Reading a Croms Score

One number for preventable waste

Croms™ is Cromus's unit of preventable AI workflow waste. Not total spend — preventable spend: model overkill, unnecessary tokens, missed caching, retry storms. Lower is better.

Severity bands make the number readable at a glance: Low (0–99), Moderate (100–299), High (300–599), Critical (600+).

The four pillars

A Croms score decomposes into four weighted pillars:

• Cost (35%) — preventable spend: model overkill, unnecessary tokens, missed caching.
• Latency (25%) — time wasted on slower routing than the task needs.
• Failure (25%) — rework from errors, retries, and unhandled edge cases.
• Structure (15%) — waste from ambiguous steps and missing output contracts.

The pillar breakdown tells you where to intervene. A workflow with a Cost-heavy score needs routing changes; a Failure-heavy one needs error handling in its steps.

From score to inbox

Once a skill is in your workspace, you don't have to remember to check its score manually. The Skill Drift Scanner runs every day across your workspace and fires a HIGH_WASTE inbox item the moment any skill crosses 300 Croms. The score you're about to read here is the same number the scanner watches — understanding the pillars tells you why an item appeared in your inbox and which intervention to reach for first.

Do It: walk a real skill

The panel below walks the same five-stage sequence used in the Cromus guided tour — Skill, Score, Pillars, Fix, Savings — using a real sample workflow (an SEO content brief generator) with pre-computed, deterministic numbers. No AI is called; everything you see derives from the skill's declared estimates.