Assured Sensing Stack
BSE cleans the sensor stream → detection (± SSO) → MoridainDefense ranks what the operator sees. One co-sell bundle from raw time series to role-tailored alert queue.
- 0.80
- Recall@10 min (Phase 1 synthetic)
- 1.00
- Cross-cuing rate (Phase 1)
- 97.8%
- BSE win rate (135 dysts benchmark)
- 0.83
- MoridainDefense Recall@10 (alerts)
The scenario
Set the picture
A shipboard radar track picture degrades under sea clutter and jamming. The fusion node still generates hundreds of candidate alerts per watch. The operator endpoint cannot host the fusion math — it needs a bounded queue that surfaces the threats that matter to that role without drowning in single-signal false positives.
Today's stack treats signal conditioning, detection, and operator triage as three separate procurements with three separate ML retraining cycles. When the AO changes, each layer drifts independently.
What it costs today
Custom denoising per sensor program. CFAR tuned per environment. On-device ML triage retrained per deployment. No shared bounded handoff between layers — each vendor optimizes locally and the operator pays the integration tax.
BSE mixed-noise win rate is 50.3% on the disclosed benchmark — co-sell with eyes open; quasi-periodic structural losses 2.2% disclosed.
What changes with SolvNum
Four Resonix SKUs compose a single Assured Sensing reference architecture — each layer validated independently, wired in the Phase 1 synthetic harness.
Bounded-influence Takens LP denoising never degrades the input signal. 97.8% win rate on 135 dysts systems in the synthetic benchmark. Optional ABFR adaptive pre-screen (τ=1.2) before OS-CFAR in the hybrid wiring path.
Fusion-node alerts compress to 17-byte bounded packets. Role/AO projection on the endpoint (32-byte static profile). Recall@10 0.83 on synthetic streams; cons-5 cross-cuing 1.000 in the 8-seed sweep.
Measurable outcome
What we'll claim — and how it survives review
Each line below maps to a captured number in the demo section. Every number is reproducible from the SolvNum validation suite.
- Phase 1 synthetic harness: recall@10 minimum 0.80, cross-cuing 1.0 (20-seed strict gate).
- G001-HYBRID-INT (job 8646): BSE→mor_adaptive(τ=1.2)→OS-CFAR→MoridainDefense — no regression vs baseline.
- Optional SSO Defense ATR: +6–15 pp precision at 50% coverage on MSTAR (classification-only; bbox SSO not shippable).
- Optional SolvFilter EKF prediction at dim=140: 74× cheaper than UKF, best NEES consistency at scale (SF-3 CONFIRMED).
- Field operational tapes: POC — partner shadow required (G-001, G-003).
The demo
What was tested. How. What the script printed.
Phase 1 extension harness (20 seeds): BSE denoised time series → detection branch → synthetic fusion alerts → MoridainDefense bounded triage. Measured recall@10, cross-cuing rate, cons-5 injection test.
Hybrid wiring branch (job 8646): same metrics with ABFR pre-screen inserted before OS-CFAR. Compared against baseline chain without hybrid pre-screen.
Illustration
In-browser diagram of what the demo proves. The numbers underneath are the captured demo output.
Upstream · raw fusion ingest
300 alerts × 12 dims~28.1 KB / cycle raw float64 — alert flood: watchlist, geofence, comms-pattern, dwell, vehicle sig, sigint, network, pattern-of-life, sensor corroboration, history…
Fusion mainframe
log-norm → per-signal cap → consequence × probability
17-byte handoff packet(+2 B AO tag · projection on)
5.6× vs raw float64300 × 17 B = 5,100 B / cycle on the wire (base) · 19 B / alert with the optional 2-byte AO tag side-band · no model weights · no calibration tables
Per-endpoint local relevance projection
same packet stream → 32 B static role/AO profile per endpoint → bounded local re-rank (≤ 1 cons-band shift / entry) → top-K heap
severe-band (cons 5) alerts surface for every role they touch, regardless of profile
Operator endpoints · one stream → four role-projected queues
Strike pilot
Knee-board
SOF ground team
Handheld
Ship CIC
Console
Vehicle TOC
Toughbook
Captured demo output
The numbers the script actually printed.
| Metric | Result | Gate |
|---|---|---|
| Recall@10 (min) | 0.80 | ≥ 0.80 |
| Cross-cuing | 1.00 | ≥ 1.0 |
| Hybrid wiring (job 8646) | PASS | No regression |
Composes with
Where this POC sits in the substrate
Every POC reinforces — and is reinforced by — others. Click through to see how each piece locks into the larger picture.
MoridainDefense — Operator-Endpoint Alert Triage
MoridainDefense operator triage is the decision layer of this stack.
Counter-UAV Kill Chain
SolvFilter EKF is the optional tracking layer inside the detection step.
Model-Free Anomaly Detection on Sensor Streams
Scale-discontinuity detection complements BSE on sudden regime changes.
Assured Spectrum — Contested EW Deinterleaving
Assured Spectrum is the parallel EW path on PDW streams.
Evidence pointers
Where the claims live in the repo
These are the files a reviewer should run, read, or grep to re-derive every number on this page.
- moridain/business/assured_sensing_stack.md
- moridain/business/g001-assured-sensing-e2e-runbook.md
- solv/17-defense-validation-gaps.md (G-001, G-003)
- phase1_extension_20260525T163757Z/
- G001-HYBRID-INT job 8646
Previous · POC 15
Fire Team Distributed Radar Dome
Next · POC 17
SolvFilter — Scalable Multi-Target EKF
Want to see this in your environment?
Brief us on a program where this POC matters.
ITAR-aware. Air-gapped delivery available. Every claim above traces back to a script in the public repo.