Glossary
TL;DR
Plain-English definitions of every Corridor platform term, grouped by area. Use this as a quick lookup when you encounter an unfamiliar word in the UI or in another doc.
Registry objects
- Table
- A registered source data table (Parquet, Hive, ORC, CSV, Snowflake, or JDBC). The starting point for everything else in the platform.
- Primary Table
- The Table that holds the exhaustive list of unique identifiers (e.g. every Application ID, every Customer ID) for an entity. Other tables join against it.
- Primary Column
- The column inside a Primary Table that uniquely identifies a row.
- Data Element
- A governed wrapper around one column or one aggregated computation, exposing it with metadata, tags, and lineage for use downstream.
- Simple Data Element
- A Data Element that maps one-to-one to a single column from a Table.
- Aggregated Data Element
- A Data Element computed by rolling up multiple rows from one or more Tables (sum, count, max, custom logic) using Python, Pandas, or PySpark.
- Feature
- A reusable, versioned variable produced by transforming Data Elements or other Features. Features become the building blocks of Models and Policies.
- Model
- A trained or imported prediction artifact (scikit-learn, PMML, ONNX, H2O MOJO, or custom Python) registered as a governed object with inputs, outputs, training data, and explainability metadata.
- Policy
- A decision workflow that turns Model scores and rules into actual outcomes (approve, decline, refer, offer terms). Policies are what get deployed to your LOS or scoring engine.
- Framework
- A reusable parameterized definition (e.g. a product framework) that Policies or Models can consume as input, so shared configuration is defined once.
- Algorithm
- A registered computational routine — typically a training procedure or optimization engine — that an Experiment or Model registration can invoke.
- Dataset
- A curated combination of Tables and Features assembled for training, validation, or scoring.
- Transform
- A reusable piece of data-shaping logic attached to a Dataset or Model that prepares inputs before scoring.
Policy building blocks
- Strategy
- A node inside a Policy that evaluates rules and produces an outcome. Strategies chain together to form the full decision workflow.
- Application Decision
- A Strategy type that decides approve / decline / refer for an incoming application.
- Derived Variable
- A calculated field defined inside a Policy and used by downstream Strategies in the same Policy.
- Offer
- A product proposal generated by a Policy (amount, rate, term, etc.) presented to a customer or prospect.
- Offer Creation
- A Strategy type that builds candidate Offers based on inputs and rules.
- Offer Filtering
- A Strategy type that removes ineligible Offers from the candidate set.
- Offer Optimization
- A Strategy type that picks the best Offer from the filtered set based on objective rules (price, take-up probability, expected value).
- Action Rule
- A terminal rule in a Strategy that fixes the outcome (e.g. "Decline"). Nothing can follow an Action Rule in a chain.
- Segment
- A branch in the policy tree that groups rules sharing a condition (e.g. by product, channel, or borrower type).
- Waterfall
- The visualization of how a population flows through the Policy stages, showing how many records were approved, filtered, or declined at each step.
Lifecycle, versions, and approvals
- Draft
- The initial status of any object. Editable, scoped to the workspace, not yet visible to broader users.
- Pending Approval
- Status when an object has been sent for approval but has not yet been signed off. Usually locked from edits.
- Approved / Published
- Status after every required reviewer has signed off. Locked from edits; a new version is needed for further changes.
- Archived
- Status of an object that has been retired and is no longer actively used.
- Version
- A point-in-time snapshot of an object. Each edit on an Approved object creates a new Draft version.
- Approval Workflow
- A configurable sequence of Responsibilities (reviewers) that an object must pass through before it can be Approved. Defined per object type by an admin.
- Responsibility
- One reviewer slot inside an Approval Workflow. Can be a user, role, or external tool, and can have Veto Power.
- Veto Power
- A flag on a Responsibility that lets that reviewer single-handedly reject an approval. While a veto reviewer is active, the object is typically locked from edits.
- Sequence
- The order in which Responsibilities review. Same-sequence reviewers act in parallel; later sequences unlock only when every earlier sequence is complete.
- Editable by Reviewer
- A Responsibility option that lets the reviewer make edits during their review phase (useful for code-style or fair-lending reviews that may require small fixes).
- Change History
- The chronological audit log on every object: author, timestamp, version, and comments for each change.
- Lineage
- The dependency graph for an object: every upstream input it relies on, every downstream object that consumes it.
- Audit Trail
- The cross-object record of actions taken (create, edit, approve, reject, deploy). Used for compliance and traceability.
Simulations and jobs
- Simulation
- A job that runs the logic of any registered object (Data Element, Feature, Model, Policy) on historical data and returns row-level outputs plus metrics dashboards.
- Job
- Any backgrounded execution: a Simulation, a Comparison, a Portfolio Analysis. Jobs progress through statuses (Queued → Compiling → Running → Completed / Partially Completed / Failed).
- What-If Analysis
- A Simulation that always runs against a chosen baseline simulation and surfaces the difference between candidate and baseline.
- Comparison
- A side-by-side run of two simulations (often two versions of the same object) on the same dataset. Used for champion / challenger evaluation.
- Single Record Review (SRR)
- A trace of one record through every block of a Policy, showing which rules fired, which Offers were generated, and why.
- Portfolio Analysis
- A custom multi-object analysis on a portfolio slice (product, region, date range) that returns a Result dataset and supporting reports.
- Data Pull Logic
- The PySpark snippet inside a Portfolio Analysis that orchestrates Policies / Features and returns the result DataFrame.
- Scheduling
- The simulation form section that lets you set a Start Date, Time, and Recurrence so the job runs repeatedly on its own.
- Sampling
- The simulation form section that selects which records the job runs against — by size, date range, rule, or an explicit ID list.
Monitoring, metrics, and alerts
- Monitoring Dashboard
- A recurring view that refreshes on a schedule, tracks metrics across a Model or Policy, and triggers Alerts on breaches.
- Alert
- A rule on a Monitoring Dashboard that fires when a condition is met (metric threshold crossed, review date approaching).
- Triggered Alert
- An Alert that has fired. Listed in the Triggered Alerts tab where it can be acknowledged, closed, or bulk-managed.
- Drift
- A change over time in the distribution of inputs or outputs that signals a Model or Policy may no longer match the population it was built on.
- PSI (Population Stability Index)
- A drift metric for continuous variables. Compares the distribution of a variable in production against a reference period.
- CSI (Characteristic Stability Index)
- PSI's counterpart for categorical variables.
- KS (Kolmogorov–Smirnov)
- A discrimination metric that measures how well a Model separates positive and negative classes. Higher is better.
- AUC (Area Under the Curve)
- A discrimination metric for classification Models. The probability that the Model ranks a random positive above a random negative.
- ROC Curve
- The plot from which AUC is derived: true positive rate against false positive rate as the threshold varies.
- Decile Lift
- The performance of the top 10% of scored records relative to the population average. A standard scoring-model report.
- Approval Rate
- The percentage of records a Policy approves on a given run.
- Bad Rate
- The percentage of approved records that go bad (default, charge-off, or another adverse outcome) over a defined window.
Data and governance concepts
- Read Logic
- Optional Python / PySpark code attached to a Table that runs before any downstream job uses it. Used for casts, joins, normalization, and filtering.
- Permissible Purpose
- Tags applied to Data Elements, Features, and Models that declare where the object is allowed to be used (e.g. underwriting only, no marketing, PII-restricted). Violations are flagged via Lineage.
- Workspace
- The scope within which Draft objects are visible. Drafts are private to the workspace until they are published through an Approval Workflow.
- Artifact
- The compiled, deployable bundle produced when an object is published — used by external systems to execute the logic.
- Job Config
- The runtime configuration for a Simulation or comparable job: cluster size, resources, environment.
Modules and UI areas
- Data Vault
- The module where Tables and Data Elements are registered and governed.
- Feature Engineering
- The module where Features are built, simulated, and approved.
- Model Studio
- The module where Models are registered, trained, simulated, and compared.
- Policy Studio / Policies
- The module where Policies are composed from Strategies, Derived Variables, and Offers, and where simulations and Single Record Reviews happen.
- Product Framework
- The module that holds reusable product definitions (loan amounts, rates, terms) consumed by Policies and Models.
- Resources
- The module that holds shared building blocks (Global Functions, Global Variables, Runtime Parameters) used across other objects.
- Monitoring Dashboard (module)
- The module that holds Monitoring Dashboards, reports, and Alerts for in-production Models and Policies.
- Settings / Admin
- The area where admins configure Approval Workflows, authentication providers, alert providers, and other platform-wide settings.
Policy types
- Prospecting Policy
- A Policy that targets prospects before they apply (pre-approved offers, mailers, lead scoring).
- Underwriting Policy
- A Policy that decides at the moment of application (approve, decline, refer, counter-offer, price).
- Customer Management Policy
- A Policy applied to existing customers (credit line management, cross-sell triggers, early-warning flags).
Common acronyms
- SRR
- Single Record Review.
- PSI / CSI / KS / AUC / ROC
- See definitions above under Monitoring, metrics, and alerts.
- DPD
- Days Past Due. The number of days a loan payment is late. Often used as the basis for Bad Rate definitions (e.g. "30+ DPD at 6 months").
- LOS
- Loan Origination System. The downstream production system that Corridor artifacts deploy into.
- MRM
- Model Risk Management. The team or function that typically owns the Approved gate on Model Approval Workflows.
- PII
- Personally Identifiable Information. A common Permissible Purpose tag.
- API-XX
- An error code returned by the platform API. See Common Troubleshooting for codes and resolutions.
See also
- Common Troubleshooting — symptom-to-fix lookup for errors.
- FAQs — answers to frequent questions.
- Getting Started — guided introduction to these concepts.