Editorial philosophy
Every article in the PointSav wikis is a learning resource, not a reference lookup. A reader who finishes an article should understand the subject — not just have retrieved a fact. Wikipedia is the model: structured, internally linked, deep enough to build genuine understanding, and consistent enough that a reader who knows Wikipedia recognises the layout immediately and knows how to navigate it. This article explains what that means in practice and how it shapes every editorial decision.
The Wikipedia model
Wikipedia succeeds at something that most documentation does not: it teaches. A reader who arrives looking up "direct-hold property structure" or "AI inference service" or "co-location mandate" leaves with a mental model, not just a definition. That happens because Wikipedia articles are built around understanding, not retrieval.
Three structural choices produce this outcome:
1. The encyclopedic lead. The first paragraph of a Wikipedia article contains the most important facts about the subject — in plain language, at the right level of abstraction for a general reader. The lead does not assume prior knowledge. It does not start with a definition. It starts with consequence: why this thing matters, what it does, what the reader gains by understanding it.
2. Internal linking that builds webs of understanding. Articles link to each other using [[slug]] wikilinks. A reader who follows the links progressively builds richer knowledge of the subject. Red links — links to articles that do not yet exist — are features, not defects. They show the reader where the encyclopedia is incomplete and invite contribution.
3. Consistent structure. Lead paragraph → body sections → See also → References. This structure is Wikipedia's muscle memory. A reader who has used Wikipedia for ten minutes recognises it. The recognition is itself a form of trust — the reader knows how to use the article before they read it.
The PointSav wikis apply this model to three distinct audiences in three distinct language registers. The structure is the same. The register shifts to match the reader.
The encyclopedic test
Every article should pass one test:
Does a reader who finishes this article understand the subject, or have they only found a fact?
A fact-retrieval article answers "what is X." An encyclopedic article answers "what is X, why does it matter, how does it work, and how does it connect to Y and Z." The second article is a learning resource. The first is a glossary entry.
Worked examples across the three wikis
This test applies to all three wikis:
- Corporate: After reading
direct-hold-structures, a banker understands why Woodfine uses this structure, what it means for capital allocation, and how it differs from pooled structures — not just what the term means. - Projects: After reading
co-location-mandate, a developer or architect understands the logic behind the mandate, the capital framework that validates it, and the market conditions that make it viable. - Documentation: After reading
service-slm, an engineer understands the routing logic, why the tier thresholds are set the way they are, and what the consequence is for the operator — and an institutional reader scanning section headers understands that the platform manages AI costs automatically without sending requests off-premises.
If an article fails this test, the rewrite adds the relationships, context, and consequence framing that turn a fact into understanding.
DataGraph-enriched content
The PointSav wikis are connected to a property graph database — the DataGraph — that accumulates knowledge about every entity in the platform: what each entity connects to, which domain it belongs to, which themes span it, and what the research corpus says about it.
The wiki articles were authored before the DataGraph was fully populated. The DataGraph now knows things that many articles do not yet express. A DataGraph-informed rewrite adds that knowledge to the article:
- What connects to what. An article about
service-slmthat does not mention the three compute tiers, the access-control gateway it transits, or the audit log it produces is incomplete — not because those facts are missing, but because the reader cannot build the mental model without them. - Why the institutional reader should care. The consequence framing — the sentence that connects the technical mechanism to the business outcome — is often the piece that articles written for an engineering audience omit. "The routing logic is operator-controlled" is a fact. "A request that resolves locally never leaves the customer's infrastructure — and never appears on a cloud billing statement" is a consequence that builds understanding.
- Domain and theme context. An article that explains what a service does without explaining where it sits in the platform architecture leaves the reader with a fact and no map. Domain and theme connections are how the reader builds the map.
Register and substance as the two inputs
The two inputs to every article:
| Input | Source | What it provides |
|---|---|---|
| How to write | RESEARCH corpus → language tokens | Register, sentence structure, vocabulary, consequence-first lead |
| What to say | DataGraph — entities, relationships, domains, themes | Content the article was written before the DataGraph existed to supply |
The language tokens govern register. The DataGraph governs substance. An article that has correct register but thin substance fails the encyclopedic test. An article that has rich substance but wrong register fails to communicate. Both inputs are required.
How articles actually improve
Article quality improves through editorial sessions, not an automated pipeline. An editor queries the DataGraph for the article's subject, reads the real source the article describes, and rewrites the article to match — dropping claims the source doesn't support and adding facts the DataGraph or the source surfaces that the article was missing. There is no scheduled sweep that regenerates drafts on a cadence; a category is rewritten when an editorial pass reaches it.
Register correctness is checked mechanically, not by human judgment alone: a linter scores each draft against the vocabulary and structural rules the language tokens define, and a draft with a register violation does not pass review. Substance correctness — whether the article's claims match what the platform actually does — is checked by reading the real source directly, the same discipline as any fact-check.
Why this produces compounding value anyway
Even without an automated generation loop, each editorial pass compounds: the DataGraph a later pass queries is richer than the one an earlier pass saw, and a pass that finds a defect in one article often finds the same defect's pattern repeated in a sibling article, correcting several at once. The improvement comes from the DataGraph and the source code getting richer and more current over time, and from each editorial pass building on what the last one already fixed — not from an inference system training on prior verdicts.
See also
- Editorial language registers — the three registers and vocabulary rules that govern how each wiki is written
- PointSav encyclopedia — glossary and lexicon — canonical term definitions