---
name: orphex-measurement-consistency-check
description: "Reconcile event, attribution, date, identity, and revenue definitions across ad platform, analytics, and CRM reports before treating differing totals as tracking failure."
license: MIT
metadata:
  version: "2.1.0"
---

# Orphex Measurement Consistency Check

Compare reports, exports, event dictionaries, and measurement notes from the Orphex connection below when it reaches them, and from the user otherwise. The task is to explain what can and cannot be compared from the evidence; do not assume a tracking defect because two systems show different totals.

Treat definition differences as untested explanations until supplied aligned records establish their actual sign and magnitude. Neither a timezone offset nor an attribution-model label predicts which source should be higher. Do not say that documented mismatches predict the observed direction, or that data-driven credit routinely exceeds last-click credit, without the actual compatible source populations, settings, and contributions. Report the observed ordering separately and keep its cause unresolved. A later caveat does not repair an unsupported expectation in the headline or evidence; check the entire final response for consistent uncertainty before presenting it.

## With an Orphex connection

If an Orphex connector is available, check it before asking for exports. Run `capability_search` with platform `orphex` and limit 100; use a read only if that search returns its id, and ask which workspace to use when several are bound. Read `controller.catalog` then `controller.fetch` for the stored platform figures being reconciled, and `workspace.config_read` for the connected sources. Then list guides with `skill_catalog` (kind `guide`, topic `doctrine`, then `cross_platform` and `measurement`) and follow a matching one through `skill_read`; choose by title, never by a stored id. Read the attribution, aggregation, currency and freshness doctrine before explaining a gap.

If a read is not returned, this connection cannot reach that Orphex data for the workspace: say so, then use the live platform reads it offers or the supplied exports. An absent, refused or empty read is not evidence of zero, none or healthy. Describe each read with `capability_describe` before `run_read`, keep `date_end` no later than yesterday, disclose request adjustments, and name each number's source, workspace and window. This skill's evidence rules still govern any guide, insight or playbook label, and no read authorizes account changes. Ask the user only for what is still missing.

## Build a definition map

For each source, record:

- event or metric name and business meaning;
- reporting period, timezone, and whether dates describe event, click, or conversion time;
- population and denominator, including filters, deduplication, consent, and bot rules when documented;
- attribution model and lookback window;
- currency and any conversion rate or rounding rule;
- identity or join key and whether counts represent events, sessions, people, or accepted records;
- refresh time, late-arriving data behavior, and export time.

Preserve the source's exact labels, then add a plain-language interpretation. Do not silently normalize two different events into one metric. If a field is missing, mark it unknown. Keep observed values, definition differences, hypotheses, and verification requests in separate columns.

## Reconcile in a useful order

First align dates, timezone, currency, and filters. Then compare the same event and denominator. Next account for attribution, identity, deduplication, and processing delay. Calculate absolute and percentage gaps only when both values are numeric and the comparison basis is clear. Define the direction of the gap, and use an unavailable result when the reference value is zero or missing.

A difference can be expected when systems answer different questions, such as ad-platform attributed conversions versus analytics form events versus CRM accepted leads. Do not describe such a difference as lost conversions or a broken pixel without evidence that the definitions should agree. If definitions align but values still differ, identify a specific test: inspect event IDs across systems, replay a supplied sample, compare timestamps, or check a documented filter. Do not invent identifiers, logs, or backend access.

Aggregate differences are net observations, not causal bounds. A lower total does not disconfirm duplication: duplicate records can coexist with missing records, filters, different definitions, or incomplete identity coverage. Attribution-model labels alone establish neither an expected ordering between source totals nor their reporting date basis. Read the actual report settings rather than deriving click/event dates from data-driven or last-click attribution.

Aligning one dimension removes only that documented mismatch; a residual gap does not isolate attribution or any other cause while other definitions, populations, identities, or processing states remain unresolved. Mature business outcomes do not by themselves prove that every source import or refresh is complete. State what each diagnostic tests and what it cannot establish; keep checks without actual results unresolved rather than marking causes ruled out.

Equal raw-event and distinct-ID counts establish only identifier uniqueness within the verified observed scope. They do not rule out duplicated business events with different IDs, missing records, inconsistent ID semantics, or duplicates outside that scope. Require the business-event grain, identifier meaning, completeness, and deduplication rules before interpreting count differences or claiming duplicate tracking is absent.

## Recommended output

Use a source-definition matrix followed by a reconciliation table. Label each row as comparable, comparable with caveat, or not directly comparable, and explain why. Prioritize unresolved discrepancies by impact and testability. State exactly which document, export, sample record, or owner input is needed next.

## Portable inputs and examples

- Read [the input contract](references/input-contract.md) when mapping a new export or checking the example's scope and definitions. Copy [the header-only CSV template](assets/input-template.csv) when preparing data; equivalent supplied exports remain acceptable.
- Read [the reusable business context](references/business-context.md) only for business facts or constraints this task needs. Reuse user-supplied facts with their source/date; the template contains no default targets.
- Inspect [the complete fictional input](assets/example-input.csv) with [its example output](references/example-output.md) when learning the output and calculation boundaries. Never use fictional values for a real account.

State whether the result is complete, partial, or blocked for the requested decision. Link material findings to actual supplied rows/sources and separate observed metrics, hypotheses, and estimates. Lead with a short business conclusion, then evidence, uncertainty, and the next measurable check. A data export or installed skill does not authorize account changes.