Seed taxonomy as SMB bootstrap

@@ -10,31 +10,29 @@ index_group: platform-mechanics
short_description: "Every tenant deployment provisions a four-part seed taxonomy — Archetypes, Chart of Accounts, Domains, Themes — as the knowledge graph bootstrap."
status: active
bcsc_class: public-disclosure-safe
last_edited: 2026-05-01
last_edited: 2026-08-22
editor: pointsav-engineering
cites: []
paired_with: seed-taxonomy-as-smb-bootstrap.es.md
---

Every tenant deployment begins with a **seed taxonomy**: a compact, hand-tunable four-part structure that forms the initial scaffold of the per-tenant [[knowledge-graph-grounded-apprenticeship|knowledge graph]]. The four parts are Archetypes, Chart of Accounts, Domains, and Themes. Each entity carries gravity keywords — explainable keyword anchors that drive [[service-content|classification of incoming content]].
Every tenant deployment begins with a **seed taxonomy**: a compact, hand-tunable four-part structure that forms the initial scaffold of the per-tenant [[knowledge-graph-grounded-apprenticeship|knowledge graph]]. The four parts are Archetypes, Chart of Accounts, Domains, and Themes. Each entity carries gravity keywords — explainable reference text an operator reads while classifying new content into [[service-content|the taxonomy]].

## The four parts

**Archetypes — who acts.** Five to seven role-by-cognitive-pattern identities. The default carries five universal roles that appear in some form in every business regardless of industry: the Executive (Strategic Direction), the Guardian (Risk and Compliance), the Fiduciary (Resource Integrity), the Architect (System Design), and the Constructor (Physical Realization). Industry-specific roles are added via [[vertical-seed-packs-marketplace]].
**Archetypes — who acts.** Eleven role-by-cognitive-pattern identities, each with a name, a signature (its core function), a "healing trigger" (the failure mode the role exists to catch), and gravity keywords — for example the Executive (Strategic Direction, triggered by Stagnation) or the Guardian (Risk and Compliance, triggered by Breach). The set spans strategic, operational, and physical-realization roles. Industry-specific roles are added via [[vertical-seed-packs-marketplace]].

**Chart of Accounts — what business this is.** Five to ten industry-specific business profiles, each with an identifier, a profile name, a sub-domain, and gravity keywords. The profiles mirror the business's actual revenue and expense categories. A property development firm's Chart of Accounts looks different from a law firm's or a regional hospital's; the form is consistent while the substance is industry-specific.
**Chart of Accounts — who holds what role, not what business this is.** Despite the name, this is not an industry or revenue/expense classification — it is a personnel-role taxonomy, organized into sub-domains (personal governance roles, compliance, finance, and several others), each entry carrying a reference number, a role type, and gravity keywords for classifying a person's title or function against it.

**Domains — macro categories of work.** Three to five macro categories that group work units. The default provides Corporate, Projects, and Documentation. Each Domain holds a Glossary and Topic collection that structure the wiki content, composed through the editorial pipeline.
**Domains — macro categories of work.** Macro categories that group work units, each backed by its own seed file. The default provides Corporate, Projects, and Documentation. Each Domain holds a Glossary and Topic collection that structure the wiki content, composed through the editorial pipeline.

**Themes — time-bound initiatives.** Four to ten active themes representing current strategic focus. Each Theme carries an identifier, a short name, and gravity keywords. Themes are the most volatile part of the taxonomy — they may change quarterly as initiatives open and close.
**Themes — time-bound initiatives.** Active themes representing current strategic focus, each with an identifier, a name, a scope (tactical or strategic), a one-line thesis, and gravity keywords. Themes are the most volatile part of the taxonomy and are genuinely tenant-specific — real deployments seed real, often commercially sensitive strategic priorities here, not a generic template set.

## The gravity_keywords mechanism
## The gravity_keywords field

Classification of new content into the taxonomy uses keyword matching rather than embedding similarity. When a document arrives, the [[service-extraction|extraction service]] counts keyword matches against each Archetype, Chart of Accounts profile, Domain, and Theme entity. The entity with the highest match count is the proposed classification.
Every taxonomy entity carries gravity keywords, but their role today is narrower than an automated matcher: they are reference text a human operator reads before making a classification decision, not an input to an automated match-counting or embedding-similarity system. In the platform's one confirmed real consumer — the [[radical-proofreader-ui|Verification Surveyor]] workflow — an operator reviewing a discovered entity is shown the list of Archetype names and picks one manually; the gravity keywords guide that human judgment rather than driving it programmatically.

This approach is deliberate. Keyword-based classification is explainable: "this document was classified as Real Estate because the terms Leasing, Office, and Industrial matched" is a statement an operator can read and verify. An embedding similarity score is not. Keyword classification is auditable — the classification path is reproducible across model versions, which matters for regulatory record-keeping. And it is hand-tunable: an operator who sees a misclassification edits the gravity keywords list. Retraining an embedding model is not a practical operation for a small business.

Embedding similarity is provided at a separate layer to augment keyword classification when explicit keywords miss. It does not replace the keyword mechanism.
The explainability goal the field is designed for still holds even under manual selection: an operator can read a taxonomy entity's gravity keywords and understand why an entity would or wouldn't belong to it, and can hand-edit the keyword list when a category needs to be redefined. Whether an automated keyword-match or embedding-based classifier is added on top of this manual step is a design the platform has not yet built.

## Provisioning a new tenant