Skip to content
Skills Library
Data & analytics · Review & audit

verify-result

/verify-result

Review analytics SQL and results against an approved plan for metric, filter, grain, join, and time fidelity, then assess plausibility and escalation needs. Use after analytics execution or when asked to audit a query/result. Do not use to define metrics or modify data.

Licence
Apache-2.0
Last repository change
27.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
2
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

Review analytics SQL and results against an approved plan for metric, filter, grain, join, and time fidelity, then assess plausibility and escalation needs. Use after analytics execution or when asked to audit a query/result. Do not use to define metrics or modify data.

Original author description

Review analytics SQL and results against an approved plan for metric, filter, grain, join, and time fidelity, then assess plausibility and escalation needs. Use after analytics execution or when asked to audit a query/result. Do not use to define metrics or modify data.

How it works

  1. Inputs
  2. Hard boundaries
  3. Review
  4. Output
Original SKILL.md headings · 93 lines, 725 words
SKILL.md preview
---
name: verify-result
description: Review analytics SQL and results against an approved plan for metric, filter, grain, join, and time fidelity, then assess plausibility and escalation needs. Use after analytics execution or when asked to audit a query/result. Do not use to define metrics or modify data.
---

Approval contract: new and changed context stays on a named working branch and
has no authored approval status. Human-reviewed protected merge grants definition
approval; consult the live authority through the runtime for the exact revision
and freshness. Missing/expired evidence or stale dependencies blocks governed
use. Explicit branch-test mode is proposed context only. Query/notebook result
verification is separate from definition approval; retain source and review
evidence. See the context-interview SPEC approval-through-merge contract.


# Verify Result

Review an analytical result the way a careful analytics reviewer examines a code
change: first determine whether the query implemented the approved intent, then
determine whether the observed result is plausible. A successful query is not by
itself a verified answer.

## Inputs

Use the approved analytical plan, executed SQL, result or result metadata, and the
context evidence used by the plan. Ask for a missing artifact only when it cannot
be recovered from the current conversation or workspace. If no approved plan
exists, reconstruct the apparent plan from the SQL, label plan fidelity
`unavailable`, and return at most `WARN`.

## Hard boundaries

- Never modify SQL, context, warehouse data, dashboards, or configuration.
- A review describes the required correction; it does not deliver a complete
  replacement SQL/Python implementation, even in a next-step section. Cite the
  relevant predicate or operation inline and hand implementation back to the
  analyst or `analytics-plan` for the approved rerun.
- Any diagnostic warehouse query must be read-only, bounded, and necessary to
  evaluate a named risk. Use only an already-authorized connection.
- Do not treat query history or typical ranges as ground truth. They support a
  plausibility finding, not a semantic definition.
- Missing evidence stays `unknown`; it is never converted into a pass.
  Do not infer data completeness from how long ago a period ended, or declare
  null handling safe without evidence about the metric. Unobserved schema risks
  are questions to resolve, not findings that those columns or defects exist.

## Review

1. **Plan fidelity.** Compare the approved interpretation and IR against the SQL:
   metric expression, source tables, joins and cardinality, mandatory filters,
   exclusions, grain, dimensions, time boundaries/timezone, comparison, ordering,
   limits, and expected output shape.
   Use the same expanded ACF definitions and canonical metric/method IR as the
   plan. Check inherited/grouped filters, entity-filter stages, allowed dimensions,
   zero-denominator policy, and effective status/open decisions. For a method,
   check selected metric branches, parameter defaults, decision IDs, declared
   overrides and stock/flow alignment against the actual SQL/Python implementation.
2. **Query risks.** Check fanout, deduplication, slowly changing dimensions,
   many-to-many joins, null behavior, divide-by-zero handling, signed values,
   mixed grains, non-additive metrics, and filters applied after aggregation.
3. **Execution evidence.** Record success/failure, row count, returned columns,
   null/duplicate symptoms, and whether the result shape matches the plan.
   For methods, verify required disclosures, output grain/keys/columns and each
   qualitative check against actual results. A generated implementation or a
   model's assertion cannot establish execution or a result relationship. Missing
   execution/check evidence requires `WARN`, even if proposed SQL looks correct.
4. **Plausibility.** When evidence exists, compare with a confirmed seed, trusted
   dashboard capture, bounded prior result, or historical range. Preserve the
   source and snapshot/filter state. Do not invent a range.
   A reviewed independent notebook comparison is also eligible evidence. Compare
   the same inputs, as-of date and bindings; compare keyed row sets, reject duplicate
   keys, preserve nulls and use only pre-agreed numeric tolerances. Equal totals
   cannot excuse wrong cohort membership, output grain or missing disclosures.
   Stale method/source/dependency revisions invalidate the prior comparison.
5. **Confidence and escalation.** Apply `references/review-rubric.md`. Post-query
   confidence may downgrade plan confidence; overall confidence is the lower of
   the two. A trusted snapshot match supports verification but does not erase an
   unresolved semantic conflict.

## Output

Lead with one outcome:

- `PASS`: implementation is faithful, required checks have supporting evidence,
  and no material anomaly was found.
- `WARN`: no proven defect, but evidence is incomplete or a plausible risk remains.
- `FAIL`: SQL departs materially from the plan, execution failed, or trusted
  verification evidence contradicts the result.

Then list findings by severity with the exact plan clause, SQL construct, and
impact. Include post-query confidence, overall confidence, evidence checked and
unavailable, and a concrete escalation recommendation. Avoid generic advice such
as “double-check the query.”

State plan fidelity explicitly as `faithful`, `unfaithful`, or `unavailable`.
With no approved plan, use `Plan fidelity: unavailable`; an apparent plan inferred
from SQL does not fill that gap or authorize upgrading the result to a pass.
Files in this package (2)
  • SKILL.md 6 KB
  • references/review-rubric.md 3 KB

Versions

RepositoryScoreLicence
This page · nodal-data/nodal-context90Apache-2.0
pandysp/claude-plugins100MIT

Database snapshot: 13.11.2026 · package version 6ec18a6fd2b3 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100review-pr
    Reviews a data/analytics PR: SQL, grain, cost and DAG risks plus data evidence from a configured non-production environment, then drafts a review the user approves before anything is posted. Use when a data pipeline PR needs review.
  • 100sql-query-optimizer
    Optimize sql query optimizer operations. Auto-activating skill for Data Analytics. Triggers on: sql query optimizer, sql query optimizer Part of the Data Analytics skill category. Use when working with sql query optimizer functionality. Trigger with phrases like "sql query…
  • 100data
    (forwward) Designs analytics systems, writes SQL queries, plans event tracking, and builds dashboards for product and user metrics. Triggers on SQL, analytics, dashboards, tracking, data pipelines, or user behavior analysis.
  • 90metrics-generation
    Atlan Analytics: generate, check, edit or save metrics and saved SQL from dashboards, BI exports, models or query history. Use when asked to build, mine, refresh, fix, propose or publish a metric. Not for answering a data question (analytics-ask).
  • 100analytics
    Chart GrowthBook product data, build Analytics dashboards, manage the metric catalog, or query the warehouse directly — run Product Analytics explorations, save charts together on a dashboard, search metrics and fact tables, create fact metrics and their fact tables, or fall…
  • 100query-optimizer
    Optimize SQL queries for performance with indexing strategies, query rewriting, and execution plan analysis. Use when queries are slow, optimizing database performance, or analyzing query execution.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.