Event data contracts

Catch it downstream.
Fix it upstream.

Agents ship your events now, and agents consume them too. Bad data doesn't get noticed anymore, it gets acted on. suky verifies event contracts against what actually lands in Snowflake, and when one breaks, opens the PR on the code that broke it.

For Product & Engineering →For Data & AI teams →
Produced by @growth-engConsumed by 14 readers
Snowflake
checkout_started
Healthy
Events / min
2,400
✓ SLA · > 500
Health
98%
✓ SLA · > 95
Freshness
4m
✓ SLA · < 30m
user_idstring
required · UUID
cart_valuenumber
>= 0 · <= 1e6
currencystring
enum · ISO 4217
items[].skustring
required · regex /^[A-Z]/
owner · @sofia.chen3 dashboards · 2 ML models · 9 dbt models
One loop · Two sides

Produced under contract.
Consumed with confidence.

One contract between the teams who ship events and the teams who rely on them.

checkout.ts
track("checkout_started", …)
SDK · WEB
segment
rudderstack
snowplow
snowflake
raw.events · checkout_started
WAREHOUSE · READ-ONLY
dbt
stg_checkout
AI agent
BI dashboard
ML pipeline
Reverse ETL
FIX
ENFORCE
VERIFY
sukyControl plane
Healthy
watching snowflake.raw · 214 contracts passing
checkout_startedENFORCED
user_idstring✓
cart_valuenumber✓
currencyISO 4217✓
suky computes in Snowflake · raw event data never leaves your warehouse · no pipeline lock-in
The case

Humans quietly compensate for drifting data. Agents can't.

01 / the semantics gap
24→86%
agent accuracy on structured data, once its meaning is curated
Snowflake's own benchmarks: agents jump from ~24% to ~86% when the data underneath is documented and consistent. The bottleneck isn't SQL; it's whether anyone verified what the data means.
Source · Snowflake benchmarks, 2026
02 / bare access isn't enough
23%
what a frontier coding agent scores with bare MCP access to the warehouse
Pointing an agent at undocumented event tables produces syntactically valid SQL and confidently wrong answers. suky's MCP serves contracted events with their schema, rules, and current health.
Source · Snowflake benchmarks, 2026
03 / the category risk
60%
of agentic-analytics projects relying on MCP alone will fail by 2028
Gartner's call, and the reason: no consistent semantics underneath. Behavioural event tables are the fastest-drifting, least-documented data an agent will ever query.
Source · Gartner, Market Guide for Agentic Analytics, 2026
Diagnosis →
Auto-generated semantics absorb instrumentation drift instead of catching it. Your semantic view is only as true as your instrumentation, and a renamed property becomes a confident wrong answer at machine speed.
01 · For Product & Engineering teams

Ship instrumentation that's right the first time.

Whether a human writes the tracking call or a coding agent does, it's checked against the contract before it ships: in the IDE over MCP, and again in CI on every pull request.

Design and version schemas like code
Changesets with draft → review → merge, breaking changes auto-classified for triage.
A CI gate that fails breaking PRs
Native GitHub check with inline annotations, plus a machine-readable block coding agents parse.
Guardrails for coding agents
The MCP server serves canonical events and SDK snippets, so generated tracking code lands on-contract.
GitHubPR #2851 · add coupon trackingby cursor-agent
✕suky / contract gateBREAKING
coupon_code · string → number
would break 14 downstream dbt models
✓fix suggested: add coupon_value instead
volume · last 28 min● live
checkout_startedhealthy98
cart_item_addedhealthy96
pricing_changeddrift62
agent pre-flight · check_event_health('pricing_changed')
→ degraded · schema_drift · violation V-1184 open · answer withheld until resolved
consumers of this table:BI dashboardsdbt modelsCortex AnalystClaude + MCPdecisioning agents
02 · For Data & AI teams

Know it broke before anyone, or anything, reads it.

Detection, monitoring and fixes run continuously against your warehouse, so drift is caught ahead of the dashboard going red, or the agent answering from it. Your newest consumers don't double-check; suky does.

Every event catalogued, owned, scored
Healthy, degraded or broken per event, with anomaly detection tuned to weekday seasonality.
Thresholds that run against your warehouse
Freshness, volume, null rate and schema drift per contract. Every breach opens a violation with a severity.
Agents check event health before they answer
check_event_health over MCP: open violations, freshness and schema stability for any event an agent is about to query.
Anatomy of a contract

Every field is a promise both sides can hold.

Hover any property to see the rule, what it catches, and the SQL that validates it against your warehouse.

contracts/commerce/checkout_started.yaml● valid
user_idstringREQ
uuid format
session_idstringREQ
uuid format
cart_valuenumberREQ
min 0 · max 1,000,000
currencystringREQ
enum · ISO 4217
itemsarrayREQ
min length 1
coupon_codestring
regex /^[A-Z0-9-]+$/
client_tsnumberREQ
unix ms · within 1h
Field · user_id

uuid format

Identifies who triggered the event. Rejected if missing or malformed.

Validates against Snowflake as
SELECT count(*) FROM analytics.events WHERE event = 'checkout_started' AND NOT (user_id ~ '^[0-9a-f]{8}-[0-9a-f]{4}-...');
Failures → Slack #data-platformSLA · null rate < 1%
suky AI

Every agent in your SDLC generates. suky verifies.

suky is the one agent that can see reality. It knows your contracts, watches your deploys, and checks what actually landed in your warehouse, then traces the violation to a commit, drafts the fix for human approval, and hands the ticket to the right owner.

Verifies every change against reality
Contracts on one side, your warehouse on the other, checked per PR rather than per quarter.
A contract registry your agents query
Cursor, Claude Code, and Windsurf look up canonical events over MCP before writing tracking code.
Drafts the fix for human approval
Corrected SDK code as a PR, scoped to the files the plan read. You review, suky does the rest.
s
suky AI
resolving · pricing_changed
Pricing

Simple pricing. Scale when you're ready.

Free for small teams. Pro when you're ready to publish, automate, and ship faster.

Free

Discover and monitor your event data with AI. 5 seats included.

$0/mo
Warehouse discovery + profiling
Up to 10 contracts
AI-drafted schemas
Basic monitoring
Slack + email alerts
5 seats, 1 workspace
100 AI credits / month
FOUNDING 25

Pro

The full platform for teams shipping on event data.

$199/mo
First 25: $149/mo locked for 12 months →
Unlimited contracts
Unlimited seats
Schema registry publishing
Teams · PagerDuty alerts
Auto-PR code generation
MCP server + API access
1,500 AI credits / month

All plans include warehouse connection, AI-drafted contracts, and Slack + email alerts. See full comparison →

Contract

Ready to put your event data under contract?