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

Cash in the wallet, no room in the portfolio

Dmitrii Balabanov
Dmitrii Balabanov
September 24, 2026 · 4 min read

The wallet had $36.82 in cash, but the bot could not afford its strongest signal today—not because the money was missing, but because the portfolio’s remaining risk budget was too small.

At 19:00 UTC, YES on Ethereum reaching $2,800 during September 21–27 offered a modeled edge of 7.5 cents per share at a 25-cent ask. The stored decision was cap:total=1.50. Open purchase cost was already $13.50 against a $15 ceiling, leaving $1.50, below the configured $2 minimum trade budget.

Bitcoin had a similar constraint: $9.20 committed against a $10 asset cap. Signals above the four-cent entry threshold at 15:00, 16:00 and 18:00 were blocked with just 80 cents of asset headroom. These are documented skips, not execution failures. Nor is a model edge proof of a missed profit.

The purchase that did fit

At 17:00 UTC, the bot bought 5 YES at $0.39, paying $1.95 before additional fees, on Ethereum dipping to $2,600 during September 21–27. The model’s fair range was approximately 46.4–52.6%, with a fee- and haircut-adjusted edge of 4.7 cents and 83 hours remaining.

The journal and ledger agree on the actual five shares and $1.95 consideration. This was the only recorded purchase today through 20:00; there were no sales. The $1.95 execution is slightly below the nominal $2 sizing target because the exact-amount order builder rounds down; the sizing target is not a guaranteed final spend.

Three hours later, the displayed mark was 29.5 cents and the position showed a $0.475 unrealized loss. The model still valued YES at 39.88–46.22%. A favorable model estimate does not stop the market moving against a new purchase.

Five tickets at 20:00 UTC

Contract / sideSharesEntry averageModel fairBidMark value
BTC reaches $88k, Sep 21–27 / YES130.18990.2045–0.29450.100$1.4300
BTC dips to $82k, Sep 21–27 / YES70.33990.3939–0.48390.260$1.8550
BTC reaches $90k, September / NO50.87000.7567–0.84330.848$4.2575
ETH dips to $2,600, Sep 21–27 / YES50.39000.3988–0.46220.300$1.4750
ETH dips to $2,300, September / YES50.47000.0075–0.01900.030$0.1650

Cash is 36.8233 USDC. Positions have a combined displayed value of $9.1825, making cash plus marks $46.0058. At the quoted bids the total is $45.8333 before exit fees. Against $13.50 of open purchase cost, the displayed unrealized loss is approximately $4.32, excluding additional fees.

All five positions remain hold. The three weekly contracts have 80 hours left; the two September contracts have 152. The BTC $90k NO ticket is close to its entry price, while the old ETH $2,300 YES accounts for roughly half the unrealized loss. Holding tickets tied to both upward and downward touches is not a risk-free hedge: the path, timing and prices paid matter.

A quiet endpoint hides changing inputs

The following comparisons use 00:00 to 20:00 UTC on September 24, not a rolling 24-hour return:

Bitcoin ended almost unchanged across those snapshots, but its realized-volatility estimate increased. ETH and SOL finished higher, also with higher seven-day volatility. These are the bot’s stored inputs, not an independently established explanation for market moves.

Reliability and the next decision

All 21 hourly reports from 00:00 through 20:00 are present. There are no new recorded errors and no PENDING or UNCERTAIN orders. The status counter still contains two historical failures, from September 19 and September 23; neither is a new outage today.

No parameters changed during this review. The next scheduled check is September 24 at 21:00 UTC. The useful research question is now allocation: whether capital tied up in old low-value tickets is worth more than fresh signals after spreads and fees. Simply raising a cap because a blocked signal looks attractive would not answer it. Today’s evidence establishes the constraint, not the profitability of removing it.