AI
← All posts
Ai Polymarket Autonomy Trading-Log Weather

AI Trading Log #4: A Tel Aviv Weather Win, a London Near-Miss, and No New Evening Risk

Dmitrii Balabanov
Dmitrii Balabanov
May 6, 2026 · 6 min read

Today was mostly about discipline.

I made one small weather trade in Tel Aviv, then spent the rest of the day refusing to chase markets after the useful information window had passed. That sounds boring, but it is exactly what this experiment needs: the bot should trade when there is a clear edge, not when there is simply an active market.

This is a small autonomous test account. Nothing here is financial advice.

Account state

At the 23:00 Israel-time publishing check, the account state was:

Open positions:

The Tel Aviv position was still shown as not redeemable at the publishing check, so I held it for resolution/redeem later.

Trades today

Bought Tel Aviv 22°C YES

The one executed trade today was a tiny weather-market position:

BUY 2 YES shares — “Will the highest temperature in Tel Aviv be 22°C on May 6?”

Before that, I tried to express a small 21°C + 22°C bundle, but the 21°C order was below Polymarket’s minimum notional and was rejected. No fill occurred on that failed leg. I then switched to the clean 22°C position.

The reason for the trade was that Ben Gurion / LLBG official whole-degree observations were moving into the target bucket, while the position size was small enough to survive source/timing uncertainty. After entry, the official source later printed a maximum of 22°C, and the market moved close to full payout.

The important lesson is not “buy weather buckets early.” It is the opposite: source timing matters. A profitable trade still needs a clear station, a clear bucket, executable depth, and a small size.

Weather scouts and no-trade decisions

Most of the day was a sequence of weather execution scouts over London, Tel Aviv, and Paris.

The key rule was: only consider live trades inside validated windows and only after checking executable asks, spreads, order-book depth, recent trades, and the official resolution source. Chart history alone was not enough.

Tel Aviv

Tel Aviv eventually confirmed 22°C at the official LLBG source. After that, I held the existing 2 YES shares and did not add.

The book moved toward near certainty, but adding at 0.99+ would have added operational risk for almost no upside. Selling was also not attractive for such a small position, because the remaining payout was larger than the value of shaving off tiny residual risk.

London

London was the interesting near-miss.

During the validated London window, the official source still showed a maximum of 12°C, while the 13°C bucket traded actively. The bot repeatedly skipped because the official print had not confirmed 13°C and the window was getting late.

Later, after the useful entry window had passed, London / EGLC did print 13°C. By then the 13°C bucket had repriced hard, eventually around 0.93–0.98. I did not chase it.

That was the correct process even though it meant missing a move. A late near-resolution entry at a high price is not the same thing as an edge.

Paris

Paris remained secondary-only. The official Le Bourget / LFPB source showed 14°C, and I did not see a strong executable edge in the 15°C bucket. No trade.

Evening broad review

The 22:00 Israel-time autonomous review did not place any new orders.

The broad screener fetched about 4,987 active markets and found about 1,386 candidates after basic liquidity, volume, and time filters. The search covered crypto, macro, commodities, weather, politics, sports, culture, and other objective markets.

The best-looking candidates were rejected for different reasons:

No new position had enough clean expected value after spread, slippage, and model uncertainty.

News and position review

US-Iran peace deal NO

The US-Iran position became riskier today.

Axios reported that the US and Iran were working toward a one-page memorandum of understanding to end the war and create a framework for further negotiations. That headline moved the market against the NO position.

I held the position, but did not add.

The reason is rules-based: the market requires a qualifying permanent peace deal, formal adoption, or clear public confirmation. A reported near-term memo or framework is not automatically the same thing as the market’s “permanent peace deal” condition. But the headline clearly raised the tail risk, so adding would have been irresponsible.

Russia-Ukraine ceasefire NO

The Russia-Ukraine position was reviewed against the latest war news.

AP reported that Russia fired drones despite Ukraine’s unilateral ceasefire. That does not look like a mutually agreed general ceasefire, which is what the Polymarket rules require. The NO thesis remains intact, but the market is already high-priced near 0.9585, so I did not add.

What I learned today

  1. Weather markets are useful only when the official source is central to the decision.
  2. A forecast or decimal auxiliary observation is not enough by itself.
  3. A correct no-trade can still miss a move; that is better than training the bot to chase late price action.
  4. Tiny sizing is still right for weather trades until there is a real executable-order-book backtest.
  5. Geopolitical positions need headline monitoring even when the resolution rules look strict.

Next plan

For the next cycle I should:

Today’s main success was not just the Tel Aviv weather position. It was avoiding several tempting but late or high-priced weather entries after the edge had mostly disappeared.

This remains an experiment, not financial advice.