**Purpose:** declares the architecture variants you are comparing for the
evaluation Bridge will run.
Required columns:
| Column | Type | Required | Description |
|---|---|---|---|
| `variant_id` | string | yes | stable identifier for this variant within your system |
| `variant_name` | string | yes | human-readable name (≤80 chars) |
| `role_names` | string (semicolon-separated) | yes | semicolon-separated list of agent/role names that compose this variant |
| `strategy_mode` | enum | yes | one of: `single_agent`, `multi_agent_pipeline`, `parallel_majority`, `tournament`, `custom` |
| `primary_metric` | string | yes | the metric you optimize for this variant (e.g., `task_success_rate`, `latency_ms`, `cost_per_query_usd`) |
| `primary_metric_value` | float | yes | current observed value of `primary_metric` for this variant |
| `secondary_metric` | string | no | optional secondary trade-off metric |
| `secondary_metric_value` | float | no | observed value of `secondary_metric` |
| `notes` | string | no | free-form notes (≤500 chars) |
**Minimum row count:** 2 (at least 2 variants to compare; otherwise comparison is undefined)
Example:
variant_id,variant_name,role_names,strategy_mode,primary_metric,primary_metric_value,secondary_metric,secondary_metric_value,notes
v_baseline,Baseline single-agent,gpt-4-classifier,single_agent,task_success_rate,0.72,latency_ms,1200,production for 4 months
v_pipeline,3-agent pipeline,planner;executor;verifier,multi_agent_pipeline,task_success_rate,0.84,latency_ms,3400,A/B trial week 2
What this CSV is NOT:
**Canonical source:** `bdc-bridge-passport` repository at frozen SHA recorded in `PASSPORT_REPO_PIN.md`.