Patient Bloodwork Protocol

Patient Bloodwork → Bachmeyer Protocol → Published Report

Full clinical-intelligence pipeline for a single patient: ingest labs → flag → build a Bachmeyer-cited peptide + supplement protocol with complete dosing → plain-language layer → de-identified published HTML report. Patient-isolated (IL-8). This is the canonical model proven on Ehsan, Bogdan, Ali Faisal, and Ion Borzea.

Source Hierarchy & HARD Rules (NON-NEGOTIABLE — bind into every run)

• BACHMEYER PRIMARY — health knowledge comes from the Bachmeyer corpus first: NLM notebooks cb9cf1fd (Primary) + 1eef6220 (Extended) + d9213daf (Peptide Dosage) via http://localhost:18790/query, and the LLM-Wiki (~/Claude/LLM-Wiki/concepts/*.md, ~/.nexus/knowledge/peptides/*.md). Cortex (research/health/llm_wiki_sources_agent) mirrors these — search it first to save latency, confirm dose-critical items against NLM/wiki.
• THRESHOLDS = BACHMEYER, NOT PAFI — every flag/threshold/dose is Bachmeyer-cited or a standard lab reference range. Never invent a number or attribute one to "Pafi". No number → "attending physician sets". Patient = clinical; logistics only otherwise.
• PATIENT CARD FIRST — on ANY new labs, update ~/.nexus/projects/health/patients/{slug}-PROJECT-CARD.md before building the report (create if missing). Group labs by collection date. Append-only.
• PLAIN-LANGUAGE — every patient report needs a 💬 In plain words line (layperson, ≤2 sentences) per red/abnormal lab item AND per peptide. New + retrofit-on-touch.
• PUBLISH + DE-IDENT — publish a de-identified copy to GitHub Pages (cryptopafi/nexusos-reports) via shared-reporter; keep a full named local copy. KEEP FIRST NAME (body + filename) for identification; scrub surname/DOB/CNP/phone/lab-IDs/sample-codes only. Verify HTTP 200 + grep-clean.
• REPORT VIA SHARED-REPORTER — never hand-roll the HTML/publish; always route through the shared-reporter skill.
• CLINICAL BOUNDARY — every report states it is informational decision-support, not a diagnosis/prescription; the attending physician decides.
• SAVE INSIGHTS — final state to Cortex (research collection) + PROJECT-CARD.

Inputs

• patient (required): name or slug. Resolve to slug firstname-lastname (e.g. ion-borzea). Map known aliases.
• --source (default ~/Desktop/Health): folder holding the lab documents.
• --version (default: auto-increment): report version vN. v1 = first pass; bump when re-running with new markers/scope.
• --add / --hold: explicit peptides to add to or hold from the protocol (e.g. --add tesamorelin,5-amino-1mq --hold ipamorelin,cjc-1295). User instructions here override defaults.

Pipeline (ordered, idempotent)

1 — Gather ALL documents for the patient

• ls/Glob the source folder; collect every file whose name matches the patient (name, surname, slug, common misspellings). Watch for mislabeled files (a file named for patient A can contain patient B — verify identity inside).
• Batched lab bulletins: Romanian/clinic labs deliver one collection in stages (part 1, part 2, "… 3", "unlocked"). Same bulletin nr + same collection datetime + same sample code = one event. The latest/most-complete (highest emit date) version supersedes earlier parts — use it as canonical; note earlier parts are subsumed. Prefer -unlocked copies for password-protected PDFs.

2 — Extract

• PDFs: python3 -c "import fitz; ..." (PyMuPDF) → fallback pdftotext -layout → fallback OCR (see pymupdf-scanned-pdf-ocr skill for old scans / RO/IT docs). Images: vision read.
• Record per marker: value, unit, lab reference range, lab flag (H/L). Note any "teste în lucru" (pending) markers — they will arrive in a later batch.

3 — Update PROJECT-CARD (FIRST, before report)

• Open/create ~/.nexus/projects/health/patients/{slug}-PROJECT-CARD.md. Log every non-locked doc by collection date; mark batched bulletins (which supersedes which). Mark resolved-vs-pending markers. Append-only; bump last_updated.

4 — Flag + triage

• Mark every out-of-range value (H/L). Identify the convergent flags (e.g. insulin resistance cluster, NAFLD/GGT, hormonal). Build a priority map (root cause first — usually metabolic).

4.5 — EXISTING PATIENT: longitudinal cross-reference (ALWAYS — non-optional)

If the patient already has a card with prior labs, you MUST cross-reference the new panel against ALL prior draws — never analyze the new panel in isolation. This produces intel a single-panel report cannot:

• Pull every prior collection date from the card (and any other lab/Rx files for this patient — e.g. an accompanying prescription explains new findings).
• Build a trend table for every marker that overlaps ≥2 draws: marker | draw1 | draw2 | …| newest | read. Use ↑/↓/✅/⚠️ to show direction and whether it resolved, worsened, or is newly out of range.
• Connect the dots → actionable intel (a dedicated section on top of the standard report): which interventions are working (marker moved toward target), which are over-shooting (e.g. TRT → supraphysiologic T / rising hematocrit; GH-secretagogue → IGF-1 over range + glucose/HOMA drift), which prior flags resolved, and what the trend implies for dose titration. Tie each to the patient's active protocol and the prescriber's titration rules already on the card.
• Flag draw-timing caveats (trough vs peak for hormones) before recommending any dose change.
• Cross-referenced trend + actionable intel are MANDATORY report sections for any returning patient; the regular flag/protocol report is necessary but not sufficient.

5 — Bachmeyer protocol WITH FULL DOSING

• For each flag, pull the matching Bachmeyer protocol from Cortex first, then confirm dose-critical items via NLM (curl -s -m 90 -X POST http://localhost:18790/query -d '{"query":"..."}') and the wiki concept pages.
• Every recommended compound MUST carry the full contract: dose · route (SubQ/IM/oral) · frequency · timing · titration/cycle length · the lab marker Bachmeyer tracks, each plus a 💬 plain-words line. A mechanism-only entry is INCOMPLETE — reject it.
• Apply --add / --hold. Default sequencing = Bachmeyer metabolic-first (Tier-1 metabolic core → Tier-2 adjuncts → hormone restoration). Honor Bachmeyer positions: prefers Retatrutide over semaglutide/tirzepatide for IR; no first-line TRT (Kisspeptin/Enclomiphene natural restoration preferred); no aromatase inhibitors if E2 low; never mix Tesamorelin + Ipamorelin; once-weekly Retatrutide (not daily); GHK-Cu always separate syringe.
• Add a supplement stack (independent clinical, complements Bachmeyer) and a pre-peptide gating workup (imaging, parasite/IgE if eosinophilia, thyroid, ApoB/Lp(a), methylation, vit D) + a phasing table (gate → start → build → reassess) + a monitoring plan with discontinuation red-lines (standard ranges).
• Reference doses (Bachmeyer corpus, start low / physician-set): Retatrutide 0.5 mg SubQ weekly → +0.5 q4wk → 2–4 mg; Tesamorelin 1–2 mg SubQ nightly pre-bed 5on/2off 12–16wk; MOTS-c 5–10 mg SubQ 2–3×/wk (mg!); 5-Amino-1MQ SubQ-only 10–20→50 mg/day belly AM 6on/2off; BPC-157 250 mcg–1 mg/day 5on/2off; Cardarine SubQ 30min pre-workout 30on/30off; KPV 300–400 mcg/day; Kisspeptin-10 100–500 mcg SubQ 2–3×/wk max 8–10wk; Enclomiphene oral ~12.5 mg; CJC-1295 no-DAC 100–300 mcg + Ipamorelin 100–200 mcg nightly 5on/2off. Always re-confirm against live NLM — corpus updates.

6 — Plain-language layer

• 💬 In plain words (≤2 sentences, layperson) per red lab item and per peptide. Mandatory.

7 — Render + de-identify + publish (via shared-reporter)

• Author the report markdown to ~/.nexus/projects/health/reports/{slug}/protocol-{version}-{date}.md (full, named).
• Dispatch the shared-reporter (Agent, Sonnet) to produce TWO outputs:
  • A (local, named, NOT published): Tier-2 health HTML (report_type=health, green accent #22c55e), saved to ~/.nexus/projects/health/reports/{slug}/{slug}-bloodwork-{period}-{version}.html.
  • B (de-identified, published): scrub surname → first name only, remove bulletin/sample/client/CNP/phone; title "FirstName (M/F, age) — …"; add noindex; publish to cryptopafi/nexusos-reports as {firstname}-bloodwork-{period}-{version}.html (NO surname in filename); verify HTTP 200 + grep-clean (0 occurrences of surname + every numeric ID); also save the de-ident HTML locally.
• Self-audit: no TODO/placeholder, toggle/print/viewport/charts present.

8 — Persist

• Update the PROJECT-CARD with report links (local full + published de-ident URL + commit + verification result) and the dosing summary.
• Store to Cortex research (metadata: type=patient_report, patient={slug}, domain=health, framework=bachmeyer, version, date).

Versioning

Re-run on new labs or scope change → bump vN, note what it adds, mark it supersedes the prior. New pending markers arriving in a later batch is the canonical trigger for a bump.

Output Contract (to user)

Concise: which docs ingested (+ which batch superseded which), the convergent flags, the full-dosing protocol table (every compound = dose/route/freq/cycle), added/held compounds, the published de-identified URL (HTTP 200 + grep-clean PASS), and the local full-named path. Doses are Bachmeyer-cited/standard-range, physician-set, start low.

Failure Handling

• NLM unreachable → rely on Cortex + wiki (both Bachmeyer-derived); do not mention the infra incident (NO INFRA INCIDENT MENTIONS rule); fix silently.
• Locked PDF → use the -unlocked sibling; never block on it.
• Unknown/unrecoverable patient identity (mislabeled file, missing source) → log as BLOCKED in the card, ask the user; do not guess a surname.
• GitHub token missing / Pages 404 → shared-reporter falls back (building/fallback_url/VPS/local); report the real status, never claim "published" without an HTTP 200 + grep-clean.