> For the complete documentation index, see [llms.txt](https://docs.slate.ceo/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.slate.ceo/products/top-telegram-traders-and-groups.md).

# Top Telegram Traders & Groups

**Goal:**

Provide you with a number that represents "how practical (good entry and exit) and profitable (high confidence of good, consistent returns) it is to follow this caller or group's token calls."

#### What ELO really is:

$$
\text{ELO};=;
\underbrace{\text{Sortino Ratio}}*{\text{quality per trade}}
;\times;
\underbrace{\text{Number of calls}}*{\text{consistency}}
$$

Find a more detailed explanation of each component and adjuments made below. 

For sure reach out to @shirtlessfounder on Telegram if you have any suggestions for improving the system. We're continuously iterating on the ELO calcs.

***

### 1. What gets counted as a “call”

A "call" in the system is defined as the first mention of a token per user in a 30 day time window.

<table><thead><tr><th width="374">Rule</th><th>Reason</th></tr></thead><tbody><tr><td><strong>First token mention only</strong> (per user, per 30 d)</td><td>Filters out echo-spam and reposts.</td></tr></tbody></table>

***

### 2. Measuring the performance of a call

The performance of a call is dictated by the mcap multiple achieved at the exit mcap (@ T0+3min) from the entry mcap (@ T0).

Return per call:

$$
r = \frac{mc\_{3m} - mc\_{0}}{mc\_{0}}
$$

Some nuances were applied to approximate for call actionability:

| Rule                                            | Reason                                                                                      |
| ----------------------------------------------- | ------------------------------------------------------------------------------------------- |
| **Get entry mcap at time-of-call + 20 seconds** | Gives realistic reaction time after seeing a call notification and entry mcap for followers |
| **3-minute holding window** (`mc₀` → `mc₃m`)    | Normalizes TP "strategy" across all calls. Average holding time among active trenchers.     |

***

### 3. Stat-checks before calculating caller ELO

* **Sample-size floor** – A **dynamic threshold of calls** is applied per the look-back window before an ELO is given to the user.&#x20;
* **Lower-bound gate** – The **lower 95 % one-sided confidence limit** of the mean return to cut out lucky moons that are statistically indistinguishable from nois&#x65;*.*

***

### 4. Calculating the two factors

| Ingredient                                                                      | How we compute it                                                                                                                                       | Why it matters                                             |
| ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------- |
| [**Sortino ratio**](https://chatgpt.com/c/681403ef-98f4-800b-ba8b-9bfc074e5353) | <p>See formula below and explanation <a href="https://chatgpt.com/share/6814de77-b0c0-800b-b337-3996de633714">here</a>. </p><p>Target = 4% returns.</p> | Rewards upside while punishing only downside.              |
| **Number of calls (n)**                                                         | Count number of first token mentions per lookback window.                                                                                               | More first mentions = higher conviction the ratio is real. |

$$
\text{sortino ratio}; = \dfrac{\bar r - \text{target}}
{\sqrt{\tfrac{1}{n}\displaystyle\sum\_{r\_i<\text{target}}!\bigl(\text{target}-r\_i\bigr)^2}}
$$

ELO is a product of these two factors. A caller with a Sortino of **2.6** over **38** valid calls ends up with an **ELO = 98.8**.

***

### 5. Continuous data refreshing and lookback windows

* **Slice size** – ratings are re-evaluated every 12 h slice using only data that existed **before** that slice.
* **Lookback windows** – on the leaderboard, we calculate 1d, 7d, 30d and all-time ELOs for each user. Each time period has its own call threshold that's dynamically adjusted according to the dataset.
* **Decay** – calls gradually lose weight (half-life ≈ 7 d) so stale edge bleeds away.

***

### 6. Why the method is reliable

| Common pitfall        | How Caller Elo avoids it                                    |
| --------------------- | ----------------------------------------------------------- |
| One-hit wonder pumps  | Needs a threshold of calls & positive lower-CI.             |
| Echo-chamber hype     | First-mention rule prevents credit for copy-paste alerts.   |
| Pump and dump callers | Downside in Sortino + optional width-test caps their score. |
| Back-test overfitting | Strictly rolling, out-of-sample scoring; no hindsight.      |

***

#### TL;DR

1. **Clean call data** → first token mention + call actionability adjustments + 3-minute return.
2. **Caller consistency** → callers must achieve a threshold of calls before ELO is assigned.
3. **Risk / downside adjustments** → Sortino ratio calculation.
4. **ELO Score** → Sortino ratio × number of calls.

That’s the ELO you see next to every alert and on the global leaderboard. Again, reach out to @shirtlessfounder on Telegram if you have a suggestion on improving the system.