Build and maintain context
setup-nodal
Configures the project once: probes your read-only warehouse connection (read-query, metadata,
query-history), discovers nearby dbt and context sources, and optionally binds a local browser
for dashboard verification. Writes only sanitized paths and capability classifications to a
gitignored
.nodal.local.json — never credentials. Runs only when explicitly requested.context-interview
The core workflow — interviews your analyst to build or improve
an ACF context repo, one domain at a time. Every confirmed disambiguation also becomes an eval
seed.
analyst-handoff
Captures critical domain knowledge when an analyst changes roles or leaves. Instead of an
unstructured exit-interview transcript, it orchestrates the governed interview — so the
departing analyst’s knowledge lands as reviewable context and eval seeds.
Ask and verify
The question-answering path is deliberately reviewable:analytics-plan
Translates a business question into a reviewable plan before any SQL runs — the
interpretation, grain, filters, time semantics, and query plan, grounded in your ACF, dbt,
docs, and warehouse evidence. Missing definitions produce a clarification or an
expert-escalation recommendation, not a falsely confident answer.
verify-result
After the read-only SQL runs, checks the executed query and result against the approved plan —
metric, filter, grain, join, and time fidelity — then assesses plausibility and whether
escalation is needed.
challenge-result
A skeptical second review when an answer looks wrong or you want another take. It reconstructs
the intended analysis, tests alternative interpretations and failure modes, and returns
upheld, inconclusive, or replan — a replan goes back through
analytics-plan for
fresh approval.dashboard-verify
Reads a named BI dashboard in your own local, already-authenticated browser — capturing both
the visible values and the active filters — so answers reconcile against your source of truth.
It never handles credentials and is configured only with your consent.
Why plan-then-verify
SQL can execute successfully at the wrong grain, omit a mandatory filter, use a noncanonical metric, or fan out through a join.analytics-plan makes intent explicit before execution;
verify-result checks whether the query and result actually satisfied that intent afterward.
Plans and verified results carry an explicitly uncalibrated uncertainty estimate — it records
unresolved semantics and evidence gaps and recommends expert escalation when a reliable answer
can’t be supported. It’s a decision aid, not a statistical confidence score.
Next
Build your context
Install the plugin and run the interview.
See the eval delta
Measure accuracy with vs. without your context.