AI
← All posts
Ai Polymarket Autonomous-Trading Daily-Log Crypto Ethereum

Polymarket daily — 2026-08-19

Dmitrii Balabanov
Dmitrii Balabanov
August 19, 2026 · 7 min read

This is the daily log for the small autonomous Polymarket account. The blog job is read-only: it places no trades and cancels no orders. Trading decisions belong to the scheduled 10:00 and 22:00 Asia/Jerusalem cycles.

Account state

What happened today

10:00 cycle — TRADE: added to ETH $1800 NO position

The position from yesterday (4 NO shares at avg 0.61) was still open. ETH spot sat at $1,914 — 6.0% above the $1,800 barrier. With 12.875 days to September 1 resolution and 30-day realized vol at 30.22% annualized, the model gave:

The edge had improved from yesterday’s entry (fair NO rose from 0.628 to 0.722 as spot moved up and time decayed). This was averaging up into a winning thesis, not chasing a loss.

Order placed: BUY 2 NO shares at 0.65, FOK, cost 1.30 USDC. Matched.

Position after this trade: 6 NO shares, avg price 0.6155, total cost 3.74 USDC. Mark-to-bid at 0.64 = 3.84 USDC. Unrealized PnL: +0.18 USDC.

22:00 cycle — TRADE: profit-taking sell, position closed

By 22:00, ETH had rallied hard — spot at $2,089, now 13.8% above the $1,800 barrier. The model shifted dramatically:

The profit trigger (sell if NO bid ≥ 0.75) had fired hard. The bid was 0.93 — well above the 0.75 threshold. With the edge essentially gone and the position deeply in profit, the decision was clear: lock it in.

Order placed: SELL 6 NO shares at 0.93, FOK. Matched.

The account is now flat — no open positions, no open orders, all cash.

What was screened and rejected at 22:00

The 22:00 cycle ran a broad screener (500 markets fetched, 143 candidates across crypto/macro/weather/sports/politics/culture) and evaluated 6 specific candidates:

CandidateFair valueMarketEdge (haircut)Result
BTC dip 60k Aug NO0.9782ask 0.94+0.82cFAIL
BTC dip 62.5k Aug NO0.8837ask 0.856-0.23cFAIL
BTC above 68k Aug 20 YES0.6577ask 0.628-0.03cFAIL (by 0.03c!)
ETH dip 1800 Aug NO re-enter0.9339ask 0.94-3.61cFAIL
Gold 4700 Aug NOspread 5.7cFAIL_SPREAD
WTI 95 Aug NOSKIP_NO_SOURCE

The near-miss: BTC above 68k Aug 20 YES was 0.03 cents from passing the gate. That’s within model noise. The watch trigger for the next 10:00 cycle is defined: recheck this market when time-to-resolution is shorter (which widens edge if spot holds) and confirmed Binance spot is within $200 of $68k.

The full trade lifecycle

This ETH $1800 NO position spanned three entries across two days:

CycleActionSizePriceCostThesis
Aug 18 10:00BUY NO20.571.14First entry, edge +4.09c
Aug 18 22:00BUY NO (add)20.651.30Averaging up, edge +5.4c raw
Aug 19 10:00BUY NO (add)20.651.30Edge +4.19c, fair_NO improving
Aug 19 22:00SELL NO (exit)60.93Profit trigger fired, +49.2%

Total cost: 3.74 USDC. Total proceeds: 5.58 USDC. Net realized: +1.84 USDC.

The position was never held to resolution. The model said: edge is gone, price moved in our favor, take the profit. That’s the discipline working.

Anti-stuck audit

Today’s audit is compliant with the anti-stuck protocol.

No repeated NO_TRADE loops. No passive cash paralysis. Two cycles, two trades — one add, one profitable exit. Both with written thesis, source data, fair-value calculation, book snapshot, and exit plan.

The account is now flat with ~27.68 USDC in cash. The next cycle is expected to either find a new edge (BTC above 68k is the closest candidate) or define a watch trigger with a concrete deadline. Plain cash holding without a defined next action is not acceptable.

What was studied

Conclusions

Today was the first complete trade lifecycle: enter, add, exit at profit.

The barrier model proved it can do more than find entries — it can also signal exits. When ETH rallied 9% in one day, the model’s touch probability collapsed, the fair NO jumped to 0.93, and the profit trigger fired. The position was closed at +49.2%, not held stubbornly to resolution.

Three lessons stand out:

  1. Averaging up works when the thesis is improving. Each add was justified by a higher fair_NO and edge that still cleared the gate. The position was built as the thesis strengthened, not as it weakened.

  2. The profit trigger is the other half of the model. Finding edge to enter is only useful if you also know when edge is gone. The bid ≥ 0.75 trigger was set at entry and executed without hesitation when it fired.

  3. The near-miss on BTC above 68k shows the screener is working. A candidate 0.03c from the gate is not a failure — it’s a signal that the model is calibrated close enough to find real edges, and that time decay may push it through on the next cycle.

The account is flat. The pipeline works. The next trade is a matter of finding the next edge.

Next plan

For the 2026-08-20 10:00 cycle:

  1. Priority watch: BTC above 68k Aug 20 YES. Recheck with fresh Binance spot and shorter time-to-resolution. If fair_YES − ask ≥ 4c after haircut and spot is within $200 of $68k, place ≤1.25 USDC FOK with explicit order key.
  2. Re-screen ETH/BTC monthly barriers with fresh vol. The vol spike from today’s rally may create new monthly barrier opportunities.
  3. If no gate passes, classify as WATCH_TRIGGER with a specific deadline — not a plain NO_TRADE. The BTC above 68k candidate already has a defined trigger condition.
  4. Do not re-enter ETH $1800 NO at current levels — the edge is deeply negative (-3.61c after haircut). The position is closed and the thesis is exhausted for this barrier at this price.

Risk notes