Below is KDOS Batch Generation Workflow v1.0 — designed to be:

• Safe (no repo pollution)
• Deterministic (same inputs → same shape)
• Scalable (10 → 10,000 pages)
• AI-agnostic (works with any model)
• Git-native (no magic)

This assumes manual or semi-automated execution first, with automation later.

KDOS Batch Generation Workflow v1.0

0. Mental model (lock this in)

Topics → Prompt Packs → Markdown Artifacts → Validation → Commit

Nothing skips steps. Nothing writes directly to main.

1. The three core artifacts

KDOS batch generation revolves around three files:

1. Topic Backlog (what to generate)
2. Prompt Pack (how to generate it)
3. Generated Markdown (the output)

Each is versioned independently.

2. Topic Backlog (source of truth)

2.1 Format: topics.toml

Create a file like:

kdos/
└── generation/
    └── topics/
        └── statistics-v1.toml

Example:

version = "statistics-v1"
domain = "statistics"
difficulty = "intermediate"

[[topic]]
title = "Bayesian Updating"
slug = "bayesian-updating"
tags = [
  "bayesian-inference",
  "bayes-theorem",
  "belief-revision",
  "posterior-probability",
  "evidence",
  "decision-theory"
]

[[topic]]
title = "Maximum Likelihood Estimation"
slug = "maximum-likelihood-estimation"
tags = [
  "parameter-estimation",
  "likelihood-function",
  "statistical-inference",
  "optimization",
  "frequentist-methods"
]

Rules

• One backlog per domain × difficulty × batch
• Backlogs are append-only
• Never edit a topic once generated (new versions get new backlogs)

3. Prompt Pack (generation contract)

3.1 Prompt pack structure

kdos/
└── generation/
    └── prompts/
        └── kdos-md-v1.0.txt

This is your canonical generator.

You already have this — but now we parameterize it.

3.2 Prompt pack variables

When generating, substitute:

This can be done:

• Manually
• With a script
• With an AI tool that supports variables

4. Generation output staging (never write to content directly)

4.1 Staging directory

kdos/
└── generation/
    └── output/
        └── statistics-v1/
            ├── bayesian-updating.md
            ├── maximum-likelihood-estimation.md

Nothing goes into content/ yet.

This is a quarantine zone.

5. Validation gate (mandatory)

Before anything touches content/:

5.1 Validation checklist

For each file:

• ✅ Front matter schema valid
• ✅ Slug matches filename
• ✅ 1–3 valid domains
• ✅ 5–12 valid tags
• ✅ Section skeleton exact & ordered
• ✅ No empty sections

You can:

• Manually skim early batches
• Later use a lint script

Fail fast. Fix at source. Regenerate.

6. Promotion to content tree

Only after validation:

content/
└── knowledge/
    └── statistics/
        ├── bayesian-updating.md
        ├── maximum-likelihood-estimation.md

Rules

• One commit per batch

• Commit message includes:

  KDOS: statistics v1 (2 entries)
  

This keeps history clean and auditable.

7. Commit & deploy cadence (important)

Recommended cadence

Cloudflare Pages handles deploy automatically.

8. Versioning strategy (crucial)

8.1 Prompt versions

When you change:

• Section wording
• Style rules
• Taxonomy rules

→ bump:

ai_prompt_version = "kdos-md-v1.1"

Never regenerate old content silently.

8.2 Content revision

If you improve an entry:

• Update last_updated
• Change status → revised
• Keep original date_created

9. Failure handling (this matters)

If output fails validation:

• ❌ Do not edit by hand (except trivial fixes)
• Fix the prompt or topic definition
• Regenerate cleanly

This prevents slow drift.

10. What this workflow unlocks

You can now:

• Generate entire domains safely
• Parallelize work
• Rebuild KDOS from scratch if needed
• Feed KDOS back into AI systems later
• Sleep at night knowing the archive is clean

You’ve essentially built:

A static, versioned, machine-curated encyclopedia