Timefire
AI forecasts for every S&P 500 stock, backed by a public history anyone can check.
Product: Timefire ranks every S&P 500 stock each trading day, locks the rankings before the market moves, and grades them in public ten trading days later. From January 2 to October 7, 2026, its paper portfolio rose 24.2% (books before April 28 are a backtest), against 14.7% for the S&P 500 (SPY) and 10.5% for its equal-weighted version (RSP), 9.5 points ahead of SPY.
Ray, our research scientist, has been building the AI model for more than two years. My job is to make a product out of it.
Show every version of the model
The model has gone through three versions. Instead of showing only the newest, the forecast page keeps all three side by side, losing years included, so readers can judge its progress for themselves.
Why an AI forecast needs a record
The product is the forecasts. Many AI stock ratings publish an accuracy figure that readers have to take on trust, which leaves an obvious question: how do I know you didn't change your record? Timefire is for skeptical readers, individuals and businesses, who want proof they can check before they believe a financial claim, and who use the forecasts alongside their own trading decisions.
The wrong first answer
In the very first version, we solved the trust problem first, and overdid it. The record became the subject, and the model behind it disappeared.
Putting the product first
I reorganized the homepage around three questions, in order. What do I get? A daily AI ranking of the S&P 500, and how a portfolio built from it has done against SPY. How does it work? A short summary of the model, with a link to the full explanation. Can I check it? Each day's ranking is sealed and timestamped to Bitcoin after the market closes, then graded in public ten trading days later.
Back every forecast with proof
The proof is what makes the forecasts believable. Every forecast goes through the same four public steps, from the moment it is sealed to the check in a reader's own browser.
Let readers test the proof
The verification page runs its checks in the reader's browser. It downloads the original locked file, calculates its unique digital fingerprint again, verifies who signed it, and confirms when it was recorded on Bitcoin. Newer books add a fifth check: the timelock that kept them sealed until their scheduled release.
Keep the data honest
One of the most challenging parts of this product is keeping the data honest. We ran into many cases where the data drifted for reasons outside our control: a spin-off made one stock look as if it had lost nearly 85%, a price vendor's missing day quietly stretched a ten-day window to eleven, and stocks went without a real closing price after a merger or delisting. Our approach is simple: never rank on a stale price, never edit the record, and publish every fix as a public note with the old figure, the corrected one, and the cause.
Brand design
The mark is a turquoise blade and its shadow: a sundial, and a flame. Time is the witness. Every forecast is locked before the market moves, then judged in public.
The tagline, “AI forecasts, sealed and graded,” says what Timefire does, not what it promises.
Design for people and software agents
Software agents are readers too: AI assistants that look up a stock and answer for a person. Every Timefire page has an agent version with the same facts in plain text, so an assistant quotes the record instead of guessing.
No one writes the agent versions by hand. One publishing step builds them from the same data as the human pages, and a release is blocked if any page is missing or the two versions disagree.