Open but Unconnected: Can the Public Research Record Be Traced from Preregistration to Outcome?

1. Introduction

Research rarely consists of a single document. An empirical study may begin with a prospective plan, proceed through materials and data collection, generate analytic code and intermediate files, undergo changes in design or analysis, appear first as a working report or preprint, and later reach a final public form. Each object records a different part of the study. Taken together, these objects can reveal what was planned, what was produced, what changed, and what was eventually reported.

Open-science reform has increased attention to making these parts of research public. Preregistration provides a time-stamped statement of planned research decisions. Data sharing makes empirical inputs available for inspection or reuse. Code sharing may expose analytic procedures. Materials permit closer examination or replication of study procedures. Public reports communicate findings whether or not they later appear in a formally reviewed publication.

These practices are commonly evaluated separately. A study may therefore be classified as having open data, shared code, a registration, or a public report. Such classifications answer an important question: Is the object available?

A different question follows once several objects exist: Can an independent reader determine that these objects belong to the same research process and follow the relationship among them?

That question concerns the structure of the record rather than the mere existence of its parts.

The distinction has become especially timely as lifecycle open science places greater emphasis on making the path of research visible from planning through outputs and outcomes. Current lifecycle language describes research as a connected process and identifies a recurring problem: plans, outputs, and outcomes may exist in different locations without being adequately connected. It therefore calls for records that are not only accessible, but linked and findable over time (Center for Open Science, 2026).

The technical possibility of building such connections is well established. The FAIR principles distinguish accessibility from interoperability and include qualified references among related data and metadata. Persistent identifiers, rich metadata, and provenance information can help establish how one digital object relates to another (Wilkinson et al., 2016). Research-object frameworks go further by providing structured descriptions capable of linking data, software, people, workflows, outputs, and contextual information across distributed locations (Soiland-Reyes et al., 2022).

These developments establish that relationships among research objects can be represented. They do not establish how often ordinary public research records permit a reader to recover those relationships in practice.

That distinction matters because public availability can coexist with relational fragmentation. A preregistration may be public at one location. Its dataset may be stored elsewhere. Analysis scripts may occupy another location. The reported outcome may reside somewhere else again. Every object may be accessible. Nothing necessarily guarantees that a reader starting from one object can identify the others.

The resulting problem may be described as open but unconnected research.

The present study treats that condition as an empirical object.

1.1. Registration-Report Comparison Does Not Settle the Traceability Question

A large prior literature has compared registrations or protocols with later reports. Such studies often ask whether registered outcomes changed, whether planned analyses were omitted, whether new outcomes appeared, whether registrations were prospective, or whether deviations were disclosed.

A systematic review identified 89 articles assessing discrepancies between registrations or protocols and associated reports across more than 7,000 studies. Among studies assessing primary outcomes, discrepancies were common. The review also found substantial variation in how registration timing, registration versions, and discrepancy definitions were handled (TARG Meta-Research Group & Collaborators, 2023).

That literature establishes an important point for the present project: linking plans with reports is not itself a new research activity.

The present question lies one step earlier.

Before a registration and report can be compared, the correct pair must first be identified.

A comparison study generally begins once correspondence has been established. The present study asks how easily that correspondence can be established by someone who begins only with the public registration.

This difference changes the dependent variable. The object of measurement is not discrepancy between two already matched documents. It is successful reconstruction of the relationship between them.

1.2. An Accidental Demonstration of the Traceability Problem

Recent inception-cohort work illustrates why this distinction matters. Ensinck and Lakens (2025) began from public registrations and attempted to determine whether the registered studies had eventually been publicly shared. Their primary purpose was to estimate public sharing rather than measure traceability.

The search itself exposed a different problem.

Among 50 registrations initially classified as lacking a public report, subsequent author responses showed that 14, or 28%, actually had a public report that the investigators had failed to identify from the available information. Most mismatches occurred where no direct link existed and only partial terminology overlapped between the registration and possible report (Ensinck & Lakens, 2025).

Other registrations produced the opposite problem: an apparently matching report turned out not to belong to the registration. The investigators therefore used researcher contact to resolve uncertain classifications. They concluded that it was often difficult to determine public-sharing status from registration information alone.

This finding is especially useful because reconstruction failure was not the intended endpoint of that study. The investigators were attempting to answer whether studies had been publicly shared. Difficulty matching registrations and reports emerged as a methodological obstacle.

The present study turns that obstacle into the primary research question.

If investigators conducting a dedicated metaresearch project can fail to identify an existing public outcome, then the fact that both the registration and outcome are publicly available does not guarantee that their relationship is sufficiently represented for an independent reader.

1.3. Availability Is Not Connectedness

The first distinction required by the study is between artifact availability and connectedness.

Artifact availability concerns whether an object can be publicly accessed.

Connectedness concerns whether its relationship to another research object is explicitly represented.

The difference can be stated formally.

A_x = 1 when research object x is publicly available.

A_y = 1 when research object y is publicly available.

Nothing follows from A_x = A_y = 1 about whether a reader can determine that x and y belong to the same study.

A relational property is therefore required.

C_xy = 1 when the relationship between x and y is explicitly represented by a usable link,
persistent identifier, or sufficiently specific relational metadata.

A research system can therefore contain two publicly accessible objects while C_xy = 0. The practical result is fragmentation.

This distinction parallels a longstanding principle in data stewardship: resources should not remain informational silos. FAIR interoperability includes qualified references between related objects (Wilkinson et al., 2016). Research-object systems likewise show how distributed resources can be represented as parts of one larger research object (Soiland-Reyes et al., 2022).

The present study does not propose another technical metadata standard. It asks whether the relations needed by an ordinary research consumer are actually recoverable from existing public records.

1.4. From Connectedness to Traceability

Connectedness and traceability are related without being identical.

A record may contain an explicit relationship that is no longer usable because a link has broken. A persistent identifier may resolve to a page containing several candidate reports without indicating which one corresponds to the registration. A project page may contain a file whose title suggests that it is the final report while providing no secure relation to the particular registered study. Multiple registrations may belong to one project. One report may contain several studies. A later version may supersede an earlier one without the relationship being clear.

For this reason, the study defines research-lifecycle traceability as follows:

the extent to which an independent research consumer can correctly reconstruct the documented path from a prospective public research plan to the associated research outcome, and where applicable to related outputs, using the public research record.

This definition is intentionally reader-facing.

A record is not classified as traceable merely because a relationship exists somewhere in a database. The relevant question is whether an evaluator beginning from the designated starting record can successfully follow it.

Traceability therefore depends jointly on the existence of relationships and their usability.

1.5. Reconstructive Burden

A binary distinction between traceable and untraceable records still leaves an important difference hidden.

Suppose two registered studies both have publicly available outcomes. For the first, the registration contains a direct persistent link to the outcome. Identification takes seconds. For the second, the outcome can eventually be found only after searching the investigators’ names, comparing several similar titles, opening multiple candidate reports, and inferring correspondence from sample characteristics.

Calling both records simply findable collapses two very different research experiences.

The study therefore introduces reconstructive burden as a secondary construct.

Reconstructive burden refers to the observable work required to recover the relationship among research objects. It includes time, search actions, candidate reports inspected, ambiguous matches encountered, and reliance on inference or direct researcher contact.

This is not a measure of scientific quality. A hard-to-trace study may contain excellent research. An easily traced study may contain weak research. The construct concerns the usability of the public record for reconstructing research lineage.

1.6. Why Traceability Matters

Many transparency practices depend on the ability to connect research objects correctly.

A reader cannot efficiently compare planned and reported analyses without locating the relevant plan. A dataset cannot easily be interpreted in relation to a report when the study relationship is uncertain. Code cannot readily be assigned to a particular analysis when several projects or versions coexist without clear relations. Null findings remain difficult to identify when a registration does not indicate whether an outcome exists elsewhere. Amendments cannot be understood historically when versions are not linked.

Lifecycle open science therefore requires more than producing public objects. It requires enough relational information to establish which objects belong together. Current lifecycle guidance makes the same move from isolated open practices toward a connected record in which plans, outputs, and outcomes can be followed through time (Center for Open Science, 2026).

The core proposition of the present study is modest:

Public availability does not entail lifecycle traceability.

Whether this distinction produces a large practical problem is an empirical question.

The study is designed to answer it.

2. Research Questions and Analytical Commitments

2.1. Primary Research Question

Among prospectively registered studies with a confirmed qualifying public outcome:

What proportion permit an independent evaluator to correctly reach that outcome using only the starting registration record and directly connected public pathways?

The study is primarily estimative.

No arbitrary benchmark such as 50% traceability will be imposed. No prior evidence establishes a theoretically justified threshold separating acceptable from unacceptable traceability.

The primary result will therefore be a prevalence estimate with uncertainty intervals.

2.2. Secondary Research Questions

• At what point do plan-to-outcome paths most frequently fail?
• How often is linkage forward, backward, bidirectional, or absent?
• How often does a confirmed outcome require external searching?
• How much additional time and search activity does external reconstruction require?
• How often do direct research paths contain broken or unusable links?
• How often are persistent identifiers present?
• How often can data, materials, code, and other applicable outputs be securely associated with the same registered study?
• Which observable characteristics of the public record are associated with successful record-only reconstruction?
• Does the distinction between artifact availability and traceability persist under a second registration architecture?

2.3. Primary Estimand

Let N_C denote registrations for which a qualifying public outcome is independently confirmed.

Let N_T denote confirmed-outcome registrations for which both independent evaluators correctly identify that outcome during the record-only stage within the preregistered time limit.

The primary estimand is:

P_T = N_T / N_C

where P_T represents strict record-only end-to-end traceability. The accompanying confidence interval will quantify sampling uncertainty.

2.4. No-Outcome Records Are Not Traceability Failures

A registration may correspond to a study that never produced a qualifying public outcome. Such a case differs fundamentally from one in which a public outcome exists but cannot be reached from the registration.

Four states will therefore remain separate throughout data collection and analysis:

State	Meaning
Confirmed outcome, record-traceable	A qualifying public outcome exists and is correctly reached during record-only reconstruction.
Confirmed outcome, externally reconstructed	A qualifying outcome exists, although record-only reconstruction fails to identify it.
Confirmed no public outcome	Available verification supports the conclusion that no qualifying public outcome exists.
Outcome status unresolved	Available evidence cannot establish whether a qualifying public outcome exists.

Only the first two states enter the primary denominator. This rule prevents nonpublication from being counted as failure of research-record linkage.

3. Methods

3.1. Study Design

The study will use a preregistered retrospective cohort audit of public prospective research registrations.

The design separates measurement of reader-facing traceability from determination of ground truth.

Stage A asks what an evaluator can discover from the record itself. Stage B permits external reconstruction after Stage A decisions have been locked. Stage C resolves remaining uncertainty through independent adjudication and, where necessary and feasible, researcher verification.

This order is required because knowledge of the correct outcome could otherwise make the starting record appear easier to navigate retrospectively.

3.2. Main Cohort

The main sampling frame will consist of public prospective registrations created during calendar year 2021.

A mature historical cohort is preferred because sufficient time must have elapsed for many registered studies to complete and produce public outcomes. A 2021 cohort also represents a period in which preregistration, persistent identifiers, public repositories, and associated open-science practices were already established.

Eligible registrations will be selected through a reproducible random sampling procedure.

The target main cohort will contain 500 eligible registrations after exclusion screening.

At a proportion near .50, a sample of 500 produces an approximate 95% margin of sampling error of 4.4 percentage points for estimates using the entire cohort. Precision for the primary confirmed-outcome subgroup will depend on the proportion for which qualifying outcomes are established.

3.3. Eligibility Criteria

A registration will be eligible when it meets all of the following criteria:

• represents a substantive empirical research study;
• was publicly registered during the prespecified sampling period;
• presents a prospective research plan rather than a purely retrospective deposit;
• contains enough information to identify a discrete intended investigation;
• remains publicly accessible during coding.

Exclusion criteria will include:

• obvious test registrations;
• empty or substantially noninformative records;
• teaching exercises without an intended research outcome;
• duplicate registrations of the same study;
• deposits created only after completion of the reported research;
• records that do not describe empirical research;
• records whose status as a distinct study cannot be established.

Duplicate and overlapping registrations will be adjudicated before Stage A. The rule used to select or combine such records will be fixed during pilot coding and frozen before confirmatory data collection.

3.4. Definition of a Qualifying Public Outcome

A qualifying public outcome is a stable public research report that communicates substantive findings from the registered study in enough detail to establish correspondence.

Eligible forms may include research articles, preprints, doctoral theses, research reports, proceedings papers, conference papers containing substantive results, and research posters containing enough methodological and outcome information to establish study identity.

The definition concerns public communication of findings, not formal publication status. This follows prior inception-cohort reasoning that a study may be publicly shared through several stable forms rather than through one publication route alone (Ensinck & Lakens, 2025).

A title, abstract notice, project description, protocol-only document, or announcement without substantive findings will not qualify as an outcome.

3.5. Pilot Phase

A pilot sample drawn outside the confirmatory cohort will be used to refine the coding manual.

Pilot testing will address duplicate-registration rules, multi-study registrations, multi-study reports, version ambiguity, linked project pages, broken identifiers, borderline outcome types, timing feasibility, coder agreement, and artifact-applicability rules.

No pilot record will enter the confirmatory dataset.

Any coding modification arising from pilot work must be incorporated before the final protocol is registered. Rules will not be altered after confirmatory outcome coding begins unless an amendment is publicly documented and affected analyses are separated.

3.6. Stage A: Record-Only Reconstruction

Two evaluators will independently begin from each sampled registration.

The task is to identify the public outcome associated with the registration using only the starting research record.

Permitted actions include reading the registration, following hyperlinks contained in the registration, following links to directly associated public project pages, opening files or resources reachable through those directly connected pages, and following persistent identifiers exposed within the permitted record pathway.

Prohibited actions include general web search, scholarly search engines, searching investigator names externally, searching registration titles externally, searching distinctive phrases externally, searching bibliographic databases, searching researcher profiles not reached through an allowed record path, and contacting researchers.

Each evaluator will have a maximum of ten minutes per registration for the primary Stage A task. The ten-minute rule creates a standardized upper bound on ordinary reconstructability. Exact elapsed time will still be recorded to permit sensitivity analyses.

3.7. Stage A Coding

For each record, evaluators will record whether a candidate outcome was reached; title or identifier of the candidate outcome; elapsed time; number of navigation steps; presence of a forward link; presence of persistent identifiers; presence of data, materials, and analytic code; presence of amendment or version information; broken links encountered; ambiguous targets encountered; and confidence in the candidate match.

Evaluators will not know whether an outcome has already been found externally. Evaluator decisions will be locked before Stage B information becomes available.

3.8. Strict Primary Traceability Criterion

A confirmed-outcome registration will satisfy the strict primary endpoint only when all of the following conditions are met:

• the Stage A evaluator begins from the designated registration;
• all navigation remains within permitted record-contained pathways;
• the correct qualifying outcome is reached within ten minutes;
• no external search is used;
• both independent evaluators independently identify the same correct outcome.

This outcome will be coded TRACEABLE = 1. All other confirmed-outcome registrations will receive TRACEABLE = 0 for the strict primary analysis. A secondary lenient analysis will count successful identification by at least one evaluator.

3.9. Stage B: External Reconstruction

Stage B begins only after Stage A decisions are locked.

External search will then be permitted for records without a securely established qualifying outcome.

A standardized search sequence will use the exact registration title, shortened or distinctive title phrases, investigator names, combinations of investigator names and study topic, distinctive intervention or population terms, investigator research profiles, public bibliographic search, and other stable public sources.

Search actions will be logged in order.

For every candidate report, correspondence with the registration will be assessed using investigator identity, research question, sample, population, intervention or exposure, measures, experimental conditions, recruitment period, study dates, hypotheses, sample size, and distinctive methodological details.

Topic similarity alone will not establish correspondence.

3.10. Stage C: Ground-Truth Determination

Where Stage B leaves outcome status uncertain, additional verification will be used.

Ground-truth evidence will be ranked approximately as follows: explicit reciprocal registration-outcome identification; explicit direct linkage from one object to the other plus matching study characteristics; matching persistent identifiers or structured relational metadata; direct confirmation from an investigator; and independent adjudication based on highly specific correspondence.

No single weak similarity will suffice.

Where public evidence remains insufficient, a standardized researcher-verification message may ask whether the registered study generated a qualifying public outcome and request a stable identifier if one exists. One initial request and one follow-up will be permitted.

Nonresponse will not be interpreted as evidence that no outcome exists. Such records will remain unresolved unless independent evidence establishes their status.

3.11. Link-Direction Coding

For confirmed registration-outcome pairs, explicit linkage will be classified as follows:

• L0 - No explicit linkage: neither object provides a usable explicit relationship to the other.
• L1 - Forward only: the registration or directly associated starting record links to the outcome.
• L2 - Backward only: the outcome identifies the registration, although the registration does not provide a usable path to the outcome.
• L3 - Bidirectional: registration and outcome explicitly identify one another.

This classification is descriptive. Bidirectionality is not assumed to be required for all research systems.

3.12. Broken-Path Taxonomy

Where Stage A reconstruction fails, the first substantive failure point will be classified.

Code	Definition
B1 - No forward relation represented	No usable route from the starting record toward an outcome is present.
B2 - Broken link	A represented relationship cannot be reached.
B3 - Ambiguous destination	A link reaches a location containing several plausible objects without enough information to determine the correct target.
B4 - Identifier failure	An identifier is missing, malformed, obsolete, or unusable.
B5 - Version ambiguity	Multiple registrations, project versions, reports, or study versions prevent secure reconstruction.
B6 - Unlinked external outcome	The qualifying outcome exists publicly elsewhere, although no usable route is present from the starting record.
B7 - Inferential match only	Correspondence can be inferred from titles, authors, methods, dates, or sample characteristics without an explicit relationship.
B8 - Other	A written explanation will be required.

The taxonomy will be tested during the pilot and frozen before confirmatory coding.

3.13. Associated Research Artifacts

The study will separately examine whether applicable research outputs can be associated with the registered study.

Artifact classes will include data, materials, analytic code, protocols, amendments, supplementary files, and later versions.

Each artifact will be coded for public availability, explicit association with the study, reachability through the permitted lifecycle path, presence of a persistent identifier, and link functionality.

The absence of an artifact will not be treated as a traceability failure when the artifact is not applicable or was never expected for the study.

3.14. Reconstructive Burden

For confirmed-outcome records, reconstructive burden will be measured through Stage A elapsed time, Stage B elapsed time, number of navigation actions, number of external searches, number of candidate outcomes inspected, number of ambiguous candidate matches, need for author-name searching, need for textual inference, need for methodological matching, and need for researcher verification.

The burden construct will remain descriptive unless later validation supports a defensible composite measure.

No unvalidated summed traceability score will be introduced.

3.15. Inter-Rater Reliability

Stage A coding will be conducted independently.

Agreement will be assessed for strict traceability classification, candidate-outcome identity, link direction, broken-path category, artifact presence, and artifact association.

Raw percentage agreement will be reported. A chance-corrected agreement statistic will also be reported where the structure and prevalence of the coding variable make such a statistic appropriate.

Original coder decisions will remain preserved after adjudication.

3.16. Primary Statistical Analysis

The primary estimate will be P_T = N_T / N_C with a 95% confidence interval.

The denominator will contain only registrations with independently confirmed qualifying public outcomes.

No primary null-hypothesis test against an arbitrary percentage will be performed. The analysis therefore asks how common record-only traceability is rather than whether it differs from an unsupported benchmark.

3.17. Secondary Analyses

Secondary estimates will include the proportion of confirmed outcomes requiring external reconstruction; proportion with forward, backward, and bidirectional linkage; prevalence of persistent identifiers; prevalence of broken links; prevalence of unresolved outcome status; median Stage A reconstruction time; median external reconstruction time; distribution of broken-path categories; and proportion of applicable associated artifacts that can be securely identified.

Reconstruction-time variables are expected to be right-skewed and will therefore be summarized using medians, interquartile ranges, and distributional displays.

Sensitivity analyses will repeat the record-only success calculation at shorter time thresholds, including five minutes.

3.18. Exploratory Predictors of Traceability

A preregistered exploratory model may estimate associations between Stage A success and observable record characteristics.

Candidate predictors may include presence of a persistent identifier, explicit forward linkage, number of directly associated research objects, existence of a linked public project page, registration type, record update history, field of research where reliably classifiable, and number of project versions.

Such associations will not establish causation. The descriptive primary estimate remains the central result.

3.19. External Validation

The main cohort provides a system-specific prevalence estimate. It cannot by itself establish the prevalence of traceability across all research infrastructures.

A smaller independent validation cohort will therefore be drawn from a second public prospective-registration architecture with stable public records and enough historical maturity for outcome ascertainment.

The target validation cohort will contain 100 eligible registrations.

The purpose is not to rank systems. Different registration systems may differ in discipline, user population, required metadata, workflow, technical design, and expectations regarding outcome reporting.

The validation question is narrower: Does the distinction between artifact availability and plan-to-outcome traceability remain observable when the record architecture changes?

Main and validation cohorts will not be automatically pooled.

3.20. Missing and Unresolved Outcome Status

Records whose outcome status cannot be established will be excluded from the primary denominator.

The number and proportion of unresolved records will be reported transparently.

Sensitivity analyses may calculate extreme bounds by assuming that all unresolved cases with eventual public outcomes would have been Stage A traceable or, alternatively, Stage A untraceable. Such bounds will show how much unresolved status could alter the main estimate.

4. Contribution and Scope

The study makes four limited claims.

First, artifact availability and relational connectedness are not equivalent.

Second, connectedness and reader-level traceability are related without being identical.

Third, the difference between what can be followed directly from a research record and what can eventually be discovered through external searching can be measured empirically.

Fourth, the additional work required when direct reconstruction fails can be described as reconstructive burden.

The study does not claim that research traceability, provenance, persistent identifiers, linked metadata, research objects, or registration-report comparison are new ideas. FAIR already requires qualified references among related objects, and structured research-object approaches already provide methods for connecting distributed resources and their provenance (Wilkinson et al., 2016; Soiland-Reyes et al., 2022). Prior metaresearch has also compared thousands of registrations with later reports (TARG Meta-Research Group & Collaborators, 2023).

The narrower contribution is empirical: the study treats correct plan-to-outcome reconstruction from the public record itself as the outcome to be measured.

Nor does traceability serve as a proxy for scientific quality.

The study does not determine whether a preregistration was well designed, an analysis was correct, conclusions were warranted, code executes successfully, data are reusable, researchers complied with the registered plan, deviations were justified, or an unlinked record reflects intentional opacity.

A directly traceable study may still contain serious methodological problems. A poorly traceable study may still contain excellent research.

The concern is narrower: Can the documented pieces be correctly followed as parts of the same research process?

This boundary is required because lifecycle traceability concerns the architecture and usability of the public record rather than the epistemic quality of the research itself.

The paper therefore begins from a simple distinction and leaves its prevalence open to empirical testing:

Research can be open in pieces without being open as a followable lifecycle.

5. Research Integrity and Reporting Commitments

No empirical results are reported in this manuscript version. The confirmatory coding rules, primary estimand, exclusion logic, and distinction between outcome absence and traceability failure are specified before confirmatory data collection.

Pilot-driven changes will be made only before the confirmatory protocol is frozen. Any later amendment will be dated, justified, and separated from the original confirmatory plan.

The analysis will preserve raw evaluator decisions before adjudication and will report unresolved records rather than forcing uncertain cases into binary classifications.

Interpretation will remain limited to the navigability and relational structure of the public research record.