SigmaLens reads public filings and public news, works out which of your positions each event actually reaches and by what route, and hands you the document it came from. When an event reaches none of them, it says that too.
Every figure on this page is one you can check, including the ones we have withdrawn.
A filing establishes that the filing was made and what it says. It does not establish what the market will do, and SigmaLens does not say that it does.
Four claims, each one checkable inside the product rather than taken on trust, and true every time you open a holding. Then a fifth claim we tested, found didn't hold, and withdrew rather than kept.
These are properties of the system, not performance figures, and that distinction is the point: each one is either true when you open a holding or it is not, and you can check it in a minute without taking our word for anything.
We used to claim the ordering beat a naive alternative. It does not. Forty-two days never used to build the engine, in three blocks, judged by 45 blinded AI runs (not human professionals), never told which system produced what they saw. Re-measured on the configuration we actually ship, the ranking comes out at −0.05, Fisher p = 0.26, winning 20 of 42 days. On the 2023 to 2024 block the plain volume sort won outright.
We published +0.12 to +0.21 and believed it. Auditing an unrelated change, we found the evaluation harness had been building its exposure graph without the entity map the product uses. The evidence had never described the shipped engine. We rebuilt all three blocks and re-ran every evaluator; the advantage disappeared. The engine now records a hash of its own configuration alongside each result so this cannot recur silently. The full working, all three blocks → Every claim and its status →
The four claims above are separate from the ordering claim, are measured separately, and none of them depends on it. Ranking was removed from the product; the withdrawal stands and the figures for it are published above and on the claim register.
Everyone knows an oil supply shock helps oil producers. The useful part is the rest of the chain, and it is the part a person reading the news does not have time to trace.
The third row is the one worth paying for. Kimberly-Clark has never bought a barrel of crude, but the resin in its products is made from one. SigmaLens carries that through with a weight and a stated reason, and scores it more faintly than the airline that burns the fuel directly, because it is a smaller share of cost and arrives with a lag. This chain is a worked example of the output format. The relationships are the curated ones the product uses; the cost share and the article count are illustrative. Live output, on a book you enter yourself, is one click away.
Most tools infer your exposures from a sector code and leave it there. We fit each holding against crude, gas, wheat, the 10-year, the dollar and gold over five years, with the market's own move removed, and report four answers: confirmed, contradicted, cannot detect, or a link the prices show that nobody modelled. On our own sample that last group included a refiner with no crude exposure on its record.
Each holding fitted against crude, gas, wheat, the 10-year, the dollar and gold over five years, with the market's own move taken out. A link is reported only when it clears both bars: distinguishable from noise, and large enough to explain something.
Your US-listed Apple is really a Taiwan bet, because TSMC makes its chips. SigmaLens surfaces the supply-chain exposure that country and sector labels miss.
Pick a scenario, a Strait of Hormuz closure or a Taiwan crisis, and see your book's reaction holding by holding before it happens.
Every signal is written to a ledger and resolved against what the market actually did, and the resolved record is published whether or not it flatters us. It is a record, not a promise that the reading improves: we have withdrawn two published figures after auditing them ourselves, and both retractions are on the evidence page with the numbers that replaced them.
Type them, paste them, or load a sample. Stocks, ETFs and funds all work.
Per-holding sentiment, your net exposure across themes, and the events driving each one, each linked to its source article.
Get alerted when a holding turns strongly bullish or bearish. In-app, by email, or to Slack.
SigmaLens is in design-partner phase. Everything below is open to the people using it, and no commercial terms are set. The tiers describe how much book each account can carry, not what anything costs.