Scientific evidence,
made inspectable.

Reading a scientific paper produces evidence: a specific passage that supports, questions or extends a research idea. In a conventional workflow that evidence lives only inside a reading session — a highlight in a viewer, a quote in a note, a sentence in a draft.

This document argues that evidence should be a first-class research object. A passage is identified by its paper, its page and its exact text, and it can be selected, highlighted, cited and connected to other passages while keeping that provenance intact.

A passage is a bounded piece of scientific text. It has a stable location inside a paper — the page it appears on and its position within that page — and it can be referenced by that location.

Because the passage keeps its provenance, a later reader can always return to the source and verify the claim for themselves. Nothing about the connection is hidden behind an opaque summary.

The workflow starts with selection. The reader marks the exact text that matters; the system stores the text together with its paper and page reference. That selection can then be highlighted, copied as a citation, sent to a note, or connected to a passage in another paper.

The critical property is that selection does not move the evidence into a separate system where the source is lost. The source reference travels with the passage everywhere it goes.

A single paper rarely settles a question. Ideas accumulate across papers: one paper provides a definition, a second provides a measurement, a third questions the measurement. The researcher who keeps these passages connected builds a map of the literature as they read.

This document argues that the connection itself is not scientific truth. It is a researcher's working hypothesis about how two passages relate. The system displays the relation transparently, with the exact source text and the exact target text side by side.

When a relation is inspected, the scientific text is visually dominant. The reader sees the source passage in a thin frame, the relation label between the passages, and the target passage in a second thin frame with its paper and page reference.

Graph abstractions are useful, but they are never a substitute for the text. A node in a graph is only a doorway back to the passage it represents.

The target passage is a live reference. Clicking the target opens the target paper as an additional panel in the desk, preserves every panel that is already open, and focuses the reader directly on the target sentence.

This behavior is identical whether the relation comes from a public demo session or from the authenticated product: the interaction is a reusable product interaction, not a special case.

Reading sessions end. Without preservation, the connections made during a session are scattered across viewer-local highlights, notes and memory. The next session starts from zero, and the effort of assembling the context is repeated.

This document argues that the research context — the selected passages, their connections and the notes around them — should be preservable as one object.

A Staple preserves the state of the desk: the papers that were open, the highlights that were made, the connections between passages, the floating note and the provenance of every passage. It is a snapshot of the reading session, not a summary of the science.

A Staple has revisions. Each revision is an immutable snapshot; a later revision never rewrites an earlier one. The reader can always return to the exact context that was preserved.

Staples are private by default. Sharing is an explicit owner action that creates a read-only capability link with an unguessable token. The link can be revoked or rotated by the owner at any time.

Sharing a Staple does not grant redistribution rights for the papers inside it. Full text is never redistributed through a share link; readers are pointed back to the source instead.