Choose the job before choosing a model
Useful jobs include finding markets that match a topic, monitoring price changes, comparing related contract wording, and summarizing source material. Write the expected output first. A briefing workflow can be useful without an order node or a model-generated probability.
The prediction-market use case shows how NickAI combines these stages. Treat any model estimate as an interpretation to evaluate, and keep the underlying sources accessible.
Follow the identifiers
Use Polymarket's discovery documentation and the NickAI data-node reference to distinguish events, conditions, and outcome tokens. A market label written by a model is not a substitute for an identifier returned by the data source.
Before comparing contracts, check their exact resolution conditions and observation times. Similar names can describe different events. A displayed price difference alone does not establish an executable cross-venue opportunity.
Inspect a first-party report workflow
The Prediction Market Edge Report is a published research template. Its design collects Kalshi and Polymarket data, screens results, adds source context, and assembles a report. Inspect the shared workflow to see the configured stages. Its name does not establish the accuracy of its interpretations.
For a smaller start, use the Polymarket Signal Scanner specification. Review one data result and one report before enabling recurring runs.
What a reviewable briefing includes
- The actual market question, identifier, and source link.
- The observation time and the source fields used in the report.
- A clear separation between quoted data and model interpretation.
- Missing inputs or mismatched contract conditions.
- The workflow run ID and notification result.
These are suggested review fields, not a claim that every template emits an identical object. Keep the original output so the summary can be checked.
Keep order placement explicit
Adding an order requires a separate authenticated action and account-specific checks. Review the order-node reference, the venue's current rules, supported order behavior, and how the workflow records rejections or incomplete fills. A data monitor is not evidence that the execution path has been tested.
The Polymarket workflow walkthrough starts with that data path. The consensus guide explains why multiple agreeing models do not establish a correct market estimate.