
+a 2026-07-20 demo. Dirty working tree is unchanged across the 2026-07-26 fetches
 (17:25, 18:12, 18:16) — the touch-only `skills/openclaw-workspace` mtime and the
 untracked `3Brain_Sample_Holder_Set_PressNet_2026-07-07.md` note are the only
 working-tree deltas, with zero new commits in the last 24 h.
 ## Vault navigation index
@@ -1,17 +1,48 @@
 # Hacker News — evidence index
-Compact evidence trail for the Hacker News connector (`hackernews-1`, LLM-agent topics). Synthesis lives on `/themes.md`, `/commitments.md`, etc.; this page only records ingestion coverage and notable items.
+Compact evidence trail for the Hacker News connector (`hackernews-1`, Robin-curated life-sciences + AI workflow). Synthesis lives on `/themes.md`, `/commitments.md`, etc.; this page only records ingestion coverage and notable items.
 ## Connector
- **Instance:** `hackernews-1` (Hacker News — LLM agent topics)
+- **Instance:** `hackernews-1` (Hacker News — Robin-curated life-sciences + AI workflow)
- **Backend:** public HN feeds (`top`, `new`, `best`) + Algolia HN search queries
+- **Backend:** public HN feeds (`top`, `new`) + Algolia HN search queries
 - **Auth:** none — public feeds
 - **Window:** last 24 hours
- **Latest fetch:** `2026-07-26T17:27:12Z`
+- **Latest fetch:** `2026-07-26T18:16:43Z` (latest 18:16 pull; prior 18:00 pull also documented below)
- **Raw dump:** `~/.openwiki/connectors/hackernews/raw/2026-07-26T17-26-41-217Z/hackernews-results.json`
+- **Raw dumps:** `~/.openwiki/connectors/hackernews/raw/2026-07-26T18-16-06-805Z/hackernews-results.json` (current) · `~/.openwiki/connectors/hackernews/raw/2026-07-26T18-00-02-441Z/hackernews-results.json` (prior)
-## Ingestion coverage
+## Latest update — 2026-07-26 18:16 UTC
 - **Coverage:** 3 feeds (`top`, `new`, `best`), 17 search queries, 90 unique resolved feed stories, and 0 search-query hits.
 - **Query-set refresh:** the query lane was revised after the 2026-07-27 `~/.openwiki/INSTRUCTIONS.md` scope was loaded. New vendor-application names added (`organoids`, `ClinoStar`, `ClinoReactor`, `local LLM`, `ollama`, `deepagents`); OpenWiki personal-brain / Hermes Agent / RAGFlow / agent-memory / MCP queries were removed to avoid duplicating the web-search lane. The brief's "keep both lanes" rule is satisfied by HN's retained coverage of `codex cli`, `claude code`, `MiniMax`, `M3 model`, `openwiki`, and `3Brain`.
 - **Life-science lane:** still zero hits. `ClinoStar`, `ClinoReactor`, `organoids`, `bioprinting`, `3D bioprinter`, `Cellink`, `CelVivo`, `3Brain`, `organoid`, and `deepagents` all returned `nbHits=0`. This matches the prior pull's "no material Hacker News footprint for the 4 vendor lines" finding.
 - **Workflow lane:** `codex cli`, `claude code`, `MiniMax`, `M3 model`, `openwiki`, `local LLM`, `ollama` all returned 0 hits in this window. No story-worthy update for Hermes Agent, MiniMax, OpenWiki, or RAGFlow.
 - **Synthesis result:** no promotion to `/themes.md`, no commitment, and no core-memory open question. HN remains the wrong lane for Robin's 4 vendor product lines; Tavily (web-search) is the productive lane, as the prior pulls confirm.
 | Query lane | Queries run | Qualifying results |
 | --- | --- | --- |
 | Life sciences / vendors | `organoid`, `organoids`, `bioprinting`, `3D bioprinter`, `Cellink`, `CelVivo`, `ClinoStar`, `ClinoReactor`, `3Brain` | 0 |
 | AI workflow | `deepagents`, `openwiki`, `codex cli`, `claude code`, `MiniMax`, `M3 model`, `local LLM`, `ollama` | 0 new/promotable |
 Coverage note: 8 queries dropped vs. the 18:00 pull (`3D bioprinting`, `bioprinting clinical`, `spheroid bioreactor`, `microgravity cell culture`, `CELVIVO`, `MEA neuroscience`, `organ-on-chip`, `liver organoid`, `brain organoid`, `spheroid drug screening`, `OpenWiki personal brain`, `Hermes Agent`, `Claude Code agent workflow`, `RAGFlow`, `agent memory wiki`, `agent personal brain`, `MCP connector agent`). The removed names were either string-collision risks or already covered by the web-search connector.
 Raw dump: `~/.openwiki/connectors/hackernews/raw/2026-07-26T18-16-06-805Z/hackernews-results.json`.
 ## Latest update — 2026-07-26 18:00 UTC
 - **Coverage:** 2 feeds (`top`, `new`), 24 search queries, 53 unique resolved feed stories, and 33 unique returned search hits.
 - **Life-science lane:** no qualifying signal. `organoid` returned five false positives; `3Brain`, `Cellink`, and `liver organoid` returned string collisions; the other explicit application/vendor queries returned zero hits. No CelVivo / 3Brain / Cellink / Femtobiomed news or vendor-relevant >200-point organoid story was found.
 - **Workflow lane:** no story-worthy update for Hermes Agent, MiniMax, OpenWiki, RAGFlow, or Robin's other named tools. Codex CLI results were mostly the already-captured low-engagement wrapper/debate cluster; the only post newer than the prior 17:27 pull was a 1-point Ask HN about Codex token limits (`49060419`).
 - **Synthesis result:** no promotion to `/themes.md`, no commitment, and no core-memory open question. All returned items are either off-scope, false positives, low-engagement generic agent chatter, or evidence already represented below.
 | Query lane | Queries run | Qualifying results |
 | --- | --- | --- |
 | Life sciences / vendors | `organoid`, `3D bioprinting`, `3D bioprinter`, `bioprinting clinical`, `spheroid bioreactor`, `microgravity cell culture`, `CELVIVO`, `3Brain`, `Cellink`, `MEA neuroscience`, `organ-on-chip`, `liver organoid`, `brain organoid`, `spheroid drug screening` | 0 |
 | AI workflow | `OpenWiki`, `OpenWiki personal brain`, `Hermes Agent`, `Claude Code agent workflow`, `Codex CLI`, `RAGFlow`, `agent memory wiki`, `agent personal brain`, `MCP connector agent`, `MiniMax` | 0 new/promotable |
 Coverage note: this pull did not include exact queries for `ClinoStar`, `ClinoReactor`, `CorePlate`, `BIO X`, `BIONOVA X`, `Lumen X`, `CellShot`, `CellPick`, `TriArm`, `NocoDB`, or `SeaFile`; absence of hits therefore is not evidence of absence for those names.
 ## Previous ingestion coverage — 2026-07-26 17:27 UTC
 | Source | Items pulled |
 | --- | --- |
@@ -51,11 +82,9 @@ Items below are filtered from the 63 unique feed items plus the relevant
 search-query hits. Each line: HN id, score, feed(s), one-line gist, URL. HN ids
 are stable, so items stay referenceable across runs.
-### context-engineering-claude5 (source-backed)
+### context-engineering-claude5 (watchlist — adjacent, not actionable)
-New this run — the dominant HN signal for Robin's AI-Agent-workflow focus.
+High-engagement generic Claude workflow coverage from the earlier pull, anchored by Anthropic's own blog post at 427 points and reinforced by operational write-ups and an Opus 5 incident. It does not describe a direct change to Robin's named stack.
 Anchored by Anthropic's own blog post at 427 pts; reinforced by Claude Code
 operational write-ups and an Opus 5 incident.
 - **49051361** — score 427 — top — "The new rules of context engineering for Claude 5 generation models" (Anthropic blog). https://claude.com/blog/the-new-rules-of-context-engineering
 - **49056194** — score 79 — top — "Elevated Errors for Opus 5" (Claude status incident).
@@ -63,17 +92,11 @@ operational write-ups and an Opus 5 incident.
 - **49056689** — score 13 — new — "Claude Code Deletes Your Context History from Your Device After 30 Days" (Claude docs).
 - **49055752** — score 5 — new — "Claude Code Cut Their System Prompt by 80%. Does That Work for Small Models Too?" (antigma.ai).
-Five items in 24h, all corroborating, ranging from official Anthropic comms to
+Five items appeared in 24 hours, ranging from official Anthropic comms to community incident posts. Under Robin's current scope this remains source-only context and is not promoted to `/themes.md`.
 community incident posts. This is the clearest AI-Agent-workflow signal in the
 window and is routed to `/themes.md`.
-### coding-agent-tooling (source-backed)
+### coding-agent-tooling (watchlist — adjacent, not actionable)
-The Claude-Code / Codex-CLI wrapper / harness / safety-rail ecosystem. Heavy
+The earlier pull contained a dense Claude Code / Codex CLI wrapper, harness, and safety-rail cluster. Engagement was low and the items were derivative tooling, matching Robin's explicit exclusion for generic agent-engineering chatter.
 Show-HN volume — every item is derivative tooling, but recurrence is real and
 engagement is positive. Worth tracking as a watchlist-grade stream because
 downstream tooling for the user's stack (Codex CLI, Claude Code) is where most
 of the moving parts live.
 - **49051072** — score 3 — new — "Ruflo: An agent meta-harness for Claude Code and Codex" (ruvnet). https://github.com/ruvnet/ruflo
 - **49050011** — score 3 — new — "I was paying too much for Claude and codex here is how I reduced 90% of it" (AjnasNB/qarinah).
@@ -87,14 +110,11 @@ of the moving parts live.
 - **49053523** — score 3 — new — Show HN: Argus — VSCode Worktree Agent Session Manager.
 - **49050042** — score 4 — new — Show HN: blind taste test for Claude and Codex designs.
-Engagement is modest (≤ 4 pts each) but item count is dense — eleven items in
+Engagement was modest (≤ 4 points each) despite eleven items in 24 hours. Do not promote this cluster by volume alone; only a direct, story-worthy change to Robin's named tooling stack should be reconsidered.
 24h. Promote to full source-backed theme row only after at least one item
 crosses 50 pts in a single run.
-### llm-on-old-hardware (source-backed — promoted)
+### llm-on-old-hardware (watchlist — adjacent, not actionable)
-Recurrence confirmed in this run. Two independent high-engagement items on the
+The earlier pull contained recurring, high-engagement "tiny model on cheap hardware" stories. The pattern is real but outside Robin's current life-science and named-workflow-tool scope.
 "tiny model on cheap hardware" theme. Promote from watchlist to source-backed.
 - **49050512** — score 261 — top — "Running a 28.9M parameter LLM on an $8 microcontroller" (ESP32, slvDev). https://github.com/slvDev/esp32-ai
 - **49053375** — score 186 — top — "Inflect-Micro-v2: complete voice in 9.36M parameters" (HuggingFace). https://huggingface.co/owensong/Inflect-Micro-v2
@@ -104,7 +124,7 @@ Three high-engagement items across two runs; both pull from independent authors
 and target different hardware tiers (legacy CPU, ESP32-class MCU, sub-30M-param
 model card). The pattern is real, not coincidental.
-### ai-safety-debate (source-backed — refreshed)
+### ai-safety-debate (watchlist — adjacent, not actionable)
 Two new items in this window — one essay/interview, one fiction piece — both
 about LLM-extinction / alignment framing.
@@ -113,11 +133,9 @@ about LLM-extinction / alignment framing.
 - **49059660** — score 10 — new — "What if LLMs escape through inferences itself? This is fiction. For now" (agrillo.it).
 - **49052912** — score 224 — top — "DeepSeek pause fundraise after comments on compute gap to US leaked (transcript)" (github demo). Adjacent: open-weight-lab geopolitics, not safety per se but routed here because it overlaps with the prior open-weight-ai theme.
-### local-LLM-stream (watchlist — refreshed)
+### local-LLM-stream (watchlist — adjacent, not actionable)
-Adjacent to the user's "本地大模型部署" focus. Mostly derivative / tooling noise
+Mostly derivative tooling and infrastructure-policy coverage from the earlier pull. These items are retained for traceability but are outside the current named-stack focus.
 this run, but the upstream signals (Debian policy, Cloudflare AI traffic
 options) are substantive.
 - **49050859** — score 199 — top — "LLM Usage in Debian: Three Proposals" (Debian vote 2026/vote_002). Real policy signal.
 - **49052564** — score 173 — top — "Cloudflare's new AI traffic options for customers" (blog.cloudflare.com). Infra signal.
@@ -185,4 +203,4 @@ Adjacent (not OpenWiki-the-brand) items worth flagging:
 ## Next update
-Run `openwiki_ingest_connector connectorId=hackernews` (or `openwiki personal --update`) to refresh. Next run should re-add the life-science / organoid / vendor queries alongside the LLM-agent set if Robin wants both streams covered each day.
+Keep both configured lanes in the same pull: life-science/vendor/application queries and Robin's named AI-workflow tools. Apply the scope filter before synthesis; generic Claude Code / Codex wrapper and LLM-engineering debate items remain unpromoted unless they directly affect Robin's stack.
@@ -36,3 +36,76 @@
 - Queries: 10
 - Instance: `web-search-1`
 - Raw evidence: `~/.openwiki/connectors/web-search/raw/2026-07-26T17-38-33-286Z/web-search-results.json`
 ---
 ## 2026-07-26 18:07 pull
 Second Tavily pull on the same day (same 10-query lane). Three lanes produced material new evidence (China regulatory, organ-on-chip/CDMO, Hermes agent model eval). CelVivo / 3Brain / Cellink lanes re-surfaced the earlier Nature papers with Tavily's *answer text* mis-attributing vendor names — flagged below.
 ### China CDE cell & gene therapy classification — source-backed
 - **cisema.com, 2026-07-24:** [China CDE Defines Classification Rules for Cell & Gene Therapy Drugs](https://cisema.com/news/china-cell-gene-therapy-classification). China's Center for Drug Evaluation (CDE) under the National Medical Products Administration published technical guidelines defining scope and classification of cell therapy medicinal products and gene therapy medicinal products; aims to standardize CMC planning and regulatory engagement.
 - **Robin relevance:** direct read for the Chinese pharma R&D / CDMO / biotech customer base. Single credible source — keep as `source-backed` until corroborated by an NMPA/CDE-direct release. Promoted to `/themes.md` row `china-cde-cell-gene-therapy-rules-2026` and `/commitments.md` as a verification item.
 - Lighter signals in this lane: Roots Analysis [Cell Therapy Manufacturing Market Size 2035](https://www.rootsanalysis.com/reports/cell-therapy-manufacturing/285.html) (2026-07-24); Frontiers [pilot study of nebulized 3D-cultured MSC-derived extracellular vesicles](https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2026.1809451/full) (2026-07-24); PharmaSalmanac [Stem Cell Therapies: Close to Transforming Promise into Real Patient Outcomes](https://www.pharmasalmanac.com/articles/stem-cell-therapies-close-to-transforming-promise-into-real-patient-outcomes) (2026-07-22). Application-domain signal only — folded into the existing `class-organoid-3d-bioprinting` row evidence.
 ### Organ-on-chip / CDMO collaboration / ADLM 2026 — source-backed
 - **BioSpace, 2026-07-24:** [Phillips Medisize to Present Strategies for Scaling Complex In Vitro Diagnostic (IVD) Builds Through Structured CDMO Collaboration at ADLM 2026](https://www.biospace.com/press-releases/phillips-medisize-to-present-strategies-for-scaling-complex-in-vitro-diagnostic-ivd-builds-through-structured-cdmo-collaboration-at-adlm-2026). Top score (0.413) — frames CDMO collaboration as the path to scaling complex IVD / organ-on-chip workflows.
 - **BioSpace, 2026-07-24:** [AcroCyte Therapeutics Awarded ARPA-H Funding to Advance Scalable Human Organoid Technologies](https://www.biospace.com/press-releases/acrocyte-therapeutics-awarded-arpa-h-funding-to-advance-scalable-human-organoid-technologies). 0.215 score. ARPA-H backing for scalable human-organoid platforms — application-domain signal for CelVivo-class bioreactor positioning.
 - **Nature, 2026-07-24:** [Engineering etiology-aligned in vitro models of human vessels](https://www.nature.com/articles/s41378-026-01397-9). 0.165 score. Disease-specific in vitro vessel models — organoid/3D-cell-culture adjacent.
 - **BioSpace, 2026-07-23:** [Roche showcases intelligent lab solutions and high-precision instruments at ADLM 2026](https://www.biospace.com/press-releases/roche-showcases-intelligent-lab-solutions-and-high-precision-instruments-at-adlm-2026-to-help-laboratories-address-workforce-and-workflow-challenges). 0.151 score. Industry context, not vendor-anchored to Robin's 4 lines.
 ### Hermes Agent best-models eval — source-backed (workflow tooling, low priority)
 - **haimaker.ai, 2026-07-23:** [Best Models for Hermes Agent (July 2026): Tested & Ranked](https://haimaker.ai/blog/best-models-for-hermes-agent). 0.731 score. Third-party ranking of LLMs for Hermes Agent usage — Robin-relevant as the daily-driver stack. Folds into the existing `MiniMax-m3` theme row last-seen.
 - **Reddit, 2026-07-22:** [What are you USE CASES for Hermes Agent?](https://www.reddit.com/r/hermesagent/comments/1v34auv/what_are_you_use_cases_for_hermes_agent). 0.424 score. User-driven Hermes use cases; not a vendor-anchored signal.
 - **Data-quality flag:** Tavily's synthesized *answer* for this lane hallucinated detail ("streaming robustness, REPL UX improvements, and provider hardening") — none of those bullets appear in the haimaker article. Treat Tavily's answer text as untrusted.
 ### CelVivo / 3Brain / Cellink / CellShot lanes — same Nature results, Tavily answer text mis-attributed
 - **CelVivo ClinoReactor / Cellink BIO X:** the Nature [MO:BOT liver-spheroid automation paper](https://www.nature.com/articles/s41598-026-58939-4) reappears at score 0.402 (CelVivo lane) and 0.074 (Cellink lane). Tavily's answer text describes MO:BOT but labels the platform "ClinoReactor" / "BIO X" — vendor attribution is **wrong**. Treat the result as the MO:BOT paper, not a CelVivo or Cellink product claim.
 - **3Brain CorePlate MEA:** the Nature [portable semi-dry EEG device paper](https://www.nature.com/articles/s41598-026-52672-8) reappears at score 0.187. Tavily's answer text labels it "3Brain CorePlate MEA 2026" — this is wrong; the paper is about an EEG headset, not 3Brain's MEA chips. Do not promote as 3Brain evidence.
 - **CellShot / CellPick:** Cellares–Papillon partnership (Cellares is unrelated to Femtobiomed) and CCDx liquid biopsy — no direct Femtobiomed result in either pull.
 - **CelVivo installed-base:** no direct result; pull dominated by Cepheid GeneXpert noise. Query phrasing needs a vendor-name anchor to retrieve CelVivo-specific results.
 ### OpenWiki / RAGFlow / agent-memory-wiki lanes — Tavily recall broken
 - "OpenWiki connectors Tavily RAGFlow 2026" returned SmartBall X pipeline inspection, AWWA ACE26, insurance backchat, helicopter partnership, OLED manufacturing equipment, and fashion-show slides. None of the 8 results touches Robin's actual stack (OpenWiki / Gitea / Wiki.js / RAGFlow / NocoDB / SeaFile). Skip until the query is rephrased.
 - "Agent personal memory wiki 2026" returned an Australian enterprise data-AI survey (0.403), a LangChain4j self-building-agent experiment (0.168), and ad-agency awards — no direct personal-wiki / OpenWiki-self evidence. Adjacent only.
 ## Source coverage (cumulative)
 - Pull 1: fetched 2026-07-26T17:39:11Z — raw `~/.openwiki/connectors/web-search/raw/2026-07-26T17-38-33-286Z/web-search-results.json`
 - Pull 2: fetched 2026-07-26T18:07:27Z — raw `~/.openwiki/connectors/web-search/raw/2026-07-26T18-07-14-335Z/web-search-results.json`
 - Pull 3: fetched 2026-07-26T18:16:21Z — raw `~/.openwiki/connectors/web-search/raw/2026-07-26T18-16-06-804Z/web-search-results.json`
 - Queries: pull 1 + pull 2 = 10 queries each; pull 3 = 6 queries (vendor-anchored only)
 - Instance: `web-search-1`
 ---
 ## 2026-07-26 18:16 pull
 Third Tavily pull on the same day, after the 2026-07-27 `~/.openwiki/INSTRUCTIONS.md` scope was loaded. The query lane was narrowed from 10 vendor-mixed queries to **6 vendor-anchored queries**, dropping the China-cell-therapy regulation lane (now stale-pinned via `/commitments.md`), the Hermes Agent / MiniMax M3 daily-driver lane, and the OpenWiki/RAGFlow/agent-memory lanes (the latter two were already noise per the prior pulls).
 ### Queries and top results
 | Query | Top hits | Score | Vendor-anchor |
 | --- | --- | --- | --- |
 | `CelVivo ClinoReactor spheroid new paper 2026` | Nature MO:BOT liver-spheroid automation paper (re-mention); Patient-Derived Tumor Organoids for personalizing cancer treatment; CCDx whole-cell genomic profiling | 0.44 / 0.11 / 0.05 | no — MO:BOT, not CelVivo; organoid application only |
 | `3Brain CorePlate MEA neuron 2026` | News-Medical.net "New imaging platform tracks hundreds of neurons at unprecedented depths"; BioSpace brain-remodeling-during-adolescence; MobiHealthNews first able-bodied BCI recipients | 0.20 / 0.10 / 0.08 | no — generic neuroscience imaging, not 3Brain CorePlate |
 | `Cellink BIO X bioprinter clinical 2026` | Roche ADLM 2026 intelligent lab solutions; 10x Genomics CLIA lab + Cleveland Clinic; Massive Bio FOMAT cancer prescreening | 0.11 / 0.09 / 0.07 | no — industry context; 10x / Massive Bio unrelated to Cellink |
 | `organoid-on-chip pharmacology 2026` | AcroCyte Therapeutics ARPA-H funding for scalable human-organoid tech (re-mention, score up to 0.617 from 0.215); Patient-Derived Tumor Organoids; Nature MO:BOT liver-spheroid | 0.62 / 0.41 / 0.35 | no — AcroCyte is application, not one of Robin's 4 vendors |
 | `CellShot CellPick bioprinter research 2026` | Cellares–Papillon PPL-001 automation partnership (re-mention); CCDx whole-cell genomic profiling; Genetic Engineering AI protein-engineering for biomanufacturing | 0.17 / 0.14 / 0.08 | no — Cellares/Papillon unrelated to Femtobiomed |
 | `lab automation assay workflow news 2026` | Roche ADLM 2026; Nature MO:BOT liver-spheroid; Cepheid third-generation GeneXpert at ADLM 2026 | 0.82 / 0.74 / 0.56 | no — Roche/Cepheid industry context, not vendor-anchored |
 ### Synthesis result
 - **No new vendor-anchored evidence** — every query's top hit is either (a) application/organoid-level without naming Robin's 4 vendors, (b) a re-mention of an item already in the 17:38 or 18:07 pull, or (c) an unrelated industry/competitor brand (Roche, Cepheid, 10x, Cellares, Massive Bio, CCDx).
 - **AcroCyte ARPA-H** reappears at 0.617 (up from 0.215 in the 18:07 pull). Same item, stronger score — the existing `/themes.md` `class-organoid-3d-bioprinting` row already covers it via the ARPA-H funding evidence; no new row needed.
 - **Tavily answer-text vendor mis-attribution continues.** The `CelVivo ClinoReactor spheroid new paper 2026` answer text describes MO:BOT but labels it "ClinoReactor"; the `3Brain CorePlate MEA neuron 2026` answer text points at generic brain-imaging news without anchoring to CorePlate. Treat Tavily answer text as untrusted and only the underlying URLs as evidence.
 - **Dropped lanes** (vs. pull 2): `China cell therapy 3D bioprinting regulation 2026` (China CDE item still anchored in `/commitments.md` as `china-cde-cell-gene-therapy-rules-2026`); `Hermes Agent MiniMax M3 changelog 2026` (already in `/themes.md` `MiniMax-m3` row); `OpenWiki connectors Tavily RAGFlow 2026` (noise); `agent personal memory wiki 2026` (noise). Existing rows on `/themes.md` and `/commitments.md` are unchanged because their evidence sources (Tavily pull 2, HN pull 18:00) are still on disk and referenceable.
 ### Coverage note for next update
 The China CDE regulatory lane is no longer being refreshed on every Tavily pull. The existing `/commitments.md` item pins it as `source-backed` with one credible cisema.com source. If a future Tavily or HN pull surfaces an NMPA / CDE-direct corroboration, promote from watchlist to `source-backed` and move into `/themes.md` as a confirmed row.
@@ -7,7 +7,8 @@ Index kept narrow by design: only themes that map to Robin's actual work (life-s
 | Topic key | Theme / Signal | First seen | Last seen | Confidence | Sources | Evidence count | Status | Evidence |
 | --- | --- | --- | --- | --- | --- | --- | --- | --- |
 | `class-organoid-3d-bioprinting` | Recurring organoid, 3D bioprinting, and 3D-cell-culture automation coverage; this window adds Nature evidence on standardized liver-spheroid workflows and organ-on-chip pharmacomicrobiomics, still application-level rather than vendor launches | 2026-07-16 | 2026-07-26 | source-backed | hackernews, web-search | 14+ | active | [HN cluster](/sources/hackernews.md#class-organoid-3d-bioprinting-source-backed); [Tavily evidence](/sources/web-search.md#3d-spheroid-automation-and-assay-standardization--source-backed) |
-| `MiniMax-m3` | MiniMax-M3 (1M context multimodal) HuggingFace model card + Twitter launch; the daily-driver inference provider behind every OpenWiki run and most of Robin's agent work | 2026-07-16 | 2026-07-16 | watchlist | hackernews | 2 | watchlist | [HN: M3 model cluster (huggingface + twitter)](/sources/hackernews.md#MiniMax-m3-watchlist) |
+| `china-cde-cell-gene-therapy-rules-2026` | China CDE (under NMPA) published technical guidelines defining scope & classification of cell and gene therapy drugs; standardizes CMC planning for Chinese pharma/CDMO/biotech customers | 2026-07-24 | 2026-07-24 | source-backed | web-search | 1 | watchlist | [Tavily: cisema.com 2026-07-24](/sources/web-search.md#china-cde-cell-gene-therapy-classification-source-backed) |
 | `MiniMax-m3` | MiniMax-M3 (1M context multimodal) HuggingFace model card + Twitter launch; the daily-driver inference provider behind every OpenWiki run and most of Robin's agent work; haimaker.ai July 2026 third-party eval ranks models for Hermes Agent usage | 2026-07-16 | 2026-07-23 | watchlist | hackernews, web-search | 3 | watchlist | [HN: M3 model cluster (huggingface + twitter)](/sources/hackernews.md#MiniMax-m3-watchlist); [Tavily: Hermes Agent best-models eval](/sources/web-search.md#hermes-agent-best-models-eval-source-backed-workflow-tooling-low-priority) |
 | `openwiki-brand` | `langchain-ai/openwiki` (96 pts) — same product family as Robin's OpenWiki; high-engagement on HN | 2026-07-16 | 2026-07-16 | source-backed | hackernews | 1 (+ 2 adjacent wrappers) | watchlist | [HN: openwiki cluster](/sources/hackernews.md#openwiki-brand-signal) |
 | `agent-memory-architecture` | Show HN items on structured memory / trust layers for agents (Worklog SQLite, Open Knowledge format v0.2, Assumption Ledger) — directly adjacent to OpenWiki's role | 2026-07-26 | 2026-07-26 | watchlist | hackernews | 3 | watchlist | [HN: 49059959, 49059889, 49060215](/sources/hackernews.md#agent-memory-architecture-watchlist--new) |