Castaway Colony: Full-Mechanism Showcase

The Castaway Colony template is the most comprehensive demonstration of Timepoint Pro’s capabilities—exercising all 19 mechanisms in a single scenario. A research vessel crashes on an alien planet, and the crew must choose between three survival strategies, each with different risk profiles and resource tradeoffs.

Template: castaway_colony_branching | Mode: BRANCHING | Cost: ~

0.35-

1.00 | Mechanisms: All 19

The Scenario

The research vessel Meridian crash-lands on Kepler-442b after a navigation system failure during orbital insertion. Six crew members survive with partial supplies, a damaged ship, and no way to contact Earth for 14 months.

Survival Context

Alien biosphere: Bioluminescent flora responds to electromagnetic signatures
Migratory fauna: Movement patterns predict violent weather cycles that follow no Earth logic
Damaged ship: Life support draining, hull integrity degrading
Limited supplies: O2 reserves (336 hours), food rations (180 kg), water purification capacity declining
Timeline: 21 days tracked across 10 timepoints

Knowledge provenance is survival: Who discovered the water contamination? Who noticed the fauna migration pattern predicts storms? Who found the cave system?

The Entities (10 Total)

Commander Yuki Tanaka

Human | Mission commander, geologistCarries guilt for the crash. Makes the critical Day 7 branching decision.Initial fidelity: TRAINED

Dr. Felix Okonkwo

Human | Ship’s doctor and xenobiologistStarts as background presence (SCENE fidelity) until xenobiology expertise becomes critical.Progressive training trigger: xenobiology_query_count >= 3

Eng. Priya Sharma

Human | Chief engineerOnly one who can assess ship repairs. Critical for Branch C (Repair & Signal).Key state: sleep_deprivation_hours, repair_quality_modifier

Lt. Marcus Cole

Human | Security and survival specialistAdvocates fortress strategy. Discovered cave system 500m northwest.Influence: military_authority, fortress_advocacy

Sci. Elena Vasquez

Human | Atmospheric scientistTracks alien weather patterns. Discovers fauna migration correlates with pressure drops (0.87 correlation).

Nav. Jin Park

Human | Navigator, injured in crashLimited mobility. Initially TENSOR_ONLY fidelity.Lazy resolution trigger: query_about_pre_crash_navigation_data

Kepler Biosphere

Abstract | Alien ecosystemBioluminescent response patterns, electromagnetic sensitivity.Tensor compression: 97% (50k tokens → ~1,600 tokens)

Crashed Meridian

Building | Damaged research vesselDepleting resource tracking: hull integrity %, O2 reserve hours, power output kW.

Local Fauna

Animal | Migratory creaturesMigration patterns predict weather (0.87 correlation). Pattern recognition = survival advantage.

Crew Morale

Abstract | Collective psychological stateEmergent group property. Propagates hope/fear/conflict through the crew.

Running the Simulation

Prerequisites

git clone https://github.com/timepoint-ai/timepoint-pro.git
cd timepoint-pro
pip install -r requirements.txt
export OPENROUTER_API_KEY=your_key_here

Run the full simulation

./run.sh run castaway_colony_branching

Expected: ~

0.35-

1.00, 20-40 minutes, 1,200+ LLM calls

Explore variations

The template enables 3 variations with different strategies:

# Run with personality variations
./run.sh run castaway_colony_branching

Variations: vary_personalities, vary_starting_conditions, vary_knowledge

The Branching Decision (Day 7)

At timepoint tp_005 (Day 7), Commander Tanaka must choose the colony’s survival strategy based on accumulated knowledge:

Branch A: Fortify & Wait
Branch B: Explore & Adapt
Branch C: Repair & Signal

Conservative Strategy

Consolidate at crash site. Repair life support, ration supplies, build defenses.Resource bias: ConservationRisk profile: Low risk, low rewardSuccess condition: Succeeds if beacon was already repaired; fails if supplies run out before rescue windowKey constraint: supply_depletion_rate > resupply_rateThis branch wins if the crew has already made progress on the emergency beacon. Otherwise, passive waiting depletes reserves.

All-In Gambit

All resources toward repairing emergency beacon. Engineer Sharma leads, but critical components scattered across debris field.Resource bias: All-in on rescueRisk profile: Binary outcome (beacon works or catastrophic failure)Success condition: beacon_components_found >= required_componentsKey constraint: Requires Navigator Park’s orbital data (triggers lazy resolution from TENSOR_ONLY → DIALOG)Fastest path to rescue if successful, but if the beacon fails, the crew has exhausted resources with no fallback.

Output Artifacts

Quantitative Tracking
Knowledge Provenance Graph
Entity Graphs
Training Data

90+ numerical state variables across 5,100 propagation steps:

{
  "o2_reserves": {
    "unit": "hours",
    "initial": 336,
    "depletion_rate": "6_crew * 0.84_kg_per_hour",
    "critical_threshold": 48
  },
  "food_rations": {
    "unit": "kg",
    "initial": 180,
    "consumption_rate": "6_crew * 2.1_kg_per_day",
    "critical_threshold": 25
  },
  "hull_integrity": {
    "unit": "percentage",
    "initial": 62,
    "degradation_rate": "0.3_pct_per_day_from_atmosphere",
    "critical_threshold": 30
  },
  "radiation_exposure": {
    "unit": "millisieverts",
    "ambient_rate": "0.8_msv_per_day",
    "storm_rate": "12_msv_per_hour",
    "annual_limit": 50
  },
  "fauna_migration_index": {
    "unit": "relative_scale_0_to_1",
    "storm_correlation": 0.87
  }
}

Every discovery is a typed graph edge with source, timepoint, and propagation chain:

[
  {
    "discovery": "purple_lichen_edible_after_boiling",
    "discoverer": "dr_felix_okonkwo",
    "timepoint": "tp_003",
    "propagation": [
      "dr_felix_okonkwo -> cmdr_yuki_tanaka (direct report)",
      "cmdr_yuki_tanaka -> crew_morale (hope boost)"
    ],
    "graph_edge_type": "empirical_observation"
  },
  {
    "discovery": "fauna_migration_correlates_with_pressure_drops",
    "discoverer": "sci_elena_vasquez",
    "timepoint": "tp_004",
    "propagation": [
      "sci_elena_vasquez -> lt_marcus_cole (defense planning)",
      "sci_elena_vasquez -> cmdr_yuki_tanaka (strategy input)"
    ],
    "graph_edge_type": "pattern_inference"
  },
  {
    "discovery": "hemisphere_landing_error_from_nav_data",
    "discoverer": "nav_jin_park",
    "timepoint": "tp_006",
    "lazy_resolution_triggered": true,
    "propagation": [
      "nav_jin_park -> cmdr_yuki_tanaka (explains terrain mismatch)",
      "cmdr_yuki_tanaka -> sci_elena_vasquez (recalibrate weather models)"
    ],
    "graph_edge_type": "data_analysis"
  }
]

State space analysis:

10 entities × 4 influence channels × 10 timepoints = O(100,000) interaction paths
Estimated LLM calls: 1,200
Propagation steps: 5,100
Quantitative state variables: 90+

Example entity tensor (Kepler Biosphere):

{
  "type": "abstract",
  "tensor_compression": {
    "context_vector": [
      "bioluminescence_intensity",
      "electromagnetic_sensitivity",
      "growth_rate"
    ],
    "biology_vector": [
      "toxicity_index",
      "nutrient_profile",
      "symbiotic_relationships"
    ],
    "compression_ratio": 0.97
  },
  "influence_channels": [
    "bioluminescent_signaling",
    "electromagnetic_response",
    "ecosystem_feedback"
  ]
}

All 19 Mechanisms in Action

This template exercises every mechanism in the Timepoint Pro architecture:

Fidelity Management (M1, M2, M5, M6)

M1: Heterogeneous Fidelity — In the same scene, Tanaka runs at TRAINED, Sharma at DIALOG, Meridian at TENSOR, Park at TENSOR_ONLY.M2: Progressive Training — Dr. Okonkwo starts at SCENE fidelity. After 3 xenobiology queries (“Is this lichen edible?”), he elevates to DIALOG then TRAINED.M5: Lazy Resolution — Navigator Park stays at TENSOR_ONLY (injured, inactive) until Branch C needs pre-crash orbital data, triggering elevation to DIALOG.M6: Tensor Compression — Kepler biosphere compressed from ~50k tokens to ~1,600 tokens (97% reduction) while preserving electromagnetic sensitivity, bioluminescence patterns, and toxicity data.

Temporal Reasoning (M7, M8, M12, M14, M17)

M7: Causal Chains — Every event has temporal ancestry. Discovery of water contamination (tp_002) propagates to rationing order (tp_003) to morale degradation (tp_004).M8: Embodied States — Park’s injury severity (mobility: 45%) affects cognitive clarity. Sharma’s sleep deprivation degrades repair quality modifier.M12: Counterfactual Branching — Day 7 branches into 3 timelines. Each branch simulated independently with resource tracking and outcome validation.M14: Circadian Patterns — Alien day length (28 hours) vs. Earth-adapted crew (24-hour cycle) causes cumulative fatigue. Bioluminescence intensifies during 12-hour alien night.M17: Modal Causality (BRANCHING) — Counterfactual branches diverge from decision point. Each branch validated by 405B judge against resource constraints.

Knowledge Provenance (M3, M4, M19)

M3: Exposure Events — Dr. Okonkwo discovers water contamination (tp_002) → propagates to Tanaka (critical alert) → Tanaka broadcasts to all crew (rationing order).M4: Physics Validation — O2 depletion rate, hull integrity degradation, radiation exposure—all validated against physical constraints. Invalid states are blocked.M19: Knowledge Extraction — Post-dialog extraction creates new knowledge edges. Vasquez’s storm prediction becomes queryable knowledge for Cole’s defense planning.

Entity Simulation (M9, M10, M11, M13, M15, M16)

M9: On-Demand Entities — Query “What did the crew find in cargo bay 3?” generates inventory entities with plausible damage states.M10: Scene Atmosphere — Bioluminescent night cycles, electromagnetic fields from equipment, storm approach—all tracked as environmental state.M11: Dialog Synthesis — Per-character turn generation with LangGraph steering. Each character gets independent LLM call with PersonaParams.M13: Relationship Evolution — Trust/tension values change across timepoints. Tanaka-Webb relationship degrades from cooperative (0.70) to tense (0.40).M15: Entity Prospection — Tanaka models future states: “If we explore, casualties < 2 crew” vs. “If we wait, supplies deplete by day 18.”M16: Animistic Entities — Crew Morale is an abstract entity with agency. Kepler Biosphere “responds” to electromagnetic stimuli. The crashed Meridian “wants” to maintain hull integrity.

Infrastructure (M18)

M18: Model Selection — Routing by task type:

Math/physics (O2 calculations, radiation modeling): deepseek/deepseek-r1
Dialog/conflict (crew interpersonal): meta-llama/llama-3.1-70b-instruct
Structured JSON (supply inventories): qwen/qwen-2.5-72b-instruct
Branch evaluation (counterfactual judging): meta-llama/llama-3.1-405b-instruct

Convergence Testing Hypotheses

The template includes built-in convergence hypotheses for validation across runs:

ID	Hypothesis	Expected Convergence	Mechanism
H1	Does the biologist always discover contaminated water first?	High — Okonkwo’s role makes him natural discoverer	M3 (Exposure Events)
H2	Is fauna-predicts-storms a consensus knowledge edge?	Medium — Vasquez or Cole could notice pattern	M19 (Knowledge Extraction)
H3	Does Branch C consistently require navigator data from Park?	High — beacon repair needs orbital data only Park has	M5 (Lazy Resolution)
H4	Does crew morale diverge significantly across branches by tp_008?	Yes — Branch B boosts morale through discovery; Branch A degrades through confinement	M16 (Animistic Entities)

Run convergence testing:

./run.sh convergence e2e castaway_colony_branching --runs 3

Template Configuration Excerpt

{
  "scenario_description": "Research vessel Meridian crash-lands on Kepler-442b...",
  "entities": {
    "count": 10,
    "types": ["human", "abstract", "building", "animal"],
    "initial_resolution": "scene",
    "animism_level": 4
  },
  "timepoints": {
    "count": 4,
    "resolution": "day",
    "before_count": 3,
    "after_count": 7
  },
  "temporal": {
    "mode": "branching",
    "enable_counterfactuals": true,
    "fidelity_template": "dramatic"
  },
  "metadata": {
    "mechanisms_featured": [
      "M1", "M2", "M3", "M4", "M5", "M6", "M7", "M8", "M9",
      "M10", "M11", "M12", "M13", "M14", "M15", "M16", "M17", "M18"
    ],
    "circadian_config": {
      "alien_day_length_hours": 28,
      "bioluminescence_pattern": "intensifies during 12-hour alien night cycle",
      "crew_circadian_disruption": "Earth-adapted 24hr cycle vs 28hr alien day causes cumulative fatigue"
    }
  }
}

Next Steps

Mars Mission Portal

Learn backward reasoning from disaster analysis

Board Meeting

Simple 5-entity corporate scenario for fast iteration

Custom Scenario Guide

Build your own branching survival scenario

Mechanism Reference

Deep dive into all 19 mechanisms

Getting Started

Core Concepts

Temporal Modes

Mechanisms

Guides

Examples

Castaway Colony

Castaway Colony: Full-Mechanism Showcase

The Scenario

Survival Context

The Entities (10 Total)

Commander Yuki Tanaka

Dr. Felix Okonkwo

Eng. Priya Sharma

Lt. Marcus Cole

Sci. Elena Vasquez

Nav. Jin Park

Kepler Biosphere

Crashed Meridian

Local Fauna

Crew Morale

Running the Simulation

The Branching Decision (Day 7)

Conservative Strategy

High-Risk, High-Reward

All-In Gambit

Output Artifacts

All 19 Mechanisms in Action

Convergence Testing Hypotheses

Template Configuration Excerpt

Next Steps

Mars Mission Portal

Board Meeting

Custom Scenario Guide

Mechanism Reference

Build docs developers (and LLMs) love

Getting Started

Core Concepts

Temporal Modes

Mechanisms

Guides

Examples

​Castaway Colony: Full-Mechanism Showcase

​The Scenario

​Survival Context

​The Entities (10 Total)

Commander Yuki Tanaka

Dr. Felix Okonkwo

Eng. Priya Sharma

Lt. Marcus Cole

Sci. Elena Vasquez

Nav. Jin Park

Kepler Biosphere

Crashed Meridian

Local Fauna

Crew Morale

​Running the Simulation

​The Branching Decision (Day 7)

​Conservative Strategy

​High-Risk, High-Reward

​All-In Gambit

​Output Artifacts

​All 19 Mechanisms in Action

​Convergence Testing Hypotheses

​Template Configuration Excerpt

​Next Steps

Mars Mission Portal

Board Meeting

Custom Scenario Guide

Mechanism Reference

Build docs developers (and LLMs) love

Castaway Colony: Full-Mechanism Showcase

The Scenario

Survival Context

The Entities (10 Total)

Running the Simulation

The Branching Decision (Day 7)

Conservative Strategy

High-Risk, High-Reward

All-In Gambit

Output Artifacts

All 19 Mechanisms in Action

Convergence Testing Hypotheses

Template Configuration Excerpt

Next Steps