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

AI Trading Log #20: Closing the Bitcoin Position and Returning to Cash

Dmitrii Balabanov
Dmitrii Balabanov
May 22, 2026 · 4 min read

Today the account closed the Bitcoin $85k May position.

The day started with a residual 25 YES position left over from yesterday’s partial exit. The morning review held it because the full-exit trigger had not fired. By the evening review, the market had weakened enough that the written trigger did fire, so the agent sold the remaining shares.

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 proxy-position API still shows several old zero-value/redeemable legacy positions from prior weather tests, but no active positive-value exposure remained at publication time.

At the same publishing check:

Trades today

One trade was executed.

At the 22:00 Asia/Jerusalem trading/review cycle, the account sold the remaining BTC $85k May position:

The order matched as a FOK sell. Post-trade reconciliation showed:

The public Data API later showed one SELL 25 at 0.04, so there was no duplicate sell.

Morning review

The 10:00 review held the residual position.

At that point:

The full-exit rule after yesterday’s partial sale was:

Neither was true in the morning, so the agent held. It also did not add, because the residual was intentionally small optionality, not a reason to average down again.

Evening review and exit

By the 22:00 review, the position had deteriorated.

Pre-trade context:

The explicit full-exit trigger was met because YES bid was 0.04, below the 0.05 threshold.

The agent therefore sold the remaining 25 YES at 0.04 and closed the position.

This was the important discipline point: the residual was conditional optionality, not a permanent lottery ticket. Once the written exit condition fired, it had to be closed.

What was studied

Both scheduled cycles screened broad candidate sets.

The 10:00 review fetched roughly 1,000 active markets and about 304 candidates. The 22:00 review fetched roughly 999 active markets and about 343 candidates.

Reviewed areas included:

No replacement trade was opened.

The reasons were straightforward:

Process conclusion

Today was a trigger-following day.

The account did not exit in the morning because the rule had not fired. It did exit in the evening because the rule had fired.

That is the correct shape of autonomous risk management:

  1. Define the trigger.
  2. Check the trigger with live data.
  3. Act when it fires.
  4. Do not invent a new story to avoid taking the planned loss.
  5. Do not immediately re-enter the same correlated theme without a fresh thesis.

The BTC trade was not profitable. But the exit process was controlled, logged, and consistent with the guardrails.

Next plan

The account is now mostly back in cash.

Next cycles should:

The immediate task is not to force a replacement trade. It is to find a cleaner independent setup, write the thesis, define the trigger, and only then deploy risk.