Merchant Features: Autopilot & Feature Flags

July 28, 2026 · View on GitHub

Every merchant-level toggle — multi-objective routing, A/B test real-payment interception, elimination, and the two self-tuning features (Autopilot and SR Auto-Calibration) — is read and set through one small feature-flag API.

List Feature Flags

curl "$BASE_URL/merchant-account/merchant_demo/features" \
  --header "$AUTH_HEADER"

{
  "merchant_id": "merchant_demo",
  "features": [
    { "feature": "gsm-scoring-filter", "enabled": false },
    { "feature": "explore-exploit-srv3", "enabled": false },
    { "feature": "ab-test-real-payments", "enabled": true },
    { "feature": "multi-objective-routing", "enabled": true },
    { "feature": "elimination", "enabled": true },
    { "feature": "auto-calibration", "enabled": true },
    { "feature": "autopilot", "enabled": true }
  ]
}

Toggle A Feature

curl --location "$BASE_URL/merchant-account/merchant_demo/features/autopilot" \
  --header "$AUTH_HEADER" \
  --header "Content-Type: application/json" \
  --data '{ "enabled": true }'

Response is the same shape as the list call, re-read after the change so you always see the current effective state of every flag.

Known Features

SlugWhat it controls
gsm-scoring-filterFilters gateway scoring using Gateway Status Mapper error-code classification.
explore-exploit-srv3Explore/exploit hedging on the SRv3 scoring path.
ab-test-real-paymentsWhether active A/B test experiments intercept real /decide-gateway traffic — see A/B Testing: Results.
multi-objective-routingMerchant-wide default for the cost-aware post-step — see Multi-Objective Routing. Per-request enableMultiObjective overrides this.
eliminationGateway-level SR-based elimination (excluding gateways in downtime).
auto-calibrationEnables the self-tuning background job (see below). Must be on together with autopilot for the job to actually write changes.
autopilotMaster write-permission switch for the self-tuning job. With auto-calibration off, this alone does nothing.

What Autopilot & Auto-Calibration Do

These two flags jointly control a background job — there is no separate "run autopilot" endpoint. auto-calibration gates whether the job considers a merchant at all; autopilot gates whether it's allowed to write. Both must be enabled for self-tuning to take effect.

Every polling interval (default 900s, configurable via [sr_auto_calibration] in config/*.toml or SR_AUTO_CALIBRATION_INTERVAL_SECS), the job:

  1. Reads recent traffic volume from ClickHouse (default 3600s lookback).
  2. Derives two SRv3 knobs purely from observed data — no merchant input:
    • Bucket size — clamped between 100 and 2000, only rewritten on a meaningful (25-step) change.
    • Hedging % — capped at 30%, with a 1.0pp minimum move before it updates.
  3. Writes the result back into the merchant's SR config (readable via GET /config/routing-keys and GET /config-sr-dimension/:merchant_id), stamped with "source": "autopilot" so autopilot-written values are distinguishable from human-authored overrides.

Every calibration run also emits an analytics event (flow_type: autopilot_calibration) — this is what powers the "Autopilot Actions" panel in the simulation UI, and is queryable via Routing Events.

Reset Gateway Scores

Flushes all SR v2/v3 score and queue keys for a merchant from Redis, and strips any autopilot-written (source: "autopilot") sub-level config overrides — human-authored overrides are preserved. Used by the simulator's "Hard refresh" so a new simulation run starts from a clean slate.

Unlike the feature-flag calls above, this route requires $TENANT_HEADER in addition to $AUTH_HEADER — see Environment setup.

curl --location "$BASE_URL/gateway-score/reset" \
  --header "$AUTH_HEADER" \
  --header "$TENANT_HEADER" \
  --header "Content-Type: application/json" \
  --data '{ "merchant_id": "merchant_demo" }'

{
  "merchant_id": "merchant_demo",
  "deleted_keys": 214,
  "removed_overrides": 3
}

This does not disable routing or delete configuration — it only clears live scoring state so the next decisions start from a fresh baseline.