Scoring engine

Raigal assigns each repository a calibrated quality score between 0 and 100. The score measures issue density rather than raw finding counts, preventing large codebases from failing solely due to codebase volume.

Scoring formula

Raigal evaluates findings across six diagnostic engines using density-normalized logarithmic scaling:

Deduction density = (Sum of capped deductions) / (Effective source files + S)

Score = max(0, round(100 - (100 * ln(1 + 5 * Deduction density)) / ln(101 + 5 * Deduction density)))

Where:

Smoothing constant (S): Defaults to 5 in configuration (internal fallback 10). Prevents a single warning on a small project from collapsing the score.
Density scale: Multiplier of 5.0 applied to deduction density.
Effective source files: Count of non-excluded source and test files in the workspace (or modified files during incremental PR scans).

Severity multipliers and engine weights

Each diagnostic finding generates a raw penalty based on its severity and parent engine:

Raw deduction = Severity multiplier * Engine weight * Tier multiplier

Severity multipliers

error: 3.0 (also triggers immediate exit code 1 in CI)
warning: 1.0
info: 0.25

Engine weights

security: 1.5 (secrets, injections, unsafe APIs, vulnerable dependencies)
architecture: 1.0 (module boundary crossings, circular imports)
ai-slop: 1.0 (swallowed exceptions, narrative comments, loose type casts)
code-quality: 0.8 (dead code, complexity, parameter limits)
lint: 0.6 (Oxlint, Ruff, Clippy, Roslynator)
format: 0.3 (whitespace, indentation, import ordering)

Dynamic rule capping

To ensure widespread style issues (such as 500 missing trailing commas) do not drown out security vulnerabilities or force score collapse on large projects, rules belong to impact tiers with dynamic baseline caps:

Impact tier	Multiplier	Baseline cap	Typical findings
`strict`	`1.0x`	Uncapped (`maxPerRule`)	Syntax errors, leaked secrets, unhandled rejections
`standard`	`1.0x`	Uncapped (`maxPerRule`)	General linter diagnostics, compiler warnings
`maintainability`	`0.75x`	24	Cyclomatic complexity, oversized functions
`mechanical`	`0.5x`	16	Dead imports, unreferenced local variables
`style`	`0.5x`	8	Formatting variances, trailing commas
`advisory`	`0.25x`	8	Informational notices, missing docstrings

Repository scale scaling

The effective cap scales with repository size:

Cap scale = max(1, Effective source files / 10)

Effective cap = min(maxPerRule, Tier cap) * Cap scale

Worked examples

Example 1: Microservice (10 files, 1 warning)

Finding: 1 ai-slop/swallowed-exception (warning, severity 1.0, engine 1.0, tier strict 1.0).
Calculation:

Deduction = 1.0
Density = 1.0 / (10 + 5) = 1.0 / 15 ≈ 0.0667
Scaled density = 0.0667 * 5.0 = 0.3333
Score = 100 - (100 * ln(1.3333) / ln(101.3333)) ≈ 100 - 6.23 = 94 (Healthy)

Example 2: Monolith (200 files, 500 style findings)

Findings: 500 missing trailing commas (format/trailing-comma, info 0.25, engine 0.3, tier style 0.5, cap 8).
Calculation:

Raw deductions = 500 * 0.25 * 0.3 * 0.5 = 18.75
Cap scale = 200 / 10 = 20
Effective cap = 8 * 20 = 160
Capped deductions = min(18.75, 160) = 18.75
Density = 18.75 / (200 + 5) = 18.75 / 205 ≈ 0.0915
Scaled density = 0.0915 * 5.0 = 0.4573
Score = 100 - (100 * ln(1.4573) / ln(101.4573)) ≈ 100 - 8.15 = 92 (Healthy)

Score thresholds and status bands

Raigal relies exclusively on numeric 0-100 scores. Letter grades (A-F) are not used.

CLI status labels

When running raigal scan or raigal ci, status labels correspond to score boundaries:

Score range	CLI label	Exit code in CI
`75 - 100`	`Healthy`	`0` (Passing)
`50 - 74`	`Needs Work`	`0` (or `1` if `ci.failBelow` is breached)
`< 50`	`Critical`	`1` (Fails quality gate)

In CI environments, raigal ci exits with code 1 if the score is lower than ci.failBelow (default: 70) or if any error-level finding is detected.

Dashboard health bands

The Raigal Cloud web interface (app.raigal.dev) categorizes repositories into calibrated health bands:

Score range	Dashboard band	Color token
`90 - 100`	`Excellent`	Lime green
`75 - 89`	`Good / Healthy`	Emerald green
`50 - 74`	`Needs Work`	Amber / yellow
`< 50`	`Critical`	Red

Quality gate verdicts (3-state)

PR status checks and CI gates evaluate pull requests into three distinct states:

Verdict	Criteria	Check run outcome
`PASSED`	Score >= 70 (or `ci.failBelow`), 0 errors, 0 warnings	`success` (Green check)
`PASSED (WITH WARNINGS)`	Score >= 70, 0 errors, non-blocking warnings present	`neutral` (Advisory pass, does not block merge)
`FAILED / BLOCKED`	Score < 70, non-zero exit code, or blocker errors present	`failure` (Blocks merge)

Getting started

Core concepts

CLI reference

Automation & CI/CD

Cloud platform

Scoring engine

Scoring formula

Severity multipliers and engine weights

Severity multipliers

Engine weights

Dynamic rule capping

Repository scale scaling

Worked examples

Example 1: Microservice (10 files, 1 warning)

Example 2: Monolith (200 files, 500 style findings)

Score thresholds and status bands

CLI status labels

Dashboard health bands

Quality gate verdicts (3-state)

​Scoring formula

​Severity multipliers and engine weights

​Severity multipliers

​Engine weights

​Dynamic rule capping

​Repository scale scaling

​Worked examples

​Example 1: Microservice (10 files, 1 warning)

​Example 2: Monolith (200 files, 500 style findings)

​Score thresholds and status bands

​CLI status labels

​Dashboard health bands

​Quality gate verdicts (3-state)

Scoring formula

Severity multipliers and engine weights

Severity multipliers

Engine weights

Dynamic rule capping

Repository scale scaling

Worked examples

Example 1: Microservice (10 files, 1 warning)

Example 2: Monolith (200 files, 500 style findings)

Score thresholds and status bands

CLI status labels

Dashboard health bands

Quality gate verdicts (3-state)