How It Works

Predict. Validate. Confirm.
Production reality, not just tests.

Legible's three-phase model validates every deployment against what your system actually does in production — building a Trusted Change Boundary before anything reaches your users.

The process

Three phases. One missing layer.

From the moment a deploy is triggered to production confirmation — Legible governs every step using evidence, not assumptions.

01
Understand

We ingest your existing OpenTelemetry traces and build versioned production fingerprints — a living map of how your services actually interact. No new instrumentation. No code changes.

From reality, not declarations
02
Predict

Before a deployment ships, Legible generates a hypothesis — the Inferred Intended Change — and validates it against staging behaviour. The result is a Trusted Change Boundary: the maximum allowed surface.

Before execution commits
03
Enforce

Production behaviour is checked against the boundary. Changes inside: explained. Changes outside: unexplained and flagged. Verdict: ALLOW, MONITOR, HOLD, or ESCALATE — with a full evidence chain.

Enforcement, not observation
💡
Phase 1
Hypothesis Generation

When a deployment is detected, Legible normalises evidence from your CI/CD pipeline, changelogs, PRs, feature flags, and config systems into a unified evidence model.

From this evidence, the system generates an Inferred Intended Change (IIC) — a structured prediction of what behavioural changes the deployment will produce.

The IIC is a hypothesis, not a source of truth. It has no governance authority on its own. It must be validated by what actually happens.

CI/CD webhook Normalise evidence Resolve identities Generate IIC
IIC hypothesis — payment-svc v2.15.0
change_type: RETRY_CHANGE affected_edge: payment → bank-integration predicted_delta: retry_amplification: 1.5x structural_change: false new_edges: [] confidence: MEDIUM # Changelog is structured but no stage evidence yet
🧪
Phase 2
Stage Validation & Trusted Change Boundary

After deployment to staging, Legible observes actual runtime behaviour and computes a Stage Behavioral Delta (SBD) — a multi-dimensional measurement across structural topology, traffic distribution, retry patterns, and latency.

The SBD is compared against the Phase 1 hypothesis. Verdicts: CONFIRMED, SUPERSET, SUBSET, DIVERGENT, or UNVERIFIABLE.

The system then constructs the Trusted Change Boundary — the maximum allowed surface — from stage-validated changes, confirmed predictions, known environment divergences, and bounded transitive dependencies.

Observe stage Compute SBD Validate vs IIC Build boundary
Stage validation — payment-svc v2.15.0
stage_validation_verdict: SUPERSET stage_behavioral_delta: retry_amplification: 3.5x # predicted 1.5x structural_change: false new_edges: [] trusted_change_boundary: retry_ceiling: 3.5x # stage-validated structural: no new nodes/edges boundary_confidence: HIGH
Phase 3
Production Confirmation

After deployment to production, Legible computes a Production Behavioral Delta and checks it against the Trusted Change Boundary.

Changes inside the boundary: explained. Changes outside: unexplained and classified by confidence tier.

Governance confidence depends on the boundary, not prediction accuracy. Runtime behaviour is always the source of truth.

ALLOW
All changes explained
MONITOR
Partial match
HOLD
Unexplained changes
ESCALATE
Invariant violation
Production confirmation — PARTIAL_MATCH → HOLD
explained_changes: retry_amplification: 3.2x # within boundary ✓ latency_shift: +12ms # within tolerance ✓ unexplained_changes: new_edge: payment → notification-service # NOT in boundary — structurally significant verdict: PARTIAL_MATCH → HOLD reason: "New dependency requires investigation"

The question isn't "is the system healthy?" — it's "did the deployment produce exactly what it was supposed to, and nothing else?"

Integration

Works with what you already have

No SDKs. No agents. No new instrumentation. Connect your existing telemetry and deployment pipelines.

📡
Telemetry
OpenTelemetry
Azure App Insights
AWS CloudWatch
GCP Cloud Logging
🚀
CI/CD & Deploy
GitHub Actions
GitLab CI
ArgoCD
Jenkins · Spinnaker
🚩
Feature Flags
LaunchDarkly
Statsig
Split
⚙️
Config & Infra
Terraform
Consul
Kubernetes · Helm
⏱️ 5 minutes to connect
🚫 Zero code changes
🔒 Read-only telemetry

Ready to see your production fingerprint?

We'll show you what Legible sees in your first 30 minutes — hidden dependencies, drift from recent deployments, and changes that are risky right now.

Get your Day-0 safety report →