See the turn before it is obvious.
AcIa Predictor scores every instrument in its universe every day with machine-learned models — the probability that a durable low or high is forming, where the name sits in its own slow cycle, and how price structure is behaving — and shows you where to look for a swing trade, one to two sessions before the crowd sees it.
Sign in
backup code
How it works — the logic in one minute
Three things are measured for every name, every day, from its own full history. BOTTOM and TOP heads: gradient-boosted models trained per instrument to answer one question — is a durable low (high) forming within days? PHASE: the instrument's slow cycle, below zero = accumulation, above = distribution, read both as a level (how much room) and as a derivative (when). Trend structure: lower-lows/lower-highs, breaks and reversals — the veto that stops any buy inside a falling structure. Cross-asset context (sector ETFs, our own volatility models on UVIX/SVIX, VIX term structure, MOVE, breadth, fear & greed) frames the whole board.
Three filters, one cycle
Early bottom is the machine-learned probability a durable low forms within 1–2 sessions — ⏳ forming on the intraday model, ● flagged on the daily, ✔ confirmed when both agree. Early momentum catches the phase rising from a recent low, the move under way but not extended. Topping out warns when the phase has risen and a top head fires. Every ticker wears its badge wherever it appears; a day slider replays any of the last 40 sessions.
One chart, every layer
Candles with the turn probabilities and phase underneath, trend structure painted across history, algorithmic support/resistance, a calibrated forward cone, options open interest with break-even-through-time, and the news that moved it. Live prices stream into the chart and the cards, so the last candle is the tape.
Trade and journal in place
Connect your own Interactive Brokers account: positions, live quotes, a two-step order ticket (what-if first, then place), a what-if → limit helper for options, and a trade journal that records every fill and your written reasoning — write-once, so the record stays honest.
Does it actually work? Measured, not claimed.
We took every confirmed bottom in the universe — a 10-day-each-side low that then paid ≥5% within 15 sessions — and asked, day by day, how often each screen was within two sessions of one. Out of sample, model trained to June 2025 and tested on the 13 months after.
One in three names in the top bucket was within two sessions of a durable low — four times the base rate. Not a crystal ball: two in three were not, and a flag says where to look, not what to do.
Why you can trust the numbers
Every score is a forecast. Models are trained on history and evaluated only on days they never saw; the app shows the score as it stood that morning, and the day slider lets you audit any past session yourself.
Nothing invented. Company figures come from the statements (EVA, ROIC vs WACC, margins, growth); short interest from the exchange; prices from a real-time feed. When data is missing, the app says so instead of filling the gap.
Actively developed. The filters are back-tested continuously against every confirmed bottom in the universe; they will sharpen, merge or be retired as evidence comes in — and you will see the change log, not a silent swap.
Built by a trader for trading. The developer runs his own book on it every day; the tools exist because he needed them.
What to expect — in numbers
Measured on the stock and ETF universe (about 295 names) over 2024–2025, out of sample:
~2,000 confirmed bottoms a year
A "confirmed bottom" is a low that holds for ten sessions on each side and then pays at least +5% within fifteen. The universe produced 1,968 of them in 2024 and 2,032 in 2025 — roughly 6–7 per name per year. The median one ran +11.3% within fifteen sessions; the top quarter +17.8%; the top tenth +27.9%.
1 in 3, at 4× the base rate
The Bottoming filter at its tight setting flags ~25 names a day. Out of sample, 34% of those flags were within two sessions of a confirmed bottom — against an 8% base rate for any random day. It catches about 39% of all bottoms; loosen the threshold and it catches more at lower precision (24% precision at 73% recall). You choose the trade-off; the numbers are on the screen.
Several real set-ups every week
Roughly 780 caught bottoms a year across the universe — several a week on your holdings and watchlist alone. Two flags in three will not be a bottom; that is why the app pairs the list with the chart, the trend veto and the phase context, and why sizing is yours. Past measurement is not a promise about the next one.
The universe
334 instruments today, expanding. 197 US stocks (large caps and the names that move — from WMT, MS and TSM to NTLA, SPOT and CORZ), 84 Nordic and European stocks (SAND, METSO, YAR, HEXA-B, CARL-B, ELUX-B, DEMANT …), the sector ETFs (XLK, XLF, XLE, XLV, XLU, XLI, XLB, XLY, XLP, XLC, XLRE, IGV) and the index ETFs (SPY, QQQ, IWM, DIA), the volatility instruments UVIX and SVIX — with their own turn models, used across the app as the vol regime — gold, silver, copper, platinum, palladium and oil, Bitcoin and Ether, plus a macro layer of 27 series that frame the board: US 10y/30y, DXY, HYG/LQD/TLT, VIX, the European indices, and thematic baskets (uranium, lithium, rare earths, steel, base metals). Every instrument gets the same treatment: full-history training, daily and 2-hour models, out-of-sample scores.
Interactive Brokers — trade where you see it
Your account, in the app
Connect your own IBKR account with a headless OAuth key (no gateway, no browser session — the guide walks you through it in five minutes). You then get your positions with live P/L, day-to-expiry and "stock theta" on options, open orders you can update or cancel, and a two-step order ticket: IBKR's own what-if preview first, then place — no single click can ever send an order.
The ticket streams the live underlying and the option's real-time bid/ask, and its what-if → limit helper reprices an option to any underlying level you type, so you can pre-place a bid for the dip you expect.
Journal and automation
Every fill accretes into a trade journal — one row per round trip, entry, exit, days held, realised P/L in € and %, and a write-once "why" per trade that can never be edited afterwards, so your record stays honest.
Automation, opt-in and bounded. A guardian can arm a trailing stop or take-profit rule per position, first in watch mode (it tells you) and only on your explicit switch in live mode (it acts). Auto-trading — signal-to-order — is being built on the same two-step, audited path and will ship as a per-user, per-rule opt-in with hard position limits. Nothing trades for you unless you turned it on, and everything it does is in the audit log.