Spectacle sits above the systems you already run, and turns what they know into what to do next.
The reason a new sector does not mean new software: only the middle layer changes.
Identical for every operation. Forecasting, simulation, computation paths, permissions, audit log. This is where the maths happens, and only here.
A dedicated object model per sector: pure configuration, no new program code. That is why a new sector means a new model on the same core, not a new product.
Your own objects, metrics and thresholds, which never touch the core, so an update cannot overwrite them.
Sectors differ in vocabulary, not in structure. An item consists of components, consumes stock, occupies capacity and costs working time, whether it is called a croissant, a room category or a class slot.
| In the core | Called in practice |
|---|---|
| Object | SKU · Work item · Case file · Lettable area · Screening · Menu line |
| Bill of materials | BOM · Cut · Pack structure · Service package · Recipe |
| Capacity | Machine · Floor area · Treatment room · Bay · Auditorium · Room |
| Resource | Shift · Route · On-call · Project role · Staffing |
| Demand signal | Order · Booking · Appointment · Season · Weather · Event |
| Constraint | Working-time law · Shelf life · Changeover · Lead time · SLA |
Expected demand per object and day, always as a range, because procurement and staffing derive from it.
A measure is computed before it runs. You see the expected effect, not only the recommendation.
Ranked by expected euro impact and limited to the most important. Twelve equally weighted hints are a second job, not help.
Questions in plain language, answered across domain boundaries, with inherited permissions and a named source per number.
Logged, reversible, released by a human. You decide what runs automatically, what needs release, and what stays manual.
Downstream of your source systems: read, never written back. Tenant separation enforced in the database, EU endpoints with no retention for personal data.
A recommendation is worth what can be traced behind it. That is the part a screenshot never shows.
Objects, attributes and relationships as a knowledge graph. Without a twin there is no simulation, only extrapolation.
Time-series models per object and day. Output as an interval, validated by backtest against your own history.
Measures are computed against the twin before they run, including knock-on effects on stock and staffing.
Run id, period, inputs, computation path. Every number traces back to its source.
Distributed tracing across the pipeline, structured logging, alerting on model drift and source-system failure.
The AI calls verified tools and phrases what they return. No model computes, and none alters a number.
Row-level security per tenant, enforced in the database. The AI inherits the permissions of whoever asked.
Every action logged, versioned and reversible, with timestamp, trigger and the person who released it.
EU endpoints with no content retention, no training on your data, subprocessors declared per service.
Spectacle fits wherever three things hold at once: demand is uncertain, capacity is fixed in the short run, and labour and material must be committed before the demand is known. That describes a distribution centre as precisely as a production line or a care home.
Batch sizes, changeover time, material requirement and line loading against a fluctuating order book.
Pick volume per shift, dock scheduling, labour requirement and stock cover across sites.
Dispatch planning, spare-part demand and SLA adherence across a distributed crew.
Occupancy, staffing by qualification, and rostering under hard working-time law.
Demand per item and branch, freshness and write-offs, staffing against footfall.
Hospitality, hotels, canteens, leisure and events: demand per part of day, stock and staffing.
Twenty minutes is enough to establish whether your object model already exists, what it would be called in your trade, and whether the effort is worth it.