Run the demo replay

The demo replay is the fastest way to verify the system works without any real network gear. It seeds a one-device, two-port topology and publishes three mac_learned events that exercise the first-sight → move classification path.

For a richer, data-populated TUI experience (3 devices, 15 hosts across realistic vendor OUIs, 48 hours of history including a host move and a cross-source disagreement), use the realistic seed instead:

make seed-realistic

That populates the bitemporal log directly with ~80 observations spanning the last 48 hours, then sets liveness pointers so the OPS FDB tree shows current state. All the screenshots in these docs were captured against this seed. The rest of this page covers the minimal replay.

Run it

make replay

You’ll see output like:

[1/3] seeding topology...
  device_id=1  Eth1=1  Eth2=2

[2/3] publishing 3 events to NATS...
  → mac_learned mac=aa:bb:cc:11:22:33 port=Eth1 event_id=019e1432-6d6f-7ba8-...
  → mac_learned mac=aa:bb:cc:44:55:66 port=Eth2 event_id=019e1432-947f-7839-...
  → mac_learned mac=aa:bb:cc:11:22:33 port=Eth2 event_id=019e1432-bb8f-726e-...

The reconciler is already running (make up started it). It picks up the three events from NATS and writes them through the state machine into mac_observation automatically.

What the events do

Event	Wire-time	Classification
`aa:bb:cc:11:22:33` learned on Eth1	T+0	first-sight
`aa:bb:cc:44:55:66` learned on Eth2	T+10s	first-sight
`aa:bb:cc:11:22:33` learned on Eth2	T+20s	move (closes Eth1, opens Eth2)

After replay, three rows land in mac_observation:

SELECT entry_id, mac, port_id, observed_at,
       upper_inf(valid_during) AS still_open
FROM mac_observation ORDER BY entry_id;

 entry_id |        mac        | port_id |        observed_at         | still_open
----------+-------------------+---------+----------------------------+-----------
        1 | aa:bb:cc:11:22:33 |       1 | 2026-05-10 22:06:53.95+00  | f          ← closed by the move
        2 | aa:bb:cc:44:55:66 |       2 | 2026-05-10 22:07:03.95+00  | t
        3 | aa:bb:cc:11:22:33 |       2 | 2026-05-10 22:07:13.95+00  | t

The first row’s valid_during was closed at T+20s when the move event landed. The third row’s valid_during opens at T+20s and stays open.

Verify in the TUI

make tui

Press Ctrl+H (HISTORY), enter aa:bb:cc:11:22:33, press Enter. You’ll see two timeline entries — the closed Eth1 row and the open Eth2 row.

Press Ctrl+O (OPS) for the FDB tree. You’ll see sw1 with Eth1 (no MACs — the move emptied it) and Eth2 (two MACs — both demo hosts).

What the demo doesn’t exercise

The demo is intentionally minimal. It does not exercise:

Late arrivals (events arriving with observed_at in the past)
Cross-source disagreements (only gnmi is used)
Quarantine events (no skew, no integrity errors)
Multi-hop topology (only one device)

For multi-source and cross-vendor scenarios, you’d point a real collector at real switches — or write a more elaborate replay script.

Reset

To replay again from a clean slate:

make nuke   # removes containers AND volumes
make up
make migrate
make oui-refresh    # optional
make replay

The volume drop is what resets the bitemporal log; make down alone keeps it.