ruvector/docs/adr/ADR-040b-microlensing-graphcut-extensions.md

ADR-040b: Microlensing Detection & Cross-Domain Graph-Cut Extensions

Status: Proposed Date: 2026-03-15 Parent: ADR-040 Related: ADR-003 (RVF Format), ADR-006 (Unified Self-Learning RVF)

Context

This sub-ADR documents the microlensing detection pipeline (M0-M3) and cross-domain graph-cut applications that extend ADR-040's core architecture. These extensions demonstrate how the MRF/mincut + RuVector pattern generalizes beyond transit-based planet detection to gravitational microlensing and to non-astronomy domains including medical imaging, genomics, financial fraud detection, supply chain monitoring, cybersecurity, and climate analysis.

Extracted from ADR-040 to keep individual files under 500 lines per project guidelines.

Microlensing Detection Pipeline (M0-M3)

Extends the transit pipeline to gravitational microlensing events for rogue planet and exomoon candidate detection.

Additional Data Sources

Source	Access	Reference
OGLE-IV microlensing events	Public FTP	ogle.astrouw.edu.pl
MOA-II microlensing alerts	Public archive	www.massey.ac.nz/~iabond/moa
Gaia astrometric catalog	ESA public	gea.esac.esa.int
Roman Space Telescope	Upcoming	High-cadence microlensing survey

Stage M0: Ingest

Microlensing light curve ingestion from OGLE/MOA archives.

• Normalize photometry across surveys (MOA-II: ~15 min cadence, 0.008sigma floor; OGLE-IV: ~40 min cadence, 0.005sigma floor)
• Segment into event windows around magnification peaks
• Store as RVF Event nodes with survey-specific metadata

Stage M1: Single-Lens Detection

Standard Paczynski curve fitting for point-source point-lens (PSPL) events.

• Two-phase fit: coarse grid search + fine refinement
• Linear regression for source flux (F_s) and blending flux (F_b)
• Parameter estimation: Einstein crossing time (t_E), impact parameter (u_0), peak time (t_0)
• Create Candidate nodes for events exceeding SNR threshold

Stage M2: Anomaly Detection via MRF/Mincut

Residual analysis after best-fit PSPL subtraction using Markov Random Field optimization with graph cut inference.

Three-statistic lambda computation:

1. Excess chi2: window fit quality relative to global reduced chi2
2. Runs test: temporal coherence of residual sign changes
3. Gaussian bump fit: localized perturbation chi2 improvement

Differential normalization: Compare each window's signal to tau-space neighbors, producing z-scores that are ~0 for uniform fit quality and positive only for localized anomalies.

Graph construction:

• Temporal chain: consecutive time windows (alpha weight)
• RuVector kNN edges: embedding similarity from residual features (beta weight)
• Edmonds-Karp BFS max-flow solver for s-t mincut

Stage M3: Coherence Gating

Dynamic mincut separates competing explanations (noise vs planet vs moon vs binary lens) using support region analysis.

Component	Role
RuVector embeddings	Cross-survey memory; retrieve similar anomalies
HNSW index	Fast similarity search across microlensing events
Dynamic mincut	Separate competing explanations
Witness chain	Full provenance from raw photometry to candidate

Important constraint: Dynamic mincut provides coherent support extraction but cannot replace lens modeling. The physics solver (PSPL/binary) provides local evidence; mincut provides spatial coherence.

Cross-Domain Graph-Cut Applications

The MRF/mincut + RuVector architecture generalizes beyond astronomy to any domain requiring coherent anomaly detection in structured data.

Implemented Verticals

Vertical	Example	Graph Topology	Anomaly Types
Medical Imaging	medical_graphcut.rs	4-connected spatial grid, gradient-weighted edges	Lesion segmentation (T1-MRI, T2-MRI, CT)
Genomics	genomic_graphcut.rs	Linear chain + kNN similarity	CNV gains/losses, LOH, cancer drivers (TP53, BRCA1, EGFR, MYC)
Financial Fraud	financial_fraud_graphcut.rs	Temporal chain + merchant edges + kNN	Card-not-present, account takeover, card clone, synthetic, refund
Supply Chain	supply_chain_graphcut.rs	Tier chain + geographic + kNN	Quality defect, shortage, price anomaly, delay, counterfeit, demand shock
Cybersecurity	cyber_threat_graphcut.rs	Source/destination chain + kNN	Port scan, brute force, exfiltration, C2 beacon, DDoS, lateral movement
Climate	climate_graphcut.rs	Spatial adjacency + gradient + kNN	Heat wave, pollution spike, drought, ocean warming, cold snap, sensor fault

Common Architecture

All verticals share:

1. Domain-specific data generator with realistic parameters from published datasets
2. Feature extraction producing 32-dim embeddings for RuVector storage
3. Graph construction combining domain topology (spatial/temporal/chain) with kNN similarity edges
4. Edmonds-Karp BFS s-t mincut solver (identical implementation), with optional QAOA quantum graph-cut solver via ruQu as a drop-in alternative (qaoa_graphcut.rs; see ADR-040 Recent Enhancements)
5. RVF integration: witness chains, filtered metadata queries, lineage derivation
6. Evaluation: comparison against threshold baseline showing graph-cut improvement

Additional Real Data Sources

Domain	Source	Access
Genomics	TCGA (The Cancer Genome Atlas)	portal.gdc.cancer.gov
Genomics	ClinVar variants	ncbi.nlm.nih.gov/clinvar
Genomics	COSMIC (somatic mutations)	cancer.sanger.ac.uk/cosmic
Medical	LIDC-IDRI (lung CT)	cancerimagingarchive.net
Medical	BraTS (brain tumors)	synapse.org
Cybersecurity	CICIDS2017	unb.ca/cic
Climate	NOAA GHCN	ncei.noaa.gov
Climate	EPA AQI	epa.gov/aqs

Measured Results

Results from running examples with cargo run --example <name> --release.

Astronomy

Pipeline	Metric	Value	Notes
planet_detection	Recall@10	8/10 confirmed	Synthetic Kepler-like targets
exomoon_graphcut	Precision	0.25	Chang-Refsdal perturbative limit
exomoon_graphcut	Recall	0.25	Limited by ~2-3sigma per-window SNR
exomoon_graphcut	F1	0.25	30 events, 12 with moons
real_microlensing	Planets rediscovered	2/4	OGLE-2005-BLG-390, OGLE-2016-BLG-1195
microlensing_detection	Events processed	20	Synthetic PSPL + anomalies

Medical Imaging

Modality	Graph Cut Dice	Threshold Dice	Improvement
T1-MRI	0.44-0.59	0.32-0.46	+27-39%
T2-MRI	0.50-0.62	0.38-0.48	+29-32%
CT	0.48-0.58	0.35-0.44	+32-37%

Genomics

Platform	Sensitivity	Specificity	Drivers Found
WGS (30x)	0.91-0.95	0.66-0.97	4/4 (TP53, BRCA1, EGFR, MYC)
WES (100x)	0.88-0.93	0.70-0.95	4/4
Panel (500x)	0.93-0.97	0.72-0.98	4/4

Rust Implementation

The pipeline examples are implemented in Rust using the RVF runtime crates:

Crate	Role
rvf-types	Core types, segment definitions, derivation types
rvf-runtime	RvfStore, HNSW indexing, metadata filters, queries
rvf-crypto	SHAKE-256 witness chains, verification
rvf-wire	Wire format serialization
rvf-manifest	Segment manifests and policies
rvf-index	Progressive HNSW index construction
rvf-quant	Vector quantization (PQ, SQ)
rvf-kernel	Kernel/initrd segment embedding
rvf-launch	Boot sequence and runtime launch
rvf-ebpf	eBPF socket/syscall policies
rvf-server	HTTP/WS server for dashboard and API

Examples location: examples/rvf/examples/

References

1. ADR-040: Causal Atlas RVF Runtime — Planet Detection & Life Candidate Scoring
2. OGLE Microlensing Events
3. MOA Microlensing Alerts
4. Gaia Archive
5. TCGA Data Portal
6. CICIDS2017 Dataset