Skip to content

Beachhead Verticals

A beachhead is a small set of domains where the same core capability — detecting and steering phase (de)synchronisation in a network of coupled oscillators — maps directly onto an expensive, well-understood operational failure mode. The three verticals below are the focus tracks: each ships lead domainpacks whose binding specs validate, whose run.py scenarios execute end to end, and which are covered by dedicated tests.

Every pack is a worked example, not a benchmarked product claim. Where a quantified baseline-vs-orchestrated comparison exists, it lives on the linked benchmark page; the commands here reproduce the scenario locally.

Run any pack two ways:

# Self-contained narrated scenario
python domainpacks/<pack>/run.py

# Through the CLI on the binding spec
spo validate domainpacks/<pack>/binding_spec.yaml
spo run domainpacks/<pack>/binding_spec.yaml --steps 500 --audit run.jsonl

Industrial predictive maintenance

Pain: unplanned downtime and catastrophic failure in rotating equipment, batch reactors, and production lines. The diagnostic signal is the phase relationship between harmonics, stages, or stations — misalignment, bearing defects, and process drift each have a phase signature before they breach an amplitude limit.

Pack Layers Oscillators Channels What it maps
rotating_machinery 4 10 P/I Shaft harmonics, blade dynamics, bearing defect frequencies, structural modes (ISO 10816-3 / API 670 boundaries)
manufacturing_spc 3 9 P/I/S Sensor → machine → line coupling, bad-layer suppression under policy rules
chemical_reactor 4 10 P/I Reaction kinetics, heat transfer, pressure vessel, feed flow with hard temperature/pressure limits

Validation: all three validate, run end to end, and carry dedicated domainpack tests (spec structure, layer map, full main() pipeline).


Critical infrastructure

Pain: cascading instability in power grids and request-driven service meshes. Generator swing-equation coherence and queue retry-storms are both synchronisation phenomena — the control objective is to hold or break phase lock before a cascade.

Pack Layers Oscillators Channels What it maps
power_grid 5 12 P/I Inertial Kuramoto swing dynamics, generator-trip transients (benchmark)
queuewaves 3 6 P/I Upstream/downstream queue phase-lock, retry-storm desync; ships a production server template

Validation: the most mature vertical — power_grid carries five demo tests and a benchmark page; queuewaves is a productised app with a server, pipeline, and a full async/e2e test suite, plus a queuewaves.production.yaml deployment template.


Biosignal / clinical

Pain: rhythm and network-state disorders where coherence is the clinical variable — arrhythmia, seizure dynamics, and sleep-stage architecture are all read out from oscillatory phase coherence.

Pack Layers Oscillators Channels What it maps
cardiac_rhythm 4 10 P/I SA node → conduction → depolarisation → repolarisation; sinus → PVC → VT → drug → pacing (benchmark)
neuroscience_eeg 6 14 P/I Delta/theta/alpha/beta/gamma bands plus a network layer; seizure and arousal boundaries
sleep_architecture 4 8 P/I/S AASM sleep-stage architecture inferred from band order parameters

Validation: this vertical previously had no dedicated tests; all three now carry end-to-end domainpack suites (spec validation, layer structure, objective sanity, boundaries/actuators, and full main() execution).

Clinical use is gated on regulatory validation (CE/MDR, clinical studies). These packs are research and engineering references, not medical devices.


Operating posture and safety tier

Each binding spec declares a safety_tier. The local runtime executes live spo run only for research-tier specs; consumer, production, and clinical tiers fail closed with a clear message and must go through the formal-export and certified-controller pipeline before any live or actuating run. The narrated run.py scenarios always execute (they demonstrate the research-mode engine), so every pack is explorable, but the higher tiers cannot be driven live from the CLI by design.

Pack Safety tier Live spo run Path to live use
queuewaves research yes runs directly
neuroscience_eeg research yes runs directly
sleep_architecture research yes runs directly
rotating_machinery consumer gated formal export → certified controller
manufacturing_spc consumer gated formal export → certified controller
chemical_reactor production gated formal export → certified controller
power_grid production gated formal export → certified controller
cardiac_rhythm clinical gated formal export → certified controller + CE/MDR

This tiering is a feature, not a limitation: it keeps non-research domains from being actuated without the evidence pipeline. See the formal-verification export in the supervisor reference for how a gated spec reaches a certifiable controller.

From demo to deployment

The operator path depends on the pack's safety tier (see the table above).

Research tier (queuewaves, neuroscience_eeg, sleep_architecture) — the full evidence chain runs locally:

spo validate domainpacks/<pack>/binding_spec.yaml          # schema + resolved config
spo run domainpacks/<pack>/binding_spec.yaml \
    --steps 500 --seed 42 --audit run.jsonl                # SHA256-chained audit log
spo report run.jsonl                                       # deterministic summary
spo explain run.jsonl --markdown-out explain.md            # regulator-oriented narrative

Before any production sign-off, confirm reproducibility for your configuration with spo replay run.jsonl --verify; the determinism profile of a specific binding spec, backend chain, and host is part of commissioning, not an assumption.

Consumer / production / clinical tier (rotating_machinery, manufacturing_spc, chemical_reactor, power_grid, cardiac_rhythm) — live spo run is gated. Explore the dynamics with the narrated python domainpacks/<pack>/run.py, then reach a controllable system through the evidence pipeline rather than the local runtime:

  1. Validate the binding spec and review the resolved configuration.
  2. Build the formal-verification package for the spec's safety properties (supervisor formal export) and run the reviewed model checkers.
  3. Bind the checker results into a non-actuating runtime-control certificate.
  4. Deploy only behind a certified controller; for the clinical tier this is in addition to CE/MDR and clinical validation.

No tier actuates without an explicit operator-approval and audit trail.

Why these three

The same engine and binding-spec workflow covers all three, so a vertical slice in one hardens the shared core for the others. They differ in go-to-market profile: industrial has the lightest regulatory drag and the clearest ROI, infrastructure is the most product-ready, and biosignal has the highest differentiation but the heaviest validation burden. See the Domainpack Gallery for the full catalogue.