How to Clone and Configure a Marketplace Strategy

A marketplace clone is a starting point for inspection, not evidence that a strategy is safe, profitable or suitable for the selected market. Preserve the exact published version and read the source before choosing an action that may start a session.

Banger's public API documents cloning into your account with an optional deployment. The dashboard's Clone & paper-trade action combines those steps. Cloning can therefore use strategy and session capacity; do not treat it as a passive preview or click repeatedly when a request is slow.

1. Read the published source and its scope

Inspect the actual callback, declared venue/universe, decisions, quantity units and risk envelope. A marketplace summary is not a substitute for that source. Check whether the rule calls place, close or cancel, and which data or external prerequisites it assumes.

Clones retain parent/version provenance rather than taking whatever edited source the author might later have in progress. Keep the version and accepted configuration beside your result. A similar title or new default is not automatically the same experiment.

2. Inspect parameters actually supplied

The public 0.1.7 Strategy default introspection scans supported scalar class values, excluding framework-reserved names. A bare annotation without a value does not automatically become an editable parameter. Hosted typed metadata applies additional validation; legacy or unsupported declarations may not expose a usable field.

from banger import Strategy


class ParameterExample(Strategy):
    label: str = "research-example"
    observation_limit: int = 3
    required_only: str  # Annotation without a default is not surfaced here.


print(ParameterExample.declared_parameters())
# {'label': 'research-example', 'observation_limit': 3}
# Metadata demonstration only; no venues, universe, orders or deployment.

This exact metadata demonstration was executed offline against the public release. It is not a complete deployable strategy. Its values illustrate introspection, not a recommended runtime duration or a trading threshold.

3. Preserve accepted configuration

Use the surfaced supported fields and read any validation or inspection warning. A field's type does not prove that its value is economically meaningful or that it changes the desired market selection. An arbitrary market_id class attribute does not rewrite a separately constructed universe.

When changing source logic, record a new version rather than describing it as a parameter-only comparison. Keep dollar notional separate from contract quantity and do not widen a limit simply to make a rejected request pass. Confirm the actual values accepted in the resulting record.

4. Check admission and the action's outcome

The current public workflow is paper-only. A source inspection, clone creation, admitted session, running worker and delivered callback are different states. If admission or startup fails, preserve the returned error and any created record before deciding the next step.

The API's optional deployment and the dashboard's combined action should be distinguished from the local banger run command. A local run requires an existing file and account setup; it does not clone the marketplace entry for you. This article did not create a production clone or session during review.

5. Observe within a defined window

Choose a deliberately short supervised acceptance question: does the expected version initialize, select the intended market and deliver usable observations? A no-order clone should have zero fills. A clone with order logic needs separate inspection of admission, requests, simulated fills and remaining positions.

Do not recommend a whole market cycle or an unlimited session irrespective of capacity. Ordinary public quotas and infrastructure bounds apply; verified internal exemptions are not unlimited compute or real execution permission.

6. Read the model and verify shutdown

Banger's packaged forward-paper path can assume a non-crossing-limit fill and uses a reported-volume cap rather than a full depth replay. Its named standard Kalshi fee model is also a simulation convention with rounding limits. A favorable result under that model does not establish real fills.

Complete the appropriate stop procedure and inspect both retained history and resource cleanup. A stopped application record can precede completed hosted cancellation. Keep errors, zero-order results and incomplete cleanup visible. There is no public promotion to live when a paper metric looks attractive; the useful result is a versioned, interpretable research record.