Production Web Analytics

Jaccard similarity across calculated metric formulas identified seven metrics with ≥ 0.85 formula overlap that all compute revenue per visit, with minor variations in attribution model and allocation. They appear to be the result of independent authors solving the same problem without coordination.

cm_revenue_per_visit              author: r.kim@      created: 2024-03-12
cm_rev_per_visit_v2               author: a.patel@    created: 2024-08-04
cm_revpv_lasttouch                author: r.kim@      created: 2024-11-19
cm_rev_visit_linear               author: m.chen@     created: 2025-01-22
cm_revenue_visit_corrected        author: a.patel@    created: 2025-04-08
cm_rpv_marketing                  author: l.gomez@    created: 2025-09-15
cm_rev_per_visit_final            author: r.kim@      created: 2026-02-03

Near-duplicate metrics produce subtly different numbers in different reports, which surfaces as “the dashboards disagree” complaints from executives. They also create maintenance debt: when the underlying definition needs to change, all seven must be updated in lockstep, and inevitably one is missed.

Designate one canonical metric, document its attribution and allocation choices, and update downstream Workspace projects to reference it. Deprecate the others by tagging them with a sunset date and reviewing downstream references before retirement.

The grader could not locate any prior snapshots of this data view in the configured snapshot directory. Without snapshot history, configuration changes cannot be diffed, drift cannot be detected, and audit trails are unavailable for compliance review.

cja_auto_sdr dv_prod_web --include-all-inventory --format json --output snapshots/snapshot_2026-04-25.json
sdr-grader snapshots/ --trend --output snapshots/trend.html

Store dated snapshots in version control and schedule the same supported export command at an interval appropriate for the project.

120 dimensions in this data view have empty description fields. Descriptions are the primary way new analysts and AI agents understand what a component measures; missing descriptions force readers to infer intent from names alone, which is frequently wrong.

Dimensions: 120 of 203 missing (59%). The strict@2.0 rubric threshold is 56%.

Use the component IDs reported above to populate descriptions via the data view API. Establish a CI check that fails new components added without descriptions.

The rubric flags segments with container nesting depth above 5 as difficult to review. Three segments in this data view exceed that threshold; the deepest reaches depth 8.

seg_qualified_lead_v3             depth: 8     containers: event/session/person mixed
seg_high_intent_returning         depth: 6     containers: session/event nested
seg_b2b_account_engaged           depth: 6     containers: event/session mixed

Deep nesting makes intent illegible. Reviewers cannot easily tell whether the segment matches what its name implies, and small definitional changes have unpredictable population effects.

Decompose each segment into composed sub-segments, named for what each layer captures. The surface segment then becomes a shallow combination of well-named pieces, each independently reviewable.

Twelve data view metrics configure an attribution override, but their descriptions do not identify the model or explain why the override is appropriate.

Update each metric description to name the configured attribution model and its purpose, or remove an override that is not intentional.

73% of custom dimensions follow the c_ prefix convention. The rubric expects ≥ 80%. Four components diverge from the established pattern.

product_category          (expected: c_product_category)
user_segment              (expected: c_user_segment)
cart_value_band           (expected: c_cart_value_band)
last_search_term          (expected: c_last_search_term)

Rename the four affected components via the data view API. Document the prefix convention in the SDR for future contributors.