A First-Week Research Checklist for Paper Trading Builders

A first week of paper research is useful when it leaves a reproducible experiment, delivered observations and a record of its assumptions. The checklist below is a suggested sequence, not a requirement to place trades or a promise about what a week will prove. Paper observations avoid real orders; provider costs and account limits are separate considerations.

The outcome is a written account of evidence and uncertainty. Keep the scope small enough to inspect each step, and stop at a missing prerequisite rather than treating the calendar as permission to bypass it.

Day 1: Pick one current contract and read its resolution rules

Pick a single live market, not a basket. One contract keeps the surface area small enough to actually understand. Read the resolution criteria in full before you look at any price. The question text is not the contract. The settlement source, the cutoff time, and the tie-breaking language are the contract.

Select a market for its documented rules and the data needed by your hypothesis, rather than an unsourced claim about a category’s share of trading volume. Preserve the reason and exact identifier so a later comparison uses a known instrument.

Check current data access and the rules of the exact venue/product. A public observation does not establish eligibility to submit a real order, and an open listing can change during a test.

Day 2: Inspect the data before you trust it

Observe bids, asks, timestamps and available quantities. Ordinary binary terminal payoffs may be complementary, but two executable offers need not sum to one dollar. Missing quote data must remain visible. A midpoint, last trade and hypothetical executable fill are different measurements.

• Top-of-book spread, and how wide it gets during quiet periods
• Depth at the best few levels, since that is what a real fill walks through
• Tick frequency: does the market update often enough for your logic to react
• Fees, which on some venues scale with price and eat small edges first

Day 3: Write the strategy as code and save a version

Begin with a no-order observer using Banger 0.1.7’s actual adapter import and on_market_tick callback. Replace the placeholder with an exact currently open ticker you have selected. Risk settings constrain exposure when orders are added; they do not create a timed stop. This observer deliberately places no orders:

from banger import Strategy, signals
from banger_adapter_kalshi import Kalshi


class FirstWeekObserver(Strategy):
    name = "first-week-observer"
    venues = [Kalshi()]
    universe = signals.kalshi_markets(
        status="open", tickers=["REPLACE_WITH_CURRENT_OPEN_TICKER"]
    )

    def on_start(self, ctx):
        self._observed = False

    def on_market_tick(self, market, tick, ctx):
        if self._observed:
            return
        self._observed = True
        ctx.log("first_week_observation", market_id=market.market_id,
                timestamp=tick.timestamp.isoformat(),
                bid=str(tick.bid) if tick.bid is not None else None,
                ask=str(tick.ask) if tick.ask is not None else None)
        # One log; no orders. The logging bound does not stop the worker.

Save the version before the run and preserve its source hash or commit with the result. A change to code, instrument, parameters or data creates a new experiment to label. A file alone is not evidence that its callback was delivered.

Day 4: Run a bounded backtest, then forward-observe on paper

For historical research, fix the window, exact universe and assumptions before interpreting the outcome. Candle-based fills are different from a replay of full depth or queue position. Inspect the actual Banger backtest model and coverage rather than promising that every simulator walks a live order book with exact current fees.

For this local observer, complete the Banger account setup in the CLI documentation, save the file as strategy.py and run the command below. Stop with Ctrl+C after a deliberately short supervised window and inspect retained history. The callback was exercised offline with synthetic ticks; this review did not launch a hosted strategy.

banger run strategy.py --paper

Historical testing evaluates the rule under past data and a particular simulator. A forward observation checks present callback behavior and delivered data. Neither proves that a hypothetical order would have filled or that the sample supports a profitable signal.

Day 5: Inspect logs and interrogate your fill assumptions

For an observer, zero simulated trades is expected. When adding order logic, inspect returned admission, fill quantity, price, fees and timestamps separately. Banger’s public 0.1.7 forward-paper path can assume a non-crossing limit fill and can cap quantity using reported volume. That is not a reconstruction of real passive queue position or a full depth ladder.

Compare simulated fills with available contemporaneous evidence and retain unknowns. The packaged standard Kalshi fee helper rounds to cents; the current venue schedule describes a finer operation. Do not label the result exact net venue performance without inspecting the implemented model.

Stop on schedule, and write down what remains unproved

Choose the supervised stop deadline before starting and verify the completed stop and resource state. A log-once flag bounds output, not lifetime. The daily-loss control rejects new exposure under its valuation baseline; it does not automatically flatten holdings or guarantee that a worker ends at the calendar deadline.

Then write the honest list. A short paragraph per item is enough. What did the week actually establish, and what is still an open question?

• Sample size: how many simulated trades, and is that enough to mean anything
• Regime coverage: did you observe only calm markets, or any stress
• Fill realism: how far your simulated fills sat from live book reality
• Fees and slippage: whether they were modeled or quietly assumed away
• Resolution risk: any ambiguity in the contract's settlement rules you haven't tested
• Venue and jurisdiction: whether the contract stays tradable where you are

Carry a versioned file, observation record, explicit model assumptions and a written conclusion into the next experiment. Missing data, failed initialization or uncertain shutdown are useful findings to resolve; they are not invitations to increase size or proceed to real-money execution. Banger’s public product remains paper-only.