INTENT — content-explorer

Project: content-explorer
Description: A user interface to dig into a field of interconnected knowledge. Inspired by onco.cc, the very well organized information collection about everything related to cancer.
Status: Initial intent v0.1 — 2026-10-12

Mission

Make any viable infospace navigable: a trustworthy, comprehensible, evolving map of a bounded field of knowledge, served through a consistent and delightful human interface regardless of the original sources, schemas or vocabulary.

Replicate the high-value user experience of OnCo through a maintainable, domain-independent adaptation of its MIT-licensed software, not through indiscriminate cloning of its cancer-specific content and pages. A connector projects heterogeneous sources into a canonical, provenance-aware information model. A dataspace manifest configures the presentation without new UI code.

Desired user experience

A visitor should be able to start with an accessible field overview; browse kinds/categories and filterable sortable lists; search across names, aliases and explanations; inspect a single knowledge object with plain-language summary first and deeper information below; explore typed incoming and outgoing relations; see evidence, freshness and uncertainty; and follow a reading path through related objects. Every aggregate is linked and every claim can identify its origin.

Core principles

Objects are addressable: stable qualified identifiers and one canonical page per object.
Connections are first-class: directed typed relations with meaningful labels and derived reverse navigation; no forced tree model.
Comprehension before expertise: TL;DR first, domain-specific technical depth progressively disclosed.
Evidence before authority: preserve primary and secondary sources, extraction lineage, review timestamps, uncertainty and conflicting claims.
Absence is data: distinguish missing, unknown, redacted, stale, contradicted, and explicitly not-applicable; never pretend completeness.
Domain vocabulary is declarative: type names, iconography, facets, widgets, reading paths and landing navigation come from manifests.
Ingestion is replaceable: no source adapter imports UI code; source identifiers, transformations, licensing and refresh policy remain inspectable.
Human and agent parity: static/exportable graph and HTTP/JSON first; MCP and richer interfaces later.
Composable and reversible: a connector can be replaced, a source re-imported, a dataspace switched and a projection replayed without editing the UI.
Reuse responsibly: preserve notices; distinguish MIT-licensed code from separately licensed data, images and brands.

In scope — first proof

Two clearly unrelated, tiny authored example dataspaces (systems thinking and software architecture).
Connectors: canonical JSON/NDJSON fixture loader and Markdown frontmatter importer; a third connector (GitHub issues or Wikidata) is a subsequent integration target.
Common types: Dataspace, Kind, Entity, Relation, Claim, Evidence/Source, Collection, Navigation and View metadata.
OnCo-inspired landing, faceted index, detail page, backlinks/related objects, search and basic graph view.
Build-time static projection and validation initially; dynamic refresh later.
Audit-grade import report with reproducible ingest inputs, warnings and provenance.

Explicit non-goals for v0.1

Copying the oncology corpus, OnCo logos, trademark, or bespoke medical decision-making widgets.
A universal ontology, automatic semantic truth determination, AI-generated citations, or unrestricted content scraping.
Distributed graph database, runtime multi-tenant writes, authentication or a production LLM answer service.
Pixel-perfect reproduction at the cost of accessibility, performance or domain independence.

Architectural wager

OnCo's generic browsing patterns are reusable, but its domain coupling is not fully abstract. Preserve reusable UX and presentation components where legally and technically feasible; insert a projection boundary between heterogeneous source records and the UI. Use profile-driven type and view registries, not oncology constants disguised as generic enums.

Proof of success / exit criteria

Switching dataspace via configuration alone shows another content field, without changes to UI component source.
Every visible entity and relationship uses canonical IDs and resolvable routes.
Every imported claim displays its source, observed/as-of date and rights metadata where present; unsupported assertions have explicit status.
Dangling references, duplicate identity conflicts and forbidden rights propagation block publication or produce explicit quarantine according to policy.
Full browse→filter→entity→neighbor→source journey works by mouse and keyboard.
Feature parity matrix demonstrates what was reused, generalized, omitted or deferred from OnCo.
Clean attribution and third-party notice report; regression tests and two-domain visual review.

Long-term vision

An ecosystem of self-improving dataspaces: connectors and human editors enrich a source-aware information substrate, quality and coverage measures reveal gaps, and a portable explorer offers reading, investigation, comparison and eventually questions/answers without becoming the system of record. The experience can be embedded in larger knowledge architectures through stable JSON, links and optional agent-facing protocols.

5.5 KiB Raw Permalink Blame History