Physical AI Action-Outcome Experiment 018 C++ Connector Replay

Date: 2026-06-23

Status: Experimental validation complete

Classification: Experimental runtime path

Goal

Connect the C++ EdgeFleetFeedConnector to the existing two-node Physical AI edge/fleet SQL replay and prove that the connector can replace the Python feed worker for bounded Action-Outcome delivery.

Hypothesis

The runtime connector plus a narrow SQL/HTTP source-sink adapter can preserve the Experiment 016 semantics:

• edge-local decisions remain materialized immediately,
• fleet-global audit converges through bounded batches,
• dropped and outage failures remain retryable,
• duplicate and late events are visible in telemetry,
• restart reloads ACK state from a checkpoint,
• fleet SQL/JOIN validation still passes after replay.

Procedure

1. Start two ZeptoDB HTTP servers with --no-auth --storage-mode pure.
2. Apply the Experiment 016 edge SQL fixture to the edge node.
3. Apply only fleet base/DDL seed statements to the fleet node.
4. Read physical_ai_edge_feed_outbox_016 from the edge node through native SQL.
5. Convert outbox rows into EdgeFleetFeedEvent values.
6. Run EdgeFleetFeedConnector passes for outage, bounded recovery with a dropped event and duplicate event, restart reload with late delivery, and final drain.
7. Materialize fleet inbox, final operational rows, ACK rows, and telemetry rows through SQL inserts.
8. Validate counts and JOIN result cardinality through native SQL.
9. Write the immutable result report.

Acceptance Criteria

• zepto_edge_fleet_replay builds.
• EdgeFleetFeedConnectorTest.* remains green.
• Live two-node replay reaches 52/52 fleet ACK rows.
• Fleet materializes 5 decision rows, 15 retrieval rows, and 32 suppression rows.
• Telemetry records outage, duplicate, late, and restart reload evidence.
• Native recovery JOIN returns 5 rows.
• Native suppression audit JOIN returns 32 rows.
• Docs keep the path classified as experimental.

Result

Result report:

• docs/research/results/physical_ai_edge_fleet_cpp_connector_replay_018.md

Verification:

cd build && ninja -j$(nproc) zepto_edge_fleet_replay zepto_tests

./build/tests/zepto_tests --gtest_filter='EdgeFleetFeedConnectorTest.*'

./build/zepto_edge_fleet_replay \
  --edge-url http://127.0.0.1:19441 \
  --fleet-url http://127.0.0.1:19442 \
  --outage-url http://127.0.0.1:1 \
  --edge-sql docs/research/results/physical_ai_edge_fleet_feed_replay_016_edge.sql \
  --fleet-seed-sql docs/research/results/physical_ai_edge_fleet_feed_replay_016_fleet.sql \
  --output docs/research/results/physical_ai_edge_fleet_cpp_connector_replay_018.md \
  --checkpoint /tmp/zeptodb_edge_fleet_cpp_connector_018.checkpoint \
  --batch-limit 12 \
  --max-inflight 12

Result: PASS.

Check	Value
Edge outbox rows	52
Fleet ACK rows	52
Fleet decision rows	5
Fleet retrieval rows	15
Fleet suppression rows	32
Outage telemetry rows	1
Duplicate telemetry rows	5
Late telemetry rows	1
Restart telemetry rows	1
Recovery JOIN rows	5
Suppression audit JOIN rows	32

Interpretation

Experiment 018 closes the gap between research harness semantics and runtime C++ connector behavior. The feed state machine is now validated against live ZeptoDB SQL endpoints, not only mocked sink callbacks.

This is still not a promoted product connector. The SQL/HTTP adapter is a standalone experiment tool, and product readiness still requires server lifecycle controls, RBAC/admin policy, catalog or documented runtime persistence, and long-running operational tests.

Next Product Or Research Step

Productize the connector lifecycle: decide how operators enable the feed, where cursor/ACK state is persisted, which RBAC roles can configure it, and how metrics appear in /metrics without running a standalone experiment tool.