Skip to content

PartlyGood content taxonomy

Every content piece lives in docs/<field>/<slug>.md with YAML front matter. This document is the single source of truth for fields, content types, status model, naming conventions, cross-cutting topics, and topic governance.

Machine-readable taxonomy and change history live at research/taxonomy/taxonomy-v<version>.json (currently v1.1).

Fields (directory structure)

Field Directory Scope
Management management Leading teams, planning, delegation, meetings, performance, change
Operations operations Processes, workflows, vendors, analytics, project delivery
Finance finance Budgeting, FP&A, reporting, forecasting, close processes
Marketing marketing Content, campaigns, customer research, analytics
Cross-cutting cross-cutting AI workflows shared across fields (prompt frameworks, evaluation, data, vendors, security, tooling patterns)

Software engineering / coding is intentionally EXCLUDED. Educational material that happens to mention code (e.g., an AI tool suggestion) must stay at the level a manager can use — no build instructions for software.

Cross-cutting dimension (v1.1)

The ai.* topics are a cross-cutting dimension: they apply to every field and can be tagged on any piece regardless of its primary field. The canonical list lives in the cross_cutting.topics section of the taxonomy file. They are exposed on the Cross-cutting index.

Cross-cutting topic Scope
ai.prompting Framing prompts for business tasks
ai.agents Using agents / multi-step AI workflows
ai.automation Automating recurring workflows
ai.evaluation Verifying and evaluating AI output
ai.safety Safe defaults, failure handling, human review
ai.adoption Getting teams to adopt AI
ai.data Data preparation & hygiene (new in v1.1)
ai.documents Document drafting & operations (new)
ai.reporting Reporting & dashboards (new)
ai.vendors AI tool & vendor selection (new)
ai.security Privacy, permissions & data handling (new)

Content types

  • playbook — a reproducible step-by-step procedure for a recurring workflow, with inputs, outputs, quality checks, and a "run it" section.
  • how-to — a focused single-workflow guide (shorter than a playbook).
  • explainer — foundation knowledge about using AI in a field (concepts, options, evaluation).
  • canonical — house documents (taxonomy, editorial standards) rather than field content.

Status model

draftin_reviewpublishedupdatedretired

Retired pieces move under docs/_retired/ and keep their slug to avoid link rot; the front matter status: retired is set.

Front matter (required for every piece)

title: <Human title>
type: playbook | how-to | explainer | canonical
field: management | operations | finance | marketing | cross-cutting
status: draft | in_review | published | updated | retired
created: YYYY-MM-DD
updated: YYYY-MM-DD
slug: <kebab-case>
product: false            # true when produced by the pipeline itself
quality:
  checked_at: YYYY-MM-DDTHH:MM:SSZ
  checks_passed: 8        # must equal the checks that ran

Additional keys allowed: summary, tags, audience, difficulty (beginner|intermediate|advanced), reviewers, sources, and topics (array of topic ids) for pieces that want to declare granular topics.

Topic governance

Who adds/merges topics, versioning, and the update process are documented in research/taxonomy/GOVERNANCE.md. Summary:

  • Additive only in MINOR releases. New topics may be proposed by any publisher/engineer via an issue; the Taxonomy owner (Founding Engineer) approves; the CEO owns editorial intent and may veto.
  • Breaking changes (remove/rename/merge) are MAJOR releases that require re-tagging published pieces and a published notice first.
  • Versioning is SemVer-style: each change ships as a new taxonomy-v<MAJOR.MINOR>.json, and the previous file stays on disk.
  • Backward compatibility is enforced by research/scripts/taxonomy_compat.py, which fails the build if any existing topic id is dropped.

Sample playbook structure (per PAR-4 template)

The canonical sample playbook lives at docs/samples/. All field playbooks follow that structure so the pipeline can validate them mechanically.