Feature: CostTracker service for per-session and per-plan spending tracking
  As a system administrator
  I want to track LLM API spending per session and per plan
  So that I can monitor costs and enforce budget limits

  Background:
    Given a temporary data directory for cost tracking
    And a CostTracker instance with default model pricing

  Scenario: Record usage for a single LLM call
    When I record usage for session "sess-001" and plan "plan-001"
      | tokens_in | tokens_out | model          |
      | 100       | 50         | gpt-3.5-turbo  |
    Then the recorded cost should be approximately 0.000125 USD
    And the cost record should have session_id "sess-001"
    And the cost record should have plan_id "plan-001"

  Scenario: Calculate cost for different models
    When I record usage for session "sess-002" and plan "plan-002"
      | tokens_in | tokens_out | model      |
      | 1000      | 1000       | gpt-4      |
    Then the recorded cost should be approximately 0.09 USD
    When I record usage for session "sess-002" and plan "plan-002"
      | tokens_in | tokens_out | model          |
      | 1000      | 1000       | gpt-3.5-turbo  |
    Then the recorded cost should be approximately 0.002 USD

  Scenario: Get total session cost
    When I record multiple usages for session "sess-003"
      | plan_id  | tokens_in | tokens_out | model          |
      | plan-001 | 100       | 50         | gpt-3.5-turbo  |
      | plan-002 | 200       | 100        | gpt-3.5-turbo  |
      | plan-003 | 300       | 150        | gpt-3.5-turbo  |
    Then the total session cost for "sess-003" should be approximately 0.00075 USD

  Scenario: Get total plan cost
    When I record multiple usages for plan "plan-004"
      | session_id | tokens_in | tokens_out | model          |
      | sess-001   | 100       | 50         | gpt-3.5-turbo  |
      | sess-002   | 200       | 100        | gpt-3.5-turbo  |
      | sess-003   | 300       | 150        | gpt-3.5-turbo  |
    Then the total plan cost for "plan-004" should be approximately 0.00075 USD

  Scenario: Cost persistence across tracker instances
    When I record usage for session "sess-004" and plan "plan-005"
      | tokens_in | tokens_out | model          |
      | 500       | 250        | gpt-3.5-turbo  |
    And I create a new CostTracker instance with the same data directory
    Then the total session cost for "sess-004" should be approximately 0.000625 USD
    And the total plan cost for "plan-005" should be approximately 0.000625 USD

  Scenario: Get session entries
    When I record multiple usages for session "sess-005"
      | plan_id  | tokens_in | tokens_out | model          |
      | plan-001 | 100       | 50         | gpt-3.5-turbo  |
      | plan-002 | 200       | 100        | gpt-3.5-turbo  |
    Then I should get 2 entries for session "sess-005"
    And each entry should have the correct session_id and plan_id

  Scenario: Get plan entries
    When I record multiple usages for plan "plan-006"
      | session_id | tokens_in | tokens_out | model          |
      | sess-001   | 100       | 50         | gpt-3.5-turbo  |
      | sess-002   | 200       | 100        | gpt-3.5-turbo  |
    Then I should get 2 entries for plan "plan-006"
    And each entry should have the correct session_id and plan_id

  Scenario: Custom model pricing
    Given a CostTracker instance with custom model pricing
      | model      | input_price | output_price |
      | custom-llm | 0.001       | 0.002        |
    When I record usage for session "sess-006" and plan "plan-007"
      | tokens_in | tokens_out | model      |
      | 1000      | 1000       | custom-llm |
    Then the recorded cost should be approximately 0.003 USD

  Scenario: Zero cost for unknown model
    When I record usage for session "sess-007" and plan "plan-008"
      | tokens_in | tokens_out | model        |
      | 1000      | 1000       | unknown-llm  |
    Then the recorded cost should be approximately 0.0 USD

  Scenario: Multiple sessions and plans isolation
    When I record usage for session "sess-008" and plan "plan-009"
      | tokens_in | tokens_out | model          |
      | 100       | 50         | gpt-3.5-turbo  |
    And I record usage for session "sess-009" and plan "plan-010"
      | tokens_in | tokens_out | model          |
      | 200       | 100        | gpt-3.5-turbo  |
    Then the total session cost for "sess-008" should be approximately 0.000125 USD
    And the total session cost for "sess-009" should be approximately 0.00025 USD
    And the total plan cost for "plan-009" should be approximately 0.000125 USD
    And the total plan cost for "plan-010" should be approximately 0.00025 USD