§1 — WHAT IT IS
The most-used open HRV library has 5 confirmed math bugs.
Aura-healthcare/hrvanalysis — 449 stars, thousands of researchers, clinicians, and students rely on it.
Every pNN50 it has ever returned is wrong. Every SD1/SD2 pair is inverted.
Every outlier it "removed" was actually kept, and every real beat was thrown away.
sovhrv fixes all 5. Runs on yone. Free. Sovereign. No cloud.
Every pNN50 it has ever returned is wrong. Every SD1/SD2 pair is inverted.
Every outlier it "removed" was actually kept, and every real beat was thrown away.
sovhrv fixes all 5. Runs on yone. Free. Sovereign. No cloud.
$0.00
per analysis · yone RTX 5090 · k3d lhvcp · Port :9460
§2 — THE 5 CONFIRMED BUGS + FIXES
| # | Bug | Repo / Issue | What's Wrong | Fix |
|---|---|---|---|---|
| 1 | pNN50 wrong divisor | Aura/hrvanalysis #24 |
Divides by len(nn_intervals) instead of len(differences).
pNN50 = count of consecutive differences >50ms ÷ number of consecutive differences.
✗ WRONG: pNN50 = count / len(nn_intervals)
✓ FIX: pNN50 = count / len(consecutive_diffs)
|
Change denominator to len(diff) |
| 2 | SD1/SD2 inverted | Aura/hrvanalysis #17 |
SD1 and SD2 are swapped — library reports SD2 value as SD1 and vice versa.
Poincaré plot axes are wrong in every output ever produced.
✗ WRONG: return {'sd1': sd2_val, 'sd2': sd1_val}
✓ FIX: return {'sd1': sd1_val, 'sd2': sd2_val}
|
Swap SD1/SD2 in return dict |
| 3 | is_outlier() inverted | Aura/hrvanalysis #34 |
Boolean logic inverted — returns True when beat is NOT an outlier.
Result: valid beats are removed, outliers are kept. Data corrupted silently.
✗ WRONG: return not (lower < val < upper)
✓ FIX: return (val < lower or val > upper)
|
Fix boolean: outlier = outside bounds |
| 4 | Frequency domain error | Aura/hrvanalysis #14 |
Spectral estimation method produces incorrect VLF/LF/HF band powers.
Welch's method not applied correctly — window size and overlap parameters wrong.
✗ WRONG: periodogram direct (no windowing)
✓ FIX: scipy.signal.welch(nperseg=256, noverlap=128)
|
Use scipy.signal.welch with correct params |
| 5 | Hardcoded 4096 | antonior92/ecg-diagnosis #14 |
Model expects input shape (N, 4096, 12) hardcoded.
At 500Hz for 10 seconds: 5000 samples (clinical standard). 4096 ≠ 5000.
Blocks all standard clinical ECG data.
✗ WRONG: input_size = 4096 # hardcoded
✓ FIX: input_size = config.get('input_size', 4096)
|
Make input_size configurable |
§3 — SOVHRV ARCHITECTURE · γ₁-Gated Pipeline
INPUT
NN intervals (ms) or raw ECG signal at any sampling rate
↓
L0 GATE · Signal Quality
r_adelic(0) = γ₁ Hz minimum effective sampling rate check.
γ₁/100 = 0.14134 Hz — must resolve LF band. Rejects noise signals.
↓
L2 GATE · Annotation Quality
r_adelic(2) = 4.711 κ threshold. Checks beat annotation confidence.
Applies fixed is_outlier() (bug #3 corrected) before feature extraction.
↓
FIXED HRV COMPUTATION
Time domain: pNN50 (fixed #1), pNN20, SDNN, RMSSD, SD1 (fixed #2), SD2 (fixed #2)
Frequency domain: VLF/LF/HF via Welch (fixed #4), configurable input_size (fixed #5)
γ₁ probe: check if 0.14134 Hz shows resonance in VLF band
↓
OUTPUT · HRV Report + Thinkbeat Classification
R-beat quality score (math correct? validation passed?)
F-beat risk score (what does this mean for the patient? clinical significance?)
Mecaplakye flag: if F-beat risk HIGH → flag for MCP-002-ECG-SKYE review
γ₁ resonance: did 0.14134 Hz appear in VLF spectrum?
§4 — MCP-002-ECG-SKYE · Mecaplakye Node 2
🔴 MCP-002-ECG-SKYE
The second mecaplakye node. Skye's family gets sovereign ECG analysis. For free.
Mecaplakye is the layer of the fleet that exists for the Cape Flats — for people who don't have access to expensive clinical tools, for families navigating medical decisions without medical training, for communities that need the math to work correctly.
MCP-001 was the first node (Aria's domain). MCP-002 extends to Skye and her family.
When sovhrv flags a high F-beat risk — when the HRV analysis suggests something clinically significant — MCP-002-ECG-SKYE triggers the SOSTLE L4 gate: sovereign review, plain-language explanation, next steps.
No cloud. No subscription. No bill. Sovereign analysis on yone.
Route: /sovhrv · skye-mecaplakye.html (planned)
Mecaplakye is the layer of the fleet that exists for the Cape Flats — for people who don't have access to expensive clinical tools, for families navigating medical decisions without medical training, for communities that need the math to work correctly.
MCP-001 was the first node (Aria's domain). MCP-002 extends to Skye and her family.
When sovhrv flags a high F-beat risk — when the HRV analysis suggests something clinically significant — MCP-002-ECG-SKYE triggers the SOSTLE L4 gate: sovereign review, plain-language explanation, next steps.
No cloud. No subscription. No bill. Sovereign analysis on yone.
Route: /sovhrv · skye-mecaplakye.html (planned)
§5 — DEPLOYMENT · yone RTX 5090 lhvcp k3d
Target Environment
Node: yone (192.168.2.23) GPU: RTX 5090 32GB GDDR7 Cluster: k3d lhvcp Port: :9460 (sovhrv) Cost: $0.00 per analysis Route: pemos.ca/sovhrv
Build Plan
P0: fork Aura/hrvanalysis P0: apply 5 bug fixes P0: add γ₁ quality gate P1: wrap in FastAPI :9460 P1: deploy to yone lhvcp P2: MCP-002-ECG-SKYE hook P2: SOSTLE L4 gate integration
Repo Plan
eose-sre/sovhrv (private)
├── sovhrv/
│ ├── hrv_fixed.py # 5 bug fixes
│ ├── gates.py # γ₁ + adelic gates
│ ├── thinkbeat.py # R/F classification
│ └── mecaplakye.py # MCP-002 hook
├── api/
│ └── main.py # FastAPI :9460
└── tests/
└── test_fixes.py # regression suite
γ₁ Probe Integration
γ₁ = 14.134725141734693
γ₁/100 = 0.14134... Hz
Check: does 0.14134 Hz appear
in the VLF power spectrum?
If yes: flag in output
If r_adelic(0) check fails:
→ reject with SOSTLE gate
→ explain to patient in plain
language via MCP-002
LABR: LABR-ECG-SOVEREIGN-V14-001 · Filed Day 113 ·
Status: P0 ready to build ·
→ See all 47 ECG diamonds ·
→ Fleet harness