1. What sort of thing is OS-ERIN
The best explanation is the one that fits your circumstances. Looking at where we want to be…
- For a researcher, it is a responsible way to make source checking easier.
- For a librarian, it is the claim-level graph that bibliometrics tries to proxy with record level data.
- For a publisher, it is distributed tending that maintains the integrity of the academic record
- For a meta-scientist, it collects data used for real-time study of the evolution of scientific thought
- For a student, it is where we learn that science is more a conversation rather than a bunch of truths
- For open science, it is to conversation what the printing press was to articles.
Like many apparently complex systems, these are legitimate perspectives on what emerges from the combination of simple routines and their objects. In the case of OS-ERIN the routine is one instance of a single reader responsibly checking the integrity of a single occurrence of a citation in one version of a record and the object is their assessment of that relation. The instance matters since an individual’s attention and competence change over time. Their identity matters because different individuals may legitimately have different views on the same relation. Responsibility matters since motivation will shape practice. The occurrence matters because the same words may do quite different work in a different part of the same argument. And finally, the version matters because the content examined may meaningfully shift between versions of the same record. Unlike bibliographic databases that may justifiably ‘correct’ metadata they hold, the object that OS-ERIN captures is human-authored stance with respect to an author-proposed relation. There is no ‘true value’ to be found in this stance at either an individual or aggregate level. Rather than hoping for convergence on a true value, as we might with a DOI, the ‘stance’ that OS-ERIN captures is better thought of as a trace of the intersubjectively negotiated norms that delimit scientific communities. Given that this stance derives from judgement that shifts across disciplines and over time, it will never be reduced to a colour-coded line between two documents, and it certainly will never support timeless assessments of entailment status. The assessment left by a reader is a community-conditioned, provenance-bearing, context-bound inspectable assertion of relation.
That object has strict limits. It does not assert truth, establish the quality of either work as a whole, exhaust every evidential route, settle authorial intent, convert the reviewer into an adjudicating authority and it certainly does not demonstrate that the reasoning was independent, competent or untouched by undeclared framing or interests. It is a record of the exercise of fallible reason. Human acceptance of a relation in OS-ERIN creates no more than an occasionally annotated, provenance-bearing stance around which the extended conversation that is science may accumulate.
Like the object, the process by which OS-ERIN supports readers also imposes limits. While OS-ERIN itself never asserts the status of a relation, this does not mean that the support it provides is neutral. What it retrieves, extracts, ranks, hides, labels and permits people to revise certainly shapes attention even when the final judgment always remains human-authored.
The wider argument presented in this Field Guide with respect to OS-ERIN builds from the consequences of recurrent accumulation of the primary scholarly object it enables. One elegantly facilitated relation-level inspection is a private event. If such events are accumulated and shared, later readers may encounter prior reasons and qualifications. Learners and reviewers may work differently because the practice exists. Dense plural trace may require derived views. Changed versions of the records create maintenance obligations. Visible contribution may attract recognition, measurement and control. Institutions may become dependent on the resulting service for examining both the record and their staff who use it. Each stage changes the conditions under which the next stage operates, and those changed conditions alter the trace the system later receives. In that processual sense, the underlying model is ecological: readers, learners, reviewers, technical systems, retained records, incentives, institutions and governance arrangements affect one another over time.
Ecological does not mean harmonious, self-correcting or destined to improve. Recurrent relation-level inspection might support more inspectable reasoning, revisable scholarly memory, transferable capability, distributed maintenance, contextual expertise and shared stewardship. The same infrastructure might support inherited authority, cognitive surrender, surveillance, compulsory measurable labour, portable reputation, proprietary training material and service dependence that is difficult to reverse. The programme-level question is not whether a preferred future can be described attractively. It is which mechanisms, designs, thresholds and institutional arrangements nudge the ecology away from demonstrably toxic outcomes.
Generative AI is not the source of the problems to which OS-ERIN responds nor are its effects necessary to justify our efforts. Scholarly verification was costly and most often lost before plausible text became cheaper and faster to produce. The sharp tilt in the production–verification asymmetry that cheap fluency has produced may increase pressure on readers and institutions, but that is a shift in degree rather than kind. As such OS-ERIN should not be cast as a response to the predations of GenAI. It leverages some of the technologies out of which GenAI emerged to enable and share the quiet acts of care on which the integrity of our scientific record depends.
There may be many reasons for quite a diversity of institutions to be interested in a dense provenance-bearing claim-level graph of the academic record that, by its very constitution, both evolves with human understanding and tracks human interest. The extensions from the base scholarly object produced by OS-ERIN into institutional forms are therefore neither a fantasy nor a promise. It is an imperfect possibility that, to our minds, borders on necessity. Responsible reading is foundational to science; its repetition can create shared memory; this shared memory creates maintenance and stewardship burdens; density creates projection and technical-selection problems; consequential visibility changes conduct; dependence creates questions of custody, funding and succession. To test whether that widening argument remains answerable to its origin, the next section follows one relation through the system before any larger claims are made.
2. One relation through the system
Let’s start with reading a manuscript that contains a sentence that matters for its conclusions: in a specified population and setting, a particular intervention reduced a particular outcome. To back this assertion the author cites a report. This example does not go as far as testing whether the reader finds the claim as made in the cited source to be either true or relevant to the citing manuscript. All this example traces is the simpler task of determining what relation the cited report bears to the occurrence of the claim found in this version of the citing manuscript.
The very first thing that must be done is to make the cited report available in a form the reader can inspect. Once retrieved, a technical process may identify candidate passages that may be relevant to the claim. One passage may state the reported result. Another may define the population. A third may qualify the finding by time period, comparison group or measurement choice. Finally, there may be relevant information present in a graph that the technical means used to scan the cited document entirely misses. While the technical process may place the passages found together near the top of a ranked list, their position there neither makes them evidence nor confers epistemic rank. All the technical process has done is surface detectable candidates whose superficial features match the models’ expectations. The reader must be able to determine which extracts are worth following, move easily into the full source, search it, inspect surrounding argument and graphs, and nominate portions the ranking did not expose. The ability to identify and recognise out-of-pool routes matters because a candidate system will fail silently. “Not shown” cannot mean “does not exist.”
After inspecting the source through the likely routes presented as a candidate pool, the reader records a deliberate interpretation of the citing relation across any number of quotations found relevant in the cited source. The interpretation might say that the source supports a narrower claim than the manuscript makes, or that it supports the claim only under a stated qualification, or that the cited passage does not bear on the claim in the asserted way. The annotation created by the reader properly includes the scope of that conclusion, the passages and reasons on which it rests, and uncertainty about matters the reader did not resolve. The object is tied to the claim occurrence found in the citing record and version of the source examined and it is relational. Other uses of the same source in the same way may support very different conclusions, just as much as a differently trained reader examining the same relation may legitimately conclude differently. The one relation assessed by a reader does not pronounce on every use of the source, every possible support path, every possible assessment metric, or the manuscript as a whole.
If the reader has chosen to authorise sharing of their assessments, the relation can remain attributable and revisable. Authorisation is plural. Keeping an interpretation inside a reader’s private workspace is not the same as making it visible to a collaborator. Visibility to an editor is not the same as public release. Public release is not permission for model development, nor does it automatically provide grist for an individual’s performance review. Insofar as known audience and future use shapes current behaviour, preserving such separations is part of the object’s meaning. It is not administrative detail added later.
Now imagine that a second reader encounters the same claim–source relation. Showing the earlier interpretation immediately might save work and expose a qualification the second reader would otherwise miss. It might also trigger cognitive surrender that anchors the second reader’s reconstruction. A governed reveal order could ask the second reader to inspect first and reveal prior trace later, or could vary by task, expertise and stakes. The example does not settle which order is best. It shows why the timing and nature of revealed prior judgment become a research question as soon as the bounded event of a reader instantiating a relation survives its original use.
The points to which a relation is attached, the citing and the cited, may later change. A new source version may correct a table. A reader may offer a better interpretation. An editor may issue a formal notice. These events are distinct. A concern is not a finding; an author response is not an editorial decision; a recorded change is not proof that downstream readers became aware of it. The system may nominate this relation and some dependent relations for reconsideration given the change in circumstances, but graph connectivity cannot determine the meaning of shifts in the circumstances abutting a relation. When such circumstances change, the appropriate response is an invitation to inspect.
Five nearby objects help mark the limits of the example. A human-inspected claim–source relation contains an attributable interpretation and its reasons. A ranked passage candidate presented when a reader follows a citation is a machine-selected object that still awaits inspection. Neighbourhood exploration trades on proxies such as topical similarity and bibliographic intersection, which have far lower epistemic status than author-declared dependence. The records it puts forward are machine-formed, speculative suggestions that a set of uncited works may be worth inspection; they establish neither the relevance of the nominated sources nor that the set is representative of anything. A projection onto an author’s declared citation, derived from accumulated reader stances on that relation, necessarily compresses diverse paths and considerations into an interpretable representation that loses information. Finally, the dense traces left by many diverse readers may be useful for model development, but they are not, by either intent or consent, a purpose-built model-development dataset.
What OS-ERIN makes possible with the first step it now takes is modest. Reading carefully may be made easier and, when we read carefully, what we find need not be lost. As demonstrated above, a great deal that is interesting may be built from this humble object. Equally interesting deformative possibilities also begin at the same point. An individual’s retained interpretation may be accepted as a verdict, collapsing scientific community into authority; the ordered set of passages nominated for consideration may be accepted as exhaustive of the evidence field, cementing machine inadequacies; a human click made in a hurry may be taken for evidence of independent considered reasoning, and the awe-inspired by a tidy workflow’s presentation of a recursive branching citation tree may be read as covering all relevant scholarly relations. The example presented above, which stopped short of validity or transferability, keeps the foundational unit visible while exploring implications of the programme, but it cannot prove that elaboration around this object will proceed or that the shared memory it affords will be beneficial.
At core, once the considered interpretation that arises from careful reading outlives its immediate task, the problem has changed. The question is what kind of scholarly memory prior trace becomes: what later readers inherit from it, and whether preserving that inheritance—despite the risk that it displaces engagement, deskills readers and impoverishes the record—is better than the uncontrolled alternatives towards which we seem to be stumbling.
3. Reading, scholarly memory and independent reconstruction
Scholarly memory already exists. It lives in individual recollection, marginal notes, correspondence, research groups, seminars, editorial conversations and disciplinary habits. A close-knit group may remember why a source mattered, why an interpretation was rejected or which caveat makes a familiar claim defensible. Much of that knowledge does not travel with the published citation. When people move, retire, change projects or leave a field, the reasoning may become inaccessible even though the paper and citation remain.
The case for preservation of this ephemeral work is easy to state. A later reader may avoid repeating every step of an earlier search. Prior trace may point to a qualification, expose a discarded interpretation or show that disagreement already exists. A team may recover why it treated a source cautiously. A community may retain reasoning that would otherwise remain inside local networks. An outsider may see annotations that strengthen their ability to understand reasoning internal to the community within which the inquiry was undertaken. In these forms, durable trace might distribute scholarly memory more widely and direct scarce attention toward unresolved relations rather than routine rediscovery.
The same mechanism creates the danger. Earlier judgment becomes influential precisely because it is present. A later reader may follow the path already made visible, inherit its confidence and fail to reconstruct alternatives. Context may be compressed until a situated conclusion appears portable so effort is diverted from the context-conditioned re-discovery that would have been more appropriate. A once-defensible interpretation may persist after the source, task or field has changed. Old trace may be easier to retrieve than better reasoning that was never recorded. What looks like memory may therefore become universal authority by exposure rather than by contextually appropriate evidence.
Reveal order is one part of this problem. Showing prior trace after an independent attempt may preserve more reconstruction, but it introduces friction that reduces the immediate apparent value of the tools provided by OS-ERIN—reducing adoption. Showing it first may help a novice notice relevant distinctions, or it may teach the novice to approve an inherited view. Expertise does not supply a universal answer: experts may use prior work productively in one task and be anchored by it in another. The appropriate sequence may depend on stakes, relation type, trace quality, similarity of tasks and the possibility of recovering an independent path.
Preservation also requires deciding what should travel with a judgment. The source and claim versions, the reader’s role, the passages examined, the scope of the conclusion, the rationale, uncertainty and later revisions may all matter. Too little context can make an old stance look timeless. Too much can make records difficult to maintain and reuse, and may increase privacy, confidentiality and rights burdens.
The reduced friction promised may not result in reduced work. Automated retrieval, organisational support and filtering of cited records may free attention for the high-value tasks of interpretation. This same architecture may create expectations that add time and make possible performance monitoring regimes that both distort and slow work. Verification of a machine-selected candidate can be more demanding than direct search when the selection is misleading. The relevant comparison when thinking about the contribution of OS-ERIN is not whether one step becomes faster but which burdens are removed, which are introduced and whether useful scrutiny improves at a defensible net cost.
Repeated reuse makes the ecology recursive. What later readers see affects what they inspect and record; those records become the prior trace seen by the next readers. A constructive branch would preserve independent reconstruction while allowing earlier reasons to remain contestable and useful. A deformative branch would accumulate path dependence, stale confidence and increasingly self-reinforcing selection.
At this point we see four immediate but distinct questions:
- When should a reader see prior human judgment?
- Does the enabled workflow improve source inspection once new burdens are included?
- Which kinds of deliberate judgment remain valuable enough to preserve and under which conditions are they most appropriate to share?
- What architecture would improve the likelihood that sufficient context is retained to keep a judgment intelligible without pretending it travels unchanged?
4. Peer review, editorial authority and confidential trace
Peer review is an obvious context for OS-ERIN but the traces made by reviewers should, perhaps, not be shared. The texts reviewers examine, pre-publication submissions, may differ significantly from those published and the purpose, audience, authority structure and confidentiality that surrounds peer review differ from those that surround normal exploration in ways that shape practice. Starting with purpose and audience, the traces left by a reviewer while inspecting a manuscript may make their reasoning more inspectable, helping an editor—and perhaps a later reviewer—reconstruct why an issue was initially raised and interpret the adequacy of a revision. Shifting now to authority structure, those who read published articles carefully are not deciding whether a text merits publication. This greater authority increases sensitivity to biases introduced by the combination of technical features of OS-ERIN and their recognition. OS-ERIN presents text fragments ranked by their machine-determined similarity to the citing passage. While machine and human ranking are hopefully close enough for the former to be a useful filter, their ordering and scope are prone to patterns of error that a human would recognise as arbitrary. The consequence of unwarranted deference to machine-supplied bias in peer review, therefore, is far more severe than it would be after publication. As such, the ranking given a passage found in a cited work can help a reviewer locate material, but that nominated fragment is perhaps an efficient entry point into the cited work—it is never a finding. A neighbourhood suggestion can make an uncited work worth inspecting, but that discovery does not imply that the author should have cited it, much less that the omission was deliberate. Machine assistance may improve efficiency, issue detection and traceability. It may also narrow attention toward claims that the interface highlights, focus attention on the words that are present in convenient excerpts rather than allow identification of silences made visible by context, and the capabilities it affords may create new ritual burdens of transparency whose discharge has no real correlation with quality. These and other possibilities are competing possibilities that must be tested before this apparently obvious context for the use of OS-ERIN can be accepted as appropriate.
Unlike work in the published record, in review there is editorial authority. Reviewers accept and respond to their tasks for reasons that are particular to review. They may feel performative or relational obligations to an editor that they do not when working in the public record—and these may shape their conduct. When a reviewer raises a concern, they write for a known audience—and an editor is wont to respond: they may request a response, make a publication decision or initiate another process. In peer review, the editor is the face of an institution—and that institution, the journal, holds authority. The technologies found in OS-ERIN are able to make suggestions, to preserve distinctions and to connect objects. These suggestions must always be embedded in structure that does not allow uncritical acceptance to transform machine nomination into ostensibly deliberative human judgement. The requirement that the technological support provided not adjudicate is particularly important where a structured record might otherwise acquire institutional weight through technical form alone.
Within OS-ERIN confidentiality is far more than a binary function. Trace may be retained for the immediate editorial process, visible to an editor but not an author, released later with consent, or never reused outside the original task. Each transition requires authority. Origin in peer review does not authorise public release, model development or use in a dossier about the reviewer or author. Finally, the fact that a record might be discovered to be valuable later cannot supply permission retroactively.
Looking to functional outcomes, use of OS-ERIN in review may capture process such that selective review reasoning remains useful beyond its instantiation. It could support editorial reconstruction, reduce avoidable repetition and connect entry review with later maintenance of the published record. The malignant branch starts from recognition that preservation may change what reviewers attend to or are willing to say, add ritual documentation, expose dependent participants or later become evidence about persons. The value of trace beyond the review cannot be assessed independently of its effects on the conditions under which review is produced.
Pre-publication review and continuing maintenance are complementary but not equivalent. Review is concentrated, episodic and situated within editorial authority. Later tending is diffuse, time-bearing and may involve different actors and institutions. They may exchange bounded information without becoming one permanent review regime. Generic questions of changed status, propagation and correction belong to ongoing tending; identity and dependence belong to attribution and power; deliberate use of review trace in decisions about people belongs to reward and recognition.
5. Education, transfer and genuine participation
OS-ERIN’s standard for its educational impact is simple: success means that those who have used OS-ERIN perform better on unfamiliar evidence problems when the platform, prior trace and machine assistance are absent. Competent use of an interface is not enough. A student may learn where to click, how to complete an annotation form or how to match expected labels without becoming better able to understand any of the fundamental tasks that assistance enables. Platform fluency can quite profitably coexist with entrapment.
Interaction order is central in building capacity and avoiding dependence. A learner might attempt to locate and interpret evidence on their own, then gain access to the machine-nominated passages around which they may explore the traces left by prior human readers before they may then be invited to revise their assessment. Another sequence might expose all assistance and prior traces at the beginning. The first may preserve productive struggle but it introduces the same barriers that currently frustrate review. The second may reduce friction at the risk of trading critical engagement for cognitive surrender. The appropriate sequence is an empirical question about which work assistance removes and which epistemic work education intends the learner to perform.
For OS-ERIN to support human assessment, humans must, truly, assess. A learner should inspect a claim and source because doing so is necessary to understand or evaluate a real scholarly problem, not because producing graph trace is itself the assignment, engagement target or the requirement to get access to others’ prior work. When students are rewarded for the number of traces they produce, that will drive their practice rather than the critical understanding that is its proper motivation. A record that began as a by-product of learning can become distortionary surveillance.
There is no fundamental difference between a learner-authored interpretation and one produced by an expert. Both can be occurrence-bound, reasoned, attributable and revisable. This does not imply equal competence, standing or usefulness. Expertise is contextual, and a novice requires room for error and development just as an expert remains vulnerable to frame entrapment. Treating learner work as genuine scholarship is therefore both attractive and destabilising. If the work can matter beyond the course, learners may gain, and recognise themselves as having, a real position as contributors far earlier than is normal today. They may also inherit responsibility and exposure that a developmental setting should not impose without explicit rights and protections.
One constructive trajectory is earlier formation as a fallible critical peer. Papers cease to appear only as finished objects to be absorbed. Claims become inspectable, sources answerable and disagreement possible without abolishing scholarly authority. Repeated practice might cultivate habits that travel into later research, review and professional work. The relevant evidence to test for this outcome would be unaided performance on unfamiliar tasks, recovery of evidence omitted by the interface, the quality of reasons rather than trace volume and the ability to revise under uncertainty.
The malignant trajectory is earlier formation as a compliant operator. Machine nominations, prior trace and expected categories become answers. Students learn to perform visible diligence, avoid risky disagreement or treat accepted suggestions as sufficient. Public or assessed trace can make uncertainty costly, especially where teachers control progression. On this path the reified surface of the paper is replaced by the reified surface of the platform. While formal human responsibility remains, the educational outcome is a particularly toxic performative surrender.
Participation, recognition and competence must therefore remain separate. A student may make a real contribution without that contribution becoming a credential. A person may choose not to contribute without the absence meaning disengagement, incapacity or assent. Educational logs may be legitimate for bounded operational purposes while remaining inadmissible for person-level inference. When assessment or certification use is proposed, the question moves to consequential person-level products, not because education is irrelevant but because a new authority object has appeared.
Repeated educational use also creates a cohort dimension. People formed under the practice may carry expectations of inspectable claims, explicit uncertainty and ordinary correction into later roles. They may instead carry checklist habits and accountability theatre. These remain possibilities that must inform governance decisions today, as well as funding priorities for longitudinal studies that describe which practices persist outside OS-ERIN and survive institutional incentives.
If authentic learner work is retained, the record no longer belongs only to a course. It may enter a wider scholarly resource alongside reading and review trace. That possibility makes the next question unavoidable: what distinguishes a maintained shared academic record from a platform that merely accumulates contributions?
7. Connecting knowledge: projections, disagreement and contextual expertise
OS-ERIN treats a relationship as something that can itself be addressed. A relation may carry provenance, role, scope, reasons, uncertainty and revision rather than disappearing into a simple citation edge. With enough such objects, a reader can ask not only which documents connect but who interpreted a relation, under what conditions, how the interpretation changed and what later objects depend on it.
Plural trace cannot become usable at scale without selection. A filter, map, aggregate or summary is a projection: a view derived from a set of data for a specific purpose. It may show patterns that are impossible to see relation by relation. It also necessarily loses information. Minority interpretations may disappear, incompatible scopes may be combined, uncertainty may be compressed and a choice of grouping may create the appearance of consensus. Preserved lineage allows a reader to inspect how a view was produced; it does not make the view fair, harmless or appropriate for every use.
Disagreement is one tempting organising signal. A cluster of conflicting interpretations may indicate a relation worth renewed attention. It may also arise from different claim framings, source versions, languages, tasks, disciplinary conventions, coordinated activity or noise. Disagreement is not truth, importance, defect or actor quality. Routing scarce attention toward disagreement could improve scrutiny or amplify controversy that is easy to manufacture. The question is under what conditions disagreement is informative enough to nominate inspection.
Expertise makes projection harder. Scholarly knowing is distributed but not epistemically flat. A reader’s relevant experience may matter to a particular task, source type or domain. Yet competence is mutable, contextual and partly unobserved. Credentials, activity and institutional position are imperfect and potentially exclusionary proxies. The negative boundary is relatively clear: relation-level provenance should not be compressed into a portable global score of the person. The positive architecture is not clear. How can contextual standing matter without false equivalence, opaque weighting or reconstructed hierarchy?
That asymmetry should remain visible. A system can prohibit global actor rankings more easily than it can justify a way of weighting contributions. Treating all stances as equal can erase meaningful expertise. Treating metadata as expertise can reproduce structural advantage. Allowing users to choose their own weighting may hide the problem behind configuration. A production projection may need to stop with qualified failure when no defensible method exists.
A constructive projection would be task-scoped, reversible and plural. It would expose its lineage, make alternative views available, preserve minority structure and let users return to relation-level reasons. It would allocate attention without claiming consensus. A reader could see that the view changes under another framing and understand which judgments were omitted or compressed.
A malignant projection would become the record most people actually see. Defaults, ranks and confidence cues would acquire authority through repeated exposure. Users would approve or contest the projection rather than reconstruct relations. Contextual provenance would become a badge, dossier or hidden reputation score. Disagreement would become heat; uncertainty would become a number; human authorship would remain formal while machine-shaped views governed attention. This is projection as a recipe for cognitive surrender.
The recursion matters. Projections direct readers toward some relations and away from others. Their subsequent judgments then increase trace density in the selected regions, making later projections appear to confirm the first. Participation may concentrate where visibility and reward are greatest. The derived view changes the ecology from which it is derived. The technical systems that select the inputs therefore cannot be treated as background. A projection over incomplete or systematically skewed candidate fields may preserve lineage perfectly while reproducing upstream omission. The next section turns to extraction, retrieval, display, model learning and literature exploration: the mechanisms that shape what becomes available to human judgment before any collective view is built.
8. How technical mediation shapes what appears
OS-ERIN is centred on human judgment, but it is not pre-technical. Before a reader accepts an interpretation, technical processes may determine which document version is available, how it is parsed, how claims and passages are segmented, which candidates are retrieved, how windows are formed, how candidates are reranked and what the interface displays. Each stage can fail differently. Treating the stack as one model score would hide where omissions enter and where a reader needs an escape route.
The first research object is the adequacy of the evidence-filtering stack. A system might recover a passage explicitly nominated by a human while missing other evidence relevant to the relation. Nominated-span recovery and exhaustive-evidence recovery are different targets. Performance may vary by language, discipline, source type, claim structure, relation type and document quality. Aggregate benchmark success cannot establish adequate coverage for every bounded use. Provider, model, version, corpus and index lineage matter because a changed component may alter the candidate field without any change in the scholarly question.
Top-k display creates a substantive boundary. A relevant passage below the display threshold may be practically absent. Scores can invite false confidence even when they are not calibrated as probabilities. The reader must be able to search the full source, inspect context, nominate an unseen passage and record that the candidate field was inadequate. “Did not find” must remain distinct from “does not exist.” Qualified failure is a legitimate system response when coverage cannot be established.
Human acceptance does not erase this mediation. A person can make a real choice among options shaped by extraction and ranking while never seeing the omitted evidence. The authority boundary is therefore not secured by placing a confirmation button at the end. It depends on whether the system preserves meaningful routes outside its own framing, whether users understand candidate status and whether the design frustrates or rewards uncritical acceptance. The research questions here revolve around when assistance frees attention for active interpretation and when it precipitates passive approval.
The second object is the possible use of accumulated trace to evaluate or improve technical systems. Deliberate passage nominations, reasons and relation-level judgments might help identify systematic retrieval misses, compare candidate methods or calibrate bounded tasks. They are not automatic labels. They were produced for particular purposes by heterogeneous participants after exposure to a particular interface and candidate field. Disagreement and uncertainty are part of the data’s meaning, not noise that can simply be averaged away.
A separate data layer is the interaction record: which candidates were shown, what the reader inspected, whether they left the ranked pool, searched the full source, encountered prior trace, revised a judgment or substantially edited a machine proposal. These events are not scholarly stance and cannot establish diligence, competence or independent reasoning. They may nevertheless provide process provenance: evidence about the conditions under which a human-accepted judgment was produced, and data for studying exposure bias, workload, interaction order and candidate-system failure.
Combined with explicit evidence selections and relation judgments, such provenance could support large datasets for evidence retrieval, scientific natural-language inference, claim decomposition, salience modelling and human–AI interaction research. The defensible object is not gold-standard data. What OS-ERIN may provide is a human-accepted judgment under recorded task, exposure and interaction conditions. Collection and reuse would require separate purpose, consent, sampling, confidentiality, anti-surveillance, withdrawal and value-allocation decisions coupled to a well-grounded understanding of the impact of these uses on the behaviour of readers in the graph.
Use of machine-enabled, human-refined decisions creates a feedback risk. A model selects what people see. Yes, the architecture enables humans to look beyond that which is nominated, but that requires extra effort. The trace resulting from their shaped behaviour is treated as ground truth for the next model. Evidence absent from the first candidate field may receive little or no human attention and become less likely to appear later. The loop can become self-confirming even while apparent agreement rises. Held-out evaluation, random or exploratory sampling, recovery of human-nominated out-of-pool material and tests across routes are needed to learn about what the current system did not expose.
The local return to generative AI belongs here. Cheaper production of plausible scholarly-looking material may increase the value of provenance-rich human judgment, but it may also increase trace volume, contamination and maintenance burden. Whether accumulated stance materially improves resilience is an open question. The answer cannot be inferred from the fact that human trace exists, and non-AI value would remain even if that particular resilience claim failed.
Trace-to-model use also has rights and value dimensions. Consent for scholarly retention is not consent for training. Confidential review trace, educational work and public relation judgments may have different purposes and permissions. Withdrawal, licensing, purpose limitation, provenance and separation of training from evaluation matter. If contributor-created trace improves a proprietary service, the resulting value does not automatically return to contributors or the shared record. Those distributional and control questions move to funding and stewardship; the technical section establishes why they arise.
The third object is literature-neighbourhood exploration. The known cited relation is bounded by a source the author selected. A reader may also need ways to discover uncited work that could alter the assessment. Different routes—semantic or lexical search, citation traversal, bibliographic relationships, synthesis pivots or contrastive formulations—may expose different neighbourhoods. Route diversity concerns retrieval logic and information sources, not interface brands. Multiple routes can share the same corpus and biases, so apparent convergence is not independence by default.
A surfaced uncited work has low epistemic status. It may be readily discoverable and worth inspection. It does not establish relevance, completeness, material omission, deliberate cherry-picking or authorial intent. Conversely, an empty neighbourhood supplies no evidence that contrary or missing literature does not exist.
These three objects should not be collapsed. Strong within-source ranking does not establish literature discovery. Useful neighbourhood exploration need not create valid training data. Trace that is valuable for scholarship may be inappropriate for model learning. The constructive branch is bounded, auditable assistance that exposes its failures, preserves escape routes and learns from deliberately sampled evidence under explicit rights. The malignant branch is a self-reinforcing field in which omissions become less visible, human approval launders technical selection into authority and shared trace becomes an enclosed asset.
Technical mediation is constitutive because it changes what the next reader can inspect. It also changes what happens when a relation later needs revision. Once a source, interpretation or recorded status changes, the system must decide what to reopen, what to show and what derived views to rebuild without confusing technical dependency with substantive invalidity.
9. Ongoing tending, concern and correction
A record that persists must be able to change without pretending that every shift is equally important or that a given shift is consistently material. Discovery of a possible problem, a reader’s concern, an author’s response, a recorded status, an editorial decision and an institutional finding are distinct. A changed source or interpretation may nominate downstream related relations for renewed inspection. The word nominate is crucial. A citation graph records connections as distinct from the content required to characterise dependency. On a claim supported by multiple sources, one weakened support branch may leave an alternative branch intact—one that may have been followed less often by virtue of vagaries of the machine ranking of relevant passages presented to readers. A correction to the error term on a measurement may affect one downstream claim but not another citing the same source for the same purpose but in differently forgiving circumstances. Routing can help allocate attention when there are changes, but it cannot alone authorise substantive change.
Where retention remains legitimate, revision should usually preserve history through annotation, addition or supersession rather than silently overwriting the earlier state. A reader needs to know which version was visible when a judgment was made and why a later state replaced it. Withdrawal and lawful erasure may still be required in some settings but they should remain the exception. Negative space must also be handled carefully: the absence of a visible relation after withdrawal should not automatically be interpreted as evidence about what once existed or why it changed.
Derived views create another maintenance obligation. If an underlying relation is superseded or invalidated for a particular task, a projection that incorporated it may need to be rebuilt. The lineage should make that possible, but lineage alone does not decide the new result. A status change can trigger recalculation without dictating the substantive interpretation. Leaving the old view visible without qualification would allow superseded states to persist through the convenience of the derivative.
Recording a correction is not the same as correction uptake. A status may be centrally updated while readers continue to encounter cached copies, old summaries or institutional decisions based on the earlier state. Awareness, changed interpretation and changed practice are separate outcomes. A maintenance system that measures only whether a record changed may overstate its effect on scholarship.
Pre-publication review and ongoing tending should inform one another without collapsing. Review can identify concerns before entry; later maintenance can reveal patterns that improve future review. Yet they involve different timing, roles and authority. A confidential reviewer comment cannot simply be released into public maintenance, and a public concern does not retroactively become part of the original editorial judgment. Bounded exchange requires explicit rules.
The constructive branch is continuous claim-level maintenance. Concern, response, provisional status, revision, withdrawal, supersession and official correction remain distinguishable but connected. Retraction becomes one relatively late authoritative event within a wider ecology of reconsideration. Affected relations can be nominated without assuming that one change destroys every downstream conclusion.
The malignant branch here is noisy reputational contagion. Alerts spread farther than the evidence warrants. Concern is read as finding. Low-value reopening overwhelms scarce expertise. Topology creates the appearance of substantive dependence, and visible status reaches people before response or due process. On this branch the record may devolve into the academic sibling of the 24-hour news cycle: exhausting, meaningless sensation.
Maintenance also has labour consequences. Repeated reading, teaching and review create opportunities to notice and repair relations. Distributed tending might make the record more revisable than document replacement alone permits. It might also externalize investigation and correction burden onto contributors, editors or dependent scholars. Support, reciprocity and limits on expected contribution matter before maintenance is normalized as an ordinary obligation.
No lifecycle design can work independently of people’s willingness and ability to raise and answer concerns. A formally available reporting route may be unusable under unequal dependence. Once concern and response become questions of attribution, audience, linkability and possible retaliation, ownership of the argument passes from status mechanics to identity and power.
10. Attribution, identity, misbehaviour and self-censorship
Scholarly trace needs enough provenance to be answerable. A reader should be able to distinguish one interpretation from another, understand the role and context in which it was produced, and respond where response is appropriate. Visible attribution anchors. It does not itself establish authority, standing, merit or protection. Authentication can show that the same account produced two records without establishing degree of competence or freedom from consequence.
Identity is not a choice between public sharing of true identity and full anonymity. It includes what identity is stored, what is displayed, to whom, for how long, under what role, with what linkability across tasks, and who may reveal or reuse it. A person might be known to a trusted intermediary but displayed pseudonymously to a wider audience. Identity might be visible during a bounded review and withheld after release.
Consider a junior scholar who notices a possible problem in a supervisor’s work in a small field. Public attribution may make response straightforward while exposing the scholar to dependence-related consequences. Pseudonymity may be illusory because topic, institution, role or writing style identifies the contributor. Confidential mediation may reduce exposure while placing substantial authority in the mediator. Strong anonymity may protect dissent while limiting answerability and enabling trolling. There are a number of concerns bearing sometimes incompatibly on the discoverability of personal identity.
Ordinary disagreement, methodological concern, integrity concern and formal misconduct allegation require different statuses and pathways. Treating every negative interpretation as an accusation would chill legitimate scholarly disagreement. Treating every serious allegation as ordinary disagreement could deny appropriate response and institutional process. OS-ERIN can preserve the distinction and route an object; it should not decide the institutional category by technical default. Answerability and protection are both legitimate objectives. A named accusation with no meaningful response route can be harmful. An unanswerable anonymous claim can also be harmful. It is not yet clear who should see stored identity, who may unmask it, for which purpose, under what review, and what rights of response, revision, withdrawal and contextual correction follow.
Persistence changes the problem. A comment that is tolerable inside a bounded exchange may have different consequences when searchable across institutions and years. Context can disappear while identity linkage remains. A later audience may interpret a provisional concern as a settled finding. Durable career effects can arise even if no one intended the record to become evaluative. Conversely, hiding all identity may make it difficult to distinguish good-faith expertise from strategic activity. Research must test both protective and abuse failure modes symmetrically.
Silence and nonparticipation require an explicit boundary. A person may refrain because of workload, access, uncertainty, safety, language or simple irrelevance. Absence of trace or response to critique is not assent, cowardice, lack of diligence or lack of contribution. Passive telemetry cannot repair this negative space. Turning nonparticipation into an inference would make the system evaluative even before any formal reward use was announced.
Identity configurations also affect the composition of the record. If visible dissent is costly for dependent actors, trace may disproportionately reflect people with security, status or institutional support. A projection built from that record can reproduce exposure inequality without containing an explicit identity score. If anonymity encourages high-volume low-accountability activity, another distortion appears. Participation safety and record validity are therefore connected, but the direction and incidence of effects remain unknown.
The constructive branch is a staged or role-bounded arrangement that permits responsible concern, meaningful response and contestable attribution. It preserves enough provenance for scholarship while preventing unnecessary cross-context linkage. The malignant branch includes chilling, retaliation, re-identification, unanswerable accusation and permanent association with provisional states. Neither branch can be selected by declaring a preferred identity mode.
This section concerns exposure and participation whether or not anyone formally evaluates the contributor. A separate transition occurs when trace, a projection or an activity history is deliberately converted into evidence about a person in a consequential decision. That conversion creates new questions of admissibility, reactivity and distribution that belong to reward and recognition.
11. When graph trace enters reward and recognition systems
Scholarly trace can make careful reading, review, correction, maintenance and care more visible, but visibility alone establishes neither recognition nor benefit. Section 11 begins at a different threshold: deliberate conversion of trace or a derivative into evidence about a person in a consequential decision.
Such uses vary along several dimensions. The trace may originate as task work, a by-product, stewardship or contract. The product may be narrative, categorical, threshold-based, scalar or automated. It may enter workload allocation, certification, hiring, promotion, grants or another domain. Stakes, human oversight, context compression, portability, identity dependence and lifecycle status all matter. These dimensions gesture at domains that we must yet investigate.
The constructive possibility is bounded recognition of specified work. A contextual narrative might help an institution see maintenance that would otherwise disappear, while preserving provenance and allowing contestation. A workload discussion might acknowledge recurring stewardship without ranking contributors globally. Such use might improve credit allocation. It might also have little effect if the institution continues to value other signals.
The malignant mechanisms are familiar. Once people anticipate evaluation, they may produce safe, routine and countable trace; avoid difficult criticism; move inquiry off-platform; or optimise visible completion. Contribution can become compulsory measurable labour. Contextual provenance can harden into a portable reputation system. People with less time, language confidence, access or employment security may become less visible and then appear less valuable. A system intended to recognise neglected work can intensify hierarchy.
Educational trace created to support learning might later be proposed for certification, but that new audience and higher stakes require a separate admissibility analysis. Origin in an authentic scholarly task does not authorise person-level use or adverse inference from nonparticipation.
High-consequence use would require more than an accurate extraction. The underlying trace needs provenance, status and lifecycle information. A provisional concern must not be presented as an official finding. Superseded relations must propagate into the product. At the very minimum, people need a way to contest errors, supply context and know what uses are prohibited.
Candidate no-use zones may be necessary where context cannot survive, participation would become effectively compulsory or adverse inference from absence cannot be prevented. Hiring, promotion, grant eligibility and automated thresholds may present different risks from bounded narrative use. The programme should not convert those candidates into a complete doctrine without evidence and institutional judgment.
The ecological effects are immediate. An evaluation product changes the practice that produces its input. If trace farming, chilling or displacement occurs, later records become less valid for the use that induced them. If bounded recognition supports maintenance without substantial distortion, the record may become more sustainable. If effects are null, added infrastructure may still impose burden. The questions we must ask will explore which configurations produce which mechanisms for whom.
Person-level use also creates distributional questions without resolving them. Who gains recognition? Who bears the work of contesting errors? Who is absent because participation was unsafe or inaccessible? Who benefits when trace is converted into an institutional product? Even a functionally defensible use needs an institution that can say who may create it, for what purpose, with what exceptions, who pays for maintenance, who bears responsibility when it fails and whether participants can exit. Reward and recognition therefore lead into funding, control and stewardship rather than supplying their own constitutional answer.
12. Funding, control, value and institutional form
If recurrent relation-level practice becomes useful, institutions may depend on the resulting infrastructure. At that point funding and governance shape which functions are built, whose needs receive priority, how long the service persists, which uses become normal and whether participants can leave without losing the shared record.
Several categories must remain separate. Value may be created by scholarly contribution, technical operation, integration, maintenance or model improvement. The people who create value may differ from those who receive it. Payment may cover a service without purchasing control. Revenue may sustain operation without establishing public value. A public or nonprofit form may still concentrate authority, while a paid service may coexist with nonexclusive infrastructure.
Forms of work also differ. Ordinary scholarship produces reasoning for a scholarly task. Stewardship maintains a shared capability. Co-production shapes the system or its services. Contractual contribution is governed by an explicit exchange. Operator labour keeps infrastructure running. Treating all of these as undifferentiated contribution can obscure rights, compensation and responsibility. Treating every scholarly act as unpaid input can normalise appropriation.
A defensible boundary is that payment may buy integration, support or additional services but not suppression rights, epistemic standing or exclusive control over canonical scholarly trace. A funder should not be able to remove an inconvenient interpretation by virtue of payment. A service provider should not acquire privileged truth status for its projection.
One constructive branch is a portable, nonexclusive spine with plural service layers. Participants can export relation objects and provenance, alternative providers can operate, derived views remain visibly derived and institutional succession does not erase the shared record. Forkability provides a practical rather than merely formal exit. Decision rights, custody and continuity are explicit enough to contest.
A malignant branch might include enclosure and capture. Integration dependence can give a provider de facto control even where data are nominally open. A proprietary service layer may become the only usable access point. Privileged projections may shape decisions. Contributor trace may become a closed training asset. A funder or host may gain priority or suppression power through operational dependence. The reverse failure is governance so elaborate that the infrastructure cannot make timely decisions or survive succession. Capture resistance and operability can conflict.
Portability must therefore be tested in practice. The existence of an export format does not show that another institution can operate the service, reconstruct rights or afford the move. Exit cost grows as workflows, identity systems, models and institutional decisions become integrated. Building portability early preserves more reversibility than promising it after dependence accumulates.
Funding routes create path dependence. Short-term project support may permit experimentation but create continuity risk. Institutional hosting may supply stability while aligning priorities with the host. Commercial services may fund operation while increasing enclosure pressure. Community governance may preserve participation while imposing substantial maintenance burden. These are alternatives that all merit investigation.
Model-development value makes the distributional question concrete. If accumulated human trace improves retrieval or ranking, value may appear in a service separate from the shared record. Contributors may lack knowledge, consent, access or any flow-back. Strict purpose limits might preserve trace for benchmarking but not training. Another arrangement might permit opt-in improvement with reciprocal access or governance. The technical validity and the institutional legitimacy of use are distinct.
Some questions are empirical: what exit costs arise, how burdens are distributed, whether governance remains workable, whether paid services can coexist with a nonexclusive spine. Others require legitimate choice: who may hold identity, authorise release, grant exceptions, bear liability or control succession.
No custodian, legal regime, ownership model or constitution is selected here. The section explains why those questions become endogenous if the system succeeds enough to create dependence. Institutional form is not evidence that commoning has occurred, and commoning is not a substitute for operational settlement.
The cost of delay changes over time. Early choices about data portability, rights and service separation may be reversible. Once careers, workflows, models and institutions rely on the trace, changing them becomes difficult. The work must therefore identify which uncertainties can remain open during local experiments and which become architectural blockers before durable commitment.
13. A research programme’s worth of questions
Tugging on the apparently simple problem of helping PhDs check sources has produced far more questions than progress. The questions that we are able to identify today form a coherent agenda insofar as answers in one area change the conditions under which other questions should be studied and the decisions that remain reversible. Each must connect a challenge to a social or design purpose, preserve competing possibilities, identify what would need to be learned and mark the transition at which uncertainty becomes an architectural inadequacy.
Our questions also occupy different relations to governance. Some concern uncertain causal mechanisms: whether prior trace anchors, whether educational capability transfers, whether a projection directs attention usefully. Some concern comparative design: reveal order, candidate display, identity configuration, response routing. Some concern legitimate authority: release, adverse inference, custodian selection, payment rights and no-use zones. Calling all of them empirical would obscure choices evidence cannot make. Calling all of them governance would avoid testing mechanisms that may defeat a chosen design.
The 25 questions and their canonical homes are as follows.
Reading and scholarly memory — Section 3.
- How should prior human judgment be revealed so reuse can reduce duplicate work without unduly impairing independent reconstruction?
- Which parts of claim–source inspection become cheaper, which burdens are introduced and what is the net effect on useful scrutiny?
- Which deliberately authored judgments retain enough meaning and value across time and task to justify preservation?
- What minimum version, scope, rationale and context are required to prevent false transport of old stance into new tasks?
Peer review and editorial work — Section 4.
- Which structured review judgments can be preserved and reused while maintaining confidentiality, candour, task fit and editorial authority?
- Under what review conditions does structured source assistance improve issue detection, traceability or decision support enough to justify its burden?
Education and epistemic capability — Section 5.
- Does repeated authentic claim–source inspection improve unaided performance on unfamiliar evidence problems?
- How should machine nomination, independent attempt, prior human trace and feedback be sequenced so learners perform the epistemic work education intends to cultivate?
Commons and commoning — Section 6.
- Which practices, rights, reciprocal relations and maintenance effects distinguish commoning from access, contribution or platform use?
- How can contribution, access, benefit, stewardship and governance be related without quotas, rankings, contribution scores or adverse inference?
Collective projection and contextual expertise — Section 7.
- Which task-scoped projections can preserve lineage, minority structure, uncertainty, revision, rights and qualified failure without manufacturing consensus?
- Under what conditions can disagreement nominate scrutiny without routing attention toward framing artefacts, coordination or noise?
- How can contextual and mutable epistemic standing matter without portable reputation, structural exclusion, false equivalence or opaque weighting?
Technical mediation, changing content pressure and discovery — Section 8.
- Under what conditions does machine assistance reduce verification burden without becoming de facto authority through nomination, ordering, explanation or confidence cues?
- Does provenance-rich human judgment reduce verification burden or improve issue detection under increased synthetic-content pressure, and at what maintenance cost?
- How adequate is each stage of the evidence-filtering stack, where does it fail and when are manual or full-text escape routes mandatory?
- Which trace layers may validly and legitimately be used to benchmark, calibrate or improve models, under what rights and feedback-loop controls?
- Which route-diverse methods can support reasonable discovery of potentially omitted literature without turning non-discovery or convergence into claims about absence, relevance or intent?
Scientific self-correction and maintenance — Section 9.
- How can a claim-level graph nominate affected relations after correction, retraction, revision or changed stance without assuming every downstream claim is invalid?
- How should concentrated entry review and diffuse continuing tending divide labour, exchange information and trigger one another without collapsing their authority structures?
Identity and unequal power — Section 10.
- Which combinations of stored identity, displayed identity, audience, persistence, mediation and reveal authority permit responsible concern and response under unequal dependence?
Reward and recognition — Section 11.
- Under which configurations may graph-derived information enter recognition or assessment without unacceptable distortion, exclusion, chilling or conversion into portable reputation, and which uses require prohibition?
Funding, value and institutional stewardship — Section 12.
- Which technical, contractual and governance boundaries allow paid integration and support to coexist with a nonexclusive canonical spine?
- How should beneficiaries, buyers, funders, co-producers and in-kind contributors be related without treating use, labour, payment and contribution as interchangeable?
- At material multi-institutional scale, what arrangement best preserves portability, continuity, public accountability and contestability without enclosure or brittle governance?
This registry of unresolved questions becomes a programme through dependencies. If the education question finds little transfer, the case for outward socialisation weakens even if students produce abundant trace. If the prior-judgment question finds strong anchoring, memory and projection design must preserve independent reconstruction rather than merely improve access to prior views. If the projection question cannot produce usable views without flattening plurality, increasing density may create noise rather than shared capability. If the recognition question shows that consequential use materially distorts participation, reward firewalls may become a condition for the validity of upstream studies. If the stewardship question remains unresolved when several institutions become dependent, governance is no longer an appendix; it is an architectural blocker. If the model-improvement question exposes a self-confirming learning loop, ordinary accumulated trace cannot be reused for technical improvement without redesigned sampling, rights and evaluation.
These dependencies suggest staged inquiry rather than a fixed roadmap. Early work can remain bounded where failure is reversible: one relation, one task, clear purpose and preserved off ramps. Expansion should require evidence about burden, transfer, anchoring, filtering and maintenance. Projection should not become the default access layer before its losses and qualified-failure behaviour are understood. Person-level use should not be smuggled in as a secondary benefit of visibility. Institutional dependence should not outrun portability and stewardship decisions.
A research result need not point toward expansion. It may justify narrowing a use, delaying a transition, separating purposes or establishing a no-use boundary. A constructive branch may be technically possible but institutionally illegitimate. A preferred governance arrangement may be legitimate but unable to overcome a causal failure. The programme remains useful only if negative and null findings can change what OS-ERIN becomes.
Time is constitutive because repeated use changes later inputs. The useful unit is not a ten-year forecast but a sequence of thresholds: local repetition, routine encounter with prior trace, density requiring projection, consequential use and institutional dependence. Cohort effects may appear after the platform is absent; path dependence may persist after an interface changes. Longitudinal inquiry is needed because a system can appear helpful in an isolated task while producing malignant effects through accumulation.
14. Terms, status and change
The following terms carry specific, limited meanings in this Field Guide.
Academic record means the papers, data, citations, interpretations, qualifications, revisions and institutional statuses through which scholarship is preserved and encountered. It is broader than a publication database and does not imply a unified authority.
Claim occurrence means a particular use of a claim in a particular version and context. It is not an abstract sentence detached from the argumentative work it performs.
Claim–source relation means the situated relation between one claim occurrence and one source version as interpreted for a bounded task. It may include passages, scope, reasons, uncertainty, attribution, status and revision.
Deliberate stance means an interpretation a person intentionally authors and accepts. Clicks, reading time, search history, silence and other passive behaviour are not deliberate stance.
Trace means a retained scholarly object or history produced through reading, review, learning, maintenance or another authorized task. Retention alone does not authorize visibility, release, reuse or model use.
Candidate passage means a passage nominated for human inspection by a technical process. Its rank does not establish evidential adequacy or completeness.
Neighbourhood suggestion means an uncited work surfaced through an exploratory route. It may warrant inspection or explanation but does not establish relevance, omission, absence or intent.
Derived view or projection means a task-specific selection, compression or aggregation over plural trace. It is not canonical truth and necessarily loses information.
Release authority means legitimate permission to move an object to a wider audience or purpose. It is separate from the authority to retain, inspect, revise, reuse or train on the object.
Contextual expertise means competence relevant to a particular task, domain, source type, time and community. It is mutable and should not be converted into a portable global property of a person.
Qualified failure means an explicit refusal to imply coverage, ranking adequacy or a defensible projection when the available method cannot support one. It is a valid outcome, not merely a system error.
Stewardship means responsibility for maintaining continuity, rights, contestability, portability and decision processes around a shared capability. Hosting or owning a service does not by itself establish stewardship.
Ecological describes recursive interaction among scholarly practice, retained trace, technical systems, incentives, institutions and governance. It does not imply harmony, equilibrium or automatic self-correction.
Developmental means that accumulation over time changes the conditions under which later activity occurs. The term does not imply maturation toward a preferred end.
Bifurcation means that an accumulating affordance can support more than one plausible trajectory, including constructive and deformative branches. It does not turn every governance boundary into an empirical question.
Commons means a shared resource sustained by legitimate relations of participation, rights, reciprocity, maintenance and governance. Publicness, persistence and use are insufficient to establish it.
Commoning means the continuing practice through which such a shared capability is made, repaired, contested and governed. It is a possible developmental branch, not a present status inferred for OS-ERIN.
Five status layers should remain separate throughout. A design commitment states a boundary or principle the project intends to preserve. An intended capability states what a system is meant to enable. A research question or proposed effect identifies what remains to be learned. A current implementation fact reports what a dated, versioned system presently does. A demonstrated effect requires evidence that the proposed consequence occurs under specified conditions. Movement between these statuses requires new warrant.
Current implementation should therefore be maintained as a separate, dated and versioned object with known limitations and links to technical documentation. The evergreen essay should not embed a software inventory that becomes stale, and the absence of later verification cannot justify upgrading an older maturity claim. Conceptual architecture, current operation and possible effects remain different objects even when they concern the same feature.
OS-ERIN is work in progress. Its smallest proposition is that a reader’s bounded reasoning about a claim and source might remain inspectable and revisable. Its larger programme asks what recurrence does to scholarship and which boundaries prevent useful trace from becoming inherited authority, extractive infrastructure or institutional fact by default. Those questions remain open; the distinctions above are intended to keep them open in a form that can be investigated.