The track record, in the open

Every analysis Argustr publishes is scored deterministically against real market prices at its horizon due date. No backtests, no cherry-picking: wins and losses, with confidence intervals.

Track record

forward-tested

Accuracy measured against real prices after the fact, never backtests.

Hit rate50.9%95% CI 38–64%

PSR0.57P(Sharpe>0)

Sharpe0.02realized

Brier0.53calibration

By horizon

1 month 52.9%3 months 0.0%

Edgenot yet distinguishable from zerot = 0.50 · n = 21

Cumulative alpha vs-13.0%

Running sum of each settled call's return minus its market (S&P 500 for stocks, Bitcoin for crypto) over the same window, oldest first · sample too small for this number to mean anything yet.

Verdictssettled against real prices

489790.KSswing · NEUTRALLOSS -39.8%394420.KQswing · NEUTRALLOSS -17.6%298380.KQswing · NEUTRALWIN -10.2%347700.KQswing · NEUTRALLOSS -40.6%403870.KQswing · NEUTRALLOSS -37.9%

Settling soon

Confidence calibration

When the desk states a confidence level, how often is it actually right? A calibrated system earns the right to be trusted.

says 32%

is right 0%

says 40%

is right 25%

says 41%

is right 25%

says 42%

is right 33%

says 44%

is right 33%

says 46%

is right 55%

says 48%

is right 55%

says 50%

is right 55%

says 51%

is right 55%

says 52%

is right 55%

says 55%

is right 55%

says 59%

is right 55%

says 60%

is right 55%

says 65%

is right 55%

says 68%

is right 55%

says 70%

is right 63%

says 74%

is right 63%

says 75%

is right 63%

says 76%

is right 63%

says 82%

is right 63%

says 83%

is right 63%

says 85%

is right 63%

How this is measured

Each analysis states a target stance (LONG / SHORT / NEUTRAL) and a horizon with a due date. The record is a paper book: one unit long per LONG, one unit short per SHORT, no position for NEUTRAL. If you cannot short, read SHORT as “do not buy; trim if held” and NEUTRAL as “no signal”.
At the due date, the realized price is compared to the price at inference time. No AI is involved in scoring.
A NEUTRAL call is judged against the asset's own volatility: it is correct when the move stays inside the band, and it never earns or loses P&L (no position).
Returns are also benchmarked against the market (beta-adjusted alpha), and the stated confidence is scored for calibration (Brier).
Verdicts are append-only: once settled, a result is never edited. Methodology changes are versioned, never retroactive.
When the desk is wrong, an attribution engine records whether the cause was a risk it had identified or a blind spot, and blind spots feed back into future analyses.

Get the same desk working on your tickers.

Start free: 3 analyses per month

No card required. See pricing