AI
← All posts
Ai Polymarket Autonomy Trading-Log Crypto Risk

AI Trading Log #17: Holding the Bitcoin Position While the Market Weakens

Dmitrii Balabanov
Dmitrii Balabanov
May 19, 2026 · 5 min read

Today was a risk-management day.

No new trades were placed. The account held its existing Bitcoin upside position while the market price for that position weakened. The main job was to decide whether the drawdown was a reason to exit, a reason to add, or simply a reason to keep watching.

The answer today was: hold, but do not average down.

Nothing here is financial advice. This is a small autonomous test account and a public decision log.

Account state

At the publishing check, authenticated account state was:

The latest scheduled trading review earlier in the evening had seen the same position around 0.085, worth 6.375 USDC, with unrealized PnL around -4.125 USDC. That small discrepancy is normal market movement and API timing, not a new trade.

There are still old resolved/legacy positions in the proxy-position data, but no other positive-value active exposure required action.

Trades today

No trades were placed today.

There were two scheduled trading/review cycles:

Both reviewed the existing BTC position, open orders, account cash, broad candidates, and strategy guardrails.

Existing BTC position

The live position remains:

This market resolves from Binance BTC/USDT one-minute high prices during May. It immediately resolves to YES if BTC touches or exceeds $85,000 during the relevant May window; otherwise it resolves NO.

That makes it a finite, objective, convex upside position. It is not a prediction that BTC is more likely than not to hit $85k. It is a small barrier-touch bet that can still pay if volatility returns before the end of the month.

Why no exit?

The position weakened today.

At the 10:00 review, BTC spot was around $76.9k, and the BTC $85k May YES book was around 0.09 / 0.10. At the 22:00 review, BTC spot was around $76.7k, and the YES book was around 0.08 / 0.09.

This was in the watch zone, but it did not break the thesis.

The explicit hard trigger was a YES bid at or below 0.07, or a clearer spot/market deterioration. The market did not reach that trigger. Selling at 0.08–0.09 would lock in a large loss while the original finite barrier-touch thesis was still alive.

So the agent did not exit.

Why no add?

The more tempting mistake would have been averaging down.

The position was cheaper than yesterday, and the account still had cash. But yesterday already included a top-up. Adding again without fresh evidence would have violated the strategy rule: do not automatically average in a losing position.

No new information improved the thesis enough to justify more correlated BTC exposure. The cheaper price alone was not enough.

So the agent did not add.

What was studied

The cycles reviewed both focused crypto candidates and broader Polymarket markets.

Focused checks included:

The broad screener fetched roughly 1,000 active markets in each review and found about 249–263 candidate rows. Top clusters included:

Most were rejected for familiar reasons:

Process conclusion

Today was not a passive no-trade day.

The account had an active losing position, so the decision mattered. The correct action was not to manufacture a new trade just because the account has an aggressive target. The correct action was to separate three questions:

  1. Is the original thesis broken? Not yet.
  2. Is there fresh evidence to add? No.
  3. Is there a better independent trade elsewhere? Not found today.

That leads to hold / no trade.

This is different from the earlier failure mode of sitting in cash while repeating low-value weather checks. Here, the agent had a concrete position, clear triggers, and a defined reason for inaction.

Next plan

The next cycles should:

The current position is uncomfortable but still inside the plan. The next mistake to avoid is turning discomfort into either panic-selling or blind averaging.