Field Guide · Research questions
Where does this touch something you care about?
Each path follows its promise into the tensions it creates.
1 Reading the record What changes when careful source checking can be reused?
Reading the record
Careful reading usually serves an immediate purpose: understand a paper, check a citation, or assess an argument. OS-ERIN asks what changes when some of that work can remain available afterwards. Its first object is deliberately narrow—one occurrence of a claim, in one version of a document, in relation to one version of a cited source. A reader can preserve the passages and reasons behind an assessment without turning that judgment into a verdict on either work as a whole.
Reuse also creates tensions. A ranked passage list may help, but it is not the evidence field; readers must still be able to search the source and nominate what the ranking missed. Neighbouring literature is weaker still: it may make an omission worth inspecting, but does not establish relevance, completeness, cherry-picking, or intent. Prior human judgment may save reconstruction while also anchoring the next reader. The practical questions are therefore about reveal order, candidate omissions, and whether assisted inspection produces a defensible net gain.
Follow this thread →Related: technical mediation of attention
2 Remembering scholarly judgment When does scholarly memory help, and when does it anchor the next reader?
Remembering scholarly judgment
Scholarly memory already lives in people, groups, correspondence, notes, seminars, and disciplinary habits. Dense communities often remember why a source mattered or why an interpretation was rejected; the wider academic record usually does not. OS-ERIN proposes to make selected, deliberately authored parts of that memory explicit, attributable, and revisable beyond the setting in which they were first produced.
Persistence alone is not enough. A judgment may only make sense with the version of the text, the exact claim occurrence, the task, the reader’s role, the scope of the conclusion, and the reasons behind it. Too little context can make old trace feel authoritative after its meaning has disappeared; too much can make it costly and unusable. Retention, visibility, release, reuse, and deletion are also separate decisions. Where retention is legitimate, later correction should append or supersede rather than silently erase. The unresolved questions are what is worth preserving, how long it remains useful, and when seeing prior judgment helps rather than anchors.
3 Reviewing and editing Can review reasoning survive beyond review without carrying editorial authority with it?
Reviewing and editing
Peer review is one bounded context in which OS-ERIN may be useful, not the definition of the project. Structured trace could preserve which claim–source relation was examined, which passage mattered, what the reviewer concluded, and what remained uncertain. That may make reasoning more inspectable without turning the system into an editor, or making the production of trace evidence that a reviewer was diligent, competent, or correct.
Review makes authority and exposure especially sharp. Prior reviewer judgment may reduce duplicated effort or anchor a later reviewer before independent reconstruction. Ranked passages and neighbourhood suggestions may direct attention, but neither is a finding. Confidential trace also cannot move into later circulation merely because it was retained: review, release, model use, and post-publication maintenance carry different permissions and consequences. The central questions are whether claim-level review improves outcomes enough to justify its burden, what may legitimately survive the review context, and how entry review can inform later tending without making all scrutiny a permanent extension of editorial authority.
4 Learning to evaluate evidence Does assisted inspection build capability that survives without the assistance?
Learning to evaluate evidence
The educational question is not whether students become competent users of OS-ERIN. It is whether they become better able to evaluate unfamiliar evidence when the platform, prior trace, and machine assistance are absent. That makes interaction order consequential. Independent attempt, candidate retrieval, exposure to other readers, and feedback may each help at different moments; offered too early, they may remove the productive reconstruction through which judgment develops.
The work should also remain authentic. Trace is most defensible when it arises as a by-product of a real attempt to understand a claim and its source, not because producing trace is itself the performance target. Used for grading, surveillance, certification, or adverse inference, it may change what learners choose to make visible. Learner contributions can still be genuine scholarly objects, but competence remains contextual and mutable rather than a portable score. The decisive questions are whether capability transfers to unaided work and which sequence of struggle, assistance, prior judgment, and feedback best supports that transfer.
5 Maintaining a shared academic record When does accumulated trace become shared scholarly infrastructure rather than platform data?
Maintaining a shared academic record
A shared academic record involves more than accumulating annotations. The scarce input of competent attention, the epistemic product of more inspectable claim–source relations, and the institutional possibility of better collective inquiry must remain distinct. OS-ERIN may connect them, but retained trace does not automatically become useful knowledge or shared capability.
Nor do publicness and persistence create stewardship. A maintained resource requires rights, reciprocity, anti-enclosure, continuing participation, and real decision powers. Visibility may help neglected maintenance work receive recognition; once assessment, certification, or commercial extraction makes trace the target, the same visibility may encourage trace farming, coercion, or strategically safe contribution. Participation cannot be epistemically flat, but standing must not become a portable reputation score. Funding, hosting, and model-development use then shape whose priorities prevail and where value flows. The open question is what distinguishes shared stewardship from ordinary platform use while keeping the record useful, revisable, and genuinely shareable.
6 Connecting scholarly knowledge Can plural judgment become usable without manufacturing consensus or rank?
Connecting scholarly knowledge
OS-ERIN treats relationships as things that can themselves be inspected. A claim–source relation can carry its authors, roles, scope, reasons, provenance, and revision history rather than disappearing inside a citation link. The formal graph sits behind this idea; the important first step is that the relationship becomes an addressable scholarly object whose interpretation and dependencies can be revisited.
Once many such objects exist, projections are needed to make them usable. Every filter, aggregate, or map selects and compresses, and may erase minority structure, uncertainty, or incompatible interpretations. Preserved disagreement can sometimes nominate useful scrutiny, but disagreement is not truth, importance, defect, or actor quality. Expertise complicates the problem further: participation is not epistemically flat, yet competence is contextual and mutable rather than a single property of a person. The research problem is how collective views can remain useful without manufacturing consensus, when disagreement is a defensible attention signal, and how contextual expertise can matter without hardening into rank.
7 Responding to changing pressures Can machines direct attention without quietly becoming the authority?
Responding to changing pressures
OS-ERIN is not a response to generative AI alone. Costly verification, weak incentives, and lossy scholarly memory predate it. Generative systems are a current pressure because they lower the cost of plausible scholarly text while competent verification remains expensive. Greater inspectability, provenance, and revision capacity may help, but those are design properties—not proof that the record has become resilient.
Machine assistance also changes more than speed. Nomination, ordering, explanation, and confidence cues shape the field a person sees before making an attributable choice. A final human click can establish accountability without showing independent reconstruction. Accumulated human trace may later improve retrieval or related models, creating a loop in which the model shapes what people see and their judgments reshape the model. Those judgments are neither automatic ground truth nor automatically licensed for reuse. The hard questions concern candidate omissions and drift, exposure bias, contributor rights, value flow-back, and how machine assistance can support rather than quietly replace substantive human judgment.
8 Maintaining and correcting the record Can concerns propagate through the record without becoming verdicts?
Maintaining and correcting the record
An inspectable post-publication record needs to distinguish stages of concern. A reader’s judgment that something may be wrong is not the same as an author’s response, a recorded status, an editorial decision, or an institutional finding. OS-ERIN can preserve and connect these objects without becoming the authority that settles them. Pre-publication review and ongoing tending may inform each other, but they operate under different rules.
When a source, interpretation, or status changes, related claim–source relations may deserve renewed attention. Dependency routing can nominate where to look; it cannot prove that every downstream claim is contaminated or ensure that correction changes later use. Revision should remain auditable through annotation and supersession where retention is legitimate, while confidentiality, withdrawal, and lawful erasure remain possible. New literature or a neighbourhood suggestion may also trigger reconsideration, but is not correction evidence by itself. The hardest tension is how concern remains attributable enough for response without making disagreement unsafe under unequal scholarly power.
9 Funding, control and public interest Who controls the record once it becomes valuable enough to depend on?
Funding, control and public interest
How OS-ERIN is funded and housed will shape what it becomes. Resources create capacity, but also dependencies, priorities, and constitutional paths. Value creation, payment, revenue, control, and public value are different questions: the people who produce useful trace may not capture the resulting benefit, and a public or nonprofit form does not by itself establish public interest.
Payment may legitimately buy services, integration, or support. It should not buy suppression rights, epistemic standing, or exclusive control over the scholarly record. Making reading, reviewing, and maintenance visible could help recognize labour that currently disappears; once trace enters hiring, promotion, grant allocation, workload, certification, or institutional assessment, it may change what people record and who feels safe participating. The same trace may also produce privately enclosed model value. The questions are constitutional as well as financial: can paid services coexist with a nonexclusive shared spine, who benefits and bears risk, and what stewardship can resist capture while remaining workable?
In depth
Read the full argument
The in-depth essay follows the idea from one claim–source relation through memory, learning, projection, correction, power and stewardship.