Watching the Market¶

Learning objectives

After reading this chapter you will be able to:

Use each browser application to watch the simulated market
Use the command-line tools inside the container for what the browser does not show: the values, the traders' activity, the logs
Read the AI traders' daily results and say which kind of trader is winning

The four windows¶

Application	Address	Watch it for
TapeDeck (terminal-gui)	http://localhost:8090	The whole market at a glance: movers, the trade tape, sessions and halts, the news
Order Book Viewer (book-gui)	http://localhost:8094	One symbol's book, price level by price level, with its tape
Trading GUI (trader-gui)	http://localhost:8093	Trading as a student desk; the instructor's screens with the `OPS01` key
Log Operator Console (log-gui)	http://localhost:8091	What every process is doing; warnings and errors

Each has its own chapter in the Participant or Operator's Guide (see See also); here is what to look at in a simulated market.

TapeDeck: the market at a glance¶

Put it on the classroom screen. The Market Overview and Movers show which of the 300 shares are moving; a share that moved a lot with no headline in the News view probably jumped (an unannounced change in its value), or is being pushed by momentum traders. The Trade Tape shows who traded: AI ids are the swarm, MM01 the market maker, TRADER ids your students. During the auctions the Session & Halt Status Board counts down to the next phase.

Order Book Viewer: one book up close¶

Pick a symbol and watch for a minute:

The market maker's quotes are the 200-share levels 20 cents apart, around the middle of the book. When one is filled, that side stays empty for about 5 seconds, until the bot posts a fresh pair.
The AI traders' orders are everything else: small passive orders from the noise traders near the touch, larger ones from value investors and institutions, an iceberg showing a fifth of its size.
After a headline the book empties on one side within seconds as news traders sweep it, then refills at the new price.

Trading GUI: trade in it¶

Students log in with their desk keys (Ten-Minute Start) and trade like any participant: their orders meet the AI traders' orders in the same books. Logged in as OPS01 you get the operator's screens — session control, halts, the kill switch — the same as on any exchange. Use them carefully on a market that is meant to keep running: a halted symbol stays halted until you resume it.

Log Operator Console: what the processes say¶

Filter by process. pm-ai-swarm writes one status line a minute and a summary after each close; pm-market-sim logs every headline it publishes; pm-scheduler every phase change. Warnings are worth a look; a steady trickle of cancel … never answered from the swarm means the exchange is overloaded (see Troubleshooting).

The command line¶

Everything below runs inside the exchange container. Either prefix each command with docker exec edumatcher-sim, or open a shell there once — docker exec -it edumatcher-sim bash — and type them as shown. (By hand, without the container, type them in any terminal with EDUMATCHER_DATA_DIR set.)

Is everything running?¶

pm-opctl-cli list --no-restart

One row per process, with its uptime and memory. dead rows are restarted by the container within a minute.

What is the market doing?¶

pm-log-cli query --process pm-ai-swarm --grep "actions=" --limit 3

[w1-AI001-AI100] t=480s session=CONTINUOUS actions=58029 (120.9/s) live_orders=2325 fills=27616 rejects=25
[w1-AI001-AI100] t=900s session=CONTINUOUS actions=91317 (101.4/s) live_orders=2368 fills=38669 rejects=25

actions counts new orders and cancels since the swarm started, and the rate in brackets is the average over that whole time, closed hours included. For the current rate, take two lines: (91317 − 58029) / (900 − 480) is 79 a second, the swarm's budget of 80. live_orders is how many orders the traders have working; rejects should stay a tiny share of the actions.

pm-board                 # a rotating board of symbols: last, change, volume
pm-viewer -s AAPL        # one order book in the terminal
pm-ticker                # a scrolling ticker of the day's prices

How far from the values?¶

pm-market-sim --status --id OPS01

lists the shares whose prices are furthest from their true values (see The True Value).

The news¶

pm-news list            # headlines as they are published (Ctrl+C to stop)

The day's statistics¶

pm-stats-cli daily --symbol AAPL                    # open, high, low, close, volume per day
pm-stats-cli trades --symbol AAPL --limit 10        # the last trades
pm-stats-cli dates                                  # the trading dates recorded

pm-stats keeps the daily rows for good and individual trades for the last seven days (Running It for Weeks).

Who made money¶

After each close (five seconds after, so the closing auction's fills are in) the swarm writes one line per trader to /data/ai_swarm/<date>-<worker>.jsonl — on the host, ./data-sim/ai_swarm/:

{"closed_at": "2026-11-03T15:05:05+00:00", "day": 2, "agent": "AI007",
 "preset": "value-investor", "submitted": 412, "fills": 233, "filled_qty": 9120,
 "cancels": 160, "rejected": 0, "killed": 0, "unseen_expiries": 0,
 "gtc_orders": 0, "gross_position": 1840, "pnl_realized": 412.10, "pnl_mtm": 87.40}

The counters are for that day; the profit and loss (pnl_realized, made on closed positions, and pnl_mtm, on the open position at the last price) since the trader last started. To total the latest P&L by kind of trader:

python3 - <<'EOF'
import glob, json
latest = {}
for path in sorted(glob.glob("/data/ai_swarm/*.jsonl")):
    for line in open(path):
        row = json.loads(line)
        latest[row["agent"]] = row
by_preset = {}
for row in latest.values():
    pnl = row["pnl_realized"] + row["pnl_mtm"]
    total, n = by_preset.get(row["preset"], (0.0, 0))
    by_preset[row["preset"]] = (total + pnl, n + 1)
for preset, (total, n) in sorted(by_preset.items(), key=lambda kv: -kv[1][0]):
    print(f"{preset:16} {n:3} traders  P&L {total:12.2f}  per trader {total / n:10.2f}")
EOF

In a market like s300-sim the informed traders — value investors, institutions, news traders — usually come out ahead, and the noise and momentum traders behind: that is what being informed is worth. If the order flips, the market is giving the uninformed an edge — worth a look at the settings (Tuning the Traders). It makes a good class discussion either way: who paid whom, and why.

For the instructor: the values through the API¶

The values are available to ADMIN keys through the API gateway, for your own dashboards and scripts:

curl -s -H "Authorization: Bearer key-ops01-3e34rs" \
    http://localhost:8080/api/v1/admin/sim

returns the model's state and every share's latest value; the WebSocket ws://localhost:8080/api/v1/admin/sim streams them. Students' keys are refused.