How Hylios Models Supply Chains

Every analyst, software vendor, and consultant claims to model your supply chain. Most do not say what that means. This page does. It covers the four-step workflow Hylios runs to produce an intelligence brief, a Pulse score, or a scenario run, so a buyer can compare Hylios to any other modeling approach apples to apples.

The four steps run in order. Each one's output is the next one's input. Skipping any of them is the difference between modeling and pattern matching.

Step 1: World modeling

A digital twin built from one customer's ERP is a single-customer simulator. It can model that customer's known lanes. It cannot reason about a fire at a contract manufacturer in Shenzhen, a strike at a port the customer does not directly use, or a tariff on a Tier-2 input. Most supply chain risk lives in the graph beyond what a single customer can see. The world model is the substrate that makes the rest of the stack honest.

Hylios continuously pulls in public sources: earnings transcripts, regulatory filings, news, port and customs data, freight-rate indexes, weather feeds, government statistics. Each event gets normalized into a shared format: an entity (company, port, region, regulator, input), an event type, a magnitude, a timestamp, and a citation. The graph holds typed connections between entities: supplier-of, customer-of, uses-input, routes-through, regulated-by, competes-with.

The graph is the same for every customer. It updates continuously, not per-engagement. This is the leverage point: every new public datum strengthens every customer's view, because every customer's overlay points into the same shared graph.

Why the world model is public-only

Mixing customer-private and public data in the world model creates a contamination problem that is hard to unwind. One customer's supplier list could leak into another customer's scenario. The audit story collapses. The system becomes hard to defend at SOC 2 time. Hylios's answer is discipline at the layer boundary: the world model is built from sources anyone could re-pull. The customer overlay never moves downstream into the world.

Step 2: Customer overlay

The overlay is the small, private set of facts that make one customer's view of the world different from another's. For most enterprise buyers this looks like: a supplier list with country and category tags, a simplified bill of materials for the Tier-1 inputs that matter, a list of lanes with carriers and modes, and a note of which regulators apply (FDA, FAA, EPA, CFIUS). The overlay arrives as files the customer uploads today or, coming soon, through direct connections to SAP S/4HANA Cloud and Oracle Fusion Cloud SCM.

Once attached, the overlay tells the platform which slice of the world model is relevant to that customer. The same Indonesian nickel mine matters to a battery customer and is irrelevant to a packaged foods customer down the street. The overlay is the difference.

Step 3: Baseline calibration

The world model plus the overlay describe what could happen. The baseline tells you what does happen for this customer when no shock is applied: on-time delivery rate, cost per pound, working capital position, disruption incidence in a normal week. Without a calibrated baseline, every what-if produces numbers with no anchor.

Hylios calibrates against 12-24 months of the customer's observed history, comparing what the model would have predicted to what actually happened. The gaps (model residuals) get inspected, and the model's connection weights get adjusted until those gaps stabilize. The calibration is presented as a transparent report: which links were over-weighted, which under-weighted, what the remaining error looks like by metric. A baseline only ships when the calibration meets a documented threshold. Below it, the report flags the gap and the customer can choose to enrich the overlay or proceed with wider confidence ranges.

A claim like "your effective tariff exposure to Vietnam is 8.4 percent of input spend" lands differently when it is anchored to a calibration the buyer can re-run against their own historical invoices. The baseline is the foundation every scenario builds on.

Step 4: What-if execution

A what-if is a deliberate change to the calibrated state. The customer describes the scenario (or picks from a library): a Section 301 expansion to a new HTS category, a 30 percent throughput loss at Long Beach for four weeks, a primary contract manufacturer's parent filing Chapter 11. The platform pushes the change through the customer's slice of the network, re-computes downstream metrics, and reports a range with a confidence interval rather than a point estimate.

Every scenario runs for 12 months by default. That horizon is long enough to capture seasonality and downstream consequences (a sourcing change shows up in working capital on a delay; a tariff shows up in margin gradually as inventory cycles through). It is short enough to stay in the range where the model is honest. Operators can shorten the horizon for fast-changing questions or extend it for strategic ones.

Range reporting is non-negotiable. Anyone who hands a CFO a single number for a future outcome is either over-confident or selling something. Hylios output for a what-if always includes: the central estimate, the 80 and 95 percent ranges, the inputs most responsible for the spread (a sensitivity table), and the public events that informed the propagation. Those events are the through-line back to the world model. Every result is auditable to its sources.

How this shows up in the product

A buyer can see all four steps in any Hylios deliverable. A daily intelligence brief is the world model plus the overlay's relevant slice, filtered to last-24-hour events. A Pulse score is the calibrated baseline summarized to a single number. A scenario run is a what-if with range output and a sensitivity table. Each artifact carries its citations and its confidence labels because Hylios publishes the methodology behind the labels too. See the confidence-labeling page for the four-tier taxonomy applied to every claim.

FAQ

How is this different from a digital twin?

A digital twin is the artifact this methodology produces. The world model is the substrate, the customer overlay is the personalization, the baseline is the validated steady state, and the what-if is the interactive layer. Other digital twin vendors build the twin from the customer's data alone, which is faster to stand up but loses the leverage of a shared world model.

Why 12 months for every simulation?

Because most of the consequences supply chain leaders care about play out over months, not weeks or years. Twelve months captures one full seasonal cycle, the working capital implications of a sourcing change, and the downstream lane and inventory effects of any disruption. Shorter horizons miss those effects; longer horizons run past the range where the model can be calibrated honestly.

Where can I see this in action?

Open any company brief at hylios.com/intelligence to see the world model's output filtered to a single company. The confidence labels and citations on every claim trace back through the four steps described here.