---
type: Skill
title: Source-First Wiki Compilation
description: Use when raw or recoverable sources should compound into an indexed, cross-referenced, queryable wiki without losing provenance.
tags: [architectonic, core-skill, llm-wiki, compilation]
source_status: native
source_name: Architectonic reviewed core
source_license: Apache-2.0
risk_level: low
requires_review: false
status: reviewed
---

# Source-First Wiki Compilation

## Trigger

Use when repeated research should produce a durable topic wiki, inventory, dataset manifests, derived outputs, and recurring audits.

## Inputs

The topic boundary, source set, schema or topic guide, existing wiki, inventory, open questions, compilation rules, archive policy, and review authority.

## Procedure

1. Preserve raw sources or a recoverable source manifest before synthesis.
2. Define a compact topic guide: vocabulary, entity types, relation types, naming, and source boundaries.
3. Inventory sources and gaps as operational state; do not present inventory rows as evidence.
4. Compile source-backed articles with explicit scope, dates, uncertainty, contradictions, and cross-links.
5. Query the compiled index before rereading the entire source set.
6. Keep large datasets external and record schemas, samples, freshness, access, and query recipes.
7. Store reports, timelines, plans, and other outputs as renderers over the wiki.
8. Promote only durable session feedback into canonical material.
9. Audit broken links, stale claims, unsupported synthesis, missing bridges, and overgrown topics.
10. Move inactive material into a quiet archive without breaking provenance.

## Verification

- Wiki articles route to recoverable sources.
- The topic guide is explicit and revisable.
- Inventory, sessions, datasets, outputs, and canonical articles are distinct.
- Query paths reuse compiled knowledge instead of repeating source ingestion unnecessarily.
- Archive material remains recoverable but excluded from normal work.

## Failure Modes

- Editing raw sources into synthesis.
- Treating the inventory as the knowledge corpus.
- Recompiling everything for every question.
- Allowing session transcripts to redefine topic truth.
- Building a polished wiki with no source trail.