Rabbit-Hole Inference Algorithm (RHIA)

I. First Principles of RHIA

At the base level, RHIA tries to answer:

“What can this user’s behavior tell us about how they approach life?”

So connection here is (or can be) reduced to a shared orientation toward struggle and effort.

The three base axioms I am looking at are:

There are others, but for now, these are my focus.

1. Behavioral patterns > stated preferences
2. Struggle rhythm > skill or speed
3. Connection = inferred parallel across multiple quests, not one

II. User Vector Model

Each user’s journey can be represented as a dynamic vector:

UserVector = {
  roadmap_id: UUID,
  quest_vectors: QuestVector[],
  inferred_traits: TraitScore[],
  last_checkpoint: CheckpointData,
  meta: {
    retries: number,
    avg_completion_time: number,
    nudge_response_rate: number,
    self-reflection_score: float
  }
}

QuestVector:

QuestVector = {
  quest_id: UUID,
  completion_time: number,
  retry_count: number,
  delay_start: boolean,
  nudge_interactions: NudgeStats
}

This gives us:

• How long they take to start
• How they bounce back from failure
• Whether they seek help (nudges)
• Their tempo and tendencies

III. The Inference Engine (RHIA Core)

1. Normalize

• All raw quest data is normalized by type (habit, social, active, reflective)
• We bucket completion times, retries, drop-offs

2. Score Traits

Map user’s data to core behavioral traits:

TraitScore = [
  { trait: "persistence", score: 0.76 },
  { trait: "reflectiveness", score: 0.84 },
  { trait: "collaborative-tendency", score: 0.42 },
  ...
]

These can be mapped with rule-based logic first, ML later:

if retry_count > 2 and eventual success: persistence += 0.3
if nudge_used and faster next time: collaborative_tendency += 0.2

Eventually, we can train a classifier to extract this from patterns.

IV. Matchmaking Logic

At checkpoint:

1. Filter users on same roadmap

2. Rank potential buddies by:

   • Similar trait score shape
   • Complementary differences (ex: low reflectiveness + high reflectiveness)
   • Mutual pain point overlap

3. Add diversity factor: avoid echo chambers, promote new insights

match_score = cosine_similarity(a.vector, b.vector)
- penalty if trait shapes are identical
+ bonus if complementary pain points

V. Optional Graph Layer (Later)

Each user = node Quest = milestone node Edges = user-quest completions (weighted by traits)

Graph search helps:

• Predict future drop-off zones
• Find latent “connector” users (influence multipliers)
• Identify isolated users (suggest nudges or warmups)

TL;DR:

• The Rabbit-Hole Inference Algorithm is a trait-based pattern inference system
• It does symbolic reasoning (Prolog-style) early, and evolves to ML classifiers with more data
• It builds human stories from struggle, not just tags or bios
• It matches not by likes, but by lived effort