Three surfaces. One question: what actually changed?

Know when the people who run the company are buying.

Ranked insider activity, not a firehose. Wands separates open-market purchases from option exercises and gifts, then scores what is left.

Form 4 filings arrive in the hundreds every day. Most of them are compensation mechanics — grants, vesting, tax withholding — that say nothing about conviction. The handful that matter are buried in the same feed.

• Open-market purchases isolatedDiscretionary buys are separated from compensation mechanics before anything is ranked.
• Cluster detectionMultiple insiders transacting in the same direction inside one window are surfaced as a single event.
• Filing-level provenanceEvery row links back to the accession number, the filing date and the source document.

See what the largest holders did last quarter.

13F changes, normalised. Compare position deltas across institutions, weight them by portfolio impact, and surface consensus before it becomes a narrative.

A 13F tells you what a fund held at quarter end. It does not tell you whether that position was a rounding error or a third of the book, and comparing two funds by raw share count is close to meaningless.

• Portfolio-weighted deltasA 2% position increase at a concentrated fund is not the same event as a 2% increase at a mega-cap index.
• Consensus and divergenceSee where holders agree, and where one desk is moving against the crowd.
• Filing lag made explicitEvery figure carries the period it describes and the date it was filed.

Test the idea before you fund it.

Backtests with the boring parts included. Define the rules, run them against history, and read the result with costs, slippage and out-of-sample splits already applied.

A backtest that ignores fees, slippage and survivorship will happily show you a strategy that cannot survive contact with a real order book. The result looks like an edge and behaves like a rounding error.

• Out-of-sample by defaultResults are reported on held-back data, not just the window the parameters were fitted to.
• Costs are not optionalFees and slippage are modelled on every fill, and the assumptions are visible next to the result.
• Reproducible runsParameters, universe and data window are stored with each result so it can be re-run exactly.