Methodology
Signal responsibility
What we answer for once a signal is published, and how a bot showcase ROI can mislead.
Showcase re-run 20 Oct 2026
A bot platform sells a tool. The user picks the coin, the settings and the timing, so the outcome belongs to the user. Sentinel Aleph publishes signals with a coin, a direction, an entry and a stop. Once a signal is published, reporting its outcome is our job.
What we are responsible for
- Every published signal stays in the signal ledger: wins, losses, break-evens, expiries and BTC veto closes.
- We never delete a row. Views can filter the ledger, and the rows stay.
- Win rate is always printed with its sample size, next to the net result after costs.
- Cost per trade: Spot 0.20% either direction. Futures 0.12% on a long, 0.08% on a short. The futures figures include 0.02% funding, paid by longs and received by shorts.
- A result enters a ratio only 96 hours after publication, so open trades cannot lift the number.
The rules, as the code runs them, are on How we grade. Monthly totals are on Reports.
How a showcase can mislead
A ledger that keeps its losses looks worse next to a showcase that lists only winners. In October 2026 we re-ran the top 10 DCA presets of a bot platform's public showcase on Binance public data. All 10 used one template: 20 averaging orders about 1% apart, a 1.4% take profit and no stop loss, mostly on small coins. The showcase ranked them by ROI over 1 Sep to 20 Oct 2026. Two presets could not be tested: one coin is not listed on Binance futures and one is delisted.
1. ROI from closed deals only
A DCA deal that keeps falling stays open, and an open deal has no realized loss. The top-ranked preset's own backtest showed +1,952 USDT realized and −1,150 USDT on a deal still open at the end. That open deal is not in the realized figure. The showcase ROI's calculation basis is not stated and could not be reproduced.
When we counted open deals and stopped trading once the capital was used up, 6 of the 8 presets that could be tested lost their whole capital inside the window they were ranked on. Over each coin's available history (up to 12 months), only one variant of one preset avoided liquidation, and it ended with an open deal at −60% of its capital.
2. Picking winners from many runs
We ran the same template on all 527 Binance USDT-M perpetuals that traded in that window. Between 0.4% and 3.6% of coins reached a realized ROI of 50% or more. The showcase coins sat at the 95th to 99.8th percentile on realized ROI. On net ROI after open deals, three of the four short presets fell to the 3rd, 5th and 11th percentile.
3. ROI divided by the order size
A grid showcase listed a realized ROI of 10,230%. That figure is the profit (2,046 USDC) divided by the 20 USDC placed per grid line. The grid needed 1,560 USDC of capital, so the return on capital was about 131%.
4. Fills at the wick with no slippage
The showcase backtest fills orders at the candle's extreme price with zero slippage. For the grid in point 3 this matters: it ran on a coin priced near 0.0000001, where one price step is about 10%, wider than its 1.34% grid step. At that scale part of the profit may come from price moving between bid and ask, a spread a real order pays. We did not re-test this grid by simulation.
Four questions to ask any showcase ROI
- Does the ROI include deals that are still open, and what are they worth now?
- How many coins, settings or runs were tried before this one was shown, and over how long a window?
- Is the ROI divided by the capital the strategy needs, or by one order?
- Would the fills in the backtest be possible at the coin's price step, after fees, funding and slippage?
Our own numbers on the same scale
Source: /api/public/ledger/monthly, read when this page loads. Months in Zurich time, by the month a signal was published. Net per entered trade is the average per-trade move after costs, not an account return.
Ledger figures as of 2 Nov 2026, 14:23 UTC
| Month | Published | Win rate, matured | Net per trade, veto closes in | Net per trade, veto closes out |
|---|---|---|---|---|
| Jul 2026 (from 28 Jul) | 100 | 50.0% (n=36) | −1.15% (n=99) | −1.07% (n=47) |
| Aug 2026 | 636 | 70.2% (n=383) | −0.12% (n=634) | +0.13% (n=498) |
| Sep 2026 | 654 | 65.4% (n=353) | −0.24% (n=652) | +0.03% (n=460) |
| Oct 2026Not yet matured, 22 still open | 938 | 59.1% (n=438) | −0.99% (n=913) | −0.47% (n=582) |
| Nov 2026Not yet matured, 56 still open | 70 | — (n=0) | +0.19% (n=14) | +0.19% (n=14) |
Across the matured months (Jul 2026 to Sep 2026): 1,390 signals published, win rate 67.1% (n=772).
Net per entered trade was −0.25% with BTC veto closes (n=1,385) and +0.02% without them (n=1,005).
A 67% win rate did not produce a positive net once every entered position is counted.
Our grading page states that no edge after costs is established.
Method
- Data: Binance USDT-M public archives (1-minute candles and funding rates), plus the public futures API for the last day.
- Assumptions: 1-minute bars, worst price first inside a bar, a take profit must trade through, slippage of at least one price step or 0.05% per market fill, 0.05% fee per fill, funding charged, 1x, liquidation when capital runs out. Capital is the amount the showcase itself listed as required.
- Check: under the showcase's own model (1-hour candles, wick fills, no slippage, no liquidation), 5 of 7 checkable presets land within 0.5x to 1.9x of its realized profit. One could not be reproduced from Binance prices.
- Window: 1 Sep to 20 Oct 2026, plus the 50 days before and up to 12 months of history.
- The analysis can be reproduced from public data. The scripts and data files are available on request from [email protected].
Risk notice
Sentinel Aleph publishes market research. It is not investment advice. Past results, ours or anyone's, do not predict future results. Crypto futures can lose more than the margin you place.