A brief goes in. A grounded, QA-gated deliverable comes out.
Kern is an agentic supply-chain engine: 50 agent-routable capabilities behind one pipeline, grounded in a knowledge graph of 26 curated external sources, adapted to each client's own cost and capacity parameters.
- BriefA plain-language request, optionally with a data file attached.
- ClassifyThe orchestrator matches the brief's intent to one of 50 registered tools.
- RunThe matched tool's own prepare -> run pipeline executes against the data provided.
- QAThe tool's own QA gate checks the result. If QA fails, nothing ships -- zero deliverables.
- DeliverA grounded, cited deliverable -- or, if execution wasn't safe, ranked options, a prepared handoff, or an escalation.
The same 50 tools, grouped two ways
Transform (production/manufacturing execution) is Kern's thinnest SCOR category by design -- Kern is a planning and decision-support engine, not a manufacturing execution system (MES).
Click a legend entry (or a donut segment) to see its tools. Each tool sits in exactly one bucket per lens.
How it's fed, and how it adapts to your context
Every deliverable carries L3 citations. On the package path they also pass citation_gate (minimum 2 citations, max 2 hops, an EXCLUDED_CONCEPTS false-friend filter): a candidate that does not resolve to a graph node near the tool's own anchor concepts is dropped, never substituted, and a batch below the minimum degrades to no citations at all rather than shipping one weak hit. Single-tool runs through the orchestrator are grounded but not gated.
Holding rate, order cost, service level, lead time, warehouse capacity -- asked once, stored under clients/<slug>/profile.json, and merged into every later run so the same brief produces a client-specific answer instead of a generic one.
Enforced in one place by the orchestrator. A result that fails its own tool's QA check is refused, not shipped.
The deterministic engine is the core. An optional LLM provider sharpens intent routing and the narrative summary when configured, but every model in the engine runs the same with or without it.
Every result is one of four outcomes
Structural -- the four possible outcome shapes -- not a measured run-frequency split.
How this maps to SCOR and five SCM certifications
SCOR Digital Standard's own Orchestrate category -- added for the "digital" layer: twins, analytics, agents, resilience -- is where Kern's agentic guarantees (the QA gate, never-unprotected, signed staged writeback) land, ahead of what most commercial suites do here.
Certification coverage
A tool can touch more than one certification's body of knowledge at once, so this is a coverage level per certification, not a tool count.
Covered
- Demand forecasting, including intermittent demand
- EOQ and volume discounts
- Safety stock from forecast-error sigma
- (s,Q) and (R,S) policies
- Newsvendor model
- Multi-echelon Guaranteed-Service Model
- DDMRP
- ABC-XYZ classification
- Aggregate S&OP
- DRP
- Policy simulation
- Cycle counting / inventory record accuracy
- Job sequencing
- Learning curve
Gaps
- Detailed MRP-II / multi-level BOM explosion
- Formal master production scheduling (MPS)
- Order-level plant capacity management
Covered
- DRP
- Facility location
- Mode selection and LTL/FTL break-even
- Vehicle routing (CVRP)
- Warehouse layout and slotting
- FEFO / expiry management
- Cost-to-serve
- Landed cost
- Reverse logistics
Gaps
- Customs / international trade compliance beyond Incoterm + tariff
- Operational WMS/TMS
- Multi-facility network optimization (only single-facility siting today)
- Global trade compliance
Covered
- End-to-end integration via the orchestrator
- SCOR process mapping
- Technology: digital twin, agent, writeback
- Metrics: cash-to-cash, cost-to-serve, DEA
- TTR/TTS risk
Gaps
- Computable sustainability (only unstructured knowledge, no calculation engine)
- Network/chain design at MILP scale
- Customer collaboration / CRM
- Strategic supplier relationship management
Covered
- Network analysis: facility location, digital twin
- KPIs and benchmarking: DEA, financial KPIs
- Problem-solving: decision support, MCDM, what-if
- Project management: earned value
Gaps
- Network analysis at scale (joint optimization)
- Live external data integration (most calculation is offline by design)
- Sustainability
Covered
- Supplier performance: OTIF/PPM
- Landed cost / TCO
- Multi-criteria supplier selection: BWM/TOPSIS
- Acceptance sampling
- Change/negotiation analysis
- Supply risk
Gaps
- Deep supplier relationship management (Kraljic segmentation, power matrix, supplier development)
- Contract lifecycle management
- Category strategy
- End-to-end strategic sourcing
- Supplier ethics/diversity
- Spend analysis
ISO 9001 alignment
A fixed, self-describing pipeline: every capability is a Tool with prepare/run/qa/deliver -- the same sequence for all 49.
An EvidenceRecord (SHA-256 of inputs, control totals, formula versions, QA attestations); a writeback AuditEntry; a Keep-a-Changelog CHANGELOG; L3 citations on every deliverable, gated by citation_gate on the package path.
Grounding, confidence, and guided options on every deliverable; personas/KPIs per mode.
Partial: data_quality validates GTIN (GS1); lots/ gives FEFO plus expiry. Lot genealogy is not yet built.
A double gate: the tool's own QA, plus a citation_gate in packages.
QA fails => zero deliverables. A package's runner computes everything first and only writes if every step passes.
Writeback stage -> signed, time-boxed approval -> apply -> rollback. Autonomy config changes only via a signed Changeset.
forecast_metrics, financial_kpis, dea, reliability.py.
A verify/backtest loop feeding a ToolReliabilityReport, which drives evidence-based autonomy promotion.
ISO 28000 alignment
risk.py: EMV, FMEA RPN, a risk heatmap, and the TTR-vs-TTS gap; risk_period for lead-time risk.
Bounded parameters, client allowlisting, path-traversal defense, a 25 MB upload cap, rate limiting, constant-time API-key comparison, CSP, formula-injection defusing.
Signed, time-boxed (HMAC) approvals tied to an idempotency + content hash; autonomy tiers with human gates; SLA-bound escalation.
A hardening flag refuses to boot without required controls; the never-unprotected contract; idempotent apply plus rollback; graceful degradation without an LLM key.
Honest gaps
What Kern does not cover yet, and which standard asks for it.
Computable sustainability
Only unstructured knowledge (a book in the graph); no carbon/GHG calculation engine yet, even though the pricing/logistics layer already names sustainability as a persona concern.
Asked for by: SCOR ES, CSCP, ISO 14001
Deep supplier relationship management
Supplier scorecarding is OTIF/DIFOT plus defect rate only; no Kraljic segmentation, power analysis, or supplier development.
Asked for by: CPSM, CSCP
Lot-level traceability / genealogy
GTIN validation plus FEFO/expiry exist; no one-up/one-down chain of custody or EPCIS event log.
Asked for by: ISO 9001 8.5.2, food/pharma
Broad regulatory compliance
Strong on pricing (EU/UK Omnibus) and customs (Incoterm); missing trade compliance, HS classification, denied-party screening.
Asked for by: CLTD, CPSM
Quantified resilience
A qualitative TTR-vs-TTS gap exists in the risk register; no network stress-test with recovery time/impact simulation.
Asked for by: SCOR AG (agility)
Network optimization at scale
Only single-facility siting (Weiszfeld); no multi-facility p-median/MILP.
Asked for by: CLTD, SCPro