CONTEXT-GRAPH AGENTBUILT ON DATAHUB

Ship schema changes
without surprises.

LineageShield turns a proposed data change into a verified blast radius, accountable owner map, and merge-ready rollout plan.

Inspect a change 10 DataHub assets in this recorded sample
LINEAGEOWNERSPOLICIESSCHEMA
CONTEXTGRAPH AGENT
01 / THE WORKSPACE

One change. Every consequence.

Choose a pull request. The public demo replays a recorded four-call DataHub MCP trace, traverses the metadata graph, and turns that context into an executable plan.

DATAHUB CONTEXT GRAPHHash customer email
ANALYSIS COMPLETE
CHANGE ORIGIN
PODATASET
raw.crm_contactsGrowth Systems
PIISource
DEPTH 1
SNDATASET
staging.customer_contactData Platform
PIICertified
DEPTH 2
SNDATASET
analytics.customer_360Customer Analytics
PIICertified
DEPTH 3
LODASHBOARD
Customer healthCustomer Analytics
ExecutivePII
FEDATASET
features.customer_valueML Platform
MLOnline
BRACTIVATION
High-value at-risk audienceLifecycle Growth
ActivationPII
DEPTH 4
MLML MODEL
Churn propensity v8Lifecycle ML
MLProduction
AFFECTED7assets
CRITICAL6need sign-off
OWNERS6teams
CHANGE RISK99/ 100 · Critical
02 / AGENT OUTPUT

A rollout teams can execute.

CONTEXT CONFIDENCE98%
IMPACT BRIEFCritical risk

Hash customer email

The proposal reaches production decisions beyond its immediate contract. LineageShield recommends a compatibility window.

01

7 assets are reachable across 4 lineage hops.

02

6 Tier 1 or policy-sensitive consumers need explicit approval.

03

6 ownership groups must coordinate the rollout.

04

5 assets retain PII semantics after the proposed transformation.

CONTEXT USED
LineageOwnershipTagsTiersSchema
SAFE ROLLOUT PLAN3 phases
01

Contract first

Growth Systems
  • Add email compatibility assertion
  • Pin current contract as varchar · plaintext
  • Block merge on downstream contract failures
02

Dual-run the change

Data Platform
  • Materialize old and new values for email
  • Compare row counts and null rates across 7 assets
  • Require sign-off from raw.crm_contacts, staging.customer_contact, analytics.customer_360
03

Progressive rollout

Customer Analytics
  • Deploy to 10%, 50%, then 100% of scheduled runs
  • Watch freshness, volume, and semantic assertions
  • Keep the compatibility view for one release window
GENERATED GUARDRAIL

Compatibility SQL, ready for review.

The agent converts graph context into a concrete guardrail—not another generic summary.

-- Compatibility view generated by LineageShield
create or replace view analytics.customer_contact_safe as
select
  * exclude (email),
  sha2(lower(trim(email)), 256) as email_hash,
  email as email_legacy
from staging.customer_contact;

-- Remove email_legacy after downstream sign-off.
03 / BUILT FOR REAL METADATA

DataHub supplies the context.
LineageShield makes it actionable.

01MCP Contextsearch + entities + schema + lineage
02TraverseTyped adjacency index maps blast radius
03ActAgent generates checks, owners, and SQL
Inspect the reproducible DataHub MCP trace →
NATIVE ARM64 · VERIFIED1,963.94×faster median traversal
BASELINE138.203 msrepeated edge scans
OPTIMIZED0.070 msCSR typed-array index
EXACT PARITY2,819 = 2,819reachable assets · view run ↗