March 26, 2026 · 4 min read
What breaks when you change your mind
“Let’s switch from NestJS to Hono.”
Sounds like a framework swap. How hard can it be?
In a project with 207 recorded decisions, that one change triggers a blast radius of 178 affected entities. 86% of the project. 4 at high severity. The answer comes back in under a second.
How blast radius works
Every decision in Graven is connected to other decisions through weighted dependencies. “Auth approach” depends on “framework choice.” “API structure” depends on “auth approach.” The chain propagates.
When you ask “what happens if I change X?”, the system traverses the dependency graph. Each edge has a weight — confidence level and dependency strength (hard vs soft). The impact score decays with depth: a direct dependency at high confidence scores near 1.0. A transitive dependency three levels deep through a soft link scores 0.3.
The result: not just a list of affected things, but a ranked list. The most critical impacts first. The low-risk ripples last.
Blast radius for “Tech Stack: NestJS”
178
entities affected
86%
of project covered
4
high severity
5
depth levels
The assumption that scares us most
More interesting than the tech stack: the assumption “LLM quality is sufficient for domain expertise delivery.”
This single assumption — not even a decision, just something we believe to be true but haven’t fully verified — has 185 entities in its blast radius. 89% of the project.
It’s marked as “high risk” with an invalidation trigger: “production accuracy falls below acceptable threshold.” If LLM quality turns out to be insufficient, we don’t need to guess what breaks. We know exactly which 185 decisions to revisit.
Why this matters for AI agents
An AI agent doesn’t have the context to know that changing the caching strategy affects 12 downstream decisions. It sees a local problem and applies a local fix.
With blast radius analysis, the agent sees the full picture before writing a single line of code. Not “I think this might affect something.” An exact list. Ranked by impact. In one second.
The question isn’t whether you’ll change your mind. You will. The question is whether you’ll know what breaks before or after.