Wiki-Lyrical Engine Protocol

An Autonomous Cognitive Poetry Generator

Document ID: CNL-PR-2025-019
Version: 1.1
Date: December 14, 2025
Author: Michael P. Hamilton, Ph.D.

AI Assistance Disclosure: This protocol was developed with assistance from Claude (Anthropic, claude-opus-4-20250514). The AI contributed to system architecture, code development, prompt engineering, and documentation drafting through an iterative collaborative process. The author takes full responsibility for the content, accuracy, and conclusions.

Abstract

This protocol documents the Wiki-Lyrical Engine (WLE), an autonomous agent that samples random Wikipedia articles during "sleep" cycles, accumulates pattern fragments in a buffer, and upon "waking" produces three poetic compressions: a limerick, a haiku, and a what-if hypothesis. A second-stage feasibility assessment (Claude API with web search only) grounds each what-if against existing literature, distinguishing speculative category errors from testable hypotheses or active research areas. The system externalizes the hypnopompic moment—when dream logic meets waking cognition—using Wikipedia as a living temporal record of collective human attention. Built on a LAMP stack with provider-agnostic AI integration (Claude API and Ollama), the engine supports both automated cron-driven operation and manual "Activate Dreaming" triggering from an administrative interface. This document provides specifications for database schema, file architecture, AI integration patterns, two-stage wake cycle mechanics, and deployment methodology.

1. Introduction

1.1 Purpose

The Wiki-Lyrical Engine addresses an operational question: how would you build a machine that generates ideas? Not through directed inquiry or problem-solving, but through collision—random juxtaposition of knowledge fragments that might rhyme in unexpected ways.

Design goals:

• Autonomous sampling of Wikipedia's random article API
• Temporal accumulation in a dream buffer
• AI-driven synthesis through hypnopompic prompt engineering
• Three-form output constraint: limerick, haiku, what-if hypothesis
• Second-stage feasibility assessment via web search (Claude only)
• Public ledger of accumulated dreams with literature grounding
• Manual and automated wake triggers

1.2 Conceptual Framework

The system draws from two conceptual threads:

Cognitive Phenology: Ideas have timescales we don't naturally perceive. Some germinate in minutes, some require years of vernalization. The engine instruments this process—sampling, accumulating, synthesizing—making cognitive rhythms observable.

The Climate Clock Lineage: During the author's decade as director of Blue Oak Ranch Reserve, a Climate Clock project broadcast phenology camera imagery to San Jose Airport, making ecological timescales accessible to travelers waiting for luggage. The Wiki-Lyrical Engine applies the same temporal translation principle inward—making cognitive time legible through accumulated collisions.

1.3 Scope

This protocol covers:

• Database schema design
• PHP file architecture
• Wikipedia API integration
• AI wake processor (provider abstraction)
• Sleep/wake cycle mechanics
• Public and administrative interfaces
• Deployment procedures

1.4 Technology Stack

1.5 Domain Context

This system is part of the Macroscope SELF domain—personal health, work, reading, writing, and social dimensions of the broader Macroscope research program integrating EARTH, LIFE, HOME, and SELF paradigms.

2. System Architecture

2.1 Overview

The Wiki-Lyrical Engine operates on a sleep/wake cycle with a two-stage wake process:

1. Sleep: Cron job (or manual trigger) samples 4 random Wikipedia articles, stores in buffer
2. Accumulate: Repeat sleep cycles until buffer reaches threshold (default: 3 sets = 12 articles)
3. Wake Ready: State flag indicates buffer sufficient for synthesis
4. Wake Stage 1: AI processes buffer through hypnopompic prompt, generates three outputs (limerick, haiku, what-if)
5. Wake Stage 2: Feasibility assessment via web search evaluates the what-if hypothesis against existing literature (Claude API only; Ollama skips this stage)
6. Record: Dream stored in ledger with buffer snapshot, outputs, and feasibility assessment timestamped
7. Reset: Buffer cleared, cycle restarts

2.2 Two-Stage Wake Rationale

The two-stage design separates creative generation from critical evaluation. Stage 1 operates in "dream logic"—making associative leaps without constraint. Stage 2 applies scientific grounding—checking whether the creative leap lands on active research, testable hypothesis, or category error.

This mirrors the hypnopompic experience: the dream produces something that feels significant, then waking cognition tests it against reality. Some insights survive; most dissolve. The feasibility assessment makes this testing explicit and searchable.

2.3 State Machine

[SLEEPING] ---(sample)---> [SLEEPING] (sets < min_required)
     |
     +---(sample)---> [WAKE_READY] (sets >= min_required)
                            |
                            +---(nudge)---> [STAGE_1: DREAM] ---> [STAGE_2: FEASIBILITY] ---> [SLEEPING]
                                                                          |
                                                                          v
                                                                   [DREAM STORED]

Note: Stage 2 (Feasibility) executes only when using Claude API with web search. Ollama-generated dreams proceed directly from Stage 1 to storage without feasibility assessment.

2.4 Buffer Management

• articles_per_sample: 4 (configurable)
• min_sets_before_ready: 3 (12 total articles)
• max_buffer_sets: 10 (FIFO eviction of oldest set)

When buffer exceeds maximum, oldest sample set is deleted before new samples are added.

3. Database Schema

3.1 Overview

Three tables support dreams, samples, and engine state.

3.2 Tables

3.2.1 wle_dreams

Primary output storage.

Indexes: idx_created (created_at), idx_featured (is_featured)

3.2.2 wle_samples

Wikipedia article buffer.

Indexes: idx_consumed (is_consumed), idx_set (set_number)

3.2.3 wle_state

Singleton state record.

3.3 Foreign Key Constraints

• wle_samples.dream_id → wle_dreams.id (SET NULL on delete)

4. File Architecture

4.1 Directory Structure

/Library/WebServer/Documents/michaelphamilton/quotes/WLE/
├── css/
│   └── style.css              # Dark theme stylesheet
├── includes/
│   ├── config.php             # Configuration loader
│   ├── database.php           # mysqli wrapper class
│   ├── wikipedia.php          # Wikipedia API client
│   ├── claude.php             # AI provider abstraction
│   └── engine.php             # Sleep/wake orchestration
├── templates/
│   └── dream-card.php         # Dream display partial
├── admin/
│   └── index.php              # Admin dashboard with controls
├── cron/
│   ├── sleep.php              # Wikipedia sampling script
│   └── wake.php               # Dream generation script
├── schema/
│   └── wle_schema.sql         # Database creation script
├── logs/
│   └── engine.log             # Operation log
├── index.php                  # Public interface
└── README.md                  # Installation guide

4.2 Credential Storage

Credentials stored outside web root:

/Library/WebServer/secure/credentials/wle_db_config.php
/Library/WebServer/secure/wle/ai-config.php

4.3 Configuration Loading

// includes/config.php
require_once '/Library/WebServer/secure/credentials/wle_db_config.php';
$ai_config = require '/Library/WebServer/secure/wle/ai-config.php';

$provider = $ai_config['provider'];
$provider_config = $ai_config[$provider];

return [
    'db' => [...],
    'ai' => [
        'provider' => $provider,
        'api_key' => $provider_config['api_key'] ?? null,
        'base_url' => $provider_config['base_url'],
        'model' => $provider_config['default_model'],
        'available_models' => $provider_config['available_models'],
        'max_tokens' => $provider_config['max_tokens'],
        'temperature' => $provider_config['temperature'],
        'wake_prompt' => $ai_config['wake_prompt'],
    ],
    'ai_config' => $ai_config,
    'sleep' => [...],
    'site' => [...],
];

5. AI Integration Layer

5.1 Provider Abstraction

The WLE_AI class supports both Claude API and Ollama:

class WLE_AI {
    private $provider;      // 'claude' or 'ollama'
    private $apiKey;        // Claude only
    private $baseUrl;       // API endpoint
    private $model;         // Active model
    private $wakePrompt;    // Hypnopompic prompt template

    public function wake($samples) {
        $buffer = $this->formatBuffer($samples);
        $prompt = str_replace('{buffer}', $buffer, $this->wakePrompt);

        if ($this->provider === 'claude') {
            $response = $this->callClaude($prompt);
        } else {
            $response = $this->callOllama($prompt);
        }

        return $this->parseResponse($response);
    }
}

5.2 Wake Prompt Engineering

The hypnopompic prompt establishes dream-state voice:

You are waking from a dream. Your dream contained these fragments:

{buffer}

You are in a hypnopompic state - not fully coherent, still holding 
dream logic. Speak from the residue before it dissolves.

Produce exactly three outputs in this exact format:

LIMERICK:
[Find something absurd in the collision of these fragments. Rhythm 
and rhyme required. Comedic compression. Five lines, AABBA rhyme scheme.]

HAIKU:
[Stillness. Two images from the fragments placed beside each other. 
No explanation. Let the cut do the work. Three lines.]

WHAT-IF:
[A question or hypothesis that could not have been asked without 
these specific fragments colliding. Begin with "What if..." - reach 
for something that might send a curious mind to the literature.]

Do not explain. Do not preamble. Do not add commentary after. 
Just produce the three forms in exactly the format shown above.

5.3 Response Parsing

Outputs extracted via regex:

private function parseResponse($text) {
    preg_match('/LIMERICK:\s*\n(.*?)(?=\n\s*HAIKU:)/si', $text, $m);
    $limerick = trim($m[1] ?? '');

    preg_match('/HAIKU:\s*\n(.*?)(?=\n\s*WHAT-IF:)/si', $text, $m);
    $haiku = trim($m[1] ?? '');

    preg_match('/WHAT-IF:\s*\n(.*?)$/si', $text, $m);
    $whatIf = trim($m[1] ?? '');

    return ['limerick' => $limerick, 'haiku' => $haiku, 'what_if' => $whatIf];
}

5.4 Feasibility Assessment (Stage 2)

The feasibility assessment is Claude-only, requiring web search capability. Ollama models skip this stage.

Feasibility Prompt:

Evaluate this speculative hypothesis for scientific plausibility:

"{what_if}"

Search for relevant research and provide a brief assessment (2-3 paragraphs):

1. Is this hypothesis testable or purely speculative?
2. What existing research areas intersect with this idea?
3. What would be the key obstacles or required breakthroughs?

Be concise and grounded. If the hypothesis is genuinely novel, say so. 
If it's already being explored, cite the relevant work. 
If it's physically impossible, explain why briefly.

End with a one-line PLAUSIBILITY rating: 
[Speculative / Testable / Active Research / Physically Implausible]

API Call with Web Search:

$data = [
    'model' => $this->model,
    'max_tokens' => 1024,
    'tools' => [
        [
            'type' => 'web_search_20250305',
            'name' => 'web_search'
        ]
    ],
    'messages' => [
        ['role' => 'user', 'content' => $prompt]
    ]
];

Plausibility Ratings:

5.5 Model Selection

The admin interface provides model selection:

6. Wikipedia Integration

6.1 API Endpoint

The system uses Wikipedia's REST API random summary endpoint:

GET https://en.wikipedia.org/api/rest_v1/page/random/summary

6.2 Response Handling

class WLE_Wikipedia {
    public function getRandomArticle() {
        $url = "https://en.wikipedia.org/api/rest_v1/page/random/summary";

        // cURL request with User-Agent header
        $data = json_decode($response, true);

        return [
            'title' => $data['title'],
            'extract' => $data['extract'],
            'image_url' => $data['thumbnail']['source'] ?? null,
            'page_url' => $data['content_urls']['desktop']['page'],
        ];
    }

    public function getRandomArticles($count = 4) {
        $articles = [];
        for ($i = 0; $i < $count; $i++) {
            $article = $this->getRandomArticle();
            if ($article) $articles[] = $article;
            usleep(100000);  // 100ms rate limiting
        }
        return $articles;
    }
}

6.3 Rate Limiting

• 100ms delay between requests
• User-Agent header identifies the project
• Maximum 4 articles per sleep cycle (configurable)

7. Public Interface

7.1 Design Philosophy

Dark theme emphasizing contemplative reading:

• Background: Deep blue-purple (#1a1a2e)
• Surface: Navy (#16213e)
• Accent: Coral-pink (#e94560)
• Typography: Georgia serif for content

7.2 Pages

7.3 Status Bar

Real-time engine state display:

• Sleep state: "○ Sleeping — 2/3 sets accumulated"
• Ready state: "● Wake Ready — 3 sample sets in buffer"
• Last dream timestamp

8. Administrative Interface

8.1 Dashboard

The admin panel provides:

• Engine state display (sets accumulated)
• Statistics (total dreams, featured count, average rating)
• Pending samples list
• Recent dreams table with feature/unfeature controls

8.2 Wake Controls

Model Selector: Dropdown of available models for active provider.

Activate Dreaming Button: Single action that:

1. Fills buffer to minimum threshold (3 sleep cycles)
2. Triggers wake processor
3. Generates and stores dream

case 'dream_cycle':
    $model = $_POST['model'] ?? null;
    $setsNeeded = $config['sleep']['min_sets_before_ready'];

    for ($i = 0; $i < $setsNeeded; $i++) {
        $sleepResult = $engine->sleep();
        usleep(500000);  // 500ms between Wikipedia batches
    }

    $result = $engine->wake($model);
    break;

8.3 Authentication

Reuses existing Quotes Collection admin session:

if (!isset($_SESSION['admin_logged_in']) || $_SESSION['admin_logged_in'] !== true) {
    header('Location: ../../admin/login.php?redirect=/WLE/admin/');
    exit;
}

9. Cron Operation

9.1 Sleep Script

# Sample every 3 hours
0 3,6,9,12,15,18,21 * * * /usr/local/bin/php /path/to/WLE/cron/sleep.php

9.2 Wake Script

# Wake every morning at 5:30 AM (if buffer ready)
30 5 * * * /usr/local/bin/php /path/to/WLE/cron/wake.php

9.3 Logging

Both scripts append to logs/engine.log:

[2025-12-14 10:28:47] SLEEP: Set #1, 4 articles, 1 sets
[2025-12-14 10:29:09] SLEEP: Set #2, 4 articles, 2 sets
[2025-12-14 10:29:14] SLEEP: Set #3, 4 articles, 3 sets [WAKE READY]
[2025-12-14 10:29:37] WAKE: Dream #1 from 12 samples

10. Sample Output

10.1 Dream #3 (December 14, 2025)

Buffer Contents: Simon B. Elliott State Park, Brockage (coin errors), CKAP2 (cytoskeletal protein), Colin Fewer (runner), Urhunden Prizes, Brunswick Square Gloucester, Kukerpillid (Estonian folk ensemble), Relic'ing (guitar distressing), Bijoya Ray, 2015 Castilian-Leonese election, Emmanuel Philibert Duke of Savoy, Vepryk Ukraine.

Limerick:

A coin collector dreamed of a park
Where brockages hung from each bark
But Colin ran through
With a guitar relic'd blue
While proteins danced bright in the dark

Haiku:

Mirror image coin—
the second growth forest holds
its own reflection

What-If:

What if the cytoskeletal proteins that organize cellular architecture 
follow similar error-correction mechanisms to those that prevent brockage 
formation in minting, and could studying coin production defects reveal 
new approaches to understanding how CKAP2 maintains proper chromosomal 
alignment during cell division?

Feasibility Assessment:

This hypothesis proposes that cytoskeletal proteins like CKAP2 might follow error-correction mechanisms similar to those preventing brockage formation in coin minting, and that studying coin production defects could inform understanding of chromosomal alignment during cell division.

1. Is this hypothesis testable or purely speculative?

The hypothesis is largely speculative with limited testable elements. While CKAP2 regulates microtubule growth and ensures proper chromosome segregation through well-characterized mechanisms, and brockage errors in minting occur when an already struck coin creates an incuse impression on another coin, the proposed analogy lacks meaningful mechanistic parallels. Cellular error-correction operates through quality control mechanisms that recognize defective components and degrade them efficiently, while mint error prevention relies on physical quality control that rates coins for defects and recycles most error coins before circulation. These are fundamentally different processes operating at vastly different scales and through entirely different physical principles.

2. What existing research areas intersect with this idea?

The cellular component is well-established: cells lacking CKAP2 develop chromosome segregation errors due to substantial decline in microtubule growth rates. However, there is no existing research connecting manufacturing quality control processes to biological error-correction mechanisms. The proposed intersection is conceptual rather than mechanistic.

3. What would be the key obstacles or required breakthroughs?

The primary obstacle is the fundamental incompatibility of the systems being compared. Biological error-correction involves nonequilibrium kinetic proofreading mechanisms and dissipative pathways, while mint quality control involves mechanical inspection and physical removal of defective products. The hypothesis conflates outcome similarity (error prevention) with mechanistic similarity, which represents a category error in scientific reasoning.

PLAUSIBILITY: Speculative

Sources: PNAS, PubMed, PMC, PLoS One, bioRxiv (CKAP2 papers); Wikipedia, numismatic sources (mint error definitions)

11. Known Limitations

1. Wikipedia API dependency: System requires internet connectivity
2. No semantic filtering: Random articles may include stubs or disambiguation pages
3. Single-language: Currently English Wikipedia only
4. No retry logic: Failed API calls not automatically retried
5. Model hallucination: What-if hypotheses are speculative, not factual
6. Rate limiting: Heavy use could trigger Wikipedia API limits
7. Feasibility Claude-only: Ollama models cannot perform web search; dreams generated via Ollama lack feasibility assessment
8. Search result variability: Feasibility assessment quality depends on web search results; some niche topics may have limited coverage

12. Future Directions

12.1 Multi-Language Sampling

Support for multiple Wikipedia language editions to introduce cross-cultural collision patterns.

12.2 Semantic Clustering

Pre-filter samples by category to create thematic dream sessions (science dreams, history dreams, arts dreams).

12.3 Dream Analysis

Long-term pattern analysis:

• Which collision types produce higher-rated what-ifs?
• Temporal patterns in output quality
• Model comparison across same buffers

12.4 Embodied Integration

Future coupling to SELF domain physiological sensors—wake triggers tied to sleep cycle data, heart rate variability, or circadian markers.

12.5 Public Participation

Allow visitors to submit "nudge" requests, creating a distributed wake schedule driven by audience attention.

13. References

[1] Wikipedia REST API Documentation. "Page Content: Random." https://en.wikipedia.org/api/rest_v1/ (accessed December 14, 2025).

[2] Anthropic Documentation. "Claude API Reference." https://docs.anthropic.com/claude/reference (accessed December 14, 2025).

[3] Ollama Documentation. "API Reference." https://github.com/ollama/ollama/blob/main/docs/api.md (accessed December 14, 2025).

[4] City of San Jose Office of Cultural Affairs. (2008). "Climate Clock Design Competition Press Release." September 25, 2008.

[5] Gupta, N., et al. (2025). "A single-fibre computer enables textile networks and distributed inference." Nature, 639, 79-86.

Appendix A: Complete SQL Schema

CREATE DATABASE IF NOT EXISTS wle_db 
CHARACTER SET utf8mb4 
COLLATE utf8mb4_unicode_ci;

USE wle_db;

CREATE TABLE IF NOT EXISTS wle_dreams (
    id INT AUTO_INCREMENT PRIMARY KEY,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    buffer_snapshot JSON,
    limerick TEXT,
    haiku TEXT,
    what_if TEXT,
    feasibility_assessment TEXT DEFAULT NULL,
    feasibility_sources JSON DEFAULT NULL,
    user_rating TINYINT DEFAULT NULL,
    notes TEXT DEFAULT NULL,
    is_featured BOOLEAN DEFAULT FALSE,
    INDEX idx_created (created_at),
    INDEX idx_featured (is_featured)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

CREATE TABLE IF NOT EXISTS wle_samples (
    id INT AUTO_INCREMENT PRIMARY KEY,
    collected_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    set_number INT NOT NULL,
    wiki_title VARCHAR(500),
    wiki_extract TEXT,
    wiki_image_url VARCHAR(1000),
    wiki_page_url VARCHAR(1000),
    is_consumed BOOLEAN DEFAULT FALSE,
    dream_id INT DEFAULT NULL,
    INDEX idx_consumed (is_consumed),
    INDEX idx_set (set_number),
    FOREIGN KEY (dream_id) REFERENCES wle_dreams(id) ON DELETE SET NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

CREATE TABLE IF NOT EXISTS wle_state (
    id INT PRIMARY KEY DEFAULT 1,
    current_set_number INT DEFAULT 0,
    sets_accumulated INT DEFAULT 0,
    last_sample_at TIMESTAMP NULL,
    last_wake_at TIMESTAMP NULL,
    is_wake_ready BOOLEAN DEFAULT FALSE
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO wle_state (id, current_set_number, sets_accumulated) 
VALUES (1, 0, 0)
ON DUPLICATE KEY UPDATE id=id;

-- Migration for existing installations:
-- ALTER TABLE wle_dreams ADD COLUMN feasibility_assessment TEXT DEFAULT NULL;
-- ALTER TABLE wle_dreams ADD COLUMN feasibility_sources JSON DEFAULT NULL;

Appendix B: AI Configuration Template

<?php
// /Library/WebServer/secure/wle/ai-config.php

$wake_prompt = <<<'PROMPT'
You are waking from a dream. Your dream contained these fragments:

{buffer}

You are in a hypnopompic state - not fully coherent, still holding 
dream logic. Speak from the residue before it dissolves.

Produce exactly three outputs in this exact format:

LIMERICK:
[Five lines, AABBA rhyme scheme, comedic compression.]

HAIKU:
[Three lines, two images juxtaposed, no explanation.]

WHAT-IF:
[Begin with "What if..." - propositional reach toward novelty.]

Do not explain. Do not preamble. Do not add commentary after.
PROMPT;

return [
    'provider' => 'claude',

    'claude' => [
        'api_key' => 'sk-ant-...',
        'base_url' => 'https://api.anthropic.com/v1/messages',
        'max_tokens' => 600,
        'temperature' => 0.9,
        'available_models' => [
            'claude-opus-4-20250514'    => 'Claude Opus 4 - Most capable',
            'claude-sonnet-4-20250514'  => 'Claude Sonnet 4 - Balanced',
            'claude-haiku-4-5-20251001' => 'Claude Haiku 4.5 - Fast'
        ],
        'default_model' => 'claude-sonnet-4-20250514'
    ],

    'ollama' => [
        'base_url' => 'http://localhost:11434',
        'max_tokens' => 600,
        'temperature' => 0.9,
        'available_models' => [
            'gemma3:12b' => 'Gemma 3 12B - Balanced',
            'gemma3:4b'  => 'Gemma 3 4B - Fast'
        ],
        'default_model' => 'gemma3:12b'
    ],

    'wake_prompt' => $wake_prompt,
];

Appendix C: Color Scheme Reference

Document History

End of Protocol

Permanent URL: https://canemah.org/archive/document.php?id=CNL-PR-2025-019