Denominator Reconciliation Checklist¶
Reconciliation checklist — instantiates Funnel Attrition Localization
A pre-analysis gate that verifies each stage's counts agree across systems, definitions, time windows, filters, and de-duplication rules before anyone trusts the funnel.
Denominator Reconciliation Checklist is a fixed set of checks run before a funnel is analyzed, confirming that the number for each stage means the same thing everywhere it appears — same event definition, same window, same filters, deduplicated the same way — so that dividing one stage by another is legitimate. Its defining purpose is not to measure the funnel or find the leak but to certify that the counts are commensurable. It exists to catch the "we compared incompatible denominators" error, which silently invents or erases attrition and is the cheapest, most embarrassing way to a wrong conclusion. It is a gate, not a gauge: fail it, and you do not analyze yet.
Example¶
A lender's loan-application funnel draws "applications" from three systems. The marketing CRM reports ≈42,000 form submissions, duplicates and bots included; the loan-origination system reports ≈38,500, deduplicated by applicant and excluding withdrawals; the warehouse reports ≈40,100 on a slightly different date window. The apparent application→underwriting conversion looks alarming — a cliff — but the checklist walks each stage against the axes that make counts incompatible: same entity grain? same dedup key? same window? same inclusion filters? It finds the numerator (underwriting, deduplicated, from origination) was being divided by the denominator (applications, not deduplicated, from the CRM). Reconciled to one definition, the "30% cliff" shrinks to a routine ≈12%. A phantom bottleneck dissolves before a team spends a quarter fixing a stage that was never broken.
How it works¶
For every stage count, confirm it along the axes that make counts disagree: entity grain (per user vs. per session vs. per event), de-duplication rule, time window and its boundaries, inclusion and exclusion filters (bots, internal traffic, test accounts), and the system of record. Trace a single count end-to-end across systems and require that the sources tie out to within a stated tolerance; flag and quarantine any stage where numerator and denominator come from different definitions. What distinguishes it from an analysis is that it never asks whether the funnel is good — only whether its numbers are the same quantity wherever they are counted.
Tuning parameters¶
- Reconciliation tolerance — the percentage gap between systems you accept before flagging; tight catches subtle leaks but raises noise on systems that naturally lag.
- Check depth — spot-checking totals versus a full row-level tie-out; deeper catches dedup and join errors but costs real effort.
- Axes covered — which mismatch sources you test (window, grain, filter, dedup, source); adding axes catches more but lengthens the gate.
- Gate strictness — hard block versus advisory flag; strict protects trust but can stall analysis over immaterial gaps.
- Cadence — once at setup versus re-run whenever definitions or pipelines change; pipelines drift silently, so periodic re-runs catch breakage.
When it helps, and when it misleads¶
Its strength is that it catches the cheapest funnel error there is — incompatible denominators manufacturing or hiding attrition — before it steers a decision, and it makes cross-system numbers trustworthy enough to compare. It borrows the discipline of accounting reconciliation[1]: independently-sourced figures for the same quantity must tie out. Its limit is that reconciled is not correct. Every system can agree on a number that is uniformly wrong because they share an upstream bug, and the checklist will happily pass it. It also cannot see biases that leave counts internally consistent. Its classic misuse is treating a passed checklist as a certificate that the funnel is right rather than merely mutually consistent — reconciliation invoked to end debate. The discipline is to reconcile against an independent ground truth where one exists, and to keep reconciliation distinct from validation.
How it implements the components¶
instrumentation_integrity_check— its core act is verifying that recorded counts are trustworthy and mean what they claim across every system that reports them.stage_entry_and_exit_criteria— it checks that every system applies the same entry and exit definition, window, and filters to each stage, so a stage boundary is identical wherever it is counted.
It does not author those definitions in the first place — the canonical ordered_stage_model and its transition rules are set by the Event Instrumentation Specification — nor test for analytic biases that leave counts internally consistent, such as survivorship, which is the job of the Survivorship Bias Audit.
Related¶
- Instantiates: Funnel Attrition Localization — it guards the archetype's precondition that stage counts be commensurable before attrition is localized.
- Consumes: Event Instrumentation Specification supplies the definitions it reconciles observed counts against.
- Sibling mechanisms: Event Instrumentation Specification · Survivorship Bias Audit · Conversion Funnel Dashboard · Cohort Transition Table · Event Trace Process Mining · Stage Drop-Off Waterfall · Segment Funnel Comparison · Loss Pareto Review · Stage Conversion Anomaly Alert · Funnel Experiment Backlog
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Denominator Reconciliation Checklist operates as a bounded evaluation of existing evidence or work that produces a finding or disposition because it a pre-analysis gate that verifies each stage's counts agree across systems, definitions, time windows, filters, and de-duplication rules before anyone trusts the funnel.
Independent corroboration: The frozen evidence defines Denominator Reconciliation Checklist as 'A pre-analysis gate that verifies each stage's counts agree across systems, definitions, time windows, filters, and de-duplication rules before anyone trusts the funnel', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Accounting & Auditing
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Accounting reconciliation cohered independent tie-out of nominally identical counts across records, definitions, and reporting periods before use.
Related originating lineages:
- Data Science & Analytics — Analytics engineering adapted reconciliation to funnel filters, identity resolution, and deduplication.
Review resolution: Accounting reconciliation cohered independent tie-out of nominally identical counts across records, definitions, and reporting periods before use. Accounting tie-out is primary, while data-schema and population alignment make the generalized checklist a cross-domain synthesis rather than a universal primitive.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] U.S. Government Accountability Office. Defense Real Property: Developing Department-Wide Instructions and Adequate Internal Controls Should Help DOD Improve Recording of Transferred Assets. GAO-22-104652 (2022). Defines reconciliation as comparing separate records for the same assets, resolving differences, and recording needed adjustments. registry ↩