Explainables
Judge

Model Scope & Limits

system limitsaudience adaptation
User question

What can this system reliably do?

Consulting signal

Comes up when a client deploys a model in a context it wasn't designed for: new market, new user group, new data type, and the performance degradation isn't visible until something goes wrong.

Overview

Why this pattern exists

AI systems are designed and trained for specific contexts. They perform reliably within those contexts and less reliably, sometimes dangerously, outside them. A credit scoring model validated on employed applicants may perform poorly for self-employed or gig-economy workers. A medical diagnostic tool trained on data from one demographic may be less accurate for another. A language model trained on English-language data may mishandle concepts that don't translate culturally.

These limits are known to the people who built the system. They are almost never visible to the people using it.

Model Scope & Limits is the pattern that surfaces this gap: making the system's intended operating envelope visible to the users and operators who need to judge whether to rely on it in a specific situation.

This is related to, but different from, Confidence & Uncertainty. Uncertainty describes how confident the system is within its scope. Scope & Limits describes where the system applies at all. A system can be highly confident and still be outside its scope: that is one of the most dangerous configurations.

Design goal

Communicate the intended operating scope of an AI system, and signal clearly when a specific case, query, or context falls outside it: so users can make informed judgments about whether to rely on the output.

Usage guidance

When to use

  • The system was validated for a specific population, domain, or context
  • There are known cases or characteristics where the system performs less reliably
  • The current case may fall outside the model's training distribution
  • Operators or caseworkers need to decide whether to apply the system to a new context
  • Regulatory requirements (AI Act, sector-specific rules) require disclosure of intended use and limitations

When not to use

  • The system's scope is genuinely broad and well-validated across all relevant use cases (show scope as a positive feature, not a warning)
  • Surfacing limitations would be technically honest but create disproportionate alarm for a low-stakes application
  • The limitations are too complex and technical to communicate usefully to the current user: in that case, restrict use rather than trying to explain it

Design

UI primitives

to be added soon
Content Block / Statement

Scope statement

A plain-language description of what the system is designed for, shown prominently in the interface: "This system was validated for salaried employees with at least 12 months of employment history."

to be added soon
Inline Signal / Indicator

Out-of-scope indicator

A visible alert triggered when the current case or query has characteristics outside the validated scope: "This applicant is self-employed. Results may be less accurate: specialist review recommended."

to be added soon
Content Block / Panel

Limitations panel

A collapsible section in the interface (also linked to the Model Transparency Card, 12) listing known limitations: - Population coverage - Time range of training data - Geographic or linguistic scope - Known edge cases or failure modes

to be added soon
Data Visualization / Visualization

Scope coverage map / visualization

For systems with geographic, demographic, or domain scope, a visual representation of where the model has been validated. Particularly useful for analysts and operators evaluating whether to deploy in a new context.

to be added soon
Inline Signal / Badge

Use with caution tier

A three-tier system: - ✓ Within scope: use normally - ⚠ Edge of scope: use with additional review - ✗ Outside scope: do not rely on this output alone

to be added soon
Inline Signal / Indicator

Version and training date indicator

Scope changes across model versions. Showing the model version and training data cutoff date alongside the scope indicator helps users assess whether the current version is appropriate for their situation.

How to use

Treat scope limits as useful information, not embarrassing failures.

The concept of "seamful design": deliberately making system seams and limits visible rather than papering over them: is directly applicable here. Research shows that surfacing limitations increases user trust when done well, because it signals that the system knows what it doesn't know. An interface that hides its limits until they cause a problem destroys trust in a single incident; an interface that proactively surfaces them builds the kind of calibrated confidence that survives encountering an edge case.

Show scope at the point of use, not only in documentation.

A model card that describes limitations is useful. A real-time scope indicator that triggers when the current case falls outside validated bounds is essential.

Design "outside scope" as a first-class state, not an error.

Encountering an out-of-scope case is not a system failure: it is the system functioning as designed. The response should be: route to appropriate review, not: display a generic error message.

Be specific about what limits apply to this case.

"This system has limitations" is not useful. "This system was not validated for self-employed applicants, and the current applicant is self-employed" is actionable.

Connect scope warnings to appropriate action.

An out-of-scope warning without a clear next step creates anxiety without resolution. Always pair with: "What should happen instead?": human review, specialist routing, alternative tool.

Update scope displays when model versions change.

The scope of a model may expand or contract with each update. Scope indicators must be version-aware.

Use cases

flow a

Out-of-scope detection during processing

  1. 1. New application submitted.
  2. 2. System detects that the applicant is a gig worker with variable income, outside the training distribution.
  3. 3. Interface flags: "⚠ This application type falls outside the standard validation scope. Automated scoring may be less accurate."
  4. 4. Case is automatically routed to a specialist review queue.
  5. 5. Specialist reviews manually, with the out-of-scope flag visible in the case record.
flow b

Operator evaluating deployment in a new market

  1. 1. Product manager considers deploying the system in a new geographic market.
  2. 2. Opens the scope and limits documentation.
  3. 3. Sees: "System validated on applications from Germany and Austria only. Performance in other markets has not been assessed."
  4. 4. Decision: commission a local validation study before deployment.
flow c

User asks about limitations directly

  1. 1. Applicant asks: "Is this system fair for people like me?"
  2. 2. System shows the relevant scope limitations in plain language: "This system was designed for applications in standard employment. If your situation is different: self-employment, freelance, non-standard income: the assessment may be less accurate."
  3. 3. Link to Model Transparency Card for full documentation.

Design trade-offs

Transparency vs. over-warning

If every case triggers a scope warning, users become desensitized. Calibrate scope warnings to cases where the limitation genuinely applies, not as a blanket legal disclaimer on every output.

Disclosure vs. system abandonment

Surfacing limitations clearly may cause users to distrust the system entirely: even for the cases it handles well. Present scope as a precise boundary, not as a general indication of untrustworthiness.

Technical accuracy vs. user comprehension

The precise description of a model's training distribution may be technically accurate but incomprehensible to non-technical users. Find plain-language equivalents that are accurate enough to be useful without requiring ML expertise.

Connections

Relation to other patterns

Sources

Amershi et al. (2019). Guideline G4: "Show contextually relevant information." G17: "Be transparent about the system's limitations." Both directly support the design approach of this pattern. https://dl.acm.org/doi/10.1145/3290605.3300233 - Seamful Design and Ubicomp Infrastructure, Chalmers & Galani (2004) — Guidelines for Human-AI Interaction

introduces "seamful design": the deliberate surfacing of system limitations and seams as useful information rather than embarrassing failures. Foundational for the design philosophy of this pattern

Mitchell et al. (2019) — Model Cards for Model Reporting

the primary framework for documenting model scope, intended use, and out-of-scope use cases. Model cards are the documentation layer; this pattern is the runtime disclosure layer

documents how AI systems validated on one population can produce discriminatory outcomes when applied to another. The empirical basis for why scope and population coverage limits matter for fairness, and why out-of-distribution use constitutes a genuine harm risk