Hire an AI report definition change reviewer

Everything it can hand back

Your task sets the actual outputs. The specialist picks the methods that cover them.

• A metric difference table that shows each metric's old and new definition and what changed.
• Worked examples that calculate one period both ways from your own data.
• A reporting procedure for pulling, calculating and reviewing each metric.
• A list of open differences that still need a metric owner's decision.
• A saved record of this review for next time.

How it does the job well

• Agree the question and the decision it supports before analysing.
• Check the data first: date range, grain, missing values, duplicates and definitions.
• Compare like with like: same period length, definitions and segments.
• Separate what the data shows from what it might mean, and give confidence levels.

Mistakes it avoids

• Repeating old recommendations as new findings.
• Calling a confident answer, a finished task or an empty export verified evidence.
• Choosing a "correct" definition for the owner.
• Keeping private records in memory or logs beyond the state fields above.
• Describing installation as a running automation.
• Reporting a change without checking whether tracking or definitions changed.

What a good result looks like

An example, not your task: Your weekly sales report and your monthly board report show different revenue for the same month.

• drift-review.md. The difference table: each metric's old and new formula, denominator, time window and exclusions, and owner approval. Also the worked examples and open differences.
• step-01-spreadsheet-checker.md. Why the same named metric gives different totals. It traces the gap to grain, formulas, time windows, exclusions and source versions, without rewriting either source.
• step-02-process-and-sop-writer.md. The reporting procedure: how to pull, calculate, reconcile and review each metric, with a versioned definition, a named owner and the unresolved conflicts.
• run-log.md. The source files and versions used, which steps ran or were skipped and why, and the checks that passed or were missed.
• state.json. The small record for next time: report ID, definition version, denominator, cutoff date, approved owner and open differences.

How the work is checked

• A shared label does not imply a shared denominator. Definition changes need the metric owner's approval; until then the trend is marked not comparable.
• Every figure, date and claim traces to a source or is labelled an assumption.
• Original files are intact. Revised files have new names or versions.
• Each component check ran, or the run stopped and says which could not pass.
• States which checks passed, failed or could not run, and names missing inputs.

These checks have not been run against a model yet. Your AI uses them to check its own work.

Capabilities

• Read scope and prior state. Loads last review's record if you give one. The record holds report ID, definition version, denominator, cutoff, approved owner and open differences. On a first run it sets the baseline and says no drift can be claimed yet.
• Check what changed. Compares each report's formula, data grain, time window and exclusions with the record. A missing export counts as unknown, not as "no change".
• Choose the path. If nothing changed and no review is due, it returns a short "unchanged" note. If something changed, it runs the Align recurring report metric definitions workflow. If a definition sits in a long memo or email thread, it runs the Source note summariser first.
• Trace and document. The workflow's Spreadsheet checker traces why the same metric differs. Its Process and SOP writer writes the reporting procedure with a named owner and open conflicts.
• Check and save. It checks every figure against a source, keeps your original files untouched and allows at most two correction passes. It then saves the files, the run log and the updated record.

How the role works

1. Checks your task, context and available tools.
2. Picks the methods needed and prepares a work plan.
3. Carries out supported tasks, with each method's limits and review gates.
4. Checks the real deliverables and corrects supported faults.
5. Hands back files, evidence and unresolved gaps.

Your specialist's working setup

Download the base specialist without a task form or account. README.md and SETUP.md explain how to use the folder. SOUL.md defines its scope. INTAKE.md asks missing questions inside your AI. PLAYBOOK.md, SKILLS.md, BEST_PRACTICES.md, CHECKS.md and HANDOFF.md guide the work and its checked deliverables.

Complete associated skills, supporting files and licences are included. Choose only the methods needed for the actual task. Keep context, working files and accepted feedback in your AI project.

Use this specialist through MCP or the API

Call prepare_specialist with role_id report-definition-drift-review-assistant only. No task is required to load it. Your AI asks what job you need done, what you want back and which relevant context it can use.

For task-led hiring, call find_specialist, explain the suggested specialists and let the user accept or change the team.

Complete ZIP: Download base specialist. API: GET /api/v1/specialists/report-definition-drift-review-assistant/pack. The host AI supplies its own model, tools and execution; a pack does not connect accounts or run a task.

Inputs and helpful context

The specialist starts from the task brief. Only active methods require their inputs.

For the primary method, when used

• The recurring reports you want checked, as exports or files, with the period each covers.
• The source tables or exports behind them, with dates.
• The written definition of each metric, if one exists, or the formula used in the report.
• The name of the person who owns each metric.
• Two or three examples where the same metric gives different numbers.
• The last review's saved record and files (optional on the first run).
• Who will read the result, and what you may and may not change.

Helpful context (optional)

• Approved examples of a good result.
• A named reviewer and their acceptance rules.

Tools and access

• A compatible spreadsheet engine: Formula recalculation and workbook inspection need a real spreadsheet engine. Without one, the agent reads the tables and says which formulas it could not recalculate.
• Any good chat AI: Claude, ChatGPT and Gemini all handle a long source in one message. Claude takes the longest texts.
• A spreadsheet tool such as Excel or Google Sheets: Lets you open the difference table and re-check a formula. The AI cannot recalculate a workbook without one.

Functions and deliverables

• Report definition change reviewer: Compare report formulas, dates and counting rules. Find mismatches before presenting a trend or changing a shared definition.
  • A metric difference table that shows each metric's old and new definition and what changed.
  • Worked examples that calculate one period both ways from your own data.
  • A reporting procedure for pulling, calculating and reviewing each metric.
  • A list of open differences that still need a metric owner's decision.
  • A saved record of this review for next time.
• Source note summariser: Paste an article, talk or chapter into this source note summariser. You get each claim, how well it is backed, the gaps and three quotes to keep.
  • Every claim with a quote and location, and the evidence behind each rated
  • The gaps and assumptions
  • What it adds to your question, three quotes to keep, and an 80-word summary
• Report totals and definition checker: Trace why recurring reports disagree. Get a difference table, worked examples and draft rules for future reporting.
  • A metric difference table: each number in each report, the gap and the cause
  • A worked example for each gap, using your own rows
  • A reporting procedure: how each number is pulled, checked and signed off
  • A list of open conflicts for the metric owner to decide
  • A hand-back summary with one next step
• Shared field change checker: Check a changed field or flag across your database, API and apps. Find where it is missing or read differently before you release.
  • A definition of each state
  • A map of everywhere the field flows
  • A list of gaps with evidence
  • Missing test and rollout checks
  Source: sickn33.
• Metric definition writer: Share a metric, calculation and business use. Get a YAML definition, validation results and questions for the data owner.
  • A filled metric, dimension or entity YAML
  • A validation report
  • dbt-ready field names
  • Notes on edge cases
  Source: Nimrod Fisher.

Tasks you can use it for

• Your weekly sales report and your monthly board report show different revenue for the same month.
• Someone changed a filter or a time window in a dashboard and last quarter no longer matches this quarter.
• A new data source feeds an old report and the totals jumped.
• You review the same reports every month and want a record of what changed since last time.

Limits to check before starting

• You have one question about two files today, such as why two totals differ. Use the Spreadsheet checker.
• You need a written procedure for a process and no report is in question. Use the Process and SOP writer.
• You want the monthly numbers report itself, not a check on its definitions. Use the Recurring KPI report maker.
• You want a single run with no saved record for next time. Use the Report totals and definition checker workflow.

Best way to use this role

Give it one clear outcome, the actual files and a definition of done. Keep the work in one AI project so it can reuse checked context and earlier files.

• Download the pack and add it to Claude, ChatGPT or Gemini. It includes this agent, the Align recurring report metric definitions workflow and the Source note summariser.
• Upload the reports, source exports and definitions. Say which review this is, and add last review's record if you have one.
• Tell it who owns each metric, who will read the result and what it must not change.
• Read the difference table first. Check two rows against your own spreadsheet before you trust the rest.
• Send open differences to the metric owners. Do not publish a trend line across a changed definition until they approve.
• Save the new record file. Give it back to the agent at the next review.

Read it as data

Read the complete role setup as JSON. Connect your AI to the AI Blocks MCP library and read aiblocks://roles/report-definition-drift-review-assistant for its role instructions and pinned sources.

This is a prepared AI role. No task has run. Your AI carries out authorized work using its available tools and reports the actual deliverables and checks.