Research

Research we run with AI, and how we keep it honest

Alongside the products, we run research desks where agents gather, check and organize evidence. The point is not the answers. It is answers you can trace.

Standards for every desk

  • Every number has a source and a date.
  • When there is no exact figure, show how the estimate was made and how confident it is. Never present an estimate as fact.
  • Record failed tests and dead ends, not just wins.
  • Missing data is reported as missing, not reused from last week.
  • Cross-check important work across more than one model.

Research areas

Previews only. Methods, prompts and templates are held back for members.

In daily use

Source-traced daily briefings

Each claim in the AM and PM Edge editions is rated for integrity and confidence, and each figure is traced to a source. A tracker breaks the briefing's improvement work into small tasks that only close with evidence.

Research · paper only

Trading research and backtests

The LOOP desk tests trading hypotheses from research to post-mortem. Backtests state data source, date range and fill assumptions, and failed ideas go in a ledger so they aren't retried by accident. Paper trading only.

In daily use

Valuation from completed sales

For cards and memorabilia, value comes from comparable completed sales, labeled with source and confidence. Different models build the same dossier independently so we can compare them.

Live

Public sentiment signals for sports

A serverless engine blends public forum posts, news feeds and prediction-market prices into a sentiment read for each golf event, refreshed on a short cache.

Research · dormant

Acquisition search research

A thesis and target-screening workspace for buying a small audiovisual and low-voltage integration business in the Northeast. Deal details stay private.

Research

Investigations ledger

A versioned ledger of conspiracy claims, each case pinned to its sources and released in numbered editions.

In daily use

Media and audience research

Weekday scans for outlier videos, podcast developments and news, plus reads on how our own posts performed and why.

Prototype

Personal knowledge from saved media

Transcribing and tagging a library of saved videos into a wiki that cites its sources, with the media kept on the owner's machine.