Give closers, opening managers, inventory staff, and finance reviewers different jobs in the record. The observer reports conditions; the owner chooses the next action; the reviewer tests whether the evidence supports closure. Do not make one checkbox stand in for all three perspectives. A note that says “resolved” hides the useful questions: resolved how, by whom, against which source, and with what remaining uncertainty?
Evidence context for this decision
Question and population The bounded question is: which fields make a daily cash-reconciliation exception interpretable at a regulated retailer? The unit is one drawer or approved cash point for one close period. The population is all closes in the stated store and date range, not only closes with variances. Including only exceptions creates selection bias and makes a rate impossible to interpret. Government recordkeeping guidance, internal-control standards, and Virginia cannabis oversight all point toward contemporaneous, attributable records. They do not provide a universal cannabis variance rate. A public source can establish why a record should be retained or why a control matters; it cannot tell a specific store whether a difference arose from timing, counting, a transaction correction, or an unrecorded event. ## Findings from the evidence model Retain opening float, tender totals by type, paid-outs, refunds, deposits, expected close, counted close, variance, preparer, reviewer, and timestamps. The timestamp of a point-of-sale close and the timestamp of a bank deposit answer different questions. Combining them can make an ordinary deposit delay look like a counting exception. Preserve the system’s original values alongside normalized values and identify any conversion or rounding. The useful denominator depends on the decision. Variances per close measures control workload. Variance dollars per tendered dollar measures exposure. Exceptions per operating day measures cadence. These rates should not be averaged together. A store with many low-value closes and a store with few high-value closes can have the same exception count but different risk profiles. | Observation | Supports | Does not establish | | --- | --- | --- | | Count difference | A reconciliation exception occurred | Why it occurred | | Late deposit | A timing gap exists | Misappropriation | | Repeated drawer pattern | A targeted review is reasonable | A person’s responsibility | | Signed review | A control was performed | The underlying count was correct | ## Interpretation, transfer, and limits Repeated exceptions at one close point may support testing the counting method, tender configuration, or handoff. Causation requires corroboration. A decline in exceptions after a system change is suggestive, but it is not a controlled experiment if staffing, hours, or transaction mix changed simultaneously. Report the change as an association and preserve the comparison window. The result does not transfer automatically across jurisdictions because tax, tender, reporting, and retention requirements differ. It also does not transfer from bank-deposit records to point-of-sale records without a crosswalk. Missing closes should be reported separately from zero variance. “No exception found” means only that the reviewed evidence did not identify one. ## How to analyze a variance series Start by defining the close event. Some operators close a drawer at the end of a shift; others close a register at a calendar boundary and reconcile the cash later. A close without a counted timestamp cannot be compared fairly with a close that has one. Keep store, register, shift, preparer, reviewer, tender type, and period as separate fields. If a correction is posted on the following day, link it to the original close rather than moving the exception into the correction date. The variance sign matters. A shortage and an overage may have the same absolute value but different review questions. Break the total into tender categories before aggregating. A cash shortage can be a counting issue, while a card-total difference may be a settlement or batching issue. A refund recorded in the point-of-sale system but settled later can create a timing difference. When the source system nets categories, retain the source total and label any reconstructed split as derived rather than observed. A useful comparison has three layers. The first is frequency: how many closes contain a non-zero variance? The second is magnitude: what is the distribution of absolute variance and signed variance? The third is resolution: how many exceptions receive a documented explanation, correction, and independent review within the defined period? These layers answer different questions. A lower count with a larger tail may increase exposure. A high closure rate may reflect administrative completion without proving the explanation was correct. Trend interpretation requires stable definitions. Note changes in register configuration, cash limits, staffing, operating hours, payment mix, tax treatment, and counting procedure. A new rounding rule can change small overages without changing control quality. A new reviewer can increase recorded exceptions by finding issues previously left undocumented. Use a change point in the report and compare pre-change and post-change periods only when the measurement process is comparable. Sampling is useful when a full review is not possible, but convenience samples are not neutral. If the sample selects only high-dollar variances, it estimates the selected group, not all closes. If it selects the first close of each day, it may overrepresent opening-float routines. State the sampling frame, selection method, exclusions, and unresolved records. For rare exceptions, show counts alongside percentages; “100% resolved” from two exceptions conveys less than the count suggests. Reconciliation is also a custody question. Identify when cash moved between drawer, safe, deposit bag, and financial institution, and who controlled each handoff. A signed handoff supports an observed transfer; it does not prove the amount was correct unless the count evidence is retained. Keep deposit slips, system close, correction history, and bank confirmation linked by identifiers. Do not infer responsibility from the last person who touched a record. Access logs, schedule records, and camera evidence may corroborate timing, but each has its own limitations. Privacy and fairness constrain interpretation. Limit the review population to people whose role is relevant, protect payroll-linked data, and avoid publishing employee-level rates. Escalate a pattern under the operator’s approved investigation policy rather than creating a new disciplinary rule from a dashboard. IRS guidance and internal-control frameworks support record quality; they do not establish a cannabis loss benchmark or determine intent. The bounded finding is that a well-defined reconciliation series can reveal workload, timing gaps, and exposure, while causation requires corroborating evidence. ## A practical comparison boundary Freeze the close population at a stated cutoff, then record late postings and reopened exceptions separately. A revised report should show what changed and why, not overwrite the first snapshot. This distinction matters when management reviews a period before all deposits or corrections have settled. Stable definitions and visible late data are more informative than a polished rate that hides the reporting lag. ## What a reviewer should request The minimum review packet is the close record, tender detail, correction history, deposit or handoff evidence, and reviewer disposition for the stated population. Requesting only the exception spreadsheet removes the context needed to distinguish a true variance from a missing or delayed record. Preserve the source export date and the query filters. If a field is unavailable, state that limitation rather than reconstructing a precise value from a secondary report. ## A final evidence check Before publishing a variance rate, ask whether the close population includes zero-variance closes, late postings, and reopened exceptions. Ask whether tender, handoff, and review timestamps can be reconstructed. If not, describe the measure as a partial operational view and identify the missing evidence instead of treating the partial series as a loss benchmark. The review should also preserve the denominator’s calendar logic. A holiday closure, shortened shift, or register opened only for a delivery should be labeled rather than treated as an ordinary close. This prevents operational availability from being mistaken for improved reconciliation performance. ## Bounded conclusion A cash-reconciliation record becomes useful when the store, close period, exposure, preparer, reviewer, and disposition remain visible. It can direct a proportionate review. It cannot, alone, establish misconduct or a benchmark for every dispensary. ## Sources 1. Virginia Cannabis Control Authority 2. IRS Recordkeeping 3. COSO Internal Control Framework 4. NIST Cybersecurity Framework ## Research question and unit What can a refrigeration alert establish about a cannabis product? The unit is one sensor interval linked to one storage zone and, where possible, one affected lot. The population is all monitored intervals and alerts in the stated period. An alert, a confirmed excursion, and a product disposition are separate events. FDA storage guidance, NIST measurement practice, Virginia cannabis
Applying the evidence to end-of-day exceptions
Review a small sample at a predictable time. Read one open case for ownership and one closed case for evidence. Look for repeated delays, copied explanations, missing witnesses, late timestamps, and actions that depend on memory. A trend is not proof of fault; it is a prompt to inspect the design. Ask whether end-of-day exceptions gives staff a clear stopping point and an easy way to surface an exception before it becomes a customer or compliance problem. Close with a legible status sentence: what happened, what is protected, what evidence is attached, what decision was made, and who owns the next review. That sentence is more durable than a long narrative that leaves the reader searching for the answer. Recheck the record after the next shift or operating cycle. Reliability comes from making the right small fact available at the right moment, not from making every record look identical.
