The graph and its research workflow
A map of the records behind the Explorer, how they connect, and how a research collection adds new material.
Reviewed 14 October 2026 · Prototype documentationThis is a conceptual schema, not an exhaustive field reference. Dates, source accounts, and editorial decisions remain separate even when the Explorer shows them together.
The graph at a glance
Each outlined shape is a navigable record. The gold treatment sits between a thinker and an idea, carrying the specific account that the broad person–idea association cannot express. An idea can bridge many treatments; a treatment can also concern several ideas or people.
Major record types
| Record | What it represents | Useful fields or context |
|---|---|---|
| Person | A thinker or other historical participant. | Stable identity; sourced birth/death dates when available; alternative source identities kept explicit. |
| Idea / concept family | A reusable topic or problem connecting distinct accounts across time. | May be broad or specific; a shared label alone does not prove the same meaning. |
| Treatment | A concise, attributed account of a thinker’s approach to one or more ideas. | Summary, participant and idea IDs, scope, contribution, source evidence, limitations, qualified dates and review state. |
| Work | A text or other work that can locate a treatment or support authorship. | Composition/publication dates have their own meanings; work context is not itself an authorship claim. |
| Other nodes | Movements, taxonomies and further source-defined types where useful. | Retained as typed records rather than forced into a person/idea binary. |
The broad “Ideas” navigation category can include concept families and treatments, but their underlying types remain distinct. Some local additions are editorial research-context nodes rather than sourced philosophical claims.
Simplified shape only: the underlying source, treatment, date, and workflow schemas carry additional fields and validation.
Links, identity, dates, and evidence
| Layer | How to read it |
|---|---|
| InPhO source links | Imported associations and taxonomy, with original identifiers and provenance. Useful for discovery; not automatically verified influence. |
| Treatment structure | Participant, idea and work-context links locate an account. “Has component” organizes a broad account into focused treatments; it is editorial composition, not a historical relation. |
| Historical relations | Typed, sourced treatment-to-treatment claims such as response, criticism or influence. Their direction and qualifications remain in data even if a view draws plain lines. |
| Inferred comparisons | Dotted blue-gray treatment links compare a specific conceptual parallel or contrast. They are undirected, attributed to the model, and include reasoning, limitations and sources for the compared accounts. They do not establish influence or historical transmission. |
| Research context | A collection groups records around a question. A direct local addition can have an editorial context link. Neither asserts historical influence. |
A source record and a local canonical identity may be linked through an explicit identity assignment. An alias helps navigation but is not created from matching labels alone. Evidence assertions, citations, source captures, and locators support a treatment without becoming separate visible graph nodes in every view.
Dates retain event and precision: birth, composition, publication, development, attestation, or an explained unknown. The timeline can use a publication date as a qualified placement for a treatment; it does not claim the idea first arose that year.
Before a treatment can be published, every person, idea, and work endpoint must resolve to an active graph record. A missing person endpoint is initialized with a stable local ID and source identity; birth and death each receive a cited date or a cited unknown after lookup. The same rule applies when an existing person is present but one side of the lifespan has never been reviewed. Treatment dates remain separate: a clearly labeled lifespan or activity range may help place an otherwise undated account without being represented as its composition date.
Inspect treatment connectivity
Collections → Treatment links compares historical relationships, inferred comparisons, and component structure. Search treatments with few recorded links, expand their existing connections and sources, or inspect connected groups. A bounded list of shared-topic pairs suggests where to look next; it creates no graph edges or research runs. An isolated treatment means no link is currently recorded within the selected collection, not that no meaningful relationship exists. The report also tracks assessed treatment pairs across collections and shows the percentage unassessed within a stated graph neighborhood. Negative and deferred reviews suppress repeat work; changed accounts or explicit reopening permit reassessment.
Research collection workflow
A collection is an ongoing question that can overlap other collections. Each bounded run has a purpose—initialization, expansion, focused research, or maintenance—and freezes its own parameters and input snapshots. A run can contain batches of graph targets and source documents.
Choose a collection and seed nodes. Traverse the active graph under a fixed neighborhood and target budget.
Check existing graph records, global drafts, and source captures. Fix root pages and allowed one-step links.
Fetch bounded source pages; save captures, versions, links followed, skips, and failures.
Read useful source sections. Draft supported treatments, components, ideas or people, including adjacent findings.
Resolve identities and initialize missing endpoints; review birth and death separately for each person, and check duplicate scope, citations, dates, and limitations.
Publish validated collection matches or save complete adjacent drafts. Record outcomes and exceptions.
Active graph
Validated, sourced additions join the atlas. They remain editorially unreviewed until assessed; a collection can include them without owning a separate copy.
Global draft library
Complete adjacent records wait outside the graph. Later collections can match and reuse them, subject to endpoint and publication checks.
The Collections page shows membership, plans, run receipts and generated stories. Coverage reports distinguish current treatments, reusable drafts and earlier assessments; treatment-link reports support bounded relationship review. A bounded run is complete when its saved batches have publication and validation receipts; the collection can remain open for another run. Earlier work is preserved as a historical snapshot, and old search proposals are identified separately from active plans. The Review page brings source-backed drafts together with records marked for removal and archived records. Individual record pages remain available from graph and source links.
How records are hidden and restored
Editorial assessment and visibility are separate. A reviewer can mark a node supported, uncertain, or removal-recommended with a reason. Pruning a removal recommendation hides that node and its incident edges from the active graph used for browsing and new research traversal. The source record, citations, historical receipts, and original edges remain saved.
The Review page’s Removed records tab can restore it. Restoration makes the node and its saved connections available again for review. A research collection’s saved historical membership is not erased, but its current active view excludes removed members.
Implementation status
Available Typed navigation, source and treatment records, dates, collection/run planning, a read-only collection dashboard, draft reuse checks, reversible editorial pruning, and source-linked story companions.
Still being built One seamless automated source-to-graph executor. Current source acquisition and publication use validated scripts and saved research artifacts, with interpretation and some source choices still guided through the research workflow. The historical Trinity backfill reports prior work; it is not proof that the new executor ran it.
Read the project guide →See the step-by-step research process →Explore the graph →Inspect collections →Review drafts and records →