Apex Evolve

Build and maintain apex articles through creation, evolution, and length management.

When to Use

• Creating a new apex article from the approved list
• Evolving an existing apex article when source content has changed
• Checking which apex articles need attention

Invocation

/apex-evolve create "Article Title"
/apex-evolve evolve [article-slug]
/apex-evolve check

Apex articles come in two shapes: synthesis (the default — weaving sources together into a coherent narrative) and applied (decision-oriented — given the positions the Map holds, what follows for a specific real-world context). The shape is recorded in the frontmatter field apex_type: synthesis | applied (default synthesis if absent) and listed in obsidian/apex/apex-articles.md. The 10-step evolution process below applies to both; the Applied Discipline section below adds extra requirements for applied apex.

Modes

Mode 1: create

Generate a new apex article from scratch.

Input: Article title from the approved list in obsidian/apex/apex-articles.md

Process:

1. Look up the article in the master list to get source articles and thesis
2. Check for slug collisions before creating:

   uv run python scripts/check_slug.py <slug> apex
   

   If the check reports a collision, choose a different filename.
3. Read all source articles to understand the material
4. Read obsidian/project/writing-style.md for apex article guidelines
5. Synthesize into a narrative that weaves sources together
6. Include the required "Evidence and Dependency" section (see below) classifying the synthesis's main lines of support
7. Target length: 2500-4000 words
8. Create with required frontmatter (see below)

Output: New apex article at obsidian/apex/[slug].md

Mode 2: evolve

Update an existing apex article when sources have changed.

Input: Specific article slug, or omit to auto-select stalest

Process: See 10-step evolution process below

Output: Improved article + review archive

Mode 3: check

Report which apex articles need evolution.

Process:

1. Read all apex articles and their apex_sources frontmatter
2. For each source, check its modified date against the apex article's effective baseline max(apex_last_synthesis, last_deep_review) (see Step 1 for why — /deep-review updates last_deep_review but not apex_last_synthesis, so the bare-field check over-reports drift)
3. Report articles with stale sources, ordered by priority

Output: List of articles needing attention with staleness scores

---

Frontmatter Schema

Apex articles use standard frontmatter plus apex-specific fields:

---
title: "Article Title"
created: 2026-01-24
modified: 2026-01-24
human_modified: null
ai_modified: 2026-01-24T00:00:00+00:00
draft: false
topics: []
concepts: []
related_articles: []

ai_contribution: 100
author: null
ai_system: claude-opus-4-5-20251101
ai_generated_date: 2026-01-24
last_curated: null

# Apex-specific fields
apex_type: synthesis  # or "applied" for decision-oriented apex
apex_sources:
  - topics/free-will
  - topics/agent-causation
  - concepts/mental-causation
apex_last_synthesis: 2026-01-24T00:00:00+00:00
apex_thesis: "One-sentence thesis statement"

# Applied-apex only:
apex_decision_context: "AI consciousness assessment"  # what real-world domain this informs
apex_positions_cited:  # the positions the decision turns on; ≥3 required
  - P-Q1
  - P-Q2
  - P-Q9
---

---

Evolution Process (10 Steps)

Step 1: Select Article

If article specified, use it. Otherwise, auto-select:

1. Read all apex articles in obsidian/apex/
2. For each, compute the effective synthesis baseline: baseline = max(apex_last_synthesis, last_deep_review) This is the fix for the stale-apex_last_synthesis drift bug — /deep-review touches an apex's body and updates last_deep_review but does not update apex_last_synthesis, so a naïve scorer keeps re-nominating apex articles whose content has already absorbed every source change via the deep-review. Using the max of the two timestamps treats a recent deep-review as a successful synthesis for staleness purposes.
3. Calculate staleness score:
   • Count sources modified after baseline
   • Score = days_since_baseline × changed_source_count
4. Select highest-scoring article. If the top candidate's baseline is within the last 7 days, declare a no-op (the work is already done) and exit cleanly without picking a target.

Step 2: Identify Changed Sources

Compare each source's modified date to the effective baseline max(apex_last_synthesis, last_deep_review). List all changed sources. (Same fix as Step 1 — guards against the stale-apex_last_synthesis drift artifact where a recent deep-review already absorbed the source updates.)

Step 3: Read Changed Sources

Read the full content of each changed source to understand new material.

Step 4: Pessimistic Review

Apply three critical personas:

Clarity Critic: Where is the prose unclear, jargon-heavy, or hard to follow? Where does the narrative lose the reader?

Redundancy Hunter: What passages repeat information? Where does the article say the same thing twice? What can be cut without losing meaning?

Narrative Flow Analyst: Where does the argument structure break down? Are transitions smooth? Does the piece build to its conclusion?

Step 5: Optimistic Review

Apply three supportive personas:

Connection Finder: What new connections from changed sources should be woven in? What synthesis opportunities exist?

Synthesis Strengthener: Where can the article better show how pieces fit together? Where is the "so what" unclear?

Human Reader Advocate: What would make this more engaging to read? Where does it feel like a reference rather than a narrative?

Step 6: Length Assessment

1. Count current word length
2. Compare to target range: 2500-4000 words
3. Note whether condensation or expansion is needed

Step 7: Apply Improvements

Always do:

• Fix issues identified by pessimistic review
• Integrate insights from changed sources
• Strengthen synthesis per optimistic review
• Install the "Evidence and Dependency" section if absent; refresh it if the synthesis's lines of support changed (see "Evidence and Dependency Section" below)

If over 4000 words (condensation required):

• Remove redundant passages identified by Redundancy Hunter
• Tighten prose: eliminate filler words, combine sentences
• Move detailed explanations to source articles; replace with links
• Remove sections that duplicate what sources already cover well
• Preserve the narrative arc while reducing word count

If under 2500 words (expansion needed):

• Expand synthesis sections showing connections between sources
• Add transitions that guide the reader through the argument
• Deepen the narrative arc with richer examples

Step 8: Update Frontmatter

ai_modified: [current UTC timestamp]
apex_last_synthesis: [current UTC timestamp]

Adjust ai_contribution if human edits were preserved.

Step 9: Create Review Archive

Create obsidian/reviews/apex-evolve-[date]-[slug].md containing:

• Article reviewed
• Changed sources identified
• Pessimistic review findings (3 personas)
• Optimistic review findings (3 personas)
• Length assessment (before/after word counts)
• Summary of changes made

Step 10: Log and Commit

1. Prepend to obsidian/workflow/changelog.md (add immediately after frontmatter, before existing entries):

   ## [current time from prompt] - apex-evolve
   - **Status**: Complete
   - **Article**: [[article path without .md extension]]
   - **Changed sources**: [count]
   - **Word count**: [before] → [after]
   - **Review**: [[reviews/apex-evolve-date-slug]]
   

2. Do not create a git commit. Leave file changes uncommitted on disk — the orchestrator (cycle_post.py for /loop runs, agent-commit for legacy evolve_loop.py runs) commits afterward. For manual invocations, run /agent-commit or commit by hand.

---

Article Structure

Apex articles follow this structure:

# [Title]

[Opening paragraph stating the integrated thesis - what this synthesis argues]

## [Narrative Section 1]

[Build the argument, drawing from multiple sources]
[Link to sources: "As explored in [[source-article]]..."]

## [Narrative Section 2]

[Continue building, showing connections between ideas]

## [Additional sections as needed]

## Synthesis

[Show how the pieces fit together in a way individual articles don't]
[This is the unique value of the apex article]

## Evidence and Dependency

[Required — see "Evidence and Dependency Section" below. A brief reader-facing
ledger classifying the synthesis's main lines of support.]

## Relation to Site Perspective

[Connect to tenets - required for all articles]

## Source Articles

This apex article synthesizes:
- [[source-1|Source 1 Title]]
- [[source-2|Source 2 Title]]
- ...

---

Evidence and Dependency Section (required)

Every apex article carries a short reader-facing section, ## Evidence and Dependency, between the Synthesis section and Relation to Site Perspective. Adopted 2026-07-16 from an outer-review recommendation (ChatGPT 5.6 Pro full-site audit, rec #29); convergent with the methodology register's independence discipline.

Why: apex pieces are the corpus's most vulnerable point for confidence amplification. When agency, altered states, quantum interpretation, subjecthood, and value all draw on the same five tenets, their convergence is one premise package producing mutually compatible outputs — not five independent reasons for those tenets. The ledger makes the dependency structure explicit so the synthesis cannot silently count coherence as evidence.

Content: classify each of the article's main lines of support as one of:

• Externally evidenced — rests on cited empirical work or external scholarship that holds independently of the Map's framework
• Independently argued — a philosophical argument that does not presuppose the tenets (state which premises it does need)
• Inherited from a tenet — follows from one or more tenets; name them
• Inherited from another synthesis — imported from another apex piece; name it so a reader can trace the chain (and so double-counting across apex articles is visible)
• Mutually coherent only — fits the rest of the framework but carries no independent weight; coherence is all it contributes

Form: natural prose or a compact list, under ~200 words. Example register: "The altered-states line is inherited from Tenets 1–3 plus the filter reading; the dissociation studies it cites are externally evidenced but do not by themselves discriminate filter from production accounts. The agency argument is independently argued, though it presupposes that reasons-responsiveness is not exhausted by physical description." Do not use the phrase "apex article" inside it (media-neutral rule) — "this synthesis" works. Do not present the five category names as a labelled scoreboard; weave them into readable prose.

Retrofit: create includes the section from the start; evolve installs it if absent and refreshes it whenever the support lines change.

---

Length Management Guidelines

Target: 2500-4000 words

Why this range:

• Under 2500: Not enough synthesis to justify an apex article
• Over 4000: Becomes unwieldy for human readers; defeats the purpose

Condensation techniques:

• "As [[source-article]] explores in detail..." (link instead of repeat)
• Combine related paragraphs
• Remove hedging language that adds words without adding meaning
• Cut examples that illustrate the same point

Expansion techniques:

• "This connects to [[other-source]] because..." (show synthesis)
• Add transitional paragraphs between major sections
• Deepen the "why this matters" throughout

---

Cross-Review Integration

When other skills create or modify content that's an apex source:

1. After /expand-topic or /research-topic completes
2. Check if the new/modified article is in any apex article's apex_sources
3. If yes, add to obsidian/workflow/todo.md:

   - [ ] P2 apex-evolve: [apex-article-slug] — source [modified-article] updated
   

This ensures apex articles stay current without manual tracking.

---

Important

• Approved subjects only: Only create apex articles from the master list in obsidian/apex/apex-articles.md
• Human-first: Unlike topics/concepts, apex articles prioritize narrative flow over LLM-atomic structure
• Media-neutral language: Never use the phrase "apex article" in the article content itself. Content may be reproduced in contexts where our internal terminology is unknown. Write as standalone prose.
• Condense actively: Evolution must maintain length discipline; don't let articles grow unbounded
• Link extensively: Apex articles are synthesis—they should link heavily to source articles
• Preserve thesis: Each apex article has a thesis; edits should strengthen it, not dilute it
• Show the dependency structure: The "Evidence and Dependency" section is required on every apex article — convergence among tenet-derived lines is not independent confirmation, and the ledger keeps that honest

---

Applied Discipline (for apex_type: applied articles)

Applied apex articles take a decision-relevant context (assessing AI consciousness, clinical-interface ethics, personal philosophy under the Map, etc.) and produce a verdict the Map's framework supports. They are subject to extra discipline beyond synthesis apex:

1. Cite ≥3 positions. The apex_positions_cited frontmatter field must list at least three position IDs from obsidian/positions/. The body should reference them explicitly: "On position P-Q1, the Map holds…" or similar. Without explicit position citation, an applied apex is just an opinion piece.

2. Surface confidence. When the article's verdict turns on a position, surface that position's confidence band in-prose, not buried in the frontmatter. A reader should be able to see what the verdict turns on and how confidently. Example: "This verdict depends on P-Q9 (moderate confidence) — were the self-concealing-interface argument to weaken, the implication for AI assessment shifts."

3. "What this implies for decisions" section. Required, named exactly that. Two-to-five concrete implications, each stated as a recommendation or evaluation the reader could act on or apply. Not a summary; an output.

4. Name the decision context. The apex_decision_context frontmatter field is required. The opening paragraph should name the decision context explicitly: "This piece addresses how to assess putative consciousness in current AI systems, given the Map's positions on…"

5. Honest verdict scope. Distinguish what the Map's positions actually license from what they merely permit. An applied apex must not over-extend the Map's commitments. If the positions support only a partial verdict, say so; if they leave a decision underdetermined, name the residual uncertainty.

6. Cascade tagged. When an applied apex relies on positions whose confidence might shift, note in the body: "If position P-XN is retired or downgraded, this verdict on [aspect] would need re-evaluation." This makes the applied piece self-flagging when positions change.

Applied apex articles are deliberately rare — the goal is 8–15 across the corpus, not 30+. Each one should answer a question a reader would actually ask of the Map.

When creating or evolving an apex_type: applied article, run the synthesis 10-step process and verify the six points above are satisfied before publishing.