Windows Preview available. Pro checkout is not open yet. Plans and capability details →
RowCoveAccount ↗

Quant / Research-result inspection

A top result row
is the start of the investigation.

Large parameter sweeps produce many plausible candidates. Use source-level queries and chronological evidence to understand the rows before relying on a ranking.

How can I investigate a multi-gigabyte backtest export?

Confirm the source population and field semantics, filter the complete supported source, inspect exact candidate records and follow them into chronological trade or equity evidence. Keep costs, losing cases and source provenance with the finding.

1. Start with a synthetic research example

This guide uses a mock research workflow. Example identifiers, records and any diagrams are synthetic, with no private strategy data, dataset sizes or research statistics.

Begin by identifying the result population and actual field semantics in your own local file. The example explains the investigation process; it is not a published benchmark or a strategy recommendation.

2. Establish what the export actually means

Inspect the schema and complete values. Match actual columns to return, drawdown, sample count, trading period and costs. Confirm whether fees, slippage and funding are represented; do not infer this from a column called profit.

Check whether the export contains every trial or only positive results. A positive-only result set cannot establish the distribution of all tested strategies. A research summary table is also different from the underlying trade ledger.

3. Filter the source, then trace the candidate

  1. Choose a small projection: strategy identifier, source grouping, period and the metrics needed for the question.
  2. Use supported full-file filters for source-wide screening. Do not treat a loaded-page search as a complete dataset query.
  3. Retrieve every page in the result scope; a first result page is not the full candidate set.
  4. Inspect the exact source record and retain its provenance.
  5. Open available chronological equity or trade data and investigate costs, drawdown periods and different time windows.

Column names and supported operations vary by source. Inspect the installed MCP tool schema before building an automated call; do not invent arguments from an example prompt.

4. Keep evidence separate from conclusions

Save reviewed results to a separate output where supported, preserving field types and source identifiers. Reopen the file and confirm numeric precision, selected rows and scope. Keep the unchanged source as the reference.

Selection from a large search is exploratory. It does not establish out-of-sample robustness or a strategy that will keep making new highs. The purpose here is a reproducible data investigation.

For market-history partitions use the Parquet guide; for local result databases use explicit table selection; for bounded AI calls follow the MCP tutorial.

Try the first page on your own data.

Start with the free Windows Preview: source information and up to 100 records, read-only. Review reader capabilities before planning a full-file Pro workflow.

Pro checkout is being prepared. The initial Windows install enables signed in-app updates for later versions.