A transparent record of decisions

OLN uses path-based URLs for Franchises and canonical entities. Subdomains are reserved for a future graduation escape hatch.

Canonical entities exist at two URLs serving two query intents — Network (who/what is X?) and Franchise (X in this world).

Most Franchises are singular; forking exists as a first-class explicit feature for irreconcilable governance disputes.

Network-level revenue flows through a transparent waterfall: infrastructure, R&D pool, Franchise distribution, then excess governed by Need-vs-Greed.

Contributors earn the right to choose their ad experience — Ad-Free or voluntary Patron Mode — through accumulated contribution.

OLN may only acquire if a clear, public before-and-after case shows the platform becomes structurally stronger — a constitutional commitment, not policy.

Open Lore Network is the entity, mission, and governance layer. LoreDoor is the consumer-facing brand the platform operates under.

HellaThis operates as a sister product under the OLN Foundation umbrella, powered by OLN's contributor-curated Fact graph.

Memory Alpha is the launch Franchise. Migration converts wikitext to structured Facts and OLN-native page templates — no wikitext support.

OLN has three first-class contribution surfaces, not two: Commons, Home, and LoreLines — the structured interpretive graph layer.

All governance Power — including founders — is earned through ongoing contribution. No member holds permanent governance authority by virtue of position alone.

Power earned in OLN is permanent but is measured as a share of total Power, which grows with the platform — sustained influence requires sustained contribution.

Voting locks Power for a cycle; conviction multipliers translate passion into influence. Locked Power returns regardless of outcome.

Constitutional changes require stacked supermajority + quadratic voting + cooldown + Foundation board ratification — a layered defense against capture.

Journal and Register entries will surface reader reactions (anonymous +1/-1 acceptable; registered viewers preferred). Registered users will accrue credits during the pre-launch phase, at a rate intentionally not pre-committed.

OLN will publish a public roadmap with project-level work items, allow viewers to signal priority, and treat completed work — including the founders' own time — as credit-earning at a single denomination. Calibration of credit values and the contribution-acceptance flow are deliberately open.

Imported facts enter the graph in a new provisional-imported lifecycle state — auto-attributed to provider, no Credits, open to validate/dispute on day one.

Four-tier citation strength model. Tier-1 and tier-2 citations modify v_for weight in the consensus formula; specific multipliers TBD.

HI-in-the-loop ingest layer with versioned mappings, multi-dimensional re-run, and drift detection. AI proposes; humans approve; code enforces — no auto-approval threshold.

LLMs handle offline batched cognitive work; classical ML handles runtime scoring; rules and code own everything load-bearing for governance. Embeddings bridge the layers.

Triple store on Postgres with Apache AGE and pgvector. Op log on Kafka/Redpanda. Composite views in Redis. OpenSearch for discovery. Neo4j deferred to read-replica only if traversal load demands.

Power-tiered rate limits, per-triple pending-op throttle, isolated provider ingestion queue, daily consensus pass as natural shock absorber. Real-time consensus carved out per-Fact via lifecycle state when needed.

theopenlore.net is the single primary domain. OLN is the neutral layer, Loredoor is the UGC product brand (not a domain), Atlas is the identity surface. Identity moves to its own repo, oln-atlas. data.* goes internal. Resolves G-031; supersedes the loredoor.com URLs in Entry 001.

Path-based URLs on one apex. Bare franchise slugs, canon as the default unmarked namespace, communities nested behind /g/, members at @handle, cross-franchise entities at reserved type paths like /person/. Extends and partly supersedes Entry 001.

Members are first-class graph entities with typed edges. Customization is a four-rung cascade, declarative-only on the main origin; raw HTML/widgets live on an isolated, PSL-listed user-content domain. Isolate by (trust x indexability).

Atlas profiles carry a linktree of external links. Ties to a franchise, entity, or Fact are optional-but-rewarded and must be proven about their target (AI-first, weighted). Verified ties enrich entity pages; links are admesh-ready but monetized later, per creator.

OLN's community wiki displaces Fandom by reimagining unstructured articles as Fact-grounded compositions over the Entity graph — adaptive to consume, owned by contributors. A gripes audit shows OLN's design already answers most top Fandom complaints; the wedge is the reimagined article/world-model, and the moat to crack is discoverability via AEO.

Resolves how AEO actually drives traffic and how OLN treats AI crawlers. AEO wins citation, not visits — so OLN plays two boards: be the most-cited source (supply) and own the world-model experiences a chat answer can't collapse (destination), with rights-holder partnerships breaking incumbency. And "AI bots" is split by class: allow citation crawlers (they attribute), block training crawlers (they launder contributor labor), and hold the structured graph — not the CC-BY-SA prose — in reserve as the licensable asset.

Making structured contribution fun is the lever on G-057's make-or-break authoring friction. The decided principle: gamify accuracy, resolution, and completeness — never raw volume — so the fun mechanic is the data-quality flywheel, not its enemy. Lean on native graph mechanics over bolted-on points, put intrinsic impact ahead of extrinsic credits, and ship every component behind the Amplitude Experiment harness paired with a quality guardrail metric.

A system-wide gap analysis found the shipped product is a civic/governance doc-site whose primitives (annotations, reactions, profiles, event tracking) are embryos of decided-but-unbuilt product surfaces. The decision: bridge to product by growing those embryos rather than greenfielding, and build the connective tissue several decided systems silently assume — a unified Fact review/confirm queue, notifications, prose↔Fact linking, search signals, the experiment harness, and payment rails.

Community source nomination (G-066) targets curated, authoritative sources (e.g. starwars.com for a Star Wars franchise), not random sites — and every source is governed by a fail-closed compliance posture: pull and archive its terms before ingest, monitor those terms daily, and auto-pause ingest the moment they change. A "No Gotchas" environment where compliance is continuous, not a one-time check.

Research found Memory Alpha is licensed CC-BY-NC 4.0 (non-commercial), which an ad-supported OLN cannot bulk-rehost. Decision: do not reuse the NC text commercially. Instead recruit the Memory Alpha community to OLN and rebuild Star Trek canon natively under OLN's license, using Memory Alpha only as a human-verified reference signal — bring the people, not the bytes.

The entity page rendered almost nothing for non-character types — a film showed one Performer and dropped its director, composer, box office, runtime, and cast — because the renderer's fact vocabulary (which keys are valid, their labels, their aliases) was hardcoded in page code and character-shaped. The facts existed in the data the whole time; the UI silently discarded any key it didn't recognize. Decision: the per-type schema registry is the single source of truth for valid predicates, labels, aliases, and display groups; the renderer derives its vocabulary from it (in-memory cached, reloads on change, read from Firestore). Adding a new fact type is now a registry edit, not a code change and deploy.

There are two economic layers, not one. The network layer — the clean, centralized canonical pages built now — carries ads whose revenue goes to a network fund that must cover all network expenses on its own before anything is asked of communities. Communities (the modernized wiki-farm layer) carry their own ads into their own treasury; the network draws a percentage from those treasuries only as a backstop, when its own fund falls short. A revolving slush fund seeds new communities with repayable advances. Exact percentages and the recalculation engine are deferred until real per-franchise numbers exist.

The CMS for Content and Community pages is a block editor you edit in place (TipTap / ProseMirror), not a WYSIWYG box and never wikitext. ProseMirror JSON is the source of truth; MDX is only a safe interchange format for export and diffing and is never compiled or evaluated. Pages render through a component allowlist, so user content cannot execute. Launch is presence + a soft single-writer lock, not real-time co-editing. Plugins and themes are marketplaces over a no-user-code substrate: a plugin configures an OLN-reviewed component, a theme is validated design tokens — arbitrary JavaScript waits for a sandbox built when demand proves it out.

Extraction is fail-closed on type confidence. Before the pipeline pulls facts from an entity, it must be confident what *kind* of thing the entity is — and that confidence is highest when independent signals agree (a Wikidata type and the source infobox say the same thing). Entities whose type is corroborated clear the bar and auto-extract; entities with a single weak signal, or a conflict, wait for human validation rather than polluting the graph with mis-typed facts. Quality is enforced by code at the gate; humans supply judgment only where the signal is genuinely ambiguous.