← Back to Docs
Architecture Integrity Index

A 0-100 score of structural health

What is the Architecture Integrity Index?

The Architecture Integrity Index is a 0-100 score of a codebase's structural health: the sum of four axis tiles, each worth 25 points and each reduced by the entropy impact of the architectural findings sitting in it. Because an axis floors at 0, the headline number and the four tiles always reconcile.

The floor is why this is not the same as subtracting every finding from 100: an axis buried under findings stops at 0 rather than dragging the other three down with it.

How is the entropy impact of a finding weighted by severity?

Each architectural finding deducts points from its own axis based on its severity: critical (P1) findings cost 5.0 points, high (P2) findings cost 2.5 points, medium (P3) findings cost 0.75 points, and low (P4) findings cost 0.1 points. Only architectural findings carry a weight — hygiene issues carry none.

SeverityEntropy weightMeaning
P1 — Critical5.0Blocks the pattern outright
P2 — High2.5Significant drift from established patterns
P3 — Medium0.75Minor inconsistency
P4 — Low0.1Cosmetic / low-impact drift

What are the four axes VibeCop scores?

VibeCop weighs four axes equally at 25% each: Consistency (similar problems solved the same way), Reuse (logic reused rather than re-implemented), Simplicity (abstractions earning their weight, no hallucinated layers), and Structure (cycles, fan-out, and coupling kept in check).

Do hygiene findings affect the index?

No. Dependency CVEs, exposed secrets, SAST findings, and infrastructure misconfigurations carry no axis, so they carry no entropy weight and never move the Architecture Integrity Index up or down. The hygiene tray on Overview counts this week's open security findings (secrets, SAST, dependency CVEs, infrastructure) and complexity warnings; dependency-health and vendored-drift findings are listed under Findings only.

Why would the index show “—” with an ⓘ instead of a number?

Because a deep agent did not run on that scan. Each axis tile is fed by specific deep agents, and an agent that is switched off — by your organization's Feature Access policy, by a project's agent toggles, or by the AI-analysis privacy switch — reports nothing, which would leave its axis scoring a perfect 25. A disabled agent can never raise the index, so VibeCop records no index for that scan and keeps the findings; the ⓘ names the switch responsible.

Related docs