How to backtest a footprint strategy and record a live tape

Footprint (order-flow) analysis needs a tape — the stream of individual trades with their aggressor side — not just OHLCV bars. This guide shows the three things you will actually do: run a footprint backtest on a synthetic tape, record a real tape from a live venue, and replay that recorded tape offline. For the design behind the feature see order-flow.md and ADR 0032.

#1. Backtest on a synthetic tape (no network, no credentials)

The default backtest path generates a synthetic tape: each candle is expanded into prints that reconstruct its OHLC, published on broker.public-trade-printed, so the whole footprint loop (order_flow → footprint-completed → strategy) runs offline.

dune exec bin/main.exe -- backtest FootprintCVDDivergence --n 300 --symbol SBER@MISX

Output (the footprint= count is the footprint-sourced signals, broken out of the total):

backtest: strategy=FootprintCVDDivergence symbol=SBER@MISX candles=300
  signals_emitted=17 (footprint=17)
  intents: planned=… approved=… rejected=…
  reservations: ok=… rejected=…
  orders: accepted=… rejected=… unreachable=…
  submissions_blocked=…

Flags:

What this proves. The synthetic delta is generated, so a synthetic backtest validates footprint mechanics — that prints are partitioned into bars, classified by aggressor, conserved, and turned into signals — not microstructure alpha. For the latter you need a real recorded tape (§2–§3).

#2. Record a real tape from a live venue

Each broker's live-smoke probe doubles as a tape recorder: pass --record FILE and every relayed print is written as one Trade_printed JSON per line — the exact wire shape the backtest replays. Probes are opt-in (not part of @runtest) and skip cleanly when their credential env var is absent. Run during MOEX continuous trading so the tape actually flows.

Finam (confirmed aggressor semantics, 2026-05-27):

export FINAM_SECRET=<portal secret>
dune exec broker/test/live_smoke/finam_public_trades_probe.exe -- --record /tmp/sber.tape

BCS (relay built; run this to confirm the LastTrades frame shape and side encoding before trusting its delta):

export BCS_SECRET=<keycloak refresh token>
dune exec broker/test/live_smoke/bcs_public_trades_probe.exe -- --record /tmp/sber-bcs.tape

Besides recording, each probe prints a per-print aggressor verdict against the L1 quote (BUY at ask / SELL at bid ⇒ side is the aggressor; the reverse ⇒ the mapping is inverted and parse_side must flip). The BCS probe also dumps the first raw frames of each channel, since the BCS tape frame shape is inferred rather than documented — if a field differs, the dump shows the truth and parser + fixtures move together. A run ends with a summary verdict:

VERDICT among L1-classifiable prints: agree=N  INVERTED=0  (ambiguous=…, no_quote=…)
=> side == aggressor CONFIRMED: of_domain mapping is correct.

#3. Replay a recorded tape offline

Feed the recorded file back through the same backtest composition with --tape. Replay reads the tape verbatim (blank lines and unparseable lines are skipped) and drives the footprint loop from real prints instead of synthetic ones:

dune exec bin/main.exe -- backtest FootprintCVDDivergence --tape /tmp/sber.tape --symbol SBER@MISX

The summary line is identical to §1; now the footprint= signals are derived from genuine aggressor data, so the run reflects real microstructure.

#4. Running it live

In a live (or paper) deployment the loop is always on — the trading host builds the order_flow factory, the broker subscribes the public tape for each watchlist instrument, and the strategy BC runs the FootprintCVDDivergence engine. There is nothing extra to enable; pick the venue with --broker and supply its credentials:

dune exec bin/main.exe -- serve --broker finam --secret <…>   # or --broker bcs / alor

#Troubleshooting

• footprint=0 on a synthetic backtest. Expected for very small --n or a flat generated series; increase --n. Divergence needs enough history to fill the lookback window.
• Probe prints [SKIP] … not set. The credential env var (FINAM_SECRET / BCS_SECRET) is absent — export it and re-run.
• Probe verdict inconclusive / no_quote high. The market was thin or quotes lagged trades; widen the window (re-run during active trading). For BCS specifically, read the raw QUOTE dump — if no bid/ask is recoverable, the L1 field names need confirming before the verdict means anything.
• backtest --tape reports fewer prints than the file has lines. Malformed or blank lines are skipped silently by design; check the file was written by a probe's --record, not hand-edited.

#See also

• Order flow — architecture.
• ADR 0032 — the decision record, including the tick-replay rationale and the synthetic-vs-recorded caveat.