---
title: Runtime and cost
summary: >-
  Keep model behavior inside latency, retry, and spend budgets without hiding
  degraded quality.
agent_summary: >
  This page explains how JudgmentKit makes latency and spend tradeoffs explicit
  so they can be reviewed like any other product decision.
canonical_url: /docs/guardrails/runtime-and-cost
page_type: guardrail
related_resources:
  - /resources/guardrails/runtime-cost.v1.json
related_schemas:
  - /schemas/guardrail.schema.json
last_reviewed: '2026-04-09'
---
# Runtime and cost

Keep model behavior inside latency, retry, and spend budgets without hiding degraded quality.

> Agent summary: This page explains how JudgmentKit makes latency and spend tradeoffs explicit so they can be reviewed like any other product decision.


## Headings
- ## Why this matters
- ## What decision is being governed
- ## What good judgment looks like
- ## What drift looks like
- ## Example in practice
- ## Boundaries
- ## How JudgmentKit responds
- ## Ownership and review
- ## Technical reference
- ## Related pages

## Why this matters

Latency and spend are not infra-only concerns. They shape the product behavior users experience. A workflow that quietly burns budget or stalls the interface has already made a judgment decision, whether the team documented it or not.

## What decision is being governed

This guardrail governs how much model work, retrieval depth, and response complexity are acceptable for a workflow.

## What good judgment looks like

- choose the cheapest path that still meets quality needs
- expose when the workflow is intentionally taking a cheaper fallback
- reserve expensive paths for documented high-value cases

## What drift looks like

1. Retry loops or unbounded refinement
2. Costly reasoning without a product reason
3. Silent fallback that fabricates certainty instead of acknowledging limits

## Example in practice

The UI generation example shows a common cost drift pattern: open-ended creative intent that silently becomes an unlimited runtime request.

## Boundaries

Allowed variation includes shorter outputs for mobile, lower retrieval depth for low-risk questions, and explicit premium paths where the business case is known.

Hard stops include unbounded retries, hidden fallback that changes risk posture, and premium-cost paths without documented intent.

## How JudgmentKit responds

Budget pressure can trim and log low-risk paths, shift to a cheaper fallback with notice, or stop the workflow entirely when the degraded path would mislead the user.

## Ownership and review

Product owns the tradeoff. Platform Engineering owns the risk and the runtime enforcement.

## Technical reference

- Resource: `/resources/guardrails/runtime-cost.v1.json`
- Schema: `/schemas/guardrail.schema.json`

## Related pages

- /docs/workflows/support-assistant
- /docs/workflows/ai-ui-generation
- /docs/examples/ui-generation-drift

## Related pages
- /docs/workflows/support-assistant
- /docs/workflows/ai-ui-generation
- /docs/examples/ui-generation-drift

## Related resources
- /resources/guardrails/runtime-cost.v1.json

## Related schemas
- /schemas/guardrail.schema.json
