Tiresias monitors every Fivetran sync, detects semantic drift the moment it lands, and quarantines bad data at the source, automatically.
The problem
A deal stage gets renamed in your CRM. Fivetran faithfully syncs it. No error fires. No alert triggers. But every downstream model filtering on the old label now returns zero rows.
Three days later, your VP of Sales asks why the late-stage pipeline shows $0. That is when you find out.
What Tiresias does
Always-on fingerprinting
On every sync, Tiresias computes a statistical fingerprint — value distributions, null rates, row counts, cardinality. It remembers what normal looks like so it can recognise what isn't.
AI classification
Gemini 3.1 Pro classifies every drift event: organic volume change, upstream schema addition, or silent semantic failure — the kind where your numbers are wrong but your pipeline shows green.
Blast radius tracing
Tiresias traces the blast radius forward through your dbt lineage graph — which staging models break, which fact tables are compromised, which executives are looking at wrong numbers.
Automatic quarantine
One approval. Tiresias disables the bad table at the Fivetran source via MCP — no more corrupt data flowing downstream while your team debugs. Re-enable when the fix is deployed.
The response
Tiresias receives the webhook, pulls the live schema config via MCP, and computes a fresh fingerprint against your BigQuery destination.
The PSI spike is classified: Silent Semantic Failure, 95% confidence. Blast radius traced to stg_deals → fct_pipeline_by_stage → VP of Sales dashboard.
One click in the Tiresias dashboard. modify_connection_table_config fires via MCP. The table is disabled at source. No more corrupt data flows downstream.
Integrations
Snowflake, Redshift, Salesforce, and more — coming soon
Tiresias watches every sync. You find out first.