NDL The evaluation layer for AI analytics

Your AI analyst can be
confidently wrong.
Nodal makes every answer accountable.

See what every agent told your business, test the questions that matter against governed ground truth, and catch answer drift when data, definitions, or models change.

dbt change 92 questions tested 4 drifted cause identified
§ 01 · Can you answer these?

Can you answer these questions about your AI analytics?

  1. 01

    What did your agent tell Finance last week?

  2. 02

    Which answer changed after yesterday's dbt commit?

  3. 03

    Which questions rely on stale or unverified context?

  4. 04

    Would Claude and Codex return the same governed number?

  5. 05

    Which business questions have no tested ground truth?

  6. 06

    Can you show why the number in the board deck came back the way it did?

§ 02 · Why it fails silently

Every failure announces itself.
Except silent SQL.

A query that returns a confident, well-formed, wrong answer and raises no error, no alert, and no objection on its way into a decision.

§ 03 · Tested in public

Tested in public. Reproduced on request.

Shorelane is our open reliability testbed: public BigQuery data, dbt models, dashboards, planted traps, and ground truth anyone can inspect. Run the same discipline on the questions your company depends on.

What others have measured Anthropic · Meta · Ramp · Lyft · Cube · Nodal

Teams and benchmarks that reinforce the same lesson: a general-purpose agent with the right context already handles most analytics questions. Anthropic found auto-generated metric definitions were net-negative on evals. The hard part is everything around the agent.

§ 04 · One loop

Observe. Test. Re-run. Catch the drift.

Observability, evaluation, and governed context are one product because they are one loop.

  1. 01

    Observe real questions

    Who asked what, which context each answer used, and where coverage is thin.

  2. 02

    Turn the important ones into verified tests

    Analyst-confirmed ground truth, harvested from the questions your business already asks.

  3. 03

    Re-run on every change

    dbt commits, definition edits, model swaps. Drift is pinned to the change that caused it. CI for analytics.

  4. 04

    Catch drift, fix the context, re-baseline

    The corrected context goes back in front of every agent, so the next answer is right for everyone.

Every corrected answer feeds the next observation.

§ 05 · One context, any agent

Two agents. One question. One governed answer.

Claude Code and Codex, the same question through the same Nodal context, landing on the same number. Swap agents without swapping truth.

Live — Claude Code and Codex answering through the Nodal MCP
Agents Bring the one you chose
Context Keep your semantic layer
Data Stays in your warehouse
Deployment Cloud or your VPC
§ 06 · Start free

No context layer yet? Start open source.

Nodal Context is optional and Apache-2.0. Already have business context your agents use? Keep it. If you do not, Nodal Context builds governed analytics context with your analyst — interview-built and reviewed by PR. Add evaluation and observability when the team depends on the answers.

Select your agent

claude plugin marketplace add nodal-data/nodal-context
claude plugin install nodal-analytics@nodal

codex plugin marketplace add nodal-data/nodal-context
codex plugin add nodal-analytics@nodal

npx skills@latest add nodal-data/nodal-context

§ 07 · Who builds this

Ron Potok

Founder, Nodal Data LinkedIn ↗

15+ years leading data science and analytics teams at enterprise scale.

Nodal is the reliability system for AI analytics: it records what agents tell your business, tests those answers against governed ground truth, and alerts your data team when something changes.

About Nodal
End of dispatch

Your AI analyst can be confidently wrong. Make every answer accountable.

Request a demo