1. A blank field is a question, not a completed investigation
A photograph of a form can show something important: an empty signature box, a number that differs from another record, or an unexplained interval. What it cannot do by itself is answer every question that follows. Was the photographed page the complete record? Was the number a container identifier or a seal identifier? Was the person signing for receipt also attesting to a physical count? Does the interval concern possession of ballots, creation of paperwork, or publication of a scan?
Those distinctions are not excuses for overlooking deficiencies. They are how an inquiry becomes capable of identifying one. Calling every incomplete record proof of changed votes skips the evidence needed to establish that conclusion. Calling every discrepancy harmless paperwork skips the evidence needed to dismiss it. Both shortcuts substitute a preferred story for a testable account.
The U.S. Election Assistance Commission’s Chain of Custody Best Practices describes a paper trail connecting transfers of election materials. The retrieved document is dated July 12, 2021, version 1.11, not a newly issued 2026 protocol. It offers voluntary practices and sample forms, expressly subject to federal, state and local law. Its organizing questions concern the item, people, place, time and condition at transfer and receipt. That is guidance for documenting custody, not a nationwide rulebook or a guarantee about a particular election. EAC chain-of-custody guide
This article examines what a reader can reasonably infer from records, using an entirely fictional exercise called Juniper Records. Juniper is not a real county, election office, incident or voting system. All identifiers, quantities, staff labels and times in its examples were invented for this article. The examples do not reproduce any jurisdiction’s prescribed workflow. They are document-reconciliation problems, not instructions for handling ballots or entering restricted facilities.
Three questions will remain separate throughout. What does the available record say happened? What additional evidence supports or conflicts with that account? What conclusion is justified at the present stage? A discrepancy can deserve urgent attention even when the last question remains unanswered. Conversely, a neat spreadsheet can conceal an unresolved identification problem.
The useful public-interest story is therefore not whether paperwork deserves unconditional trust. It is whether the evidence can be connected well enough to distinguish an omitted field, an incomplete production of records, a procedural departure and an established change to election materials. Those are different findings. A responsible account identifies which one it has reached and what prevents it from reaching the next.
2. First establish which document governs which event
A custody form cannot be evaluated against an unspecified idea of what election workers should have done. The relevant procedure has a jurisdiction, version, material category and period of application. A requirement for unvoted stock may not describe voted-ballot transport. A local implementation document may add detail that a statewide manual leaves to administrators. A rule found through a search result may have changed since the search engine indexed it.
Colorado supplies a concrete version-control warning. The Secretary of State’s retrieved unofficial Rule 20 copy is headed as adopted and temporarily effective October 9, 2026. Section 20.7.1 addresses unvoted ballots, including container logs, seal verification on receipt and daily reconciliation of unvoted, issued and spoiled paper ballots by judges of different affiliations. Section 20.7.2 separately addresses voted ballots and their transport. Section 20.7.3 requires an incident report for a discovered violation of Rule 20.7. These are Colorado provisions in that dated copy, not requirements this article attributes to every state. Colorado Rule 20
Minnesota’s 2026 Election Judge Manual takes the reader to local implementation: its custody section explains that counties and specified local jurisdictions maintain plans, and directs judges to follow their local election official’s procedures. The introduction also says the manual does not replace state law or local guidance. A state manual alone therefore may not answer every question about a particular assignment. Minnesota Election Judge Manual, pages 2 and 11
For Juniper, imagine receiving three files named instructions, revised instructions and final instructions. The filenames establish almost nothing about which applied to an event. One PDF might be a draft circulated before approval; another might cover a different building. A reader needs the issuing authority, adoption or approval evidence, effective period and relevant subject. A later upload date does not necessarily mean a later operational rule.
Suppose a fictional transfer occurred on Day Four and a revised instruction took effect on Day Six. A comparison with the revised instruction can identify a difference, but it cannot establish a Day Four violation merely by applying the later wording backward. Conversely, an older form remaining in a folder does not establish that the older procedure still applied. Version selection is part of the factual argument, not administrative decoration.
It helps to write the proposition before reading the form: the applicable instruction required field X for this class of transfer, and the supplied record lacks field X. Each component needs support. That narrow proposition is much stronger than a screenshot caption announcing that the chain was broken. It also makes a meaningful response possible: the office can identify another record, dispute the applicable version, explain an exception or acknowledge a deficiency without being asked to answer a conclusion broader than the evidence.
3. Join the records before adding their numbers
Juniper’s first packet contains a dispatch sheet for container J-14, a receipt for J-14 and a scanned copy of that receipt. The dispatch lists 240 folded paper sheets. The receipt lists 240. The scan also shows 240 because it depicts the same receipt. Adding the three entries would produce 720, but the packet describes one claimed movement of 240 sheets, not three deliveries. The problem is duplication of evidence, not multiplication of physical objects.
Now introduce a transfer reference, T-08, appearing on the dispatch and receipt. The scan’s file name is upload-003. A hurried spreadsheet importer treats upload-003 as a third transfer identifier. This is a data-model mistake invented for the exercise. The file identifier describes a digital object; T-08 describes the asserted movement. Keeping both is useful, but placing them in the same column makes a false total look mechanically calculated.
A sensible analytical view has separate entries for source document, represented event and physical grouping. It preserves the originals rather than silently deleting the apparently redundant scan. The scan could matter if it differs from the later copy. Deduplication for counting should not become destruction of evidence about document versions. The reader wants to know both how many events are represented and how many versions of each account exist.
Juniper’s second packet complicates the join. Container J-14 arrives with receipt R-21, while the dispatch refers to receipt R-12 in a handwritten note. Both receipts exist. R-12 concerns another container and another quantity. Matching solely on the handwritten reference creates a contradiction; matching solely on the convenient quantity could hide one. The appropriate conclusion at this point is that the cross-reference is inconsistent and requires explanation, not that the analyst may choose whichever join makes the columns balance.
The fictional office later supplies an original register whose row links T-08 to R-21 and an amendment dated the following morning correcting the handwritten reference. That evidence supports a clerical-reference explanation more strongly than an unsupported assurance would. It does not make the earlier error disappear. A report can retain the original discrepancy, describe the supplied correction and say exactly which link is now supported.
This approach makes corrections visible without treating them as automatically suspicious or automatically decisive. A correction made contemporaneously, one made during routine review and one created after a public query may have different evidentiary contexts. The date alone does not settle credibility. The question is whether the correction identifies its basis and whether that basis can be examined independently of the corrected statement.
In this exercise the final inventory is still 240 claimed sheets for T-08. What changed was the confidence in the relationship between documents. That is an important result even though no physical total changed at all.
4. A container’s identity is not the identity of every seal it has used
A record can name both a reusable container and a seal associated with a particular interval. Confusing those identifiers makes a legitimate change look like an impossible history, or makes an unexplained change vanish into a generic container label. The first task is to read the field definitions, not to assume that every number on a bag serves the same purpose.
New Jersey’s Early Voting Guidelines, revised May 11, 2026, illustrate why a packet can contain several related records. The guidelines distinguish a Seal Audit Log and Daily Ballot Reconciliation Form retained at the early-voting site from a Ballot Transfer Form accompanying the bag. They describe recording replacement seals and provide for documented opening and consolidation of returned batches by designated staff. A different seal number is therefore not inherently evidence of an unauthorized opening; the relevant documents and circumstances matter. New Jersey early-voting guidelines, pages 9–13
The following Juniper example is original and is not a reenactment of those instructions. Its packet says container J-14 was associated with label S-410 at departure and S-419 at a later storage entry. The packet initially contains no explanation of the difference. Three accounts remain logically possible: the entry was copied incorrectly; an authorized event changed the association; or something occurred that the available records do not authorize or explain. The labels alone do not choose among them.
To test the copying account, imagine that a contemporaneous source image clearly reads S-419 while an exported spreadsheet reads S-410. The mismatch can then be localized to transcription in the export. This finding concerns the represented identifier. It does not establish every fact about the container’s physical history, but it removes one alleged contradiction between the original records.
To test the documented-event account, imagine instead that the original departure image clearly reads S-410 and a separate event note records an intervening inspection involving J-14, followed by S-419. The note cites a specific instruction and names responsible roles. Its existence supplies an explanation to investigate; its compatibility with the applicable procedure, chronology and other evidence determines how much that explanation supports. A later note is not self-authenticating simply because it tells a coherent story.
In the third version, no additional explanation is available. The report should leave the interval unresolved. It should not fill the blank with either an invented benign event or an invented malicious one. An unexplained identifier transition is a finding about the evidence; an unauthorized opening is a claim about conduct; alteration of votes is a further claim about consequences. Establishing one does not automatically establish all three.
The reverse caution matters too. Equal identifiers do not prove that everything inside a container remained unchanged. They establish agreement between the recorded identifiers. The inference being sought must be narrower than the evidence being used, or the log becomes a substitute for the investigation it was supposed to support.
5. Count sheets, voters and selections in different columns
Juniper’s next exercise begins with 1,000 blank paper sheets assigned to a fictional location. At the chosen reporting cutoff, the packet classifies 620 as cast sheets, 18 as spoiled sheets retained separately, 350 as unused sheets and 12 as sheets transferred elsewhere. These categories are stipulated to be mutually exclusive and exhaustive for this exercise. Their sum is 1,000. This is an inventory identity, not a model of every jurisdiction’s ballot accounting.
A second report says 638 sheets were issued, including the 18 later spoiled. Adding issued to cast and spoiled would count overlapping categories. The invented issuance figure is a flow through the process; the four terminal categories are the sheets’ stipulated positions at the cutoff. Neither number is inherently wrong. The error would be treating them as disjoint pieces of the same total.
Now remove the transfer entry from the first packet. The visible terminal categories sum to 988, leaving a difference of 12. If a later-produced transfer record accounts for those 12, the arithmetic difference closes. That does not mean the analyst should rewrite the earlier report as though no question arose. The useful finding is that the initial production was incomplete for the reconciliation, and an identified record subsequently accounted for the difference. Whether the underlying transfer complied with applicable procedures remains a separate question.
Consider a different version in which no transfer record appears. Writing a balancing adjustment of 12 into the spreadsheet makes the formula return zero, but produces no new evidence. A balancing entry may be useful as a clearly named unresolved difference. It is misleading when placed among documented categories as though it describes an observed disposition. Arithmetic can reveal the size of a problem without explaining its cause.
The 620 cast sheets also do not imply 620 votes for candidates in every contest. For a wholly fictional single-choice contest printed on those sheets, stipulate 590 valid candidate selections, 24 blank responses and six overvotes. Those categories total 620 sheets. Comparing 590 candidate selections with 620 sheets and announcing 30 missing ballots would compare different units. This example does not specify how any real jurisdiction adjudicates a mark; it simply demonstrates why a selection count and a physical-sheet count are different quantities.
Nor should a reader assume that one voter always corresponds to one sheet in an unfamiliar system. The counting unit has to come from the actual form and ballot design. A multipage ballot packet, an envelope, a sheet and a scanner record are not interchangeable labels. The analyst should state the unit in the sentence containing the number. That small discipline prevents a surprising amount of confusion without making any claim about whether a particular election result is correct.
6. Distinguish the time of an event from the time somebody uploaded its record
Juniper’s fictional dispatch is recorded at 20:14 and receipt at 20:39. On their face, those entries describe a 25-minute interval. A document-management system shows the receipt scan arriving at 22:05. That later time does not turn the stated transport interval into 111 minutes. It describes a different event: availability of a digital copy in that system.
The distinction also runs in the other direction. A scan uploaded quickly does not prove that the event time written on the paper is accurate. The upload supplies an upper bound on when that particular digital copy entered the observed system, assuming the system’s time and record are reliable. It does not retrospectively verify every handwritten statement. A reader who separates the two clocks can ask a much better question than one who labels all timestamps arrival time.
For a second example, suppose one fictional device’s clock was four minutes fast and another’s was three minutes slow. A displayed dispatch of 20:14 corresponds, under those explicit assumptions, to 20:10; a displayed receipt of 20:39 corresponds to 20:42. The corrected interval is 32 minutes, not 25. That calculation is not a license to adjust inconvenient times. Without evidence of the offsets, the correction is merely a hypothetical explanation and must remain labeled as such.
A real investigation might have evidence about clock settings, synchronized logs or contemporaneous observations. This exercise claims none. Its purpose is to show what information is missing from a confident duration calculation. Precision to the minute in a form does not establish accuracy to the minute in the event it describes.
The order of records can also differ from the order of events. Imagine a batch upload containing yesterday’s receipt alongside today’s dispatch, sorted alphabetically by filename. A screenshot of the folder displays them in that arbitrary order. Calling the order impossible confuses an interface’s sorting rule with a historical sequence. Sorting by asserted event time helps, but the analyst should still retain the upload and amendment times because they answer different questions.
Finally, a time gap needs a defined boundary. An interval between signatures may include waiting under documented custody; an interval between photographs may simply reflect when photographs were taken. Neither automatically establishes an interval of uncontrolled access. Conversely, an explained journey length does not establish that custody was adequately documented throughout it. The meaningful finding identifies the precise interval, the records bounding it and the proposition those records do or do not support. That is more informative than a dramatic duration without a defined clock.
7. Missing from a response is not the same as never created
A public-records response is a collection assembled in answer to a request. To interpret an absence within it, the reader needs to know what was requested, the search or production scope, any stated exclusions and whether further installments are expected. A folder labeled complete is useful evidence about the producer’s representation; it is not a substitute for understanding what complete means in that exchange.
North Carolina’s numbered memo 2022-01, updated July 16, 2026, distinguishes searching existing records from creating or compiling new ones. It also addresses supervised access and information that must be withheld or redacted. Those are North Carolina instructions, not a general right to obtain every election document in every state. They illustrate why the contents of a public response and the total underlying record collection can differ for stated reasons. NCSBE public-records guidance
Suppose Juniper receives a request for dispatch sheets from Building East on Day Four. The response contains the requested dispatch sheets but no destination receipts, which are indexed under Building Central. An analyst may accurately report that the production lacks receipts. Reporting that no receipts exist would go beyond the stated request and response. A follow-up asking for the corresponding receipts is not a concession that the first concern was illegitimate; it is the next evidence-gathering step.
Change the facts. Suppose the fictional records custodian expressly states that the search covered both buildings, that all located responsive records have been supplied, and that no receipt for T-08 was located. That is stronger evidence of an unresolved records gap. Still, no record located is not identical to no event occurred. The response concerns the outcome of a search; the underlying physical history requires additional evidence. The report should preserve the custodian’s precise statement rather than inflate it.
A third response says a document exists but a portion is withheld under an identified legal basis. That is neither an empty database nor an affirmative account of the concealed content. A reader can scrutinize the stated basis through lawful channels without inventing what the unavailable portion says. The absent text may be important, but importance does not supply its content.
These distinctions can be captured in separate analytical statuses: not requested, requested but not yet produced, reported not located, identified but withheld, supplied and unreadable, or supplied and legible. The statuses describe the researcher’s evidence, not a rating of the election. Collapsing them into one missing category loses the information needed to resolve them.
There is also an ethical boundary. The desire to check custody does not justify publishing voters’ personal information, harassing named workers or entering restricted areas. A useful public explanation can identify a record category, an event reference and an unresolved question without turning a document review into a campaign against an individual. Accountability improves when claims are specific enough to answer, not when uncertainty is personalized.
8. An exception rate needs a unit and a denominator
Imagine Juniper’s synthetic dataset contains 200 containers, each represented by three expected transfer events, for 600 expected events in total. Twelve event records lack a required field under the invented exercise specification. Those twelve are associated with eight containers: four containers have two deficient records each, and four have one each. The numbers are stipulated; they are not observations from an election.
The event-record deficiency rate is 12 divided by 600, or two percent. The share of containers associated with at least one deficient record is eight divided by 200, or four percent. Both calculations are correct. Neither is a percentage of ballots altered, a probability that votes were counted incorrectly, or an estimate of the election’s outcome. They summarize different aspects of the same fictional documentation problem.
Now stipulate that the 200 containers together are listed as holding 12,000 sheets, and the eight identified containers are listed as holding 960 of them. Eight percent of listed sheets are associated with a container having at least one deficient event record. That is a useful measure of the scope requiring documentary follow-up. It is not an assertion that 960 sheets were affected physically. The words associated with are doing necessary analytical work, not hiding a conclusion.
A count-weighted measure can also obscure where the problem occurs. If one container holds many sheets, it may dominate that measure even though only one transfer record is deficient. If the aim is improving form completion, event counts might be more useful. If the aim is locating materials connected with an unresolved interval, grouping by container may be necessary. A statistic is informative only in relation to the question it was constructed to answer.
Suppose eight of the twelve deficient records are later supplemented with legible contemporaneous copies, leaving four unresolved records across three containers. A before-and-after report should retain the original review date and describe what changed. It should not quietly replace two percent with two-thirds of one percent and suggest that the first number was fabricated. Twelve divided by 600 is two percent; four divided by 600 is approximately 0.67 percent. They describe different evidence states at different times.
Nor should the remaining four be dismissed because the percentage is small. An exception’s significance depends on its nature and evidence, not merely its prevalence. A common formatting omission and an unexplained physical discrepancy could share a count while demanding different inquiries. This is why an exception register needs a short description of each issue, not just a red cell and an aggregate percentage.
No statistical sampling inference is being made here. The exercise describes the entire stipulated dataset. It cannot support a risk limit, a national rate, or a conclusion about unexamined jurisdictions. Those would require different methods and evidence, outside this article’s custody-record question.
9. Agreement is stronger when the accounts are genuinely separate
Juniper’s dispatch spreadsheet, summary dashboard and public PDF all display the same total: 240. That looks like three confirmations until the data lineage shows that the dashboard imports the spreadsheet and the PDF is exported from the dashboard. They are three presentations of one entry. Their agreement can confirm that the copying pipeline preserved the value. It does not provide three independent observations of the underlying quantity.
Contrast that with an exercise in which a dispatch entry and a separately recorded receipt both state 240, and the sources identify distinct observations rather than one copied field. The agreement is more informative about the represented transfer. It still depends on what each observer actually checked. A signature acknowledging a container’s arrival is not necessarily a recount of its contents. The form’s wording and the author’s basis matter more than the number of documents carrying a signature.
The strongest-looking record can be misunderstood if its assertion is never read. Juniper’s fictional form has one checkbox for package received and another for quantity verified. The first is marked, the second blank. The receiving record supports receipt of a package under the example’s definitions; it does not attest to the quantity check. Treating it as a full count confirmation expands the statement beyond what the form says.
Consider a further wrinkle: the receipt quantity is preprinted from the dispatch database. A handwritten acknowledgment appears beside it. Without an explanation of what the acknowledgment means, the analyst cannot infer that the printed quantity was independently measured. This is an original example of evidence dependence, not a claim about any real election office’s software or forms.
A contradictory independent observation can be more useful than a thousand repeated copies. If a separately documented count says 238 rather than 240, the difference deserves reconciliation. It could reflect a counting error, a unit mismatch, an excluded category or a physical discrepancy. Those are candidate explanations, not findings. The existence of several agreeable exports does not outvote the contradiction. Evidence is not a popularity contest among files.
The next analytical step is to locate the first point where the accounts diverge. Did the original handwritten sheet say 238, with 240 introduced during transcription? Did both original accounts disagree? Did a later correction identify a specific reason? Each answer localizes a different problem. The aim is not to produce a tidy majority view but to explain the relationship among observations, entries, transformations and published summaries.
This distinction also helps readers evaluate official assurances and public allegations symmetrically. Ten posts repeating one screenshot are not ten separate incidents. Ten official displays drawing from one database are not ten independent measurements. The same discipline should apply regardless of whether the repeated number supports the reader’s preferred conclusion.
10. Separate a documented departure from a proven consequence
California’s current posted recount regulations define a custody break in terms of an interval when control is uncertain and actions involving the materials are unaccounted for or unconfirmed. That definition concerns the evidence of control; it does not define every such interval as proof of altered votes. The same regulations identify relevant records and provide safeguards for examining materials. Their scope is California recounts, not every election process nationwide. California recount regulations, sections 20351 and 20353
For Juniper, consider three distinct findings. First, an applicable fictional instruction required a completed field, and the original record lacks it. Second, available independent evidence does not establish who possessed the specified container during a particular interval. Third, a comparison of authenticated records establishes that a stated quantity changed. These findings require different evidence and need not occur together. A missing field alone cannot supply the second or third finding.
Suppose the record deficiency is acknowledged but a separate contemporaneous account resolves possession during the interval. A report can state both facts without contradiction: the documentation requirement was not met in the specified form, while additional evidence supports a stated custody history. Whether that additional evidence satisfies an applicable legal standard is a question for the authorized decision-maker, not something this hypothetical article determines.
In another version, possession remains unresolved but independent quantity records agree. That agreement is relevant, yet it does not establish every property of the materials. In a third version, quantities differ but the evidence locates the difference in transcription. Here the documentary error is established without a demonstrated change in the physical quantity. Treating these versions as interchangeable would prevent an investigator from identifying what actually needs attention.
Consequences should therefore be expressed with the same care as causes. A statement that an issue affected a container may mean only that the container appears in a questioned record. A statement that it affected ballots may be read as saying the physical ballots changed. A statement that it affected the outcome makes a much larger claim. Reports should define the intended meaning rather than relying on a word that can slide between these propositions.
The practical value of this separation is not merely rhetorical restraint. It makes corrective actions more precise. A transcription problem calls for a documented correction and an examination of the transformation that introduced it. An unresolved account of possession calls for evidence about that interval. A substantiated physical discrepancy calls for an authorized investigation of that discrepancy and its implications. This article does not prescribe operational remedies or decide who has legal authority to apply them.
A broken seal, standing alone, is not proof that votes were altered. Nor is it a reason to stop asking questions. The defensible response preserves the observation, identifies the material and interval, and seeks the records needed to determine what the observation means. Certainty should grow from the evidence, not precede it.
11. A preserved record and a visible record are different things
The custody question can outlast the first public discussion. A useful investigation may depend on an original form, a later correction, a record of who produced a copy and contextual material explaining a timestamp. Keeping only the final spreadsheet can erase the path by which an initial contradiction was resolved. That would make future readers dependent on an assurance rather than the underlying explanation.
Washington’s Managing Election Records page, last updated February 2026, provides a specific classification lesson. For recordings of election operations that are created, it directs retention under election-administration records series rather than the generic security-monitoring, no-incident series. The page distinguishes federal and non-federal election categories. This is a Washington records-management example, not a universal retention deadline or permission to destroy any particular record. Washington Secretary of State, managing election records
Juniper’s fictional video index covers a three-hour review window, or 180 minutes. It lists 169 minutes of available recordings and an eleven-minute unavailable segment. The numerical coverage is approximately 93.89 percent. That is a measure of recorded time available to the reviewer. It is not 93.89 percent confidence in custody, and it cannot be multiplied by a ballot count to estimate the number of ballots with an uncertain history.
Why not? The significance of the eleven minutes depends on what happened during them and what other evidence exists. If no relevant transfer occurred then, the missing segment has a different relationship to the custody question than if the disputed event fell entirely within it. Conversely, continuous video might fail to show the identifier or action at issue. Time coverage alone does not describe evidentiary coverage.
A still image extracted from a recording introduces another boundary. It can depict a visible condition at one moment, subject to authentication and interpretation, but it does not contain the intervening movement. An index entry can show that a file was cataloged, but not that every frame was reviewed. A statement that video exists should not quietly become a statement that the video corroborates a specific event.
For this exercise, the analyst records which segment was actually examined, the relevant visible feature and the limitation. If a timestamp cannot be related reliably to the form’s clock, that remains an explicit uncertainty. If an object’s label is illegible, the analyst does not identify it by assuming that the expected container must be the one shown. These are limits on an inference, not evidence that something improper occurred off camera.
Preservation, access and interpretation are therefore three different tasks. A record can be retained but unavailable in the current production; available but unreadable; readable but irrelevant to the disputed proposition. Keeping those states separate prevents both an unsupported accusation of destruction and an unsupported claim that the existence of an archive settles the issue.
12. Write a finding another reader can actually test
The EAC’s July 30, 2026 explanation of election results, canvass and certification distinguishes election-night reporting from final official results. It describes the canvass as a process for checking and bringing together election data. A public dashboard’s completion label therefore should not be substituted for the underlying records or the status of that process. Custody documentation and a results webpage answer different questions. EAC results, canvass and certification
A final Juniper report can be concise without being simplistic. It might say that the initial production contained a dispatch for T-08 and two copies of its receipt, representing one claimed transfer of 240 sheets. It can identify the erroneous handwritten receipt reference, describe the register and dated correction subsequently supplied, and explain why the documentary link is now supported. It should also state that this reconciliation did not independently establish the physical contents of the container.
For the separate identifier issue, the report can distinguish the versions examined. In the version where the source image and export differed, the error was localized to transcription. In the version where no explanatory record was supplied, the identifier transition remained unresolved. Combining those alternative hypothetical outcomes into one narrative would create a false resolution. The article’s scenarios are alternatives designed to test reasoning, not successive discoveries in an actual investigation.
The inventory report can state that the stipulated terminal categories sum to 1,000 sheets and show the twelve-sheet transfer entry needed to complete that sum. It can explain that 638 issued sheets overlap with the cast and spoiled categories, and that 590 candidate selections are not the same unit as 620 cast sheets. These are concrete, reproducible findings. None requires inventing a candidate’s motive or speculating about workers’ intentions.
The exception summary can preserve both review dates: twelve deficient event records at first review, four still unresolved after additional contemporaneous copies were supplied. The report can say which fields were deficient and which evidence changed the classification. Readers then have a way to challenge the analysis: point to an omitted record, a mistaken unit, an unsupported join or an incorrect calculation. A slogan that everything was secure or everything was compromised offers much less to test.
This standard is demanding in both directions. Officials should not expect a procedural assurance to substitute for the records relevant to a specific concern. Critics should not expect an unanswered question to substitute for proof of their proposed answer. A publication can maintain an unresolved finding without turning it into a permanent insinuation; it can also report a documented explanation without pretending the original defect never mattered.
The ballot’s paper trail is valuable because it makes particular propositions examinable. Its gaps matter because they identify where evidence or compliance may be deficient. Its corrections matter because they can show how a discrepancy was resolved. Its limits matter because the history of possession, the accuracy of a count and the truth of an allegation are not interchangeable questions. Good scrutiny follows those distinctions all the way to a conclusion the available evidence can bear—and stops short of a conclusion it cannot.