Skip to main content
Codluma

How the Impact Analyzer Works#

The AI Review Impact Analyzer answers the question every engineer faces before merging: "What does this change actually affect?" It runs a deep, multi-layer analysis on every pull request to map out all areas touched — including areas not obviously connected to the changed files — then scores the risk and recommends next steps.

Dependency graph

Traverses the full import/dependency graph outward from changed files up to 20 levels deep.

Security impact

Flags changes touching auth, session handling, encryption, and payment flows with recommended reviewers.

Database awareness

Detects migrations, identifies affected tables, and assesses rollback safety.

8 parallel analyzers

Multiple specialised analyzers run concurrently — results are merged into a single risk score.

The analysis pipeline

01

Trigger

A PR is opened or pushed (automatic), or a team member runs the analyzer manually from the dashboard.

02

Code graph analysis

The engine builds a dependency graph of the repository and traverses it outward from changed files up to the configured depth (1–20 levels).

03

Multi-layer analysis

8 specialised analyzers run in parallel: generic file-level, language syntax, API contracts, database migrations, security, test coverage, dependency/SBOM, and historical failure correlation.

04

Risk scoring

A composite risk score (Low / Medium / High) is calculated from weighted evidence across all analyzers.

05

AI summary

A natural-language narrative is generated explaining the full impact in plain English (Pro plan and above).

06

Results delivery

Published as a PR comment (auto-comment mode) and visible in the dashboard. Merge Readiness Card shows Ready / Needs Attention / Blocked.

07

Ask Impact AI

Engineers can open a chat session grounded on this PR's specific analysis data to ask follow-up questions.

Risk Scoring#

Every analysis produces a composite risk score from 0–100 and a three-level classification. The score is calculated by weighting evidence from all active analyzers.

High

Score ≥ 70. The change touches sensitive areas, has breaking changes, or has a history of failures in the affected code. A security or senior review is strongly recommended before merge.

Medium

Score 40–69. Meaningful impact detected — downstream consumers affected, partial test coverage, or non-breaking API changes. Review recommended.

Low

Score < 40. Well-contained change with good test coverage, no breaking changes, and no sensitive area involvement. Standard review is sufficient.

Contributing risk factors

Factor
Effect on score
Notes
Auth / payment / config file touched
+High weight
Detected via path pattern matching
Breaking API contract change
+High weight
Endpoint removed, required param added, schema change
Database migration with no rollback
+Medium weight
Rollback safety assessed from migration file
Low test coverage on changed paths
+Medium weight
Below 60% on directly changed files
Historical failure correlation
+Variable
Past failures in same area increase score
Large dependency graph blast radius
+Low weight
Scaled by number of affected downstream modules
SBOM change (new/updated package)
+Low weight
Any dependency addition or version bump

Analysis Layers#

Eight specialised analyzers run in parallel. Each can be independently enabled or disabled per repository in Settings.

Syntax & Symbols

Detects which functions, classes, and interfaces changed and which files import them.

Dependency Graph

Full graph traversal showing all files and modules reachable from the changed code.

API Contracts

Endpoint changes, breaking change detection, schema diffs, and affected downstream consumers.

Database Changes

Migrations, affected tables, migration risk badge, rollback safety assessment.

Security Impact

Changes to auth, session handling, encryption, payments — with recommended security reviewers.

Test Coverage

Which tests cover the changed files, missing coverage warnings, and test history badges.

Dependencies (SBOM)

Package additions, removals, version changes, and supply chain summary.

Historical Failures

Past PRs that changed the same areas and the failures they caused — with correlation badges.

Dependency graph depth

The graph traversal depth controls how many hops outward from changed files the analyzer follows import edges. Higher depth catches more indirect impact but increases analysis time.

Max depth
Typical use case
Analysis time impact
1–3
Small, self-contained modules
Minimal
4–8
Most production codebases
Moderate (+1–3s)
9–15
Monorepos, large shared libraries
Noticeable (+3–10s)
16–20
Full blast-radius audit
Significant (10s+)

Tip

The default depth is 5. Increase it when touching core shared libraries. Set it low (1–2) for leaf-node UI components where deep traversal adds noise.

API contract analysis

The API contract analyzer compares the current branch to the base branch and flags:

  • Removed endpoints or operations
  • Added required request parameters or fields
  • Changed response schemas (field removal, type change)
  • Status code changes that could break consumers
  • Authentication requirement additions

Note

The API analyzer works with OpenAPI / Swagger specs, GraphQL schemas, and infers REST contracts from framework routing (Express, FastAPI, Spring, Rails, NestJS).

Database change analysis

The database analyzer scans migration files and ORM model changes to produce a risk badge:

Migration type
Risk badge
Notes
Add nullable column
Low
Safe — existing rows unaffected
Add NOT NULL with default
Low
Safe if default covers existing rows
Add NOT NULL without default
High
Will fail on tables with existing data
Drop column
High
Irreversible — verify no code reads this column
Drop table
High
Irreversible
Add index (non-unique)
Medium
Locks table on some engines during creation
Add unique constraint
High
Fails if duplicate data exists
Rename column / table
High
Breaks queries referencing the old name

Ask Impact AI#

Ask Impact AI is an embedded conversational AI grounded exclusively on this PR's analysis data. It cannot hallucinate from general knowledge — every answer cites the specific evidence collected during the analysis run.

Built-in suggested questions

Why is this risky?
Which tests should I run?
Why is this file affected?
Who should review this?
Did this area fail before?
Is this safe to merge?
What evidence supports this finding?
What downstream services are affected?

Note

Ask Impact AI is available on Pro and Enterprise plans. Chat sessions are scoped to a single PR analysis — the AI has no memory of previous PRs or conversations. All answers include a "Confidence" indicator and cite the specific analyzer layer that produced the evidence.

Configuration & Plan Tiers#

All Impact Analyzer settings are configured per repository under Settings → Repositories → [repo] → Impact Analyzer.

Per-repository settings

Setting
Options
Default
Enable / disable
On / Off per repo
Off
Auto-comment on PR
On / Off
On
Manual-only mode
On / Off
Off
Comment verbosity
Minimal / Normal / Detailed
Normal
Max graph depth
1–20 levels
5
Enabled analyzers
Individual toggles per layer
All enabled
Visual dependency graph
On / Off (Enterprise)
Off
Retention
7–365 days
90 days

Plan tiers

Feature
Starter
Pro
Enterprise
Core impact analysis
Risk score & affected areas
Auto PR comments
AI natural language summary
Ask Impact AI chat
Visual dependency graph
Custom analyzer selection
Custom retention policies
Monthly run limit
Limited
Higher
Unlimited

Tip

The Starter plan includes a limited number of Impact Analyzer runs per month to let you evaluate the feature. Upgrade to Pro to remove the monthly cap and unlock AI summaries and Ask Impact AI.

Can't find what you need?

Contact support

We use cookies and analytics to understand how you interact with Codluma and improve your experience. We never sell your data. See our Privacy Policy and Terms for details.