Constant SCHEMA_V6 

pub const SCHEMA_V6: &str = "
    CREATE TABLE IF NOT EXISTS ai_experiment_assignments (
        id          INTEGER PRIMARY KEY AUTOINCREMENT,
        ts_ms       INTEGER NOT NULL,
        request_id  TEXT NOT NULL,
        experiment  TEXT NOT NULL,
        variant     TEXT NOT NULL,
        model       TEXT NOT NULL,
        consumer    TEXT NOT NULL DEFAULT ''
    );
    CREATE INDEX IF NOT EXISTS idx_ai_exp_assign_ts ON ai_experiment_assignments(ts_ms);
    CREATE INDEX IF NOT EXISTS idx_ai_exp_assign_experiment ON ai_experiment_assignments(experiment);

    CREATE TABLE IF NOT EXISTS ai_eval_results (
        id              INTEGER PRIMARY KEY AUTOINCREMENT,
        ts_ms           INTEGER NOT NULL,
        eval_name       TEXT NOT NULL,
        model           TEXT NOT NULL,
        variant         TEXT NOT NULL DEFAULT '',
        prompt_version  TEXT NOT NULL DEFAULT '',
        case_index      INTEGER NOT NULL,
        input           TEXT NOT NULL,
        expected        TEXT NOT NULL,
        actual          TEXT NOT NULL,
        passed          INTEGER NOT NULL,
        scorer          TEXT NOT NULL,
        latency_ms      REAL NOT NULL DEFAULT 0.0
    );
    CREATE INDEX IF NOT EXISTS idx_ai_eval_results_ts ON ai_eval_results(ts_ms);
    CREATE INDEX IF NOT EXISTS idx_ai_eval_results_eval ON ai_eval_results(eval_name);

    CREATE TABLE IF NOT EXISTS ai_feedback (
        id          INTEGER PRIMARY KEY AUTOINCREMENT,
        ts_ms       INTEGER NOT NULL,
        request_id  TEXT NOT NULL,
        label       TEXT NOT NULL,
        comment     TEXT NOT NULL DEFAULT '',
        consumer    TEXT NOT NULL DEFAULT '',
        model       TEXT NOT NULL DEFAULT ''
    );
    CREATE INDEX IF NOT EXISTS idx_ai_feedback_ts ON ai_feedback(ts_ms);
    CREATE INDEX IF NOT EXISTS idx_ai_feedback_request_id ON ai_feedback(request_id);
";

Schema v6 (DW-086): three tables for prompt experimentation.

• ai_experiment_assignments — one row per request that was served by an A/B test alias, recording which variant was selected. The request_id ties the assignment to the access log and spend rows; the experiment and variant labels are config-bounded (the experiment and variant names from config).

• ai_eval_results — one row per eval case run, recording the input, expected, actual output, pass/fail, scorer, and attribution (eval name, model alias, variant, prompt version). The admin /analytics/eval-results endpoint reads these rows.

• ai_feedback — one row per feedback record ingested via the admin API, carrying the request id, label, comment, and attribution. The admin /analytics/feedback endpoint reads these rows.