Medical Research System Upgrade Plan

Goal: Build a positive, cure-oriented, evidence-aware research system that continuously searches the internet for new information about ulcerative colitis/proctitis and related health issues, records useful findings into this Obsidian knowledgebase, and sends Paul a weekly summary of what changed.

Design stance

This system should be hopeful and healing-oriented. It should actively look for cure/remission/healing claims, mechanisms, protocols, coaches, clinicians, and patient success patterns. It should not default to the assumption that ulcerative colitis is incurable or strictly autoimmune.

At the same time, the system must preserve intellectual honesty:

• Clinical evidence, mechanistic theories, clinician protocols, coach claims, and patient anecdotes must be clearly labeled.
• “Cure,” “healed,” and “remission” claims should be captured rather than dismissed, but the system should record whether objective markers were provided: colonoscopy, histology, fecal calprotectin, CRP, medication status, relapse duration, etc.
• The output should be encouraging and possibility-oriented, while still surfacing safety flags and clinician questions.

Current state observed on 2026-06-16

• Existing cron jobs: only e50d059a0af0, a silent daily backup of this vault.
• No active medical research digest cron was found in Hermes cron.
• The wiki currently has schema, index, dashboard, and a primary ulcerative colitis condition page, but little synthesized content.
• Google Workspace/Gmail skill is not authenticated yet, so direct weekly email delivery requires setup or another email route.
• Guava appears to support CSV export of manually logged entries via Settings → Download / Export, and specific data-type export as PDF/CSV. Guava also has an API page, but access/details need review.

Target architecture

0. Condition-first notebook model

The notebook should be organized around Paul’s main medical issues as top-level condition objects, with supporting research, coaches/influencers, methods, mechanisms, anecdotes, and personal history beneath each condition.

The journal is the chronological discovery layer: it captures the newest interesting findings from internet scrubs. Good journal findings can later be promoted into the organized knowledgebase for the relevant condition.

Initial condition objects:

• conditions/ulcerative-colitis/
• conditions/ulcerative-proctitis/ if useful as a distinct object.
• conditions/silent-reflux/
• conditions/high-cholesterol/

Each condition object should eventually include:

• index.md — main condition hub.
• personal-history.md — Paul-specific timeline, tests, meds, symptoms, experiments.
• research.md — clinical/mechanistic research map.
• methods.md — protocols, diets, supplements, lifestyle methods.
• coaches-and-influencers.md — clinicians, coaches, creators, communities, protocols.
• journal-highlights.md — best recent finds promoted from the chronological journal.

Cron outputs should be stored under:

• journal/weekly-digests/YYYY-MM-DD-weekly-medical-research-digest.md
• journal/finds/YYYY-MM-DD-short-topic.md

Then high-value items are promoted into condition pages and cross-condition canonical pages.

1. Personal medical profile layer

Create a structured, private profile from Paul’s Guava data and selected manual notes.

Proposed files:

• _meta/personal-health-profile.md — human-readable summary and current research priorities.
• raw/personal-notes/guava-export-YYYY-MM-DD/ — immutable raw Guava exports.
• _meta/guava-ingestion-map.md — mapping from Guava fields to wiki fields.
• conditions/silent-reflux.md
• conditions/high-cholesterol.md
• conditions/ulcerative-proctitis.md if distinct from UC page.
• symptoms/ pages for major recurring symptoms.

Profile sections:

• Diagnoses and suspected diagnoses.
• Symptom history and timeline.
• Medications tried and response.
• Supplements tried and response.
• Diet/lifestyle experiments and response.
• Labs and objective markers.
• Procedures/imaging/colonoscopies/endoscopies.
• Current medications/supplements.
• Current goals and concerns.
• Research priorities.
• Clinician questions.

2. Cure-oriented topic map

Create index pages for the people, mechanisms, and protocols Paul wants tracked.

Initial pages to create/update:

• mechanisms/hydrogen-peroxide-redox-uc.md — Dr. Pravda / RDLA / STS / redox model.
• clinicians-and-sources/jay-pravda.md
• clinicians-and-sources/dr-snow.md — identify exact Dr. Snow/protocol and track sources.
• clinicians-and-sources/ray-peat.md — gut health, endotoxin, serotonin, thyroid/metabolism context.
• interventions/whole-food-plant-based-high-carb-uc.md
• interventions/carnivore-keto-low-carb-uc.md
• interventions/stress-trauma-nervous-system-uc.md
• interventions/probiotics-fermented-foods-uc.md
• interventions/turmeric-curcumin-uc.md
• interventions/nicotine-uc.md
• interventions/qing-dai-uc.md
• interventions/fmt-uc.md
• interventions/rdla-sts-reducing-agents-uc.md
• anecdotes/youtube-uc-proctitis-cure-comments.md

Each page should include:

• What the claim/protocol says.
• Why it might work mechanistically.
• Best evidence found.
• Best anecdotal reports found.
• Reported harms/worsening.
• Confounders.
• How it relates to Paul’s profile.
• Clinician questions.

3. Raw source ingestion layer

All valuable sources should be saved before synthesis.

Proposed raw paths:

• raw/youtube/YYYY-MM-DD/{video_id}-metadata.json
• raw/youtube/YYYY-MM-DD/{video_id}-transcript.md
• raw/youtube/YYYY-MM-DD/{video_id}-comments.json
• raw/reddit/YYYY-MM-DD/{thread_id}.md
• raw/social/YYYY-MM-DD/x-{post_id}.md
• raw/forums/YYYY-MM-DD/{source-slug}.md
• raw/articles/YYYY-MM-DD/{source-slug}.md
• raw/papers/YYYY-MM-DD/{pmid-or-doi}.md

Raw source frontmatter should include source URL, source class, ingestion date, topic tags, and a body hash when possible.

4. Weekly research cron

Create a new weekly Hermes cron job. Do not replace the backup cron.

Suggested schedule:

• Weekly, Monday morning.
• Job name: Weekly Medical Research Internet Scrub.

Suggested skills:

• medical-research-and-anecdote-kb
• agent-reach
• youtube-content
• obsidian

Suggested toolsets:

• web/search
• terminal
• file
• skills

The cron prompt should be self-contained and should:

1. Read SCHEMA.md, index.md, log.md, _meta/personal-health-profile.md if present, and this upgrade plan.
2. Search the last 30 days for tracked topics.
3. Search older sources only when discovering a new expert/protocol not yet in the wiki.
4. Prioritize new, high-signal material:
   • Cure/remission/healing claims.
   • New protocols or protocol updates.
   • Expert/coach interviews and testimonials.
   • YouTube comments where people report what helped/cured them.
   • Reddit/social/forum clusters with detailed anecdotes.
   • New papers/case reports/trials/mechanistic reviews.
   • Safety signals and contradictory reports.
5. Save raw sources.
6. Update synthesized pages only when the finding adds something meaningful.
7. Update index.md for new pages.
8. Append to log.md.
9. Produce a weekly digest with sections:
   • “Most hopeful new finds”
   • “New cure/remission claims worth tracking”
   • “Anecdotal patterns”
   • “Mechanisms and theories”
   • “Experts/coaches/protocols indexed”
   • “Sources browsed and new takeaways”
   • “Reviewed but no major new data”
   • “Risks or contradictions to keep in view”
   • “Wiki pages updated”
   • “Questions to ask a clinician”

Source-audit requirement for the weekly digest:

• Include a readable list/table of all sources browsed that were interesting or relevant enough to inspect.
• For each source, include: source title/name, URL/platform, source class, why it was browsed, the main new takeaway if any, and what wiki page/condition/theory it affected.
• If a source did not add genuinely new information, put it in a lower section titled “Reviewed but no major new data” and state whether it reinforced an existing idea, duplicated already-known information, was low-quality/SEO/commercial, or was not relevant enough to promote.
• Make novelty explicit: distinguish new_to_wiki, reinforces_existing, contradicts_existing, safety_signal, and discarded_low_signal.
• This source-audit section is meant to help Paul see what the system is regularly ingesting, not just the polished conclusions.

Tone instructions:

• Optimistic, proactive, and cure-oriented.
• Assume healing/remission is possible and worth investigating.
• Do not use dismissive phrasing like “just anecdotal.” Prefer “anecdotal but potentially useful signal.”
• Still label evidence levels and safety flags.
• Never recommend stopping prescribed medication; frame as research and clinician discussion.

5. Email delivery

The weekly summary should be sent by email.

Options:

1. Configure Google Workspace/Gmail OAuth for Hermes and send with Gmail.
2. Configure a simpler SMTP/himalaya route if Paul only needs email sending.
3. If email setup is delayed, deliver to the current Hermes conversation first and save the digest in _meta/weekly-digests/ until email is configured.

Suggested digest archive path:

• journal/weekly-digests/YYYY-MM-DD-weekly-medical-research-digest.md

6. Search strategy

Core recurring searches should include at least:

• ulcerative colitis cure
• ulcerative colitis healed
• ulcerative colitis remission naturally
• ulcerative proctitis cure
• ulcerative proctitis healed
• ulcerative proctitis remission
• proctitis cured diet
• RDLA ulcerative colitis
• R-dihydrolipoic acid ulcerative colitis
• sodium thiosulfate ulcerative colitis
• Jay Pravda ulcerative colitis
• Dr Pravda ulcerative colitis
• Dr Snow ulcerative colitis protocol
• Ray Peat ulcerative colitis
• Ray Peat gut health endotoxin serotonin
• Qing Dai ulcerative colitis
• nicotine ulcerative colitis remission
• curcumin ulcerative colitis remission
• carnivore ulcerative colitis remission
• high carb health ulcerative proctitis
• ulcerative colitis comments cured
• site:reddit.com/r/UlcerativeColitis cured remission proctitis
• site:reddit.com/r/UlcerativeColitis RDLA
• site:reddit.com/r/UlcerativeColitis Pravda

Platform-specific channels:

• YouTube search + transcript + comments.
• Reddit via Agent Reach/OpenCLI.
• X/Twitter via Agent Reach/twitter-cli where available.
• Forums/blogs via web search and Jina Reader.
• PubMed/PMC via web/PubMed script.
• Optional: RSS/blogwatcher for recurring expert blogs/channels.

6b. Community/forum watchlist

Standing watchlist page: conditions/ulcerative-colitis/community-watchlist.md.

The weekly scrub should read this page and prioritize:

• new posts/threads from Tier 1 communities;
• updates to Tier 2 high-signal cure/remission/protocol threads;
• new search results matching the page’s site-specific queries;
• detailed reports with diagnosis context, dose/duration, objective markers, medication status, follow-up, and harms/worsening.

If Reddit/OpenCLI or a forum is blocked, record the access failure and use browser/Jina/search-engine snippets as fallback rather than silently skipping it.

7. Quality scoring for findings

Each finding should be scored along these axes:

• novelty: new to the wiki or meaningful update?
• relevance_to_paul: directly related to UC/proctitis, reflux, cholesterol, or profile?
• healing_potential: claims remission/cure/healing or plausible mechanism?
• evidence_strength: guideline/review/RCT/case report/mechanistic/anecdotal.
• anecdote_quality: diagnosis clarity, intervention detail, timeline, outcome clarity, confounders.
• safety_priority: potential risk or clinician review needed?
• actionability: does it produce a good question or tracked experiment idea?

8. Implementation phases

Phase 1 — Notion/Guava inventory and personal profile foundation

• Use ntn/Notion API to inventory current health docs and meetings.
• Export Guava data.
• Save raw export to raw/personal-notes/guava-export-YYYY-MM-DD/.
• Create _meta/personal-health-profile.md.
• Create/update top-level condition objects for UC/proctitis, silent reflux, high cholesterol, and major symptoms.

Phase 2 — Knowledgebase expansion

• Create the initial expert/protocol/mechanism/intervention pages listed above.
• Import the YouTube/proctitis comment analysis already performed in Hermes into anecdotes/youtube-uc-proctitis-cure-comments.md.
• Add index links and log entries.

Phase 3 — Weekly cron prototype

• Create a weekly cron that runs a focused internet scrub and saves a digest to _meta/weekly-digests/.
• Initially deliver summary to the current Hermes channel until email is configured.
• Run manually once and inspect quality.

Phase 4 — Email delivery

• Configure Gmail/SMTP delivery.
• Update cron to email Paul the digest and also archive it in Obsidian.

Phase 5 — Improve recall and reduce noise

• Add watchlists for specific experts, YouTube channels, Reddit queries, blogs, RSS feeds, and PubMed queries.
• Use conditions/ulcerative-colitis/community-watchlist.md as the standing forum/thread watchlist for Reddit, HealingWell, Smart Patients, Mayo Connect, MyCrohnsAndColitisTeam, Low-Toxin Forum, Bluelight, Crohn’s Forum thread leads, and cure/remission query patterns.
• Add deduping and per-source quality filters.
• Track whether findings later proved useful or low-value.

Open setup questions

1. What email address should receive the weekly digest?
2. Should the digest be sent from Gmail, another SMTP account, or just delivered via Hermes until email is configured?
3. Can Paul export Guava CSV/PDF data and place it somewhere Hermes can read?
4. Which exact “Dr. Snow” should be tracked? Need canonical website/YouTube/book/protocol source.
5. Should the system include only UC/proctitis at first, or immediately include reflux and cholesterol in the weekly search scope?
6. What tone should the weekly email use: concise research memo, enthusiastic health scout, or detailed scientific briefing?

Acceptance criteria

The upgraded system is working when:

• A weekly cron runs successfully.
• It saves raw sources and a digest into the vault.
• It updates synthesized pages for meaningful new findings.
• It emails Paul a readable weekly summary.
• It maintains a positive, cure-oriented tone while preserving source labels and safety notes.
• The personal health profile is available to guide relevance filtering.
• New experts/protocols are indexed and not lost in chat history.