Multiple Analyses of One Structure
Configure different inputs, follow concurrent calculations and examine their results against a common financial structure.
- One structure is analysed with three input configurations, A, B and C.
- Each configuration has a separate result stream; the square marks represent arriving batches.
- Illustrative metric values are aligned by model node and configuration. The ellipsis indicates a result not yet available.
What It Does
Analysts can examine how a financial structure behaves under different inputs and analysis options. Each configuration produces its own calculation and result stream. The interface listens to those streams and places the arriving metrics against the relevant model components and configuration.
Several analyses can remain active together. Users can inspect partial results without waiting for every calculation to finish, and compare the resulting views of the structure.
The Design Problem
An analysis is more than a request to run a calculation. It joins a structure, its parameters, input datasets and options such as the selected calculation model or historical version. Different configurations can require different data even when they concern the same structure.
Results arrive in batches and out of order. The interface has to retain the relationship between an input configuration, its running calculation and the metrics being displayed, including when a user changes options or interrupts the work.
Design Decisions
Input configuration
The client prepares a calculation representation for each configuration. Historical analyses retrieve the corresponding model snapshots and dataset information. Uploaded data must have finished processing.
The structure’s connections, its components’ parameters and the chosen inputs are assembled together. This keeps configuration attached to the model it will evaluate.
Result streams
Submitting a calculation returns a run identity. The client associates that identity with its analysis options and subscribes to its result stream. Each run has its own completion and timeout handling.
A slow calculation does not have to finish before another can contribute results. Incoming batches are accumulated under their configuration.
Result mapping
Metrics are associated with the model components they describe, rather than displayed as an undifferentiated stream of messages. The configuration and model identities provide the coordinates for placing each result in the analysis view.
This separates the timing of computation from how analysts read its output: several streams can produce different analytical views of the same structure.
Cancellation and cleanup
Interruption requests remote cancellation and releases the client’s subscriptions. Changes to relevant analysis options also end the previous listening lifecycle, so its streams no longer update that view.
The interface can stop listening immediately; the calculation service remains responsible for stopping execution. Those are separate operations.