# How to test an AI or Telegram trading signal

Version: 1.0
Format: caption-led explainer with no voiceover
Duration: 8:30
Canonical method: https://kostrade.pro/en/trading-signals

## 00:00 — A screenshot is not an audit

A screenshot can show a selected result. It cannot prove when the idea was created, whether losing posts were deleted, or whether the displayed payout was available. This is a method demonstration, not a live signal and not performance evidence.

## 00:30 — Define the record before the result

Give every alert a unique signal ID. Record provider, asset, quote mode, direction, expiry, entry rule and invalidation before the outcome is known. An immutable timestamp is the first provenance field, not the entire proof.

## 01:20 — Preserve Telegram provenance

Store the public channel username, direct post link, message ID, original timestamp, edit status and forwarded status. A copied screenshot loses this chain. If the original message changes or disappears, keep the preserved record and label the evidence limitation.

## 02:15 — Measure delivery latency

Generated time and received time are different. Delivery latency equals received time minus generated time. Also record the valid-until time and executable entry time. If the alert arrives after the window, the honest outcome is SKIP, not a theoretical chart entry.

## 03:10 — Record executable terms

Capture the asset identifier, regular or OTC quote mode, direction, expiry, observed payout and entry price or range. Do not replace the actual platform value with a later chart price. Differences in quote source, payout and execution can change the result.

## 04:05 — Define every skip reason

Write the skip rule before testing. Examples include late delivery, payout below the minimum, invalidation reached, wrong quote mode, missing field, market unavailable or risk limit reached. A skip reason is part of the sample and must not be silently deleted.

## 05:00 — Keep every outcome

Use a fixed status vocabulary such as PASS, SKIP, PROFIT, LOSS and RETURN. Preserve every alert in sequence. Do not rewrite the original entry after the result, replace missed trades with ideal prices, or show only a winning streak.

## 05:55 — Use payout-aware metrics

Report total alerts, executed alerts, skipped alerts, win rate, net return, longest losing sequence and maximum drawdown. The break-even formula is 1 / (1 + payout). At an 80 percent payout the threshold is about 55.56 percent before other friction, so win rate alone is incomplete.

## 06:50 — Separate tuning from testing

Choose the provider, rules, asset list and test dates before the forward period begins. Do not tune on the same outcomes used to report performance. Repeat on an untouched demo period and disclose the complete sample. AI analysis is decision support, not a guarantee.

## 07:45 — Decision rule and risk

Trust the process only as far as the preserved evidence supports it. If timing, provenance, payout or execution fields are missing, mark the result unresolved or SKIP. Download the audit template from the canonical guide. Binary options can result in the loss of the full stake.