corbett3000 · corbett3000 · Apr 19, 2026 · Apr 19, 2026
diff --git a/README.md b/README.md
@@ -6,7 +6,7 @@
 [![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/downloads/)
 
-**OpenDNA** parses a 23andMe / AncestryDNA / MyHeritage raw DNA file against 8 curated SNP panels (cardiovascular, methylation, pharmacogenomics, athletic, dietary, HFE, cognition, stimulant sensitivity), joins findings with ClinVar + PharmGKB/CPIC annotations, and renders a self-contained HTML report. Optional BYOK LLM synthesis (Anthropic Claude or OpenAI GPT) adds a prose interpretation layer.
+**OpenDNA** parses a 23andMe / AncestryDNA / MyHeritage raw DNA file against 8 curated SNP panels (cardiovascular, methylation, pharmacogenomics, athletic, dietary, HFE, cognition, stimulant sensitivity), joins findings with ClinVar + PharmGKB/CPIC annotations, and renders a self-contained HTML report. It now also scores match confidence, shows panel coverage / blind spots, and rolls multi-marker genes like APOE, HFE, MTHFR, CYP2C19, and warfarin PGx into composite calls. Optional BYOK LLM synthesis (Anthropic Claude or OpenAI GPT) adds a prose interpretation layer.
 
 **What leaves your machine:** nothing, by default. No account, no upload, no telemetry. The raw DNA file is parsed, analyzed, and rendered entirely offline. If — and only if — you paste an API key and opt in to AI synthesis, OpenDNA sends the filtered *findings* (rsid + gene + genotype + tier + short note; roughly 40–50 lines of structured text) to the provider you chose. Your raw DNA file is never transmitted. See the [Privacy](#privacy--what-leaves-your-machine) section for the full rules.
 
@@ -86,6 +86,8 @@ opendna --version
 pytest -v                           # (only if you installed [dev])
 ```
 
+Current contributor baseline: `51` tests passing.
+
 ---
 
 ## Quickstart — the web UI
@@ -135,13 +137,13 @@ The JSON schema is stable and suitable for ingestion by another pipeline stage.
 
 ## What OpenDNA interprets
 
-Every finding is tier-scored (`risk` / `warning` / `normal` / `unknown`) and, where available, cross-referenced with ClinVar clinical significance and PharmGKB/CPIC dosing guidelines.
+Every finding is tier-scored (`risk` / `warning` / `normal` / `unknown`), annotated with match confidence, and, where available, cross-referenced with ClinVar clinical significance and PharmGKB/CPIC dosing guidelines. The report also separates `not on chip`, `no-call`, and `ambiguous` markers so missing data is not misread as a negative result.
 
 | Panel | Focus | Representative genes |
 |---|---|---|
-| **Cardiovascular & Longevity** | CAD risk, lifespan | APOE, 9p21, FOXO3, LPA |
+| **Cardiovascular & Longevity** | CAD risk, thrombosis, lifespan | APOE, 9p21, FOXO3, LPA, F5, F2 |
 | **Methylation & Detox** | Folate/B12 cycle | MTHFR, COMT, MTR/MTRR, FUT2, CBS, VDR |
-| **Pharmacogenomics (PGx)** | Drug metabolism | CYP2C19, CYP2C9, VKORC1, SLCO1B1, TPMT, CYP3A5 |
+| **Pharmacogenomics (PGx)** | Drug metabolism | CYP2C19, CYP2C9, VKORC1, CYP4F2, SLCO1B1, TPMT, CYP3A5 |
 | **Athletic Performance & Recovery** | Fiber type, recovery | ACTN3, PPARA, PPARGC1A, COL5A1, IL6 |
 | **Dietary Sensitivity** | Lactose, caffeine, alcohol, omega-3 | LCT, CYP1A2, ALDH2, FADS1, TAS2R38 |
 | **Iron Metabolism (HFE)** | Hemochromatosis | C282Y, H63D, S65C |
@@ -150,6 +152,12 @@ Every finding is tier-scored (`risk` / `warning` / `normal` / `unknown`) and, wh
 
 The analyzer automatically handles both **allele-order variations** (`AG` == `GA`) and **reverse-strand reports** (panel `CC` matches file `GG` for C/T SNPs), so genotypes from any major consumer-testing vendor should resolve correctly.
 
+### Source-file caveats, now surfaced in the report
+
+- `Not on chip` means the marker was not present in the source file. It does **not** mean the user has a reassuring genotype there.
+- `No-call` means the vendor included the marker but did not make a confident genotype call.
+- Composite calls are only made where the source file type can support them. OpenDNA still does **not** infer rare variants, structural variants, HLA types, CYP2D6 star alleles, or methylation from array data.
+
 ---
 
 ## Troubleshooting

diff --git a/docs/superpowers/handover-2026-04-17.md b/docs/superpowers/handover-2026-04-17.md
@@ -16,15 +16,16 @@ Local-first web utility that parses a raw 23andMe-style DNA file, matches agains
 
 ### Shipped features
 
-- 8 curated SNP panels (cardiovascular, methylation, pharmacogenomics, athletic, dietary, HFE, cognition, stimulant sensitivity) — ~50 SNPs, each with ≥2 genotype interpretations.
-- ClinVar (14 rsids) + PharmGKB (7 rsids, CPIC drug recs for TPMT / CYP2C19 / CYP2C9 / VKORC1 / SLCO1B1) shipped subsets.
+- 8 curated SNP panels (cardiovascular, methylation, pharmacogenomics, athletic, dietary, HFE, cognition, stimulant sensitivity) — now expanded with F5 / F2 thrombophilia markers and CYP4F2 warfarin PGx support.
+- ClinVar (16 rsids) + PharmGKB (8 rsids, CPIC drug recs for TPMT / CYP2C19 / CYP2C9 / VKORC1 / CYP4F2 / SLCO1B1) shipped subsets.
 - FastAPI server with five endpoints: `/`, `/api/panels`, `/api/analyze`, `/api/analyze-stream` (SSE for live progress), `/api/update-db`.
 - Vanilla-JS SPA: DOM APIs only (no `innerHTML` on dynamic data), report rendered into sandboxed iframe. Live progress bar with per-stage messages. Optional browser `localStorage` persistence of form values behind a *Remember* checkbox + *Clear saved values* button.
 - CLI: `opendna serve`, `opendna scan <file>`, `opendna update-db`.
 - Jinja2 HTML report (autoescape on) styled after the original sandbox synthesis — stat tiles, color-coded card grid per panel, critical-findings alert box for risk tier, AI synthesis block when LLM is configured.
+- Report now includes source-file QC, vendor/build heuristics, panel completeness, confidence labels, gene caveats, and composite summaries for APOE, HFE, MTHFR, CYP2C19, and warfarin PGx.
 - LLM provider layer: pluggable ABC, Anthropic default (`claude-sonnet-4-6` + prompt caching), OpenAI (`gpt-4o` using `max_completion_tokens`). API keys held per-request in memory only.
 - Analyzer normalizes allele order (`AG == GA`) AND reverse complement (panel `CC` matches file `GG` for non-palindromic C/T SNPs).
-- 44 tests passing (parser, panels, analyzer, annotations, report, LLM providers, server, CLI, SSE stream, e2e). `ruff check` clean. CI runs on Python 3.11 / 3.12 / 3.13.
+- 51 tests passing (parser, panels, analyzer, summaries, annotations, report, LLM providers, server, CLI, SSE stream, e2e). `ruff check` clean. CI runs on Python 3.11 / 3.12 / 3.13.
 - Apache 2.0 license.
 - README with full setup guide, per-provider raw-DNA download instructions, sample report + screenshot, troubleshooting, author links (`@corbett3000`, `stillrush.co/bio`).
 - Sample report: `examples/sample-report.html` (served live via raw.githack; preview image at `docs/assets/sample-report.png`).
@@ -58,6 +59,8 @@ pytest -v                          # 44 tests should pass
 opendna serve                      # localhost:8787
 ```
 
+The current baseline is `51` passing tests.
+
 ## Key design decisions (for context)
 
 | Decision | Choice | Why |

diff --git a/src/opendna/analyzer.py b/src/opendna/analyzer.py
@@ -6,6 +6,16 @@
 from opendna.models import Finding, Panel, SnpDef
 
 _COMPLEMENT = {"A": "T", "T": "A", "C": "G", "G": "C"}
+_NO_CALL_GENOTYPES = {"--", "-", "00", "NN", "NC"}
+_CONFIDENCE_SCORES = {
+    "exact": 1.0,
+    "normalized": 0.97,
+    "reverse_complement": 0.9,
+    "unmatched": 0.35,
+    "not_tested": 0.0,
+    "no_call": 0.0,
+    "ambiguous": 0.15,
+}
 
 # A/T and C/G are palindromic sites: forward and reverse complements collide,
 # so RC inference is unreliable. Panels whose allele set is exactly one of
@@ -50,7 +60,30 @@ def _is_palindromic_site(snp: SnpDef) -> bool:
     return _panel_alleles(snp) in _PALINDROMIC_PAIRS
 
 
-def _match_interpretation(snp: SnpDef, genotype: str) -> tuple[str, str]:
+def _classify_call_status(genotype: str | None) -> str:
+    if genotype is None:
+        return "not_tested"
+    genotype = genotype.upper()
+    if genotype in _NO_CALL_GENOTYPES:
+        return "no_call"
+    if not genotype or len(genotype) not in {1, 2}:
+        return "ambiguous"
+    if any(base not in "ACGT" for base in genotype):
+        return "ambiguous"
+    return "called"
+
+
+def _confidence_label(score: float) -> str:
+    if score >= 0.9:
+        return "high"
+    if score >= 0.6:
+        return "medium"
+    if score > 0:
+        return "low"
+    return "none"
+
+
+def _match_interpretation(snp: SnpDef, genotype: str) -> tuple[str, str, str, str | None]:
     """Match a genotype to a panel interpretation.
 
     Fallback chain:
@@ -63,21 +96,21 @@ def _match_interpretation(snp: SnpDef, genotype: str) -> tuple[str, str]:
     # 1. exact
     if genotype in snp.interpretations:
         i = snp.interpretations[genotype]
-        return i.tier, i.note
+        return i.tier, i.note, "exact", genotype
     # 2. allele-order normalization
     norm = _normalize_genotype(genotype)
     for key, interp in snp.interpretations.items():
         if _normalize_genotype(key) == norm:
-            return interp.tier, interp.note
+            return interp.tier, interp.note, "normalized", key
     # 3. reverse-complement (only for non-palindromic sites)
     if not _is_palindromic_site(snp):
         rc = _reverse_complement(genotype)
         if rc is not None:
             rc_norm = _normalize_genotype(rc)
             for key, interp in snp.interpretations.items():
                 if _normalize_genotype(key) == rc_norm:
-                    return interp.tier, interp.note
-    return "unknown", f"Genotype {genotype} not interpreted in panel."
+                    return interp.tier, interp.note, "reverse_complement", key
+    return "unknown", f"Genotype {genotype} not interpreted in panel.", "unmatched", None
 
 
 def analyze(
@@ -92,29 +125,84 @@ def analyze(
             continue
         for snp in panel.snps:
             genotype = parsed.get(snp.rsid)
-            if genotype is None:
+            call_status = _classify_call_status(genotype)
+            if call_status == "not_tested":
                 findings.append(
                     Finding(
                         panel_id=panel.id,
                         rsid=snp.rsid,
                         gene=snp.gene,
                         genotype=None,
+                        interpreted_genotype=None,
                         tier="unknown",
-                        note="SNP not present in raw DNA file.",
+                        note="Marker not present in this source DNA file.",
                         description=snp.description,
+                        call_status="not_tested",
+                        match_method="not_tested",
+                        confidence_score=0.0,
+                        confidence_label="none",
                     )
                 )
                 continue
-            tier, note = _match_interpretation(snp, genotype)
+            if call_status == "no_call":
+                findings.append(
+                    Finding(
+                        panel_id=panel.id,
+                        rsid=snp.rsid,
+                        gene=snp.gene,
+                        genotype=genotype,
+                        interpreted_genotype=None,
+                        tier="unknown",
+                        note=(
+                            "Marker is present in the file, but the source genotype "
+                            "was a no-call."
+                        ),
+                        description=snp.description,
+                        call_status="no_call",
+                        match_method="no_call",
+                        confidence_score=0.0,
+                        confidence_label="none",
+                    )
+                )
+                continue
+            if call_status == "ambiguous":
+                findings.append(
+                    Finding(
+                        panel_id=panel.id,
+                        rsid=snp.rsid,
+                        gene=snp.gene,
+                        genotype=genotype,
+                        interpreted_genotype=None,
+                        tier="unknown",
+                        note=(
+                            "Marker is present, but the source genotype format is "
+                            "ambiguous for this panel."
+                        ),
+                        description=snp.description,
+                        call_status="ambiguous",
+                        match_method="ambiguous",
+                        confidence_score=_CONFIDENCE_SCORES["ambiguous"],
+                        confidence_label=_confidence_label(_CONFIDENCE_SCORES["ambiguous"]),
+                    )
+                )
+                continue
+
+            tier, note, match_method, interpreted_genotype = _match_interpretation(snp, genotype)
+            confidence_score = _CONFIDENCE_SCORES[match_method]
             findings.append(
                 Finding(
                     panel_id=panel.id,
                     rsid=snp.rsid,
                     gene=snp.gene,
                     genotype=genotype,
+                    interpreted_genotype=interpreted_genotype,
                     tier=tier,
                     note=note,
                     description=snp.description,
+                    call_status="called",
+                    match_method=match_method,
+                    confidence_score=confidence_score,
+                    confidence_label=_confidence_label(confidence_score),
                 )
             )
     return findings
diff --git a/src/opendna/annotations/clinvar.json b/src/opendna/annotations/clinvar.json
@@ -82,5 +82,17 @@
     "condition": "Simvastatin-induced myopathy",
     "review_status": "reviewed by expert panel (PharmGKB)",
     "source": "ClinVar 2026-03 snapshot"
+  },
+  "rs6025": {
+    "clinical_significance": "pathogenic / risk factor",
+    "condition": "Factor V Leiden thrombophilia; venous thromboembolism",
+    "review_status": "multiple submitters",
+    "source": "ClinVar 2026-04 snapshot"
+  },
+  "rs1799963": {
+    "clinical_significance": "pathogenic (low penetrance)",
+    "condition": "Prothrombin-related thrombophilia; venous thromboembolism",
+    "review_status": "multiple submitters",
+    "source": "ClinVar 2026-04 snapshot"
   }
 }
diff --git a/src/opendna/annotations/pharmgkb.json b/src/opendna/annotations/pharmgkb.json
@@ -106,6 +106,18 @@
       }
     ]
   },
+  "rs2108622": {
+    "gene": "CYP4F2",
+    "recommendations": [
+      {
+        "drug": "warfarin",
+        "diplotype": "*3 carrier",
+        "recommendation": "Use a CYP2C9/VKORC1-guided dosing algorithm and recognize that CYP4F2 carriers can need a modestly higher dose.",
+        "evidence_level": "2A",
+        "source": "CPIC 2017"
+      }
+    ]
+  },
   "rs4149056": {
     "gene": "SLCO1B1",
     "recommendations": [

diff --git a/src/opendna/cli.py b/src/opendna/cli.py
@@ -11,8 +11,9 @@
 from opendna.annotations import annotate, load_clinvar, load_pharmgkb
 from opendna.annotations.updater import refresh
 from opendna.panels import load_panels
-from opendna.parser import parse_23andme
+from opendna.parser import parse_source_file
 from opendna.report import render_report
+from opendna.summaries import build_analysis_summary
 
 
 def build_parser() -> argparse.ArgumentParser:
@@ -56,12 +57,16 @@ def cmd_scan(args: argparse.Namespace) -> int:
     if not input_path.exists():
         print(f"error: file not found: {input_path}", file=sys.stderr)
         return 1
-    parsed = parse_23andme(input_path)
+    parsed = parse_source_file(input_path)
     panels = load_panels()
     selected = set(args.panels) if args.panels else None
-    findings = analyze(parsed, panels, selected_panel_ids=selected)
+    findings = analyze(parsed.genotypes, panels, selected_panel_ids=selected)
     findings = annotate(findings, load_clinvar(), load_pharmgkb())
-    bundle = render_report(findings)
+    bundle = render_report(
+        findings,
+        source_file=parsed.source,
+        analysis_summary=build_analysis_summary(findings, panels),
+    )
 
     html_out = input_path.with_name(input_path.stem + ".opendna.html")
     json_out = input_path.with_name(input_path.stem + ".opendna.json")

diff --git a/src/opendna/llm/base.py b/src/opendna/llm/base.py
@@ -18,7 +18,10 @@ def findings_to_prompt(findings: list[Finding]) -> str:
     """Render findings as a compact structured block for the LLM."""
     lines = []
     for f in findings:
-        line = f"[{f.tier}] {f.gene} {f.rsid} = {f.genotype or '--'} ({f.panel_id}) — {f.note}"
+        if f.call_status != "called":
+            continue
+        display_genotype = f.interpreted_genotype or f.genotype or "--"
+        line = f"[{f.tier}] {f.gene} {f.rsid} = {display_genotype} ({f.panel_id}) — {f.note}"
         if f.clinvar:
             line += f" | ClinVar: {f.clinvar['clinical_significance']}"
         if f.pharmgkb: