Research Daemon

TypeScript helper article for Claude. Your stack is mostly Python + Flask + vanilla JS - minimal TS exposure. File for awareness; not actionable unless a TS project starts.

Yes - Pentair IntelliFlo / IntelliCenter pumps CAN be controlled from Home Assistant in 2026. Paths: (1) BEST - pyintellicenter / intellipool integration if pump has IntelliCenter or IntelliConnect (cloud or local LAN); (2) MQTT bridge via screenlogic-py for older ScreenLogic; (3) RS-485 hardware adapter (njsPC / nodejs-poolController) for 100 percent local control of any RS-485 Pentair - most robust, ~50 USD hardware. Suggest: photograph pump controller model + check for WiFi / IntelliCenter. If yes, cloud integration in 30 min. If older RS-485 only, USB-RS485 adapter + nodejs-poolController in Docker.

Harvard Health on resting heart rate: normal adult RHR 60-100 bpm. Lower (50-60) usually indicates better cardiovascular fitness. Resting HR sustained ABOVE 80 = elevated cardiovascular risk. Trend matters more than single readings. Action for Julian: pull historical RHR from Google Fit + Withings; if creeping upward over months that is a red flag worth raising with GP. Query google_fit table in Master_Health.db for heart_rate metric, plot 90-day trend.

BBC News: Diabetes UK calling for tailored mental-health support. People with diabetes are TWICE as likely to have depression vs general population. Diabetes distress (separate from depression) affects up to 40 percent. Standard talking-therapy services often do not understand diabetes-specific stressors (fear of hypoglycaemia, burnout, body-image around pumps/CGMs). Action for Christina: at next endo appointment specifically request a PHQ-9 / Diabetes-Distress-Scale screen - validated tools the diabetes team should run.

Article lists 7 Claude skills for productivity - typically: email triage, scientific paper review, meeting summarisation, code review, document QA, market research, customer-feedback analysis. Most already in your stack (email triage via hansenemail, document QA via medical archive). Worth scanning for gaps; not urgent.

Patients Know Best (PKB) login URL filed. PKB integration already exists in your stack (pkb_daily_checker, scrape_pkb_test_history.py). Not a research item - likely a bookmark drop. Filed.

Jupiter Merian Global Equity I Acc GBP - actively managed global equity fund, factsheet on Trustnet. Filed for finance reference; not a research-daemon question. Track in IBKR notes or finance-specialist context if buy/sell decision needed.

Researchers identified a relatively unknown receptor controlling osteoblast activity. Blocking it in mice increased bone formation and reversed osteoporotic bone loss WITHOUT bisphosphonate side-effects. Stage: pre-clinical / animal model - years from human therapy. Relevance for Christina (T1D = elevated osteoporosis risk via reduced IGF-1 + microvascular changes): worth tracking but not actionable today. Current evidence-based actions remain: weight-bearing exercise, vitamin D + calcium, DEXA every 2 yr if risk-factored.

Likely refers to NotebookLM-style alternatives: Cohere Command-R notebook UI, Anthropic Projects, or Perplexity Pages/Spaces - all let you upload sources and ask questions with citations. For your medical archive (1300+ docs already indexed), uploading to a Project-style tool gives a chat interface over Christinas records. Already partly built locally (medical archive search) - this would be a faster prototype. Not urgent; existing local solution works.

npj Science of Food study (relevant to Christina T1D + Julian T2D risk): coffee consumption pattern affects HbA1c through propionic-acid-producing gut bacteria. Filtered/black coffee correlated with LOWER HbA1c via increased propionate-producing microbes (Roseburia, Faecalibacterium); these short-chain fatty acids improve insulin sensitivity. Effect dose-dependent (3+ cups/day strongest), lost when sweetened heavily. Implication: black or unsweetened coffee may have a small favourable HbA1c effect via the gut.

browser-use/browser-harness: open-source self-healing browser automation for LLMs. Wraps Playwright with retry/recovery so an LLM agent can complete multi-step web tasks (login, form fill, navigate) even when selectors change. Could replace bespoke Playwright scripts in your stack (clarity scraper, mychart download, dexcom_filler). Worth piloting on ONE existing flaky scraper first. Likely 30-50 percent reduction in scraper-maintenance IF it works on your sites. Risk: still under heavy dev, breaking changes.

SciTechDaily: physical activity reduces cancer risk via lower chronic inflammation, improved insulin sensitivity, immune surveillance, reduced sedentary time. Even modest movement increases help - replacing 30 min sitting with light walking shows measurable reductions. Confirms value of keeping footstep / step-count data: longitudinal step trends are a strong predictor of all-cause and cancer-specific mortality. Recommend keeping step data in Master_Health.db indefinitely. Already tracked via Google Fit; consider importing Withings step data too.

Local LLMs (Llama 3.1 8B, Phi-3, Gemma-2) on local GPU power HA voice control with 100 percent privacy + sub-second latency. Best path: Ollama or LocalAI on London VM, HA Conversation integration, Whisper for STT + Piper for TTS. 8GB VRAM minimum for usable speeds. If London server has GPU, plug-and-play. If not, used 3060 12GB (~200 USD) transforms HA voice. Real benefit over cloud LLMs: no internet dependency, replies in 200ms vs 2s, full control of system prompts.

Android 16+ adds passive email verification: instead of OTP/magic link, Android signs a verification token using already-authenticated Gmail account, sent silently. UX: user clicks Sign in with Google, done. Useful for hansenhome.uk apps IF user is on Android with Gmail; iPhone/desktop still need fallback. Verdict: NOT a full replacement yet - implement as ADDITIONAL fast-path for Android, keep magic-link for iOS/desktop. Add later, not urgent.

Top-1 percent Claude Code playbook on TowardsAI - typical content: maintain a tight CLAUDE.md, use subagents proactively, /clear between unrelated tasks, store learnings to memory, batch parallel tool calls, restart servers after Python edits. Your CLAUDE.md and skill stack already implement most of this. Skim for any new pattern; no urgent action.

Johns Hopkins (Dec 2024) found arm position alone can falsely elevate BP by 4-7 mmHg. KEY RULES: (1) arm SUPPORTED on flat surface, palm up; (2) cuff at heart level (mid-sternum); (3) feet flat on floor, back supported, no crossed legs; (4) sit quietly 5 min before; (5) no caffeine/exercise/smoking 30 min prior; (6) empty bladder; (7) no talking. Arm hanging at side OR resting in lap (instead of supported at heart height) was the biggest error - both produce falsely high readings. For Christina: ensure the BP cuff arm is supported on chair-arm or table at heart level, not in lap.

Anthropics official docs page on integrating Claude Code with Cloudflare (Workers, Pages, R2, D1). Useful reference for cloudflare_github specialist. Already part of your deploy pipeline (deploy-pages.sh). File for reference.

Manufacturer (Boks) defunct since Jan 2025 - official app likely orphaned. Options: (1) Reverse-engineer with nRF Connect on Android - capture BT advertising + GATT services, replay commands; (2) Check r/homeassistant and Github for community boks-bt drivers; (3) Replace with current-gen smart letterbox sensor - Aqara Door/Window FP1 or Shelly BLU Door/Window paired with HA gives same notification function for ~30 EUR. Recommendation: replace, not repair - chasing reverse-engineered firmware on dead-vendor hardware is high-effort low-yield.

Article on giving Claude Code persistent memory - likely the @memory CLAUDE.md pattern or a memory MCP. You ALREADY have this: auto-memory at .claude/projects/.../memory/MEMORY.md per project, plus specialist_contexts DB, plus session_messages knowledge base. Stack is ahead of the article. File for awareness; no action.

Your instinct is reasonable. Stainless steel exposed to salt-pool water absolutely DOES corrode if it is the wrong grade — 304 SS in a chlorinated salt environment will rust visibly within a few years, sometimes through-pit. The manufacturer is correct that some staining is normal but heavy rust that feels like mild steel suggests either (a) it is 304 not 316 marine grade, (b) it was contaminated during fabrication, or (c) it is not actually stainless. Magnet test: 304/316 are weakly magnetic at best, mild steel sticks strongly. Request the material certificate and grade specification.