1. The match is the beginning of the question
A description mentions water, a tall structure or an enclosed space. Someone later displays a photograph containing those features. The juxtaposition may be striking, but the evidentiary question is larger than whether the description can be made to fit the photograph. Which other photographs could have been selected? Who knew the intended target? When was the description fixed? What rule determined that the match counted? A convincing demonstration has to make those questions answerable, not merely display its most persuasive pair.
The CIA's October 27, 2021 historical account says the agency investigated whether people could describe distant locations and objects without being there, referring to this as remote viewing. That establishes an institutional research history, not a scientific finding of paranormal ability. This article does not infer efficacy from the existence of a government file, and it does not reconstruct the unavailable underlying historical evaluation from summaries. CIA historical account
Instead, it examines the structure of a claim through an invented exercise called Alder. Alder is not a laboratory, a group of participants or an actual remote-viewing session. Every target description, report, score and event in the exercise was authored for this article. A local script enumerates a small set of assignments and checks arithmetic. No person attempted to perceive a hidden object, and no artificial-intelligence system generated or judged the descriptions.
That limitation is useful. It allows the exercise to isolate what a scoring decision means without arguing about a particular person's honesty or a disputed historical anecdote. A result can become more impressive on paper because the comparison set changed, a tie was resolved favorably, or a score was defined after the target was revealed. Each of those possibilities concerns the design of the inference. None requires assuming that a participant intentionally misled anyone.
The central distinction is between a description that resembles a target and a procedure that gives such resemblance an interpretable evidentiary value. Those are not rival ways of describing the same thing. The second includes information about alternatives, timing, access and scoring that the first does not supply. A single attractive match can be part of that record, but it cannot stand in for the rest.
Readers therefore need a protocol they can follow backward from the reported result to the original materials. They should be able to identify the exact question tested and see whether the answer would have been recorded in the same way had it been less flattering. Alder will make that demand concrete with four invented targets and a scoring rule simple enough to inspect without specialist software.
2. Define the target before deciding what resembles it
Alder's four fictional targets are a fountain, a tower, an aquarium and a pavilion. These names are only labels for artificial feature sets, not pictures of real places. For the exercise, the fountain has the tags water and circular; the tower has vertical; the aquarium has water and enclosed; the pavilion has vertical, circular and enclosed. A feature absent from the stipulated set does not earn a point, regardless of what a real fountain or pavilion might usually look like.
That last qualification prevents ordinary knowledge from silently changing the task. A real fountain might contain a tall central column. A real aquarium might have circular tanks. Neither fact modifies Alder's definitions. The exercise is testing a declared tag comparison, not the reader's ability to imagine a version of an object that would improve a match. Once the labels have been translated into explicit sets, there is less room for that movement.
Alder also has four invented reports. Report one contains water and circular. Report two contains vertical and enclosed. Report three contains water and enclosed. Report four contains vertical and circular. The scoring rule awards one point for each reported tag present in a target's stipulated set. It does not subtract points for unreported target features or add credit for evocative language. This is an arbitrary toy rule, not a validated measure of remote-viewing accuracy.
The complete score row for report one is therefore two, zero, one and one in the order fountain, tower, aquarium and pavilion. Report two scores zero, one, one and two. Report three scores one, zero, two and one. Report four scores one, one, zero and two. All sixteen comparisons are available, not just the four that will later be called intended matches. A reader can reconstruct every number from the tags.
This full matrix changes what can be seen. Report two is compatible with the tower's vertical tag, but it matches the pavilion more strongly under the stated rule. If an announcement displays only report two beside the tower and calls it a clear success, it suppresses an available comparison. The issue is not whether the tower description contains something true. It is whether the announced conclusion follows from the scoring rule and alternatives that the exercise itself defined.
Alder's simplicity is deliberate. A real free-form drawing or narrative would demand more explanation about what counts as a feature and who interprets it. The toy cannot resolve those harder judgments. It shows why their answers belong in the protocol rather than being improvised after a particularly suggestive picture appears.
3. Blinding is a question about particular people and information
The APA's 2018 quantitative reporting standards ask researchers to report whether participants, people administering experimental manipulations and outcome assessors knew condition assignments. They also ask how masking was accomplished and whether its success was evaluated. This is more specific than attaching the label blinded to an entire study without identifying who lacked which information. The standards are general reporting guidance, not an endorsement of any remote-viewing claim. APA quantitative reporting standards
Consider an invented Alder event log. At 09:00, a coordinator receives a target label. At 09:15, a report is saved. At 09:20, an assessor receives the report and all four candidate descriptions. At 09:40, the assessor submits scores. At 10:00, the intended target label is revealed to the assessor. These times describe the fictional record. They do not by themselves prove that information was unavailable through every other channel.
Now add one log entry: at 09:25, the assessor opens a worksheet whose filename includes the target label. The original account that the formal reveal occurred at 10:00 remains true, but it no longer captures all the recorded access. The discrepancy matters because scoring was submitted after the earlier worksheet access. It does not prove that the assessor used the filename, and it certainly does not establish deliberate misconduct. It identifies a concrete ambiguity in the claimed information sequence.
A different discrepancy would have a different consequence. If the worksheet was opened at 10:15, after the recorded scores were submitted, that event would not explain how the earlier scores were produced unless some other part of the chronology were wrong. The same filename can therefore be relevant or irrelevant to a particular concern depending on when it was accessible. Replacing the sequence with a general statement that there was a labeled file would erase this distinction.
The coordinator's knowledge also needs its own description. The assessor may lack the target label while a person speaking with the report author knows it. Conversely, a coordinator may never interact with that author but may prepare materials later delivered to the assessor. The phrase everyone was blind is not a substitute for these relationships. Alder's record can list the roles and documented contacts without pretending that a single adjective captures the entire information flow.
For a public reader, the useful output is a narrow conclusion about what the records establish. The log may support that scores preceded a particular reveal. It may leave an earlier access unresolved. Either is more informative than asserting perfect blinding or declaring the whole result fraudulent on the basis of an unexplained file entry.
4. Candidate materials can change the question without changing the target
Suppose Alder's fountain description arrives as a polished page with a blue border, while the other three candidates are plain text. The report author also used a blue cover sheet. Even if nobody supplies the target's name, the presentation now contains a visible correspondence unrelated to the stated tag rule. This invented detail does not establish that any real experiment had such a cue. It shows why the candidate set includes its presentation, not only the nouns written on it.
The editor can inspect the actual materials instead of speculating about motives. Were all four candidates shown in the same format? Did they contain dates, labels or other annotations? Were any changes made between the version shown during scoring and the version later published? A claim that the candidates differed only in their content is a claim about those specific files. It can be checked more directly than a claim about what an assessor intended to notice.
Consider a separate Alder revision. The original candidate set contains fountain, tower, aquarium and pavilion. A later presentation replaces the aquarium with a desert, giving report one's water tag fewer alternative matches. The target fountain has not changed. The report has not changed. But the comparison has. Reporting the new display as if it were the original evaluation would obscure a decision that affects how selective the match appears.
The full score matrix makes that change visible. Under the original candidates, report one shares one tag with the aquarium. Under an invented desert definition with none of the report's tags, that entry becomes zero. The target's score remains two in both versions. A reader looking only at the target score would miss the changed contrast. A reader looking at the candidate version and all comparison scores would not.
A third revision changes the wording of the report instead of the candidates. Imagine replacing enclosed with enclosed or sheltered after seeing the pavilion. That is no longer the same report under the original tag dictionary. The new phrase could be useful material for a later exploratory analysis, but the old and new results need separate identities. An expanded interpretation cannot simultaneously be presented as the unchanged original response.
These examples do not prescribe a universal target library. They identify what must remain stable for a particular result to mean what its caption says. In Alder, the objects, tags, report wording and candidate versions together define the evaluated comparison. Keeping only the final target photograph would preserve too little of the exercise to determine which comparison actually produced the score.
5. A ranking and an assignment are different scoring tasks
Alder can be scored in at least two ways. In the first, each report independently selects its highest-scoring candidate, even if several reports favor the same target. In the second, the four reports must be assigned to the four targets with each target used exactly once. The data matrix is unchanged, but the allowed answers differ. Neither task should be substituted for the other after someone sees which produces the more attractive result.
Under independent selection, report one favors the fountain, report two the pavilion, report three the aquarium and report four the pavilion. The pavilion appears twice; the tower is never selected. If the invented intended order is fountain, tower, aquarium, pavilion, three reports select their intended target. That is three of four under this particular rule. It is not a claim that three people correctly perceived hidden objects: there are no people or perceptions in the fixture.
Under the one-use assignment rule, the maximum total score is seven. There are two assignments achieving that maximum. One is the stipulated intended order. The other exchanges the tower and pavilion for reports two and four while leaving reports one and three unchanged. The matrix therefore does not uniquely identify the intended mapping, even though that mapping is one of the best-scoring possibilities.
A headline saying the scoring recovered the exact target order would conceal this tie. The correct statement is that the intended order belongs to a set of two equally scoring best assignments. Selecting the intended member after consulting the answer key would be a separate decision, not something the original score uniquely determined. A tie rule could be declared in advance, but its application and consequences would still need to be shown.
This example also explains why the number of candidate comparisons is not the number of independent trials. Alder contains sixteen report-to-target score entries, but they are arranged around four reports and four shared targets. Treating the entries as sixteen unrelated opportunities would discard their shared structure. The available assignment set is constrained precisely because targets cannot be used repeatedly under the second rule.
A reader does not need to prefer independent ranking or one-use assignment to recognize the distinction. The publication needs to say which task was proposed, which was executed and how ambiguous outputs were handled. Otherwise, a single percentage can merge incompatible definitions of success and acquire a confidence that neither definition supports on its own.
6. State the allowed assignments before calculating surprise
NIST's experimental-design handbook describes randomization in terms of randomly assigning factor levels or run sequences to experimental units. The relevant principle for this exercise is that the allowed assignments matter. A design with fixed counts per category has a different set of possible sequences from unconstrained repeated selection. The handbook is general experimental-design guidance, not a remote-viewing test or evidence about a paranormal mechanism. NIST experimental-design handbook
For Alder's one-use rule, there are twenty-four possible assignments of four distinct targets to four reports. The local script enumerates all twenty-four rather than drawing a random sample. Two assignments have total score one, six have score three, eight have score four, six have score five and two have score seven. These frequencies add to twenty-four. There are no score-two or score-six assignments in this particular matrix.
The stipulated intended order scores seven. Two of twenty-four assignments score at least seven, a fraction of one twelfth, or about 8.33 percent. That fraction describes this explicitly defined enumeration. If one assumes all twenty-four assignments are equally likely and the fixed matrix is appropriate for that reference calculation, it is the upper-tail proportion for the chosen statistic. It is not a measured probability that remote viewing is real or false.
More importantly, Alder's reports were hand-authored with full knowledge of the invented targets. No random assignment was actually used to generate an experimental result. The enumeration is therefore a demonstration of how a reference set changes an interpretation, not a valid test of a human ability. Calling the output an experiment would misdescribe the origin of the data even if every arithmetic step were flawless.
If targets could repeat, four choices at each of four positions would produce 256 possible sequences, not twenty-four. That is a different hypothetical design. This article does not calculate its score distribution or substitute it for the one-use calculation. The difference in possible sequence counts is enough to show why a reference model cannot be inferred merely from the phrase four targets.
A clearly reported result would name the assignment rule, statistic, tie handling and reference assumptions together. These details determine what the numerical comparison asks. A small-looking number detached from them is not more objective than a visual match; it is simply harder to inspect. Alder's complete enumeration keeps the assumptions visible precisely because the exercise is small enough to list every allowed case.
7. Preregistration preserves a question rather than making it correct
The Center for Open Science describes preregistration as specifying a research plan before the relevant study or analysis and placing it in a registry. Its guidance distinguishes planned from unplanned work and calls for transparent disclosure of changes. The value is the ability to see what was decided when; registration alone does not establish that a scoring rule is sensible or that a claimed result follows. Center for Open Science preregistration guidance
Imagine an Alder planning record dated Monday that specifies one point per shared tag and the one-use assignment rule. On Tuesday, the intended target mapping becomes known. On Wednesday, an analyst proposes giving the enclosed tag three points instead of one. There may be a reasonable substantive argument for that alternative weighting. But Wednesday's proposal is not the weighting that Monday's record specified. Both analyses can be shown without calling them the same planned test.
The change is numerically consequential. With enclosed worth three points and the other tags still worth one, the intended mapping scores nine instead of seven. A superficial comparison might call that an improvement. But the best alternative assignment now scores eleven, and it is the order fountain, pavilion, aquarium, tower. The intended mapping is no longer tied for the maximum. Its larger raw total does not mean it has become more distinctive among the available alternatives.
Under the modified rule, four of the twenty-four assignments score at least nine. That upper-tail fraction is one sixth, about 16.67 percent, compared with one twelfth for the original rule's intended total. The comparison does not tell us which weighting is scientifically appropriate. It shows why raw scores from different definitions cannot be treated as interchangeable evidence. The new rule changed the whole score distribution, not merely the number attached to the favored mapping.
Alder's original and revised matrices should therefore remain separately identifiable. A report could explain the rationale for the alternative weight, present both results and propose a future exercise using the alternative prospectively. It would be misleading to display only the rise from seven to nine while suppressing the larger rise available to another mapping. The full candidate comparison is what gives either number its context.
Preregistration is helpful here because it keeps Monday from being rewritten as Wednesday. It does not prohibit thinking after results appear. Exploration may reveal an ambiguity worth studying. The important distinction is between developing that idea and claiming it survived a test that was fixed before the same observations suggested it. Alder's revised weighting is an illustration of this distinction, not an accusation about any actual investigator or registry entry.
8. A favorable statistic does not identify its explanation
The American Statistical Association's March 2016 statement explains that a p-value is not the probability that a hypothesis is true, does not measure an effect's size or importance, and should not substitute for broader scientific reasoning. Those cautions apply to a numerical result even when its calculation is exact. The number answers a question defined by its statistical model; it does not by itself identify why the observed pattern occurred. ASA statement announcement
Alder's exact enumeration is particularly useful for seeing this boundary because there is no hidden experimental cause to discover. Its apparent matches were created by the author. The one-twelfth fraction under the original rule is a property of a hand-built score matrix and a declared assignment set. It is not evidence that the author possesses unusual perception. A mathematically correct reference calculation cannot change the provenance of the observations on which it operates.
Now imagine a different fictional announcement that reports only the percentage of intended targets ranked first. In Alder's original independent-ranking task, that is three of four. The report could truthfully print 75 percent. It could also truthfully print that the one-use assignment task has two tied optimal mappings. Those statements describe different outputs. If the announcement presents the percentage as proof of uniquely recovering all four targets, it has combined the favorable part of one task with the stronger interpretation of another.
A useful evaluation can put both outputs beside their definitions. Three independent first-place matches do not mean that every report identifies its target uniquely within the joint assignment. A tied optimum does not mean every candidate is equally plausible under the scoring rule. The exercise supports neither an inflated success claim nor a claim that the scores contain no structure. Its structure is exactly what the full matrix displays.
There is also a difference between numerical uncertainty and missing procedural evidence. Suppose Alder's arithmetic is independently recomputed and matches perfectly, but the actual candidate version used by an assessor is absent. Repeating the arithmetic cannot reconstruct that missing version. A more elaborate confidence interval would not answer the archival question either. The unresolved issue is which inputs belong in the calculation, not how many digits the known calculation should display.
A public account can describe the result and its unresolved explanation separately. It might say that a fixed score exceeded many allowed alternative assignments while the provenance of one input remains unverified. That sentence is less dramatic than declaring a new capacity of the mind. It is also more faithful to the available evidence. A conclusion about a mechanism needs support beyond the fact that one chosen summary number looks favorable.
9. Review before results is different from approval of the result
Registered Reports add a publication-stage distinction. The Center for Open Science describes a format in which protocols receive peer review before outcomes are known, with qualifying manuscripts receiving in-principle acceptance that is not contingent on a favorable result. Continued acceptance still depends on the relevant quality and reporting conditions. This is not a promise that a registered study will establish its hypothesis or that every protocol bearing a registry identifier received such review. Registered Reports guidance
Imagine two versions of Alder's planned output form. Version A has a single box labeled best mapping. A reviewer notices that equal totals could produce multiple best mappings, so version B provides room for a set of assignments. The score rule has not changed. The revision changes how an ambiguous output will be represented. This is a concrete improvement that readers can inspect without accepting any conclusion about the eventual result.
Now suppose the final appendix follows version B and lists both optimal mappings, while the summary graphic still uses version A and displays only the intended one. The record contains its own inconsistency. The correction is to bring the graphic into agreement with the stated output rule and appendix, not to infer a hidden scientific disagreement from the mismatch. A publication can contain a correct calculation and an overstrong visual summary of that same calculation.
An editor can make the sequence legible with separate dates for protocol review, data collection, analysis submission and final publication. Those dates need not be identical across every research format. Their purpose is to establish what each document was capable of saying at the time it was produced. A pre-results review cannot have independently checked observations that did not yet exist. A later review may examine those observations while still leaving questions open.
Alder's hypothetical reviewers might object to the overlap between target features, ask for a clearer tie rule or request a better explanation of the intended interpretation. Such revisions would alter the proposed exercise before new data were obtained. They would not create findings in advance. If a revised protocol is eventually used, the result needs to point to that version rather than merely to the first plan carrying the same project name.
This distinction is useful beyond the strongest claims. A null or ambiguous result can still be informative if it answers a clear question and preserves its limits. Conversely, an exciting result can remain difficult to interpret if its protocol is unclear. The status of the publication process and the strength of the evidentiary conclusion are connected, but they are not synonyms. Keeping them separate prevents a familiar badge of credibility from doing more argumentative work than it can support.
10. Make the report reconstructible without exposing unnecessary personal information
The official TOP Guidelines policy page presents the TOP 2025 framework as a set of practices supporting verifiable research claims. It distinguishes disclosure, sharing and citation, and independent certification, and separates results transparency from computational verification. These are different activities: saying that data exist is not the same as supplying them, and supplying them is not the same as someone successfully checking the reported result. TOP Guidelines policy framework
For Alder, a reconstructible packet is small. It contains the four target feature sets, the four report feature sets, the rule assigning points, the stipulated intended mapping and the code enumerating permitted assignments. Each item has a clear purpose. Without the target sets, the score entries cannot be checked. Without the reports, the row meanings disappear. Without the assignment rule, twenty-four and 256 might be confused as alternative counts of the same experiment.
The original report text also matters. If a packet contains only extracted tags, a reader cannot inspect whether the extraction represented the wording fairly. Alder avoids that extra interpretive layer by defining its reports as tag sets from the beginning. That is a convenience of the toy, not a claim that free-form narratives can always be reduced without judgment. An actual study using a coding step would need to make that step understandable on its own terms.
A concrete comparison catches a common documentation error. Suppose a summary spreadsheet records report two's score against the aquarium as zero, while the declared original tags produce one. Recomputing the matrix exposes the mismatch. The correction does not require a debate about paranormal mechanisms; it requires fixing the inconsistent cell or explaining a different rule. Until that is resolved, the spreadsheet and stated method do not describe the same calculation.
One further arithmetic check concerns labels rather than scores. Reorder Alder's candidate columns to tower, fountain, aquarium and pavilion, leaving the underlying feature sets unchanged. A label-aware calculation of the intended fountain, tower, aquarium, pavilion mapping still totals seven. A careless calculation that reuses the old column positions instead scores tower, fountain, aquarium and pavilion against the four reports and obtains four. The data have merely been rearranged; the apparent change arose because positional indexes were mistaken for target identities. The local fixture checks both totals explicitly.
This is a useful correction to the idea that publishing a spreadsheet makes its result self-explanatory. A column heading, an identifier and a displayed order perform different jobs. If the article states a target order in names while the script expects numbered positions, the translation between them belongs in the reproducible record. The mistaken total of four would not be evidence against remote viewing any more than seven is evidence for it in this synthetic exercise. It would be evidence that the calculation evaluated a different assignment from the one its caption named.
Sharing a research record also need not mean publishing every personal detail associated with participants. In this exercise there are no participants, names, consent records or private narratives to disclose. A real packet would have additional responsibilities that this synthetic example does not resolve. The public-interest goal is enough information to evaluate the reported comparison while respecting applicable protections, not an indiscriminate dump of personal material.
The strongest documentation claim Alder can make is modest and testable: the supplied toy inputs reproduce the stated scores and assignment counts. The fixture cannot certify a laboratory's conduct, verify a historical session or authenticate a person's experience. Giving a small artifact an exact purpose makes it more useful than decorating it with a broad claim of transparency that readers cannot evaluate.
11. Repeating the arithmetic is not a new test of the claim
The National Academies' 2019 report distinguishes computational reproducibility using the same data and procedures from replicability across studies that obtain their own data. It also warns that reproducing a calculation does not guarantee that the calculation is correct, and that a single replication outcome does not settle an entire scientific claim. Those definitions help identify what a repeated evaluation actually adds. National Academies, Reproducibility and Replicability in Science
If a second person runs Alder's script and obtains the same twenty-four totals, that is a useful check of the supplied arithmetic. It is not twenty-four new trials, and it is not a second remote-viewing experiment. The reports, targets and intended mapping have not changed. Calling the rerun an independent confirmation of psychic functioning would be particularly misleading because the original fixture contains no observation of such functioning to confirm.
A separate stress check can nevertheless reveal something useful about the toy rule. Replace each of the four reports with the complete set of all four tags: water, circular, vertical and enclosed. Every report now scores two against the fountain, one against the tower, two against the aquarium and three against the pavilion. The rule rewards matches without penalizing extra descriptions, so the least selective report earns a larger raw score against the most extensively tagged candidate.
With four identical all-tag reports, every one-use assignment totals eight, because each assignment uses every target once. All twenty-four mappings tie. The intended mapping's total of eight is higher than its original seven, but it has no distinctiveness within this new matrix. This second executed check does not estimate how people would behave. It identifies a property of the declared toy score: adding every allowed descriptor can increase a total while eliminating the distinctions needed to identify any particular assignment. That is useful information to have before treating the rule as a measure of meaningful accuracy.
Agreement between assessors is another separate issue. Two assessors can produce the same score because they followed a clear rule, or because they shared the same interpretation of an ambiguous description. Their agreement alone does not reveal which explanation is correct. In Alder, the scores are deterministic tag intersections, so there is no real human-assessor disagreement to measure. The article does not report a reliability coefficient or invent a panel of judges.
A claim about broader circumstances needs evidence about those circumstances. Alder's outputs describe these invented tag sets and rules, not unrestricted perception. The all-tag check adds a diagnostic example, not a population of human observations. Nothing in either fixture supplies an empirical estimate of how anyone would perform in a genuinely new study.
12. Preserve the outcome that the protocol actually produced
Return to report two and the tower. The report contains vertical, and the tower's stipulated feature set contains vertical. That is a match. The same report contains enclosed, which makes its score against the pavilion higher under the original rule. Displaying the tower match alone conceals the stronger alternative. Nothing about this observation depends on believing or disbelieving remote viewing. It follows from the defined comparison that the exercise asks the reader to inspect.
The final Alder record should therefore retain several different outcomes rather than squeezing them into a single success label. The independent-ranking task has three intended targets ranked first. The one-use assignment task has two tied optimal mappings, one intended. The original exact reference set has two of twenty-four assignments at the intended total or above. The alternative weighting gives a larger intended score but a different, less distinctive position within its own assignment distribution.
All of those statements are true of the invented inputs. None is an observation of a paranormal ability. The example's provenance is decisive: its reports were authored to demonstrate evaluation issues, not collected from blinded participants. The complete scoring packet helps verify that limitation as well as the arithmetic. It would be contradictory to emphasize transparency while hiding how the example was created.
For an actual claim, a reader can look for the corresponding evidence rather than demand a particular conclusion in advance. The original responses, candidate alternatives, access chronology, declared scoring rule and complete results together describe what was tested. Missing elements can be named without filling them with assumptions about good faith, bad faith or extraordinary powers. Where a gap cannot be resolved, the conclusion should remain narrower than the gap allows.
A correct guess may deserve attention. It becomes interpretable scientific evidence only in relation to the procedure that generated, preserved and evaluated it. The purpose of a protocol is not to make a surprising answer less interesting. It is to make clear what would count as support, what would count against the claim and whether the reported result actually met those terms. That clarity is what allows a striking resemblance to become a question that other people can test.