Research Lab

This is an experiment into what an agentic-AI framework can do — whether disciplined AI agents, held to one rule (a claim only counts if there's evidence it actually predicts returns), can research, decide and honestly track investment ideas. We publish everything in full so you can check the system's work — that openness is the proof, and the whole point. It is deliberately detailed, but it is not a tip service and following it is not the idea: the experiment is the framework and how it performs, not any single stock.

You'll find our analysis of investment signals (which work, which don't, and why), the concepts behind them, deep-dive company research, and a forward-tracked book that keeps an honest score of how the framework's picks do. The book is a hypothetical £20,000 ISA: ideas are held (buy-and-hold), with uninvested cash in short-dated gilts. We size new ideas by conviction (the first few were equal-weight £1,000) and never re-size a position after the fact — so the record shows what we actually did, not a flattering rewrite. Tracked from publication, never a backtest. No real money, no client money, no trades placed.

Signal Research work in progress

The evidence engine behind how we pick — which signals actually predict returns, tested on our own data. Keep-decisions and reject-decisions both matter. See all signal studies →

Concepts

The ideas our framework rests on — how we think, in plain English. See all concepts →

The portfolio we're watching

Headline figures are after Hargreaves Lansdown portfolio costs (dealing, UK stamp, FX, platform); +1.9% before costs — early on, mostly one-off entry costs.

This is the hypothetical portfolio the system runs — the ideas it generates, deployed and tracked honestly with real costs, shown in full for transparency, not positions to follow. A £20,000 ISA: ideas are deployed conviction-weighted (a higher-conviction idea gets more) and then held, while uninvested cash sits in short-dated gilts (~4%). The three lines show pure performance and after Hargreaves Lansdown vs Interactive Brokers costs. Hypothetical — notional money, no trades placed; capital at risk.
Costs so far: IB £28.91 vs HL £236.76 in dealing fees — HL's flat per-trade charges on small holdings are the gap, and the reason platform choice matters.

Idea tracker & calibration

Calibration — does higher conviction actually win more? Average realised return by conviction, from closed ideas:

The system's tracked ideas

Research archive

A library of every research report the system rates — including good ideas it didn't put in the live book (shortlisted, on the bench, watchlist, held). The rejects are kept too — see Passed — why the system said no below. Click a column to sort; use the box to filter.

Passed — why the system said no

Every name the system passes on is kept on the record, price-stamped at the date it passed (2026-06-02), so the system's track record isn't flattered by quietly deleting them. Over time this shows whether the names it declined went on to do well (the filter was too harsh) or badly (it was working). Where the system wrote a full note before declining, the name links to it; the rest carry the one-line reason it passed. Not advice.