How Sienna works
Sienna's memory graph stores definitions and data-quality observations. Business-context records are coming soon. The analyst harness reads the graph before it queries the warehouse, verifies the result, and returns an answer. Saved charts retain the records, checks, source tables, and SQL used to produce them.
Four ways an answer goes wrong
Each system addresses a different class of failure. This example uses the wrong definition, contains a known data-quality issue, fails a query check, and omits relevant business context. Each section shows the evidence Sienna uses to identify the problem.
Wrong definition: raw invoice total instead of recognized revenue.
This question requires recognized revenue as defined in the dbt project. Summing raw invoices returns $3.16M instead of $3.47M. Sienna resolves the definition before writing SQL and flags unresolved business choices such as refund timing.
Data-quality issue: missing FX rates are replaced with 1.00.
The SQL replaces missing FX rates with 1.00 on 14 days, so euro amounts are treated as dollar amounts. Sienna records the affected field, evidence, and preferred treatment, then assesses the issue for each request that uses those values.
Query error: subscription rows are counted twice.
The query uses the correct definitions but counts each subscription twice. Sienna checks joins, filters, row counts, and accounting identities before delivery. A failed check triggers a revised query. If verification still fails, the run stops without an answer.
Missing context: Q3 is compared only with Q2.
Q3 is lower than Q2, but this business follows the same seasonal pattern every year and revenue is 6.1% higher than the previous Q3. Business-context records will attach that timing note to the affected definition and period so the answer includes the relevant comparison.
Corrected answer
Result after definition resolution and verification
The corrected answer uses recognized revenue, applies the recorded FX treatment, removes the duplicate subscription rows, and includes the seasonal comparison. The FX treatment and unreviewed refund-timing definition remain visible as caveats.
Interface
What you work with directly: answers in chat, the dashboards you keep, and the verification every answer passes through.
Memory graph
The records Sienna uses during analysis and the integrations that create or update them.